|
1. |
Time-resolved microspectroscopy and interferometry of organic mesoscopic materials† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 531-536
Hiroshi Masuhara,
Preview
|
PDF (111KB)
|
|
摘要:
Time-resolved microspectroscopy and interferometry of organic mesoscopic materials† Hiroshi Masuhara*, Keiji Sasaki, Hiroshi Fukumura and Hiroshi Furutani Department of Applied Physics, Osaka University, Suita 565, Japan The characterization of mesoscopic materials and analysis of their physical and chemical processes were performed by advanced laser and optical techniques. By using a laser manipulation technique, a micrometre-sized particle in solution was chosen and fixed at the focal point of a microscope by the photon pressure of a focused near-infrared laser beam.The laser trapping potential exerted on the single microparticle was thus analysed. A picosecond time-resolved fluorescence microspectroscopic system was applied to measure picosecond lasing dynamics of a single particle, and the results are discussed. The electric charge on a single microparticle in solution and the adhesion force between the particle and a quartz plate were also determined by the laser manipulation method.Mesoscopic thin films undergo morphological change and ablation on intense laser pulse excitation. The photothermal expansion–contraction and photodecomposition behaviour of a surface layer was interrogated by means of time-resolved interferometry developed by combining a nanosecond time-resolved laser and Michelson interferometry. Keywords: Mesoscopic materials; microparticle; thin film; fluorescence microscopy; lasing dynamics; laser trapping; nanosecond time-resolved interferometry; expansion–contraction; ablation The characterization of bulk materials and analysis of their physical and chemical processes have been extensively performed in the last few decades and the relevant research field is now well established.More recently, scanning probe microscopy and related techniques have been developed extensively such that observation, manipulation, and analysis are being made possible for individual atoms and molecules.The trend is very clear, and indeed the chemistry of single molecules is now one of the most interesting topics. This suggests that bridging the gap of understanding between bulk materials and single molecules will be necessary and important as the next step in relevant research fields. From this viewpoint, we consider that studies on chemistry in mesoscopic domains will become an active area. Mesoscopic dimensions have received much attention in solid-state physics, particularly in semiconductor technology, where the range from 1 to 10 nm is extremely important.On the other hand, organic molecular systems have not been examined and considered in detail from the point of view of mesoscopic science. Electrons are confined to a single molecule, whereas the association of molecular aggregates involves a distribution and cannot be well controlled. Consequently, meaningful data cannot be obtained by examining an assembly of mesoscopic systems, which is typically the case for microparticles.The physical and chemical properties of an assembly of particles with different sizes, morphologies and inner structures have been measured and understood as an average of the assembly. However, some properties cannot be explained as an average, for example, an assembly of microparticles with different relaxation time constants gives a non-exponential decay curve on excitation, whereas a single exponential decay giving the time constant should be shown for each particle.1 Hence, the real nature of microparticles can only be ascertained when a single microparticle is interrogated as a function of its size, morphology and inner structure. Similarly, the individual surface and interface layers of a solid or liquid/solution can only be revealed by a novel, sensitive methodology.The development of modern lasers and optical techniques has been extremely rapid and influential and has led to great advances in the spectroscopic analysis of mesoscopic systems.For example, we have combined pulse lasers, confocal microscopic methods and reflection optics, and proposed various experimental approaches, viz., picosecond time-resolved fluorescence and absorption microscopy, picosecond time-resolved total internal reflection fluorescence spectroscopy, nanosecond time-resolved attenuated total reflection spectroscopy, picosecond time-resolved specular reflection spectroscopy and femtosecond time-resolved diffuse reflectance spectroscopy.2–7 By focusing the laser beam onto a small particle under a microscope, photon pressure is exerted on the particle.With a suitable optical set-up, it is possible to catch, transfer and fix a single microparticle at a certain position in solution against its Brownian motion—laser trapping. By combining the trapping technique with time-resolved microspectroscopy and microelectrochemistry systems, which we have developed separately, various laser trapping methods have been proposed, viz., laser trapping/fluorescence spectroscopy, laser trapping/absorption spectroscopy, laser trapping/photochemistry, laser trapping/ electrochemistry, and laser trapping/fabrication.2 As reviewed in detail in the literature, the spectroscopic and electrochemical analysis, photochemistry and fabrication of individual particles in solution have been made possible just as in the bulk state.2 Furthermore, we have succeeded in developing the laser trapping method for manipulating individual particles arbitrarily in three-dimensional space.2 The dynamic fluorescence spectroscopic analysis of surface/ interface layers of solutions and polymer films has been made possible by time-resolved total internal reflection fluorescence spectroscopy.2,3 Not only the association and distribution of molecules but also relaxation dynamics have been confirmed to be different from those observed in the bulk state.Therefore, organic mesoscopic materials are an interesting and important research target, which is comparable to mesoscopic science in semiconductors. In this work, new developments in the characterization and analysis of mesoscopic materials are described.One development is a more quantitative determination of the properties of individual microparticles in solution which is based on a laser micromanipulation technique. The laser trapping potential exerted on a single microparticle in solution, particle electric charges and the adhesion force between a particle and a quartz plate are discussed.For morphological dynamic studies on surface layers, nanosecond time-resolved interferometry was applied to some polymer films. Expansion/contraction dy- † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997. Analyst, April 1998, Vol. 123 (531–536) 531namics and the subsequent ablation process were directly measured in the nanosecond to second time region.These new aspects of microspectroscopy and interferometry for organic mesoscopic materials are discussed and summarized. Results and discussion Optical manipulation of a single lasing microparticle Micrometre-sized spherical particles such as liquid droplets, polymer latexes and microcapsules, which form genuine spheres due to the surface tension, can act as optical cavities in air or liquid. Under the total internal reflection condition at the interface between the microspheres and the surrounding media, light propagates in a circumferential manner so that the resonant field is formed just inside the surface of the sphere.8 Theoretical studies based on the Mie–Debye scattering theory established that microparticles possess high quality factors (104–108), which are sufficient for inducing laser oscillation.9 A dye-doped polymer microsphere is irradiated by intense light so that an inverted population of dye molecules is induced in the particle.When the emission returns to the starting point with the same phase except for some integer multiple of 2p, oscillation is induced so that the emission is amplified by the populationinverted dye molecules–laser oscillation. In order to investigate precisely the physical properties of microspherical lasing without any disturbances, such as thermal Brownian motion, gravity and convection, an advanced technique that makes it possible to levitate and trap individual lasing microspheres is essential. This can be achieved by using a system that combines a laser manipulation technique with fluorescence microspectroscopy; a schematic diagram of the system is shown in Fig. 1. Photon pressure is induced on photon momentum transfer from refracted and reflected laser light to scattering particles, and is sufficiently strong to levitate and fix microparticles against gravity and a viscous drag.A lasing microparticle has the advantage of high sensitivity owing to intracavity enhancement of the tunneling loss. The resonant field in the microspherical cavity penetrates just outside the sphere as an evanescent field, so that the lasing emission can leak to an object placed in the vicinity of the particle, affecting the microspherical lasing process.10 Based on this photon tunneling phenomenon, the distance between the microsphere and the object can be measured with nanometre resolution by monitoring the change in the lasing emission spectrum.11 Indeed, the dependence of the lasing peak intensity on the distance is theoretically calculated on the basis of variations in the microspherical cavity.Fig. 2(a) shows the emission spectra of a rhodamine B (RhB)-doped poly(methyl methacrylate) (PMMA) microsphere (20.8 mm diameter) observed in water on changing the distance between the particle surface and a microscope glass plate. Distinct resonant peaks at 584.3 and 585.0 nm, which exhibited a non-linear dependence on the pumping intensity, can be ascribed to lasing emission, whereas the broad and structureless background corresponds to spontaneous emission of RhB.As the separation distance was reduced, the intensities of the resonant peaks changed. The change in the emission intensity of the lasing peak at 585.0 nm is plotted as a function of the distance in Fig. 2(b). The curve obtained could be well fitted to an exponential function with a decay constant of Å 260 nm.This value is comparable to the penetration depth of an evanescent field produced by Å 90°-incident light on the interface between PMMA (n = 1.49) and water (n = 1.33). Thus, the marked change in the lasing peak intensity can be explained by the fact that leakage caused by photon tunneling through the evanescent field leads to a reduction of the quality factor of the microspherical cavity so that resonant modes would be quenched as the particle approached the glass.The optically manipulated lasing microsphere described here is a highly sensitive light source in the mesoscopic dimension. For nanometre-based technology, the lasing microsphere makes it possible to control precisely the separation distance by monitoring the lasing spectrum in addition to spectroscopic analysis. Laser trapping potential of a single microparticle Experimentally, photon pressure exerted on microparticles has been measured by use of viscous flow.2,12,13 If the sample cell is moved with an appropriate velocity, an optically trapped particle experiences a viscous drag in the medium. On increasing the velocity, the photon pressure should also increase Fig. 1 Schematic diagram of the laser manipulation–fluorescence spectroscopy system. Fig. 2 (a) Lasing emission spectra of a RhB-doped PMMA microsphere (20.8 mm diameter). The separation distance between the particle and the glass plate was (A) 770, (B) 510, (C) 260 and (D) 0 nm.(b) The difference obtained by subtracting the emission intensity of the lasing peak at 585.0 nm from its maximum value at far distance in (a) is plotted as a function of the distance. The solid curve is an exponential function fitted to the plot, whose decay constant is 260 nm. 532 Analyst, April 1998, Vol. 123to keep the viscous drag constant. On the basis of Stokes's law, one can quantitatively determine the photon pressure from the flow velocity, viscosity of the medium and radius of the particle. Unfortunately, a microparticle undergoes thermal Brownian motion so that the random fluctuation of the position and velocity degrades the accuracy of this technique.Here, we describe a new system which makes it possible to observe precisely and instantaneously a three-dimensional potential profile of the photon pressure exerted on a single microparticle.14 The system is based on a thermodynamic analysis of the Brownian motion evaluated with total internal reflection microscopy and nanometre position sensing.No assumption of the potential profile is necessary, and the only physical parameter that should be known is the temperature of the sample. A schematic diagram of the system is shown in Fig. 3. A microparticle dispersed in water is optically trapped by a focused 1064 nm beam ( Å1 mm spot size). The particle is positioned close to the surface of a microscope glass slide (n = 1.52) by laser manipulation, while an s-polarized beam from a power-stabilized He–Ne laser is introduced into a prism that is optically coupled to the glass slide through immersion oil having a refractive index close to that of the glass.The incident angle on the surface of the glass slide is adjusted by a rotating mirror to be 62.0°, which is larger than the critical angle at the glass/water interface (61.6°). Thus, the laser beam provides the evanescent field under the total internal reflection condition, which illuminates the trapped microparticle.The scattered light from the particle is collected by the objective lens and detected by a quadrant photodiode (QPD). A sum of four output signals (Z-signal) from the QPD and two differential outputs normalized by the sum (X- and Y-signals), as shown in Fig. 3, are used for data acquisition and analysis. Since the particle is randomly moved with thermal energy, the position distribution p(x,y,z) is determined by the potential energy profile V(x,y,z) exerted on the particle via the Boltzmann distribution p(x,y,z) ª exp[2V(x,y,z)/kT] (1) where k is the Boltzmann's constant and T is the absolute temperature.Thus, a three-dimensional potential V(x,y,z) is given by V(x,y,z) = 2kT ln[p(x,y,z)] (2) Curves (A)–(D) in Fig. 4(a) show x-directional potential profiles observed for a polystyrene (PS) latex particle (4.2 mm diameter, Polyscience, n = 1.59) trapped in water at room temperature (293 K).Curves (A)–(D) are well fitted to parabolic functions, and the width of the potential well is reduced on increasing the trapping laser power. The spring constants of the harmonic potentials were calculated with deconvolution of the instrumental response function, and plotted against the laser power. The result for the 4.2 mm particle clearly showed a linear dependence of photon pressure on laser power and the gradient was 1.6 31024 N m21 W21. This value is smaller than reported spring constants of 4.3 3 1024–1.8 3 1023 N m21 W21 for Å 0.6 mm particles,12,15 and larger than that of 3.7 31025 N m21W21 for a 10 mm particle.2 Geometrical optics shows that the spring constant of optical trapping will be inversely proportional to particle size, which can qualitatively explain the difference between the observed and reported values. Determination of absolute charge of a single microparticle One application of the potential analysis of the photon pressure exerted on a single microparticle is the determination of its Fig. 3 Schematic diagram of the potential analysis system. Scattered light of the evanescent field by a microparticle is measured with a quadrant photodiode detector, whose differential outputs correspond to x and y displacements and total intensity depends on the distance z between the particle and a glass plate. Fig. 4 (a) x-Directional potentials exerted on a polystyrene latex particle (4.2 mm). The particle was trapped in water by a focused laser beam with a power of (A) 25, (B) 50, (C) 100 and (D) 160 mW. (E) Instrumental response function observed by measurement of a particle adhered to a glass plate.(b) Electric field effect on potential curve of a polystyrene latex particle (4.2 mm). The distance between the two electrodes is 600 mm. (A) Without and (B) with applied voltage of 2 V. (C) Difference between (A) and (B). Analyst, April 1998, Vol. 123 533absolute charge. A set of microelectrodes was placed in a sample cell under a microscope, and a polymer latex particle was again trapped between the electrodes by a focused infrared laser beam, after which the electric field was adjusted to shift the position of the microparticle.As the particle is usually charged positively or negatively in water, the particle receives electrostatic potential in addition to the laser trapping potential. The minimum position in the sum of both potentials is different from that of the trapping potential; hence, from its shift the electric field intensity was calculated.This is shown schematically in Fig. 4(b), where two potential curves are given for cases with and without an electric field. The difference between them gives a linear relation, which confirms that our analysis is reasonable. For the polystyrene microparticle in water described here, the negative surface charge density was estimated to be 1.8 3 1026 C m22. This is ascribed to the –SO3H substituent, which dissociates in water giving SO32.By adding NaCl, the effective charge of the surface is varied and the shift in the position of the trapped microparticle was clearly demonstrated to give a change in the surface charge density. Adhesion force between a single microparticle and a surface The photon pressure effect provides a new method for analyzing the adhesion force, making non-contact and non-destructive measurement possible. Since the adhesion force exerted on micrometre-sized particles in air is of the mN–nN order, an instantaneous and strong radiation force induced by the intense pulsed light of a Q-switched laser was used.16 The strength of the force is determined on the basis of the Mie scattering theory with the critical laser intensity at which the photon pressure and the adhesion force are balanced.The glass plate to which the microparticles adhere faces down and is placed on two microscope cover glass plates (separation distance between the two plates Å 5 mm).The second harmonic pulse of a Q-switched Nd3+:YAG laser is introduced into a microscope and irradiated onto a single particle through the glass plate to exert photon pressure on the particle. While monitoring the behaviour of the particle using a CCD camera, the laser power is gradually increased shot by shot until the particle is removed from the glass plate by the laser irradiation. It was confirmed that the released particle was deposited onto the other microscope cover glass plate facing the left-hand plate, and was not damaged.Based on the light scattering theory, the radiation force Fpr exerted on a particle at the threshold laser intensity Ith is given by Fpr = c21IthCpr (3) where c is the velocity of laser light and Cpr represents the crosssection for the photon pressure, defined as Cpr = (1 2 cosq) Csca + Cabs (4) where Csca and Cabs are the scattering cross-section and the absorption cross-section of a particle, respectively, and cosq is an asymmetry factor that is the mean of cosq with an angular intensity distribution as a weighting function. By separate experiments it was confirmed that the contribution of the thermomechanical force is negligible compared with the photon pressure Fpr.Fig. 5 shows the adhesion forces observed for PMMA particles of various sizes doped with 5 3 1024 and 5 3 1023 m rhodamine 101. The forces were determined from eqns.(3) and (4) with the threshold laser intensity. The plots for both concentrations can be well fitted with the same line, which also supports the conclusion that the thermomechanical force is negligible. The linear relation between the adhesion force and the particle size is consistent with the Lifshitz theory of van der Waals forces.17 Nanosecond–nanometre measurement of expansion/contraction of polymer films The interaction between polymers and intense laser light leads to rapid and large morphological changes in the entire polymer matrix, and it is possible to determine the dynamics.Several methodologies with a time resolution of picoseconds to milliseconds and a spatial or vertical resolution of submillimetre to micrometre dimensions have been developed for their direct observation, viz., ultrafast imaging, Schlieren photography and holographic interferometry. Here, we describe the use of time-resolved interferometry, which interrogates nanometre expansion and contraction behaviour during or immediately after the excitation pulse.18 The experimental set-up is shown schematically in Fig. 6. A 248 nm KrF excimer laser was used for inducing photothermal expansion or photochemical ablation. The laser beam was focused onto a small spot of about 0.1 cm2 on a PMMA film surface at a slightly tilted angle. A Q-switched Nd3+:YAG laser Fig. 5 Adhesion forces exerted on Rh101-doped PMMA particles of various sizes with concentrations of 5 3 1024 m (closed circles) and 5 3 1023 m (open circles).Fig. 6 Schematic diagram of a nanosecond time-resolved Michelson interferometer: PF, polymer film; QS, quartz substrate; M1–M3, mirrors; L1, imaging lens of 200 mm focal length; L2, 250 mm focal length lens to adjust the spot size of the excimer laser; BS, beamsplitter; PG, pulse delay generator; OSC; sampling oscilloscope; PD1 and PD2, fast photodiodes; and CCD, video camera. 534 Analyst, April 1998, Vol. 123(a) (b) (A) (B) (532 nm, 10 ns FWHM) was used as the probe light source of the Michelson interferometer.Time-resolved interference patterns were imaged by a CCD camera, stored in frame memory and then analyzed. Some representative interference patterns of a PMMA film doped with 2% m/m pyrene on a quartz substrate are shown in Fig. 7(a). Deformation of the fringe pattern can be easily recognized at Dt = +24 ns, which clearly shows the transient morphological change of the PMMA film below the ablation threshold.A shift by 0.6 of a fringe spacing to the right is observed. As the 532 nm light of a Nd3+:YAG laser is used as the probe, a shift of one fringe spacing to the right corresponds to an expansion of 266 nm, half the wavelength of the probe laser light. Thus, the fringe shift observed here is ascribed to a film expansion of 160 nm. The shift of the fringe pattern decreased by a few tens of milliseconds, indicating contraction of the expanded film, i.e., the initial flat surface was restored completely.The dependence of the maximum value of the expansion on the fluence is plotted for PMMA and poly(methyl acrylate) (PMA) films doped with 2% m/m pyrene in Fig. 7(b). At about 65 mJ cm22, a deviation was observed for the PMMA film and the slope increased by a factor of about 4.2, whereas that of the PMA film did not alter. It is interesting that only the PMMA film shows a discontinuous fluence dependence. As the thermal expansion of polymer films was proposed above, the surface temperature of the PMMA film at the critical fluence of 65 mJ cm22 was estimated, assuming that the Beer–Lambert law holds and all absorbed photons are converted to heat completely.A surface temperature of about 410 K was obtained, which is close to the glass to rubber transition temperature (Tg) of PMMA, viz., 378 K. It is worth noting that the PMMA film shows delayed expansion behaviour until Tg is attained in the course of laser excitation.When PMMA and PMA films were heated to about 550 K by the excimer laser irradiation at a fluence of 150 mJ cm22, expansion of the PMMA film is expected to be delayed compared with the PMA film because of the incorporation of the phase transition. Indeed, the PMMA film showed a slight delay at the end of the excitation laser pulse, continued to expand and reached a constant value at Dt Å +60 ns, whereas the PMA film ceased to expand and reached a constant value almost at the end of the excitation laser pulse (Dt = +40 ns).The delayed expansion behaviour may be explained by the temperature dependence of molecular motions of a processes, microscopic structural changes in the polymer backbone, and volume relaxation in the glass to rubber transition of supercooled glass-forming materials.19 Nanosecond time-resolved measurement of photodecomposition and ablation dynamics of polymer films A reactive nitrocellulose (NC) sample film, in which a light absorber, savinyl blue (SB), and a polymethane binder (PU) are incorporated, was examined and its novel decomposition dynamics were shown as a representative example of laser ablation.NC is a highly reactive material and has been widely used in explosives and propellants. It is well known that NC can be thermally decomposed and that it releases heat during the reaction (exothermic decompositon), i.e., its reaction is selfaccelerated by the decomposition itself and self-sustained after reaching the explosive decomposition.The chemical nature of NC is different from that of other polymers; hence, an NC film is expected to show specific dynamics of thermal decomposition and macroscopic morphological changes on laser irradiation. Its decomposition dynamics have received considerable attention in the field of mechanistic studies of combustion and burning processes of reactive explosives and energetic propellants. Nanosecond time-resolved interferometry can directly monitor the solid to gas transition process during decomposition.Expansion and contraction dynamics were analysed and are plotted against delay time in Fig. 8. Expansion in both the central and peripheral (border) areas was initiated during the excimer laser pulse and continued until 50 ns. The expansion behaviour in both areas was very similar, but the absolute displacement was different because of the inhomogeneous fluence distribution.After 50 ns, the increase in the film thickness in the central area ceased and was followed by a decrease, whereas the expansion in the border area continued. The latter area then showed a slow decrease in the microsecond time region, and reverted to the original value (zero displace- Fig. 7 (a) Nanosecond time-resolved interferometric images of a PMMA film doped with 2% m/m pyrene on a wedged quartz plate. The images were taken at the same area of the sample film. (A) Dt = 2H (before irradiation with the excimer laser).(B) Dt = +24 ns. The laser fluence was 150 mJ cm22 below the ablation threshold ( Å 180 mJ cm22). (b) Fluence dependence of the maximum expansion of PMA (2) and PMMA (©) films below the ablation threshold (180 and 330 mJ cm22, respectively). Both films were doped with 2% m/m pyrene. Fig. 8 Expansion–contraction and ablation dynamics of the NC sample film at 110 mJ cm22. Examined area: (:) central and (½) border. The broken line represents the etch depth measured by a depth profiler.Analyst, April 1998, Vol. 123 535ment), i.e., no permanent etching was observed. Thus, the behaviour of the border area, irradiated below the ablation threshold, can be interpreted in terms of expansion and contraction dynamics. As the behaviour of the central area is different from that of the border area in the 50–500 ns region and as permanent etching was observed, it is considered that ablation (etching) takes place in this time region.On the other hand, the decrease in the central area in the several tens of microseconds region is due to thermal contraction caused by slow heat dissipation, as the observed behaviour was similar to that of the border area. The measured decomposition rate of the NC–SB–PU film was about 6.3 3 1021 m s21, which is higher than that in conventional combustion of NC–nitroglycerol composites or glycerin azido polymers ( Å 1022 m s21).20 Laser-induced decomposition of the NC sample film can be attributed to an explosive combustion because of the higher decomposition rate.The large difference in the decomposition rate seems to arise from the rapid attainment of the combustion conditions by excimer laser irradiation which cannot be attained by conventional heating methods. In the present case, a surface temperature of 1400–1800 K was estimated, whereas the initial temperature in conventional combustion is less than 400 K.21 Furthermore, it was reported that a rapid pressure jump ( Å 1 GPa) occurred on laser ablation,22 which might also accelerate the decomposition (burning) rate.It is well known that an increase in the initial temperature or pressure causes an increase in the burning rate, which accelerates decomposition and subsequent evaporation of decomposed gaseous products in the solid phase. Thus, the extreme conditions realized by excimer laser irradiation are considered to be the cause of the unique decomposition and ablation behaviour observed here.The authors express their sincere thanks to H. Fujiwara, M. Tsukima, K. Horio and K. Wada for their contributions. References 1 Koshioka, M., Misawa, H., Sasaki, K., Kitamura, N., and Masuhara, H., J. Phys. Chem., 1992, 96, 2909. 2 Microchemistry, ed, Masuhara, H., De Schryver, F. C., Kitamura, N., and Tamai, N., Elsevier, Amsterdam, 1994. 3 Masuhara, H., in Photochemistry and Solid Surfaces, ed. Anpo, M., and Matsuura, T., Elsevier, Amsterdam, 1989, pp. 15–29. 4 Fukazawa, N., Fukumura, H., Masuhara, H., and Prochorow, J., Chem. Phys. Lett., 1994, 200, 461. 5 Ichikawa, M., Fukumura, H., Masuhara, H., Koide, A., and Hyakutake, H., Chem. Phys. Lett., 1995, 232, 346. 6 Asahi, T., Matsuo, Y., and Masuhara, H., Chem. Phys. Lett., 1996, 256, 525. 7 Fukumura, H., Yoneda, Y., Takahashi, H., and Masuhara, H., Chem. Lett., 1996, 509. 8 Kerker, M., The Scattering of Light and Other Electromagnetic Radiation, Academic Press, San Diego, CA, 1969. 9 Barber, P. W., and Chang, R. K., Optical Effects Associated with Small Particles, World Scientific, Singapore, 1988. 10 Sasaki, K., Misawa, H., Kitamura, N., Fujisawa, R., and Masuhara, H., Jpn. J. Appl. Phys., 1993, 32, L1144. 11 Sasaki, K., Fujiwara, H., and Masuhara, H., J. Vac. Sci. Technol. B, 1997, 15, 2786. 12 Kuo, S. C., and Sheetz, M. P., Science, 1993, 260, 232. 13 Finer, J. T., Simmons, R. M., and Spudich, J. A., Nature (London), 1994, 368, 113. 14 Sasaki, K., Tsukima, M., and Masuhara, H., Appl. Phys. Lett., 1997, 71, 37. 15 Svoboda, K., Schmidt, C. F., Schnapp, B. J., and Block, S. M., Nature (London), 1993, 365, 721. 16 Sasaki, K., Horio, K., and Masuhara, H., Jpn. J. Appl. Phys., 1997, 36, L721. 17 Lifshitz, E. M., Sov. Phys. JETP (Engl. Transl.), 1956, 2, 73. 18 Furutani, H., Fukumura, H., and Masuhara, H., Appl. Phys. Lett., 1994, 65, 26. 19 Furutani, H., Fukumura, H., and Masuhara, H., J. Phys. Chem., 1996, 100, 6871. 20 Urabanski, T., Chemistry and Technology of Explosives, Pergamon Press, Oxford, 1983, vol. 2. 21 Bennett, L. S., Lippert, T., Furutani, H., Fukumura, H., and Masuhara, H., Appl. Phys. A, 1996, 63, 627. 22 Ben-Eliahn, Y., Haas, Y., and Welner, S., J. Phys. Chem., 1995, 97, 6010. Paper 7/07104C Received October 10, 1997 Accepted February 9, 1998 536 Analyst, April 1998, Vol. 123
ISSN:0003-2654
DOI:10.1039/a707104c
出版商:RSC
年代:1998
数据来源: RSC
|
2. |
Quantitative X-ray fluorescence analysis of geological materials using partial least-squares regression† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 537-541
M. J. Adams,
Preview
|
PDF (81KB)
|
|
摘要:
Quantitative X-ray fluorescence analysis of geological materials using partial least-squares regression† M. J. Adams* and J. R. Allen School of Applied Sciences, University of Wolverhampton, Wulfruna Street, Wolverhampton, UK WV1 1SB A computerised algorithm to select automatically appropriate wavelength (or 2q) variables for subsequent multivariate calibration modelling was applied to the determination of iron, manganese, potassium, calcium, titanium, silicon, aluminium, magnesium and sodium in a range of certified geological materials by XRF spectrometry.The application of partial least-squares (PLS) regression is shown to be superior in terms of predictive performance to univariate linear regression modelling and multiple linear regression analysis. The combined process of automated variable selection and PLS modelling is amenable to providing an automated XRF quantitative analysis software system. Keywords: Geological materials; X-ray fluorescence spectrometry; partial least-squares regression; multivariate calibration modelling XRF spectrometry is a routinely applied technique for both qualitative and quantitative multi-elemental analysis, particularly for solid samples which often require minimal pretreatment prior to analysis. The technique exhibits wide elemental coverage, good detection limits (typically mg kg21) and an extensive working range (to high percentage or pure materials).Compared with optical emission spectra, XRF spectra are relatively simple and well defined rules or heuristics can be developed to assist in the qualitative interpretation of the spectra.To this end, we have developed and reported a computerised expert system (AXIS) for automated XRF qualitative analysis.123 From a digitised spectrum the emission peaks are extracted and, in decreasing order of intensity, are identified according to the elements present in the sample. Fuzzy logic and fuzzy operators are employed to account for uncertainty in the data and to mimic human interpretation behaviour.Results from the expert system on a wide range of sample types of widely different chemical composition have demonstrated the robust and accurate operation of the system.3 A useful and valuable extension to this software would be an automated quantitative analyser, providing the system with the ability to report not only the presence of an element but also its concentration in the sample under investigation. In the simplest case, using the so-called univariate model, a linear regression analysis is performed between analyte concentration and fluorescence emission intensity recorded at a single wavelength.Where interference effects (spectral or matrix effects) are present, however, such a simple model may not be valid and other, additional, wavelengths can be introduced into the model in order to improve its predictive performance. A wide variety of these multivariate models are readily used in XRF analysis4 and selection of appropriate ‘best’ model relies on the experience of the user.In many branches of spectroscopy in recent years, considerable attention has been devoted to developing and applying orthogonal multivariate regression models in which the independent variables are uncorrelated. These uncorrelated variables are formed from linear combinations of original, recorded variables. Many reports and discussions in the analytical and chemometrics literature have reviewed the relative merits of multivariate calibration models for quantitative analysis, and the advantages of orthogonal models such as principal components regression and partial least-squares (PLS) regression in terms of stability and robustness are well documented.5,6 Swerts et al.7 employed PLS regression for the determination of sulfur in sulfur–graphite mixtures using energy-dispersive XRF data.Although traditional least squares models failed to correct adequately for interferences and abnormal scattering of the excitation radiation, the PLS model was able to predict sulfur concentration with an accuracy of better than 5% in the concentration range 2–60% sulfur.Swerts et al.8 also successfully applied a similar PLS technique to the determination of 15 elements in Chinese porcelain, again using energy-dispersive XRF spectral data. PLS calibration models have also been studied by Urbanski and Kowalska9 for quantitative energy-dispersive XRF analysis of a range of sample types.They reported that the use of the PLS model not only gave greater robustness to interference effects but also provided new insights into aspects of the measurement methodology. Wang et al.10 compared wavelength-dispersive XRF fluorescence calibration using PLS regression with a conventional explicit calibration model employing multiple linear regression (MLR). With spectral data known to suffer from both absorption and enhancement interference effects, obtained from a set of nickel alloy samples, Wang et al.demonstrated the superiority of the PLS implicit model. One area of concern noted by Wang et al. referred to the need for further work devoted to the study of the selection of appropriate spectral regions for subsequent calibration model development for each analyte. Hence an automated quantitative analysis software package should be capable of performing wavelength selection in addition to the application of a suitable calibration model, and in addition provide a chemically sound basis for its use and application.We have recently reported the preliminary results from a study of automated variable selection methods for wavelength-dispersive XRF data and subsequent application of a PLS regression algorithm.11 The results presented here demonstrate the application of the technique employed to a study of a series of geological samples analysed for nine elements using three dispersion crystals in the spectrometer. Experimental Twenty-five geological samples representing a wide range of such materials were examined, including basalt, granite, feldspar, soils and sediments (Table 1).Obtained from a variety of sources, including the US Geological Survey, the Geological Survey of Japan and the Canadian Reference Materials Project, no attempt was made to group samples into similar classes or sub-samples. The only criterion for selection and inclusion in the study was that each sample should have a certified value for each of the nine elements determined. Supplied as powders, † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.Analyst, April 1998, Vol. 123 (537–541) 537each sample was accurately weighed and mixed with Bakelite powder in the proportions 85 + 15 to provide about 10 g of homogeneous mixture, which was pressed at 10 t in22 in an aluminium cup and then cured overnight at 60 °C to provide a permanent reference sample.All spectra were recorded and digitised using an ARL Model 8410 wavelength-dispersive XRF spectrometer (Thermo-Fi, Crawley, Sussex, UK) measuring emission intensity (counts s21) as a function of scanning angle, 2q. For the nine elements determined, three dispersing crystals were employed in the spectrometer to provide suitable wavelength coverage. Iron, manganese, potassium, calcium and titanium were determined using an LiF (200) crystal in the range 17–154° (digitised at 0.1° intervals), aluminium and silicon using a PET crystal (18–154°, at 0.1° intervals) and magnesium and sodium with an AX06 crystal (18–45°, at 0.05° intervals). All data processing and manipulation software was developed and evaluated using Mathematica V3.0 (Wolfram Research Europe, Oxford, UK) on a Pentium PC.Results and discussion The averages of the 25 sample spectra using each spectrometer dispersing crystal are illustrated in Fig. 1. Both the LiF (200) and the PET spectra are characteristic sharp-band spectra superimposed on a non-linear background typical of that obtained from a relatively low density matrix sample such as the geological materials examined here. The AX06 spectra appear broad-band owing to the greater digitisation frequency and the narrow range of angles scanned with this crystal. No pretreatment, filtering or smoothing was performed on the data prior to numerical analysis.All calibration models were developed and evaluated using mean-centred data, i.e., the average spectrum using each crystal was subtracted from each sample spectrum recorded using that crystal. Some of the problems associated with independent variable selection for subsequent calibration modelling have been discussed elsewhere,11 and an efficient and effective technique for XRF analysis is the application of a modified algorithm due to Brown et at.12,13 A linear relationship between emission intensity and analyte concentration is assumed and for each mean-centred digitised variable the following equation is applied: �y = ak·xk + ek (1) where �y is the vector of predicted analyte concentrations, xk the vector of emission intensities at 2q angle k(k = 1,...,K), ak the sensitivity (i.e., the slope of the intensity versus concentration regression line) using this variable for the analyte and ek is the error vector associated with assuming and fitting the linear model, i.e., the residuals.For iron as the analyte in the geological samples, a plot of ak versus angle is shown in Fig. 2(a). The spectral regions of high linear sensitivity to iron concentration are evident and are located about the major Ka and Kb fluorescence lines of iron. In addition, the Ka lines from calcium and potassium can also be identified, indicating a linear correlation (positive for calcium and negative for potassium) between the concentrations of these elements and the concentration of iron.In order to enhance the selectivity of the 2q variable selection algorithm, the ak values can be scaled by dividing each value by the variance, sk 2, of the residuals, ek. This serves to lessen the importance and influence of spectral regions with non-linear response and regions of spurious correlation due to high noise levels.13 Brown13 proposed this wavelength selection method using near-infrared absorbance spectra. For emission techniques, such as XRF, the dynamic range of the recorded response (i.e., emission intensity) can be several orders of magnitude greater. By performing a logarithmic transformation of the absolute sensitivity values, ak, the wide dynamic range characteristic of XRF emission intensities is suitably compressed, and the selection process is not overly dominated by any single most intense line.Hence our selection term, s, can be expressed as s a k k k = log 2 2 s (2) Table 1 Certified reference samples used.All standards corrected for elemental content from certified oxide content Standard Origin* Fe (%) Mn (%) K (%) Ca (%) Ti (%) Si (%) Al (%) Mg (%) Na (%) JB-2 GSJ 10.03 0.155 0.349 7.068 0.713 24.87 7.74 2.81 1.506 DT-N ANRT 0.44 0.006 0.099 0.029 0.839 17.04 31.25 0.02 0.030 JF-1 GSJ 0.05 0.0007 8.36 0.658 0.003 31.15 9.50 0.004 2.630 JF-2 GSJ 0.04 0.0007 10.88 0.060 0.002 30.48 9.73 0.002 1.825 JGb-1 GSJ 10.60 0.132 0.199 8.560 0.970 20.31 9.32 4.72 0.912 SO-3 CCRMP 1.45 0.052 1.16 14.64 0.198 15.76 3.06 5.08 0.749 SARM-3 MINTEK 6.64 0.596 4.57 2.300 0.288 24.50 7.20 0.169 6.210 UB-N ANRT 5.59 0.093 0.017 0.860 0.066 —† —† —† —† STM-1 USGS 3.32 0.178 3.54 0.786 0.078 27.88 9.71 0.061 6.63 RGM-1 USGS 1.30 0.028 3.60 0.829 0.162 34.34 7.24 0.166 3.02 JA-3 GSJ 4.61 0.082 1.17 4.488 0.407 29.11 8.22 2.20 2.25 BIR-1 USGS 7.88 0.136 0.025 9.520 0.576 22.33 8.10 5.84 1.298 JG-2 GSJ 0.64 0.011 3.918 0.572 0.024 35.97 6.55 0.024 2.63 MAG-1 USGS 4.76 0.077 2.988 1.070 0.420 32.54 8.64 1.809 2.84 SGR-1 USGS 2.03 0.026 1.328 5.989 0.158 13.20 3.44 2.677 2.218 G-2 USGS 6.10 0.115 0.794 8.210 0.290 32.29 8.12 0.452 3.027 QLO1 USGS 2.66 0.069 2.997 2.300 0.360 30.64 8.54 0.603 3.116 SDO-1 USGS 4.42 0.093 2.656 1.000 0.588 —† —† —† —† JG-1A GSJ 1.43 0.046 3.329 1.520 0.150 33.75 7.51 0.416 0.74 SO-1 CCRMP 6.0 0.085 2.64 1.760 0.520 25.70 9.29 2.310 2.000 AGV-1 USGS 4.53 0.075 2.400 3.530 0.623 27.48 9.05 0.923 3.160 SO-4 CCRMP 2.27 0.062 1.727 1.107 0.340 31.96 5.40 0.537 0.987 DR-N ANRT 6.78 0.170 1.410 5.040 0.650 24.71 9.25 2.650 2.218 FK-N ANRT 0.06 0.004 10.63 0.079 0.010 30.39 9.83 0.006 1.910 JA-2 GSJ 4.30 0.085 1.490 4.631 0.402 26.26 8.09 4.63 2.285 * CCRMP, Canadian Certified Reference Material Project; ANRT, Association Nationale de la Recherche Technique, Paris; GSJ, Geological Survey, of Japan; USGS, United States Geological Survey, Reston; MINTEK, Council for Mineral Technology, South Africa.† Not used in calibration model. 538 Analyst, April 1998, Vol. 123A plot of sk versus 2q angle is presented in Fig. 2(b). With spurious correlations now removed, the prominent peaks in this spectrum are due to the Ka and Kb (first and second-order) lines of iron. It now remains to identify and select the regions about these peaks for inclusion in the calibration models. Brown13 has proposed a novel means for the selection of the optimum number and identity of the spectral regions for calibration. The selection process is achieved using the equation Z C Sk k K = = ¢å ( ) h 1 (3) where 12h is some selected confidence level and C X k ( ) ( [ ]) h h = - 1 2 (4) is the square root of tabulated chi-squared values on k degrees of freedom and KA is the reduced number of variables, KA < < K.To choose the value for KA, and the corresponding spectral regions, the computed s values [eqn.(2)] are ranked from largest to smallest and, by eqn. (3), the function Z is minimized. From the geological samples and for iron, the form of eqn. (3) is shown in Fig. 3. The number of 2q variables to be selected and used is given by the location of the minimum for the curve and the identity of the corresponding variables obtained from the ranked list. For any spectral data set, the location of the minimum from eqn. (3) is dependent on the chosen significance level, 12h.Hence, for iron and h values of 10, 1 and 0.1% the sets (A) of spectral 2q variables from the geological sample spectra are as follows: h = 10%: A0.9 = {57.5, 57.6, 57.7, 57.8} h = 1%: A0.99 = {51.8, 51.9, 52.0, 57.5, 57.6, 57.7, 57.8, 57.9, 148.8, 148.9} h = 0.1%: A0.999 = {51.8, 51.9, 52.0, 57.5, 57.6, 57.7, 57.8, 57.9, 58.0, 148.6, 148.7, 148.8, 148.9} (5) Decreasing the significance level increases the number of 2q variables selected, and each larger set contains the elements of a previous set, i.e., A0.9 … A0.99 … A0.999 (6) Reference to look-up XRF reference tables for the LiF200 dispersing crystal identifies the four variables in set A0.9 as being on the Fe Ka line.The A0.99 set has added the Fe Kb line and the second-order Fe Ka line. The final set, A0.999, merely adds more points from these three lines. This process of selecting 2q variables was undertaken for each of the nine analytes using all 25 reference samples, at each of the three significance levels indicated.In every case the technique resulted in a dramatic reduction of the number of variables and a chosen set of variables that can be readily Fig. 1 Average XRF spectra from 25 geological samples using (a) an LiF (200) crystal, (b) a PET crystal and (c) an AX06 crystal. Fig. 2 (a) The linear sensitivity coefficient, ak, for iron in geological samples as a function of scan angle and (b) the linear selectivity coefficient, sk, as a function of scan angle.Fig. 3 The selectivity function, Z, for iron in geological samples as a function of number of variables chosen. Analyst, April 1998, Vol. 123 539checked automatically for chemical consistency. Table 2 summarises the results of variable selection for the nine analytes. In all cases the technique has selected the appropriate Ka line for each element and, with the exception of iron, the algorithm indicates that only this line is required for subsequent calibration.Compared with simple univariate regression analysis, however, the technique identifies a range of wavelengths (2q values) about each line, thus providing an averaging effect and reducing the influence of spurious noise at one particular wavelength. For the three sets of 2q variables (h = 0..01, 0.001) selected for each analyte element, calibration models were developed using PLS regression and MLR. In addition to these multivariate models, univariate ordinary least-squares (OLS) regression was also undertaken using the characteristic Ka line of each element. The PLS algorithm employed was based on that reported by de Jong14 and referred to as SIMPLS.The algorithm avoids deflation of the full data matrix, and for single component analysis it has been shown to to yield equivalent results to the more common NIPALS algorithm described by Martens and Næs.15 Calibration model performance was compared using a full miss-one-out strategy on all 25 samples.For each model evaluated all samples less one were used to derive the model coefficients with the odd sample used for prediction or validation. The process was repeated 25 times, with each sample in turn used for prediction purposes. Numerical comparison of models was achieved by calculating the rootmean- square error of prediction (RMSEP): RMSEP y y n i i i n = - = å ( � )2 1 (7) where �y 2 y represents the residuals between predicted and known concentrations and n is the number of samples examined.Table 3 presents a summary of the results obtained. The RMSEP values indicated for PLS and MLR are those obtained using the set of variables giving the lowest RMSEP values with the PLS model. RMSEP values provide a useful quantitative summary of the performance of calibration models and it is evident from Table 2 that in every case the PLS model is as good as or superior to either of the more classical approaches. A more detailed picture of model performance can be obtained by visual inspection of residuals plots.Fig. 4 presents such a plot obtained from the determination of calcium. One feature of Fig. 4 is particularly worthy of note, viz., the calcium concentration values of the samples are not evenly distributed across the range encountered. This is not surprising given that the samples were not selected for any single, particular element but merely as representatives of the class of materials.For calcium, one sample has a considerably higher concentration than the others, hence with a miss-one-out validation strategy this sample will be determined by extrapolation from the calibration model developed using the remaining samples. It is not surprising in these circumstances that OLS and Table 2 Results of the variable selection algorithm at three significance levels for the nine elements studied Element and crystal Fe Mn K Ca Ti Si Al Mg Na LiF(200) LiF(200) LiF(200) LiF(200) LiF(200) PET PET AX06 AX06 h = 0.1— No.of variables 4 3 2* 3* 2* 11* 5* 18* 16* Identity Fe Ka Mn Ka K Ka Ca Ka Ti Ka Si Ka Al Ka Mg Ka Na Ka h = 0.01— No. of variables 10* 3 3 4 5 14 10 23 20 Identity Fe Ka Mn Ka K Ka Ca Ka Ti Ka Si Ka Al Ka Mg Ka Na Ka Fe Kb Fe Ka (2nd order) h = 0.001— No. of variables 14 6* 6 4 6 15 14 28 22 Identity Fe Ka Mn Ka K Ka Ca Ka Ti Ka Si Ka Al Ka Mg Ka Na Ka Fe Kb Fe Ka (2nd order) * Indicates the selection providing the lowest RMSEP value by PLS regression. Table 3 RMSEP values for validation using univariate ordinary leastsquares (OLS), multiple linear regression (MLR) and partial least-squares (PLS) regression. The figures in parentheses indicate the number of factors used in the PLS model RMSEP value Element OLS MLR PLS Fe 0.798 1.060 0.521 (4) Mn 0.017 0.013 0.013 (4) K 0.177 0.187 0.178 (1) Ca 0.924 1.446 0.553 (1) Ti 0.045 0.049 0.039 (1) Si 1.471 2.016 1.482 (1) Al 0.495 0.506 0.499 (2) Mg 0.543 2.685 0.530 (3) Na 0.446 0.968 0.465 (1) Fig. 4 Residuals from the determination of calcium with a miss-one-out strategy using (2) OLS, (8) MLR, and (Æ) PLS models. 540 Analyst, April 1998, Vol. 123MLR fail to predict accurately the concentration of this sample. It is noteworthy that PLS regression performs well and is better for extrapolation. Other elements in these samples exhibiting a similar heterogeneous distribution of concentrations with a single outlying sample are manganese and aluminium.For manganese, both the PLS and MLR models provide similar, good prediction values for the outlying sample, and for aluminium PLS and OLS perform equally well. Conclusion The results presented here, obtained from a diverse range of geological samples, demonstrate the potential for an automated variable selection and calibration procedure for quantitative XRF analysis. By providing an automatic means of selecting the spectral regions suitable for calibration, and which can be demonstrated as chemically relevant, and then applying PLS regression, a general calibration scheme is afforded that requires minimal input from the user and does not rely on the user’s prior experience or knowledge of the samples.Results for PLS regression are as good as or better than those obtained using more conventional methods, indicating that the technique may be used as a general model for XRF analysis. It remains to evaluate the performance of the procedure on a wider range of sample types and more analytes, including its ability to correct for matrix, physical and spectral interference effects, and to integrate the software with the qualitative expert system analyser. References 1 Abbott, P. H., and Adams, M. J., Lab. Autom. Inf. Manage., 1995, 31, 115. 2 Abbott, P. H., and Adams, M. J., Lab. Autom. Inf. Manage., 1996, 31, 211. 3 Abbott, P. H., and Adams, M. J., X-Ray Spectrom., 1997, 26, 125. 4 Lachance, G. R., and Claisse, F., Quantitative X-Ray Fluorescence Analysis, Wiley, Chichester, 1995. 5 Brereton, R. G., Chemometrics, Ellis Horwood, Chichester, 1990. 6 Phatak, A., and de Jong, S., J. Chemom., 1997, 11, 311. 7 Swerts, J., van Espen, P., and Geladi, P., Anal. Chem., 1993, 65, 1181. 8 Swerts, J., Aerts, A., De Biscop, N., Adams, F., and van Espen, P., Chemom. Intell. Lab. Syst., 1994, 22, 97. 9 Urbanski, P., and Kowalska, E., X-Ray Spectrom., 1995, 24, 70. 10 Wang, Y., Zhao, X., and Kowalski, B., Appl. Spectrosc., 1990, 44, 998. 11 Adams, M. J., and Allen, J. R., J. Anal. At. Spectrom., 1998, in the press. 12 Brown, P. J., Spiegalman, C. H., and Denham, M. C., Philos. Trans. R. Soc. London, Ser. A, 1991, 337, 311. 13 Brown, P. J., J. Chemom., 1992, 6, 151. 14 de Jong, S., Chemom. Intell. Lab. Syst., 1993, 18, 251. 15 Martens, H., and Næs, T., Multivariate Calibration, Wiley, Chichester, 1996. Paper 7/07073J Received September 30, 1997 Accepted December 15, 1997 Analyst, April 1998, Vol. 123
ISSN:0003-2654
DOI:10.1039/a707073j
出版商:RSC
年代:1998
数据来源: RSC
|
3. |
Development of a photoacoustic gas sensor forin situand on-line measurement of gaseous water and toluene† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 543-545
A. Beenen,
Preview
|
PDF (66KB)
|
|
摘要:
Development of a photoacoustic gas sensor for in situ and on-line measurement of gaseous water and toluene† A. Beenen and R. Niessner* Institute of Hydrochemistry, Technical University of Munich, Marchioninistr. 17, D-81377 Munich, Germany, Developments towards a portable sensor system for the detection of benzene, toluene and xylene for screening are described. The monitoring system is based on the resonant photoacoustic technology. As light sources, NIR laser diodes at wavelengths of overtone vibrations of the alkyl CH bonds are used.The resonator is of cylindrical shape with diameter 2R = 5 cm and total length L = 10 cm and is operated in its first azimuthal mode. The acoustic wave is detected by an electret microphone. Test measurements of water and toluene vapor are presented. For the detection of water vapor a 1.31 mm NIR laser diode was used. Water concentrations down to 0.5 mg l21 in air could be measured. For the detection of toluene vapor a wavelength of 910 nm exciting the third overtone vibration of the alkyl CH bonds was used.The sensitivity of 3.2 mg l21 in air may be increased by several orders of magnitude as soon as suitable laser diodes with wavelengths fitting the absorption lines of the first overtone become available. Keywords: Near-infrared laser diodes; photoacoustic spectroscopy; sensor system; overtone; benzene; toluene; xyleve; water Fast and on-line screening methods in analytical chemistry play an ever-increasing role in industry, medicine and environmental sciences.1-5 In addition to fluorescence, photoemission, plasma emission and Raman spectrometry, photoacoustic spectrometry belongs to those techniques where signal formation is related to the incident light intensity.Hence the use of laser sources is extraordinarily advantageous. In particular, trace gas detection with photoacoustics has found a wide range of applications.6 The photoacoustic spectrometric technique measures the energy deposited by absorbed light, and differs in this respect from conventional optical techniques which measure light attenuation or extinction, i.e., scattering and absorption.Photoacoustic spectrometry measures light absorption directly via sound waves produced by absorption of the modulated light. The theory of the photoacoustic signal production in a relaxing gas was described by Da Silva.7 In particular, the infrared range is widely used, because most pollutants show absorption lines in this region.A typical light source for trace gas analysis is the CO2 laser.8,9 It has a high optical power and emits in the mid-IR (MIR) region. However, this laser system is expensive, heavy and space consuming. Therefore, a cheap and portable sensor system for field use requires a different solution. Unfortunately, laser diodes with wavelengths in the MIR and IR regions have to be cooled very efficiently. Another problem is the small output intensity of lead–salt diode lasers ( < 1 mW), commonly used.10 In contrast, laser diodes with wavelengths in the NIR region are easy to operate at room temperature.Another advantage of NIR laser diodes is the availability of robust glass fibers for these wavelengths. Recently, several photoacoustic applications and developments using NIR laser diodes have been presented.3,5,11,12 Owing to their small size, long lifetime, easy operation and low price, their application in sensitive trace gas detection systems seems to be especially promising.Many air pollutants are of concern owing to their potential carcinogenic properties, especially benzene, toluene and xylene (BTX). These substances are caused by traffic exhaust, but are also used in industry. To correct for the interference between the analyte to be detected and other substances in a photoacoustic measurement, different wavelengths have to be used. Developments towards a portable sensor system for application for screening using NIR laser diodes are described in this paper.Experimental Laser diode wavelength and analyte One of the most important issues in designing a spectroscopic sensor system is the choice of the detection wavelengths. As the fundamental of vibrational modes in BTX are located in the MIR region, the use of NIR laser diodes requires the excitation of overtones. The absorption coefficients at overtone wavelengths are at least one order of magnitude lower compared with the fundamental of vibrational modes.However, this can be compensated for by increasing the intensity of excitation. For benzene and toluene it is favorable to use the absorption line of the C–H stretch vibration at 1.67 and 1.68 mm, respectively. Laser diodes with a wavelength of 1.67 mm can be produced, but they are not commercially available. Therefore, the measurements were performed at the third overtone of the alkyl CH bonds of toluene at 910 nm [EG&G Princeton Applied Research (Princeton, NJ, USA) C86135ECD Instrument].13–16 Fig. 1 shows the absorbance of benzene, toluene and xylene at the third overtone vibration. It may be argued that using the CH † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997. Fig. 1 Absorbance of BTX vapor at the third overtone vibration. Analyst, April 1998, Vol. 123 (543–545) 543Air and water vapor or toluene vapor from mixing nozzle Electret microphone Adsorption at active charcoal Power meter Resonator PC Lock-in amplifier Collimated and modulated laser diode light Current source and temperature stabilized vibrations leads to a loss of specificity for the sensor system because CH bonds are present in virtually all organic molecules.However, Fig. 1 shows that the frequencies of the CH vibrations depend on the chemical environment and, for example, the aromatic and the alkyl CH vibrations are considerably shifted with respect to each other.Although the absorption of water is low at the wavelength chosen, its comparatively high concentration in ambient air may lead to considerable cross-reactivity. Therefore, the concentration of water has to be measured independently. The wavelength for water detection was chosen as 1.31 mm [Mitsubishi Electric (Tokyo, Japan) ML7781A-01, Instrument]. In the case of only two analytes, namely water and toluene, the problem of determining the concentrations reduces to two equations with two unknowns: A910 = ctat910 + cwaw910 A1310 = ctat1310 + cwaw1310 (1) Where A910 and A1310 denote the measured photoacoustic signals at the respective wavelengths, ct is the toluene concentration, cw is the water concentration, at910 is the photoacoustic response of for toluene at 910 nm, and aw910 the response for water at 910 nm; at1310 and aw1310 are defined correspondingly. For the wavelengths chosen at1310 is negligible and aw910 is small.With more analytes, we need more laser diodes and we have to use the multivariate calibration method. Apparatus and procedure Fig. 2 shows a schematic view of the sensor system, which was used to test the laser diodes for the photoacoustic measurement. The acoustic resonator is of cylindrical shape with a diameter 2R = 5 cm and total length of 10 cm. The material of the cylindrical resonator is brass and all supply is of high-grade steel. This resonator is similar to that used by Adams.17 It is operated in its first azimuthal mode (off-axis excitation; see Fig. 2) at 4050 Hz. The windows for the laser beam have a diameter of 8 mm and consist of quartz glass with an anti-reflection coating. The laser diode light was collimated and focused in the middle of the resonator. Intensity modulation of the laser radiation was performed using a sine-wave generator for modulation of the operating current. The laser diodes were temperature stabilized by a Peltier element [Profile (Karlsfeld, Germany) LD 1000] and the operating temperature was 15 °C.The laser intensity is monitored by a power meter. The quality factor was determined as Q = 250. An electret microphone [Sennheiser (Wedemark, Germany) KE 4-211-2] was mounted at the inner cell wall at positions of maximum pressure variation. The sensitivity of the microphone was 40 mV Pa21. The sine-wave generator both modulates the operating current and serves as the trigger for the lock-in amplifier.System control and signal processing was performed with a dual-phase lock-in amplifier [Stanford Research System (Sunnyvale, CA, USA) SR 830]. A low-noise preamplifier [Stanford Research System SR 550] was used. The gas inlet ports were placed on the axis of the cylinder at a nodal plane of the azimuthal modes to minimize the signal perturbations when floating sample gas through the cell. The maximum gas flow rate was 1.5 l min21.Gas generation For water vapor generation, air was metered into a chamber of hot, refluxing, distilled water, and subsequently cooled to the required temperature with coolant. The flux of the carrier gas through the liquid was kept constant. The surplus of this gas was drawn off. Different humidities were adjustable with a ring-gap mixing nozzle with a determined ratio of saturated water vapor and dry air.18 To allow for stable humidity conditions, it was necessary to wait at least twice the time required for changing the whole volume (200 ml) of the resonance cell.For the preparation of toluene vapor with a well defined concentration a vessel containing the solvent was kept in a constant temperature bath. Nitrogen gas was flowing slowly over the surface of the solvent and subsequently mixed with a second adjustable nitrogen gas flow in a ring gap mixing nozzle. Conventional analysis To verify the different concentrations, conventional analysis was required.For the measurement of water vapor, a humidity sensor was used. Below 5% relative humidity the measurements were performed gravimetrically using phosphorus pentoxide in a glass tube. The measurement of toluene vapor was performed by gas chromatography. The sample was taken directly from the flowing gas with a gas-tight injection syringe. Results of measurements To find the resonance point of operation, the modulation frequency was scanned over the acoustic resonance in steps of 2 Hz.One advantage of this scanning technique is that it is insensitive to shifts of the resonance frequency which may be induced by temperature variations. Fig. 3 shows eight different acoustic resonance spectra at three different concentrations of water vapor. A Lorentzian function could be fitted to the Fig. 2 Experimental set-up. Fig. 3 Eight different acoustic resonance spectra at three different concentrations of water vapor. 544 Analyst, April 1998, Vol. 123resonance profile. The frequency of maximum amplitude was chosen for the measurements and the amplitude was recorded as a function of the concentration of the analyte. Fig. 4 shows measurements of water vapor. With a resonant photoacoustic cell and an NIR laser diode, it is possible to detect water concentrations down to 0.5 mg l21 in air. Fig. 5 shows the results for toluene vapor in air. The photoacoustic amplitude at each toluene concentration was measured three times. The detection limit was calculated using three times the value of the standard deviation of the signal at zero concentration.For toluene vapor the detection limit is 3.2 mg l21. However, it must be taken into account that using the wavelength of the first overtone the sensitivity may be increased by a factor of up to 1000. Conclusion The development of a simple sensor system for trace gas detection based on NIR laser diodes has been described. A setup for the detection of toluene and water vapor in a continuous air flow led to a sensitivity of 3.2 and 0.5 mg l21, respectively.An increase in sensitivity may be achieved by using laser diodes at lower overtones. The sensor system could be used for a variety of other trace gases. References 1 Brand, C., Winkler, A., Hess, P., Miklos, A., and Bozoki, Z., Appl. Opt., 1995, 34, 3257. 2 Bijnen, F. G. C., Harren, F. J. M., Hackstein, J. H. P., and Reuss J., Appl. Opt., 1996 35, 5357. 3 Spanner, G., and Niessner, R., Fresenius’ J.Anal. Chem., 1996, 354, 306. 4 Beenen, A., Spanner, G., and Niessner, R., Appl. Spectrosc., 1997, 51, 51. 5 Petzold, A., and Niessner, R., Appl. Phys. B, 1996, 63, 191. 6 Sigrist, M. W, Analyst, 1994, 119, 525. 7 Da Silva, R. M, Can. J. Phys., 1986, 64, 1098. 8 Meyer, P. L., and Sigrist, M. W., Rev. Sci. Instrum., 1990, 61, 1779. 9 Trushin, S. A., Ber. Bunsenges. Phys. Chem., 1992, 96, 319. 10 Tacke, M., Infrared Phys. Technol., 1994, 36, 447. 11 Feher, M., Jiang, Y., Maier, J. P., and Miklos, A., Appl. Opt., 1994, 33, 1655. 12 Bechara, J., Karecki, D. R., Mackay, G. I., Schiff, H. I., and Nadler, S., SAE (Soc. Automot. Eng.) Tech. Pap., 1994, No. 940824, 1. 13 Gough, K. M., and Henry, B. R., J. Phys. Chem., 1984, 88, 1298. 14 Reddy, K. V., Heller, D. F., and Berry, M. J., J. Chem. Phys., 1982, 76, 2814. 15 Reddy, K. V., and Berry, M. J., Chem. Phys. Lett., 1977, 32, 111. 16 Henry, B. R., J. Phys. Chem., 1976, 80, 2160. 17 Adams, K. M., Appl. Opt., 1988, 27, 4052. 18 Nelson, G. O., in Controlled Test Atmospheres, Ann Arbor Science Publishers Inc., Ann Arbor, MI, 1980, pp. 185–186. Paper 7/07113B Received October 1, 1997 Accepted January 5, 1998 Fig. 4 Results of measurements of water vapor. Fig. 5 Results of measurements of toluene vapor. Analyst, April 1998, Vol. 123 545
ISSN:0003-2654
DOI:10.1039/a707113b
出版商:RSC
年代:1998
数据来源: RSC
|
4. |
Photoacoustic depth profiling of layered samples† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 547-550
C. Kopp,
Preview
|
PDF (76KB)
|
|
摘要:
Signal generator Digital storage oscilloscope Computer Sensor head Preamplifier Dye laser Excimer laser Photoacoustic depth profiling of layered samples† C. Kopp and R. Niessner* Institute of Hydrochemistry, Technical University of Munich, Marchioninistr. 17, Munich, D-81377 Germany. E-mail: C.Kopp@ws.chemie.tu-muenchen.de A photoacoustic sensor for the indirect detection of acoustic waves on the basis of a prism was developed. This wide-band piezoelectric transducer allows temporal resolutions of better than nanoseconds.Owing to the finite acoustic velocity it is possible to distinguish between sound waves generated in different layers. Suitable samples were built up to estimate the depth profiling capabilities of the developed sensor. For a simple experiment the lower limit for the depth resolution in an aquatic system is about 30 mm, corresponding to a temporal resolution of 20 ns. Only qualitative results were obtained with more complex layered samples.Keywords: Photoacoustic spectrometry; pulsed excitation; depth resolved; layered samples; waveform reconstruction The analysis of absorption spectra is a common technique in different fields of science to identify specific ingredients. Because of the inhomogenities in almost every relevant sample, depth-resolved measurements offer an important gain of information. An interesting aspect of this approach is the biological film in sewage treatment systems, used for the removal of organic and inorganic pollutants. For the purpose of reliable system operation, the thickness and the effectiveness of the biofilm have to be monitored.1 Both requirements can be achieved with depth profiles of specific components obtained by absorption spectra. One optical technique capable of the depth-resolved determination of absorption spectra is photoacoustic spectrometry (PAS).When a short laser pulse interacts with condensed matter, the absorbed energy is converted into heat by fast nonradiative relaxation processes.Subsequently, the thermal expansion of the instantaneously heated sample causes an acoustic shock wave in the irradiated volume. The corresponding photoacoustic pressure rise PPAS of the shock wave is directly proportional to the absorption coefficient ma (cm21) of the sample and the energy fluence W (J cm22) at the sample surface: P c C PAS s p a = b m W (1) where cs (cm s21) is the speed of sound, Cp (J g21 K21) is the heat capacity at constant pressure and b (K21) is the thermal coefficient of volume expansion in the sample.2 A more detailed mathematical derivation of PA signal generation can be found elsewhere.3–6 Therefore, the profile of an acoustic signal is determined by the physical properties of the sample and by the characteristic laser irradiation parameters such as pulse duration, beam diameter and energy fluence.If these parameters are known, it is possible to determine the absorption coefficient ma of the sample.2 The profile of the photoacoustic pressure distribution can be detected with microphones, fiber-optic sensors and piezoelectric transducers depending on the light source.3 For the detection of acoustic pulses generated by excitation with pulsed radiation, piezoelectric transducers are widely used because of the limited bandwidth of microphones. Owing to the better acoustic impedance matching between piezoelectric materials and liquids, piezoelectric detection is preferred for liquid samples.An important aspect, especially for the design of the transducer, is the distortion of the acoustic shock wave during propagation from the excited volume to the pressure-sensitive element. Such phenomena distorting the profile of the acoustic shock wave include diffraction, attenuation, stress relaxation and transmittance through interfaces. Additionally, the transducer response must be taken into account.2 Another advantage of PAS, in contrast to conventional absorption spectrometry, is the ability to obtain absorption spectra of opaque samples.3 Using time-resolved detection techniques, it is possible to distinguish between signals generated in different layers.2,6 The depth profiles achieved by the analysis of the time delay are examined in this paper.Experimental Apparatus Our experimental set-up for time-resolved PAS is presented in Fig. 1. A dye laser (FL 3002; Lambda Physik, G�ottingen, Germany) pumped by an XeCl excimer laser (EMG 201 MSC; Lambda Physik) as the light source is used.With the laser dye coumarin 153 the wavelength can be tuned between 520 and 600 nm. The laser pulses have a symmetrical shape with a duration of about 10 ns, and are transferred via an optical fiber to the sensor head.7 The signal originating from the detector is fed directly into a preamplifier (Femto Messtechnik, Berlin, Germany). The amplified output signal is recorded with a digital storage oscilloscope (9310 AM; LeCroy, NY, USA) for a period of about 1–10 ms.For the purpose of synchronization both the laser and the digital storage oscilloscope (DSO) are triggered with a signal generator. To improve the signal-to-noise-ratio, the time-resolved PA signal is averaged by the DSO over 30–100 pulses. The averaged PA signal is transmitted to a personal computer for further data treatment. † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.Fig. 1 Experimental set-up for PA depth profiling. Analyst, April 1998, Vol. 123 (547–550) 547Fiber coupler PVDF foil Preamplifier Prism PVDF-foil Glass prism Seal Glass plate Air Absorbing liquid Photoacoustic sensor head Similarly to Karabutov et al.,6,8 a sensor head for the indirect detection of acoustic waves on the basis of a prism was developed (Fig. 2). The surface of the medium under study is irradiated through a side and the base of a transparent prism.A wide-band piezoelectric poly(vinylidene fluoride) (PVDF) film is coupled to the opposite side of the base. The preamplifier is positioned directly behind the piezoelectric foil. Both the preamplifier and the PVDF film are shielded from the electrical background caused by the laboratory surroundings. To obtain a nearly uniform cross-sectional intensity distribution, the excitation beam was enlarged to a diameter of about 10 mm.This is essential to achieve the generation of plane acoustic waves in the medium under study.9 The parallel orientation of the wideband piezoelectric film and the excited acoustic wave phase front is guaranteed by the fabrication of the prism. This transducer design allows the detection of acoustic waves with temporal resolutions of better than nanoseconds. The intensity of the laser light (the peak intensities were in the range 0.3–1 mJ depending on the wavelength) was adjusted by means of wire screens attenuating the pump beam of the dye laser.Results and discussion A simple way to estimate the capability of the experimental setup in relation to depth profiling is the following experiment. A black absorber was positioned at a well defined distance da from the sensor head with water as coupling medium (similarly to the experiment reported by Niessner and co-workers7,10). A micrometer was used for the precise adjustment of the distance da. Therefore, the error in da is in the order of a few micrometers.Under these experimental conditions it is necessary to measure the largest distance da first, otherwise the formation of air bubbles between the surface of the sensor head and the black absorber cannot be avoided, increasing the distance da from 100 to 200 mm. This measurement procedure results in a small offset, when the distance da approaches zero. Fig. 3 shows the dependence of the time delay of the leading acoustic wavefront on da.These data points represent the average of three measurements. The offset of the regression line is a direct consequence of the offset of da. Good linearity of the experimental data is expected for this kind of transit time measurement. This allows one to deduce the acoustic velocity va in the intermediate medium. In this case, an acoustic velocity of 1.52 km s21 was determined, in good agreement with literature data11 for water. For this periment, the depth resolution is about 30 mm (equivalent to a time resolution of about 20 ns).In contrast to the previous experiment, we now inserted an absorbing layer between the sensor head and a nearly transparent layer. This configuration can be attained with an absorbing liquid confined between two glass plates (100 mm thickness). One glass plate was replaced by the sensor head, improving the coupling between the absorbing layer and the detector (see Fig. 4). The air behind this sandwich represents the transparent layer.An aqueous solution of CuCl2 was chosen as a light-absorbing fluid. The absorption coefficient of this solution determined with a UV/VIS spectophotometer (Beckman, Fullerton, CA, USA) was 9.1 cm21 at an excitation wavelength of 550 nm. Three such sandwiches with different thicknesses td of the absorbing liquid (0.5, 1 and 2 mm) were built up. Fig. 5 shows the time-resolved PA signals of these samples. A comparison of these three signals reveals some promising features.The first pressure pulse arriving at the detector is almost independent of the thickness td of the absorbing layer. The full width at half maximum (FWHM) is identical for the different thicknesses. Small deviations in the amplitude of the pulses caused by the variations of the laser’s pulse intensity are observable. However, the second pulse contains some information about the thickness td of the layer. The time delay of the second pulse with respect to the first agrees very well with the time t = td/va (where va = acoustic velocity) that sound needs to propagate over a distance td.This means that the second pulse was generated in the vicinity of the glass plate. Because of the phase reversal, this pulse must have been generated at a boundary to a layer with a smaller acoustic impedance.12 The only possibility is the boundary layer between glass and air (or liquid and air if the thin glass plate has no impact on the pressure pulse).These results confirm that the laser pulse causes an expansion of the whole absorbing layer nearly simultaneously, resulting in pressure gradients only at the boundary layers. The cavity effect suggests that such a Fig. 2 Sensor head for indirect photoacoustic signal detection. Fig. 3 Dependence of the time delay on the distance da. Fig. 4 Schematic diagram of the coupling between a liquid layer and the sensor head. Fig. 5 Time-resolved PA signal of three different sandwiches consisting of an absorbing fluid confined between two thin glass plates.Excitation wavelength: 550 nm. 548 Analyst, April 1998, Vol. 123sandwich consisting of two glass plates confining an absorbing liquid is not suitable for further estimations of the depth resolution. However, this is only true for a thickness td > 100 mm. However, in the range 30–40 mm (corresponding to the FWHM of the pressure pulse) or less, the two pulses mentioned above merge together. One reason for this assumption is the work of Karabutov et al.6 on measurements of arrays of such sandwiches.They observed only one pulse. Because of the lack of information about the thickness of their liquid layers, more detailed comparison is not possible. To achieve the condition of a varying absorbing layer followed by a relatively transparent layer, thin adhesive plastic films were chosen. These films have some advantages: (1) the known thickness makes it very easy to build up well defined samples; (2) owing to the lack of boundary layers such as glass plates, there are only small variations in the acoustic impedance of the sample reducing the waveform distortion by reflections; the inhomogeneities in the acoustic impedance of the sample are caused by the adhesive on the film; and (3) the reproducibility with such samples is much better than that with sandwiched solutions because of the very simple sample preparation. Table 1 summarizes the experimentally determined physical properties of the films used.The thickness was measured with a micrometer, the density was examined by gravimetry and the absorption coefficient was determined with a spectrophotometer. The time-resolved PA signal of a black absorber behind a layer of several films was used to derive the acoustic velocity. This value was checked against the eigen resonances of the film. Fig. 6 shows the time-resolved photoacoustic signals S1(t) of a single layer of a colorless [S1c(t)] and a yellow film [S1y(t)] excited at a wavelength of 570 nm.A comparison of these signal amplitudes S1(t) with the absorption coefficients ma of the corresponding films (Table 1) confirms the proportionality [see eqn. (1)] between S1(t) and ma. This proof is supported numerically by the agreement of the ratio may/mac = 2.48 ± 0.18 (see Table 1) with the ratio S1y(tmy)/S1c(tmc) = 2.63 ± 0.21 [tm indicating the time of the first four maxima and minima of S1y(t) and S1c(t)]. A simple division of S1y(t) and S1c(t) is not possible, because the locations of the maxima and minima of S1y(t) and S1c(t) deviate.This deviation is caused by the different eigen resonances, which are dependent on the film thickness (see Table 1). The shapes of acoustic signals excited (l = 570 nm) in samples consisting of three equal layers (the colorless and the yellow film) are presented in Fig. 7. There is also a reconstruction of each PA signal (thick curve) added.The acoustic transit time Di through the ith layer of the corresponding film is marked. The reconstruction is based on the PA signal S1(t) of a single layer sample of each film. Hence the PA signal of the second [third] layer is to a first approximation just the signal of a single layer shifted in time S1(t-D1) [S1(t-D1-D2)]. The reconstruction of the PA signals for the whole sample is the superposition of the signals S1(t), S1(t-D1) and S1(t-D1-D2).The agreement between the original and the reconstructed signal up to the time D1 + D2 + D3 is surprising considering the simple treatment. This result supports the assumption that the detector geometry employed generates plane waves. A greater problem is the explanation of the deviation in the later time period. This effect only concerns the ringing of each layer. However, it seems that the ringing of the first layer is only affected after a delay time of 2D1 corresponding to one complete oscillation (one dilatation and one compression) of the film.If the ringing had been affected at an earlier time the deviation should also have been shifted to earlier times. This suggests that each layer oscillates almost independently of the adjacent layer, but the eigen resonance of the whole sample is superimposed on these oscillations (in analogy with a system of weakly coupled pendulums). A more complex sample consists of a combination of colorless (c) and yellow (y) films.Fig. 8 shows two examples of Table 1 Summary of the experimentally determined physical properties of the adhesive plastic films used Property Colorless Yellow Thickness/mm 85 ± 5 105 ± 5 Density/g cm23 1.27 ± 0.08 1.29 ± 0.07 Acoustic velocity/km s21 2.06 ± 0.12 2.10 ± 0.10 Absorption coefficient (570 nm)/cm21 1.25 ± 0.07 3.11 ± 0.15 Fig. 6 Shapes of acoustic signals S1(t) of single-layer samples. Excitation wavelength: 570 nm. Fig. 7 Shapes of acoustic signals excited in samples of three equal layers. Excitation wavelength: 570 nm.Fig. 8 Shapes of acoustic signals excited in three layer samples consisting of combinations of colorless (c) and yellow (y) films. Excitation wavelength: 570 nm. Analyst, April 1998, Vol. 123 549such combinations. The PA signal was reconstructed in the same way as described above. However, the agreement between the original signal and the reconstruction is poor, especially for the sample y–c–y.These samples demonstrate the need for a more sophisticated reconstruction scheme including parameters such as acoustic attenuation and transmission and reflection coefficients of each layer. Conclusions The photoacoustic sensor for the indirect detection of acoustic waves presented here allows the determination of the depth of a black absorber immersed in water with a resolution of about 30 mm (corresponding to a temporal resolution of 20 ns). The reconstruction of the shape of the PA signal of more complex samples consisting of thin adhesive plastic films was in good agreement with the original signal only for three equal layers.Combinations of different films resulted in strong deviations, demonstrating the need for a more sophisticated reconstruction scheme. According to the measurements mentioned above, it seems possible to resolve layers of at least 100 mm thickness reliably, with minimum absorption coefficient of about 1-2 cm21. References 1 Wilderer, P. A., Characklis, W. G., in Structure and Function of Biofilms, ed. Bernhard, S., Wiley, New York, 1989, pp. 5–17. 2 Oraevsky, A. A., Jacques, S. L., Esenaliev, R. O., and Tittel, F. K., in Proceedings on Advances in Optical Imaging and Photon Migration, ed. Alfano, R. R., Optical Society of America, Bellingham, Washington, 1994, pp. 161–165. 3 Tam, A. C., Rev. Mod. Phys., 1986, 58(2), 381. 4 Liu, G., Appl. Opt., 1982, 21(5), 955. 5 McDonald, F. A., Appl. Phys. Lett., 1989, 54(16), 1504. 6 Karabutov, A. A., Podymova, N. B., and Letokhov, V. S., Appl. Phys. B, 1996, 63, 545. 7 Spanner, G., and Niessner, R., Anal. Methods Instrum., 1994, 1, 208. 8 Karabutov, A. A., Podymova, N. B., and Letokhov, V. S., J. Mod. Opt., 1995, 42(1), 7. 9 Puchenkov, O. V., Malkin, S., Rev. Sci. Instrum., 1996, 67 (3), 672. 10 Beenen, A., Spanner, G., and Niessner, R., Appl. Spectrosc., 1997, 51, 51. 11 Braslavsky, S. E., and Heibel, G. E., Chem. Rev., 1992, 92, 1381. 12 Eggers, F., and Kaatz, U., Meas. Sci. Technol., 1996, 7, 1. Paper 7/07114K Received October 1, 1997 Accepted January 19, 1998 550 Analyst, April 1998, Vol. 123
ISSN:0003-2654
DOI:10.1039/a707114k
出版商:RSC
年代:1998
数据来源: RSC
|
5. |
Method for detecting information in signals: application to two-dimensional time domain NMR data† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 551-559
D. N. Rutledge,
Preview
|
PDF (439KB)
|
|
摘要:
Method for detecting information in signals: application to two-dimensional time domain NMR data† D. N. Rutledge* and A. S. Barros Laboratoire de Chimie Analytique, Institut National Agronomique, 16 Rue Claude Bernard, 75231 Paris Cedex 05, France. E-mail: rutledge@inapg.inra.fr Time domain (TD) NMR is used in industry for quality control. Like near-infrared (NIR) spectrometry, it has many advantages over wet chemistry including speed, ease of use and versatility. Unlike NIR, TD-NMR can generate a wide range of responses depending on the particular pulse sequences used.The resulting relaxation curves may vary as a function of the physico-chemical properties or even the biological and geographical origin of the product. The curves are usually decomposed into sums of exponentials and the relaxation parameters are then used in regression models to predict water content, iodine number, etc. The diversity of possible signals is both an advantage and disadvantage for TD-NMR as it broadens the range of potential applications of the technique but also complicates the development and optimisation of new analytical procedures.It is shown that univariate statistical techniques, such as analysis of variance or chi-squared, may be used to determine whether a signal contains any information relevant to a particular application. These techniques are applied to 2D TD-NMR signals acquired for a series of traditional and ‘light’ spreads. Once it has been demonstrated that the signals contain relevant information, partial least-squares (PLS) regression is applied directly to the signals to create a predictive model.The Durbin–Watson function is shown to be a means characterising the signal-to-noise ratio of the vectors calculated by PLS to select the components to be used in PLS regression. Keywords: Time domain NMR; chemometrics; analysis of variance; partial least squares; Durbin–Watson Time domain nuclear magnetic resonance (TD-NMR) is often used to quantify major proton-containing constituents in agrofood products or to monitor their evolution during processing.Like Raman, FT-IR and NIR spectrometry, it has many advantages over wet chemistry, including speed, ease of use and versatility. TD-NMR has two major advantages over these other instrumental techniques: it is possible to obtain non-invasively a signal from the whole of the sample, not just a superficial layer; and the nature of the signal observed depends on the particular radiofrequency pulse sequences applied to the sample to excite the protons.One can generate very different signals depending on the particular pulse sequences, Carr–Purcell– Meiboom–Gill (CPMG), inversion–recovery (I-R), progressive saturation (PS), Hahn spin echo (HSE), free induction decay (FID), Goldmann–Shen (GS), spin locking, Jeener–Broekhaert, etc., used to excite the sample.1–3 The resulting relaxation curves may vary depending on the water content, hydration state, solid fat content, iodine number or even biological or geographical origin of the studied product. This diversity of responses is both an advantage and a disadvantage for NMR compared with other instrumental techniques, as it broadens the range of potential applications of the technique but also complicates the development and optimisation of new analytical procedures.As far as TD-NMR is concerned, where the objective is usually the development of a rapid instrumental method of quantification or characterisation,4 this apparently unlimited number of possible signals means that it is often difficult, or at least time consuming, to determine which pulse sequence, if any, produces a signal with information content. The objective of this study was to demonstrate that chemometric techniques, which are already widely used in NIR spectrometry,5 can facilitate the detection of information in a signal.These techniques may also be used for the analysis of other signals, such as spectra,6 chromatograms or sensor responses.There has been a small number of recent examples of the application of chemometrics to the analysis of TD-NMR signals. Rutledge et al.7 used ANOVA and factor analysis to examine the effects of factors such as fat and moisture content, pH and temperature on the one-dimensional relaxation curves of spreads and gelatines, and to determine the stoichiometry of a complexation reaction.Davenel et al.8 applied multivariate statistics to the relaxation curves of doughs during cooking, Gerbanowski et al.9 compared partial least squares (PLS) regression applied to relaxation curves with PLS and multiple linear regression (MLR) applied to estimated relaxation parameters. Vackier and Rutledge10,11 applied univariate and multivariate statistical techniques to relaxation curves to study the influence of the physico-chemical characteristics of gelatines on their relaxation properties.Clayden et al.12 applied factor analysis to 19F TD-NMR FID signals acquired for a set of PTFE samples of varying crystallinity. In this study, the univariate techniques ANOVA and chisquared were used to determine whether the two-dimensional TD-NMR relaxation surfaces acquired for a series of traditional and ‘light’ spreads contain information on their moisture content. Once it is demonstrated that the signals contain relevant information, the multivariate regression technique PLS is applied directly to the signals to create a predictive model.The Durbin–Watson function is shown to be a means for characterising the signal-to-noise ratio of the loadings and B-coefficient vectors calculated by PLS to select the components to be used in PLS regression. As the term indicates, univariate techniques analyse only one variable at a time—each of the N points in a set of M TD-NMR relaxation curves, NIR spectra or gas chromatograms is analysed separately.These univariate techniques therefore require little computer memory and can be very rapid. However, they have the disadvantage of neglecting any correlations or interactions between the variables and only revealing proximities between samples within the uni-directional space of the individual variable. Multivariate techniques, such as factor analysis or PLS regression, analyse the total N 3 M data matrix at one time, with the advantage of giving information about interactions between variables and similarities between samples within the multi-dimensional space of all the variables.This, however, makes much greater demands on computer memory and calculation time. † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997. Analyst, April 1998, Vol. 123 (551–559) 551Experimental Samples Twelve samples of different traditional and ‘light’ butters and margarines were studied.The moisture contents, determined in triplicate by Karl Fischer titration, ranged from 12 to 25% for the traditional spreads and from 31 to 60% for the light spreads. The fat content given by the manufacturers ranged from 25 to 82% for the butters and from 60 to 80% for the margarines. The number of samples used is small, but the objective here is to demonstrate the methods of detecting information in signals, not to develop a robust multivariate predictive model.The very strong anticorrelation between the fat and moisture content of the samples (Fig. 1) makes it impossible to separate their effects on the signal. Any characterisation based on moisture content necessarily parallels a characterisation due to fat content. NMR measurements NMR tubes of 10 mm od were filled to a height of 10 mm. The samples were stabilised at 20 °C before being transferred to the NMR apparatus thermostated at 20 °C. The measurements were performed in triplicate on a 20 MHz TD-NMR apparatus (QP20+, Oxford Instruments, Oxford, UK) with phase quadrature detection and a maximum acquisition frequency of 10 MHz.A VISUAL BASIC program, developed in collaboration with Oxford Instruments, was used to generate the pulse sequences and to acquire the data. This program can be used to acquire two-dimensional relaxation signals and a set of relaxation curves was acquired by inserting a CPMG sequence into an I–R sequence: [(180°)x–tvar–CPMG–RD]n where CPMG = (90°)x–t–[(180°)y–2t–(180°)y–2t– (180°)y–2t–(180°)y–t–Acq–t]m tvar(i) = 2000 3 1.55(i21) ms for i = 1 to n; n = 20; t = 1000 ms; m = 100; RD = 3 s; and four scans are performed with phase cycling.This sequence gave a 2000 variable vector of 20 T1-weighted 100-point CPMG curves (Fig. 2). This vector can be folded back, by putting each CPMG curve on a separate line, to produce a 20 3 100 matrix corresponding to a T1–T2 relaxation surface made of transverse relaxation curves, weighted by longitudinal relaxation (Fig. 3).It should be noted that the exponential form of the I–R curves is distorted by the equal spacing of the data points in the graphic, resulting in a sigmoid shape. Different statistical techniques were then applied to the set of 36 vectors, using a program developed in the laboratory, and validated with Matlab routines.13 Statistical analyses Analysis of variance If the samples being examined can be classed into definite groups, it is possible to calculate the amount of the variation in the signal intensity due to the samples belonging to particular groups.14 By difference from the total variability of the measurement, it is possible to calculate the amount of variability that is not due to the groups.For each group one can then calculate the group variance: V n x x g j j j g G = - ( ) - ( ) = å 2 1 1 (1) and the residual variance: V n x x n j g j ji i n j j g j R = - ( ) - ( ) = = = å å å 1 2 1 1 1 (2) where g = number of groups, nj = number of samples in group j, xij = value for sample i in group j, �xj = mean value for group j, =x = grand mean value and N = total number of samples.If the variables studied are in fact points in a signal, such as a TD-NMR relaxation curve, it is interesting to plot these variance values as a function of their position in the signal. Regions in the signal that vary systematically from one group to another will give high VG values.In addition, if there are no important differences between the samples, other than those due to the groups, the VR will be low. A high VG associated with a low VR indicates a significant influence of the factor used to create the a priori grouping on the signal intensity. Fisher’s F value is commonly used to estimate the level of significance of the influence of the factor studied on the variable x. The F value for the variable x may be calculated as Fig. 1 Correlation between fat and moisture contents of the spreads. Fig. 2 Series of T1 weighted CPMG relaxation curve for a ‘light’ butter. 552 Analyst, April 1998, Vol. 123F V V x = G R (3) Other statistical functions may also be of interest to characterise the effect of the factors on the signal intensity. The group standard deviation (SDG), given by SD V n x x g j j j g G G = = - ( ) - ( ) = å 2 1 1 (4) has the advantage of giving values which are in the same units as the signal, making comparisons easier.Similarly, the group coefficient of variation (CVG) (relative standard deviation), given by CV SD x n x x g x G G j j j g = = - ( ) - ( ) = å 2 1 1 (5) has the added advantages of being signed and of correcting for the distortion introduced by differences in grand mean signal intensities. Chi-squared criterion This test is commonly used to determine whether the observed frequencies for a particular set of events differ significantly from their expected frequencies: c2 2 1 = - ( ) = å O E E j j j j g (6) Chi-squared (c2) may also be used to test the significance of the grouping by using the grand mean as the expected value and the means for each group as the observed values: c2 2 1 G = - ( ) = å x x x j j g (7) The absolute value of this statistic may then be used, like Fisher’s F, to give a level of significance for the groupings.It also has the advantage of being in the same units as the signal. Durbin–Watson criterion For most, if not all, instrumental techniques, the intensities of adjacent points in signals containing information are strongly correlated.Therefore, when the variables being studied are points in a signal, it is interesting to plot the above statistics as a function of their position. Regions in the signal that vary systematically from one group to another will give high absolute values for SDG, F, VG, c2 G and CVG. In addition, the values of SDG, F, VG, c2 G and CVG will have a structured distribution as a function of position of the point, as they are not random but evolve as a function of the information in the signal.It is possible to have an objective measure of this structure or non-randomness. The Durbin–Watson D statistic is commonly used to determine whether the residuals after a regression are randomly distributed.15 This statistic is given by D x x x x i i i n i i i n = - ( ) ( ) - = = å å d d d d 1 2 2 2 (8) where dxi and dxi21 are the residuals for successive points in a series.For n > 100, the distribution is random with a 95% confidence interval for D between 1.7 and 2.3. Fig. 3 T1–T2 relaxation surface for a ‘light’ butter. Analyst, April 1998, Vol. 123 553Partial least-squares regression PLS regression is a least-squares regression procedure based on regressing a reduced set of uncorrelated, linear combinations, T, of the original independent variables, X, on to the dependent variable, Y.It is very similar to principal components regression, where the T are simply the principal components, but in PLS the T are calculated iteratively, maximising their covariance with Y.16 This procedure helps to avoid or reduce the collinearity problems associated with strongly correlated variables and, as a consequence, to have a better predictive model. This predictive regression model is of the form : Y = X·B + e (9) where B, the vector of B coefficients, is calculated using the loadings of the X variables on the T vectors.As before in the case of the univariate statistics, the structure of both the loadings vectors and B coefficient vectors may be used to give indications on the information content of the signal analysed. The Durbin–Watson statistic can help in determining the optimum number of components to be used in the model. Results and discussion Univariate analyses Effect of moisture content The ANOVA procedure was first applied to the set of 36 I–R weighted CPMG curves for the butters and margarines using the factor ‘light’/‘traditional’ as a grouping criterion.To eliminate signal intensity variations due to differences in sample size, the vectors were first normalised to unit maximum intensity. Fig. 4(a) presents the group variance surface based on this grouping of the samples and Fig. 4(b) presents the corresponding residual variances. The values close to zero at the beginning of the CPMG and at the ends of the I–R in Fig. 4(a) are due to the normalisation of the signals. As these points usually have the highest intensity, they are all set close to unity and their variability is therefore almost zero. It is clear that both the VG and VR surfaces are ‘structured’. The VG surface indicates that the relaxation surfaces contain information on the moisture content, whereas the VR surface shows that there is another factor which influences the relaxation to a certain extent. The regions which are most sensitive to the ‘Light’/‘Traditional’ nature of the samples are, on the one hand, the beginning of the CPMG curves and the middle of the I–R, near its null point, and on the other hand, the extremities of the I–R and all along the CPMG.One could therefore suppose that the discrimination was based on a variation in the properties of the constituents of the spreads with a fast T2 relaxation rate (early part of the CPMG) and whose T1 varies (null point of the I–R).This hypothesis will be confronted below with the resultsf the other calculations. From these figures, it can be seen that the highest VG is 100 times greater than the highest VR. The Fisher’s F values calculated from these two sets of values are plotted in Fig. 5(a). It is clear from this graph that the F plot is much ‘noisier’ than the VG plot because of the contribution from the VR, and that the F value is high in some areas simply because the corresponding VR values are small. This problem will become all the more important when the factor used to group the samples explains more of the variability in the signal, resulting in small, ‘noisy’ VR values.For this reason, despite its usual interest as a measure of the level of significance of the effect of a factor, the Fisher’s F plot will rarely be of use as a means to detect information in a signal. The group standard deviation surface was plotted [Fig. 5(b)] in order to be able to compare more easily the evolution of the variability due to the factor with the evolution of the original relaxation surface.As expected, the topography of this surface is less pronounced than that of VG and the evolution of the surface in the CPMG direction is more like that of the relaxation surface (Fig. 3). The coefficient of variation is not only on the same scale as the original relaxation surface, but has also been corrected for differences in the grand average signal intensity at each point on the relaxation surface.If the variability in the signal intensity due to the factor is proportional to the average signal intensity, then the CVG surface should remain almost constant, making it easier to pinpoint regions of unusual variability. However, this plot introduces a major distortion when the signal changes sign or the signal intensity approaches zero, as can be seen here when the I–R signal goes through the null point. In this figure it is possible to see that the position of this passage through the null point shifts to a later position in the I–R direction, going from point 11 to point 15, at the same time as it moves down the CPMG direction. Given the fact that for a monoexponential T1 relaxation, the null point corresponds to T1ln2, and given the delays used in the I–R pulse sequence, the observed shift would correspond to T1 changing from about 230 to 1300 ms.Fig. 4 (a) Group variance and (b) residual variance surfaces based on the factor ‘light’/‘traditional’. 554 Analyst, April 1998, Vol. 123These approximate values are not incompatible with the T2 values calculated for these samples. A detailed analysis of the CPMG relaxation curves was performed using CONTIN17 to decompose them into a continuous distribution of relaxation times, and MARQT18,19 to have a discrete sum of exponentials. The whole butters and margarines, and also the separated aqueous and lipid phases, were studied (results not shown). The T2 values of the two relaxation components in the butters were about 30 and 200 ms, and those in the margarines were close to 50 and 500 ms.In the light spreads, the values were close to 100 and 1000 ms for the margarines and largely unchanged at about 30 and 200 ms for the butters. An expansion of the plot [Fig. 5(c)] gives information on the zones of the relaxation surface which are most sensitive to the factor ‘light’/‘traditional’. Except for a small region at the beginning of the CPMG and the zone where the I–R passes through zero, the CVG is almost constant, meaning that the effect of the factor on the relaxation surface is proportional to the intensity of the signal.The low CVG values at the beginning of the CPMGs are due to the sample sizes being very similar. When the grand mean signal intensity approaches zero at the I– R null point, the CVG becomes undefined, making difficult the interpretation of the surface.Because of the change in sign of the signal intensity, the CVG values change sign at the null point. The chi-squared test also introduces a distortion when the signal intensity approaches zero. The expanded chi-squared plot [Fig. 5(d)], like the CVG plot, shows a uniform effect of the factor on the rest of the relaxation surface. The CVG and chisquared plots show that the discrimination ‘light’/‘traditional’ is in fact due to relaxation components present throughout the CPMG curve, and not just the fast relaxing component as appeared to be indicated by the VG plot.By calculating the Durbin–Watson D from successive values in the vectors of SDG, F, VG, VR, c2 G and CVG values, it is possible to have an indication of the non-randomness of the values or, in other words, of the information content of these vectors. For each of these statistics, the D values were calculated for the vectors in the I–R direction and plotted as a function of the corresponding CPMG point [Fig. 6(a)]. The same calculation was done in the CPMG direction and the results were plotted as a function of the I–R point [Fig. 6(b)]. Because of the null point in the I–R curve, the c2 G and CVG vectors give D values close to 2, as can be seen in Fig. 6. The VR Fig. 5 (a) Fisher’s F, (b) group standard deviation, (c) expanded plot of the coefficient of variation and (d) expanded plot of chi-squared for the factor ‘light’/ ‘traditional’. Analyst, April 1998, Vol. 123 555values are higher than SDG, F and VG for the I–R vectors and higher than SDG and VG for the CPMG vectors, outside the nullpoint zone, confirming the lower signal-to-noise ratio of this vector. Effect of type of fat The ANOVA procedure was applied to the same set of normalised T1–T2 relaxation surfaces, this time using the factor ‘butter’/‘margarine’ as the grouping criterion. Fig. 7(a) presents the group variance surface and Fig. 7(b) the corresponding residual variance, based on this grouping of the samples.It can be seen from these figures that the highest VG is only three times greater than the highest VR. Although this VG surface is also ‘structured’, the signal-to-noise ratio is much lower than in the previous case, indicating that the relaxation surface contains less information about this factor. The VR plot for this factor has a higher signal-to-noise ratio and has a structure similar to the VG for ‘light’/‘traditional’. This confirms the predominant effect of this latter factor in determining the relaxation properties of the spreads.The general complementarity of the two VG and VR plots indicates that the two factors ‘light’/‘traditional’ and ‘butter’/‘margarine’ are responsible for nearly all the variability in the relaxation surfaces. The Fisher’s F values are plotted in Fig. 7(c), and confirm that the grouping based on ‘butter’/‘margarine’ is much less significant than that based on ‘light’/‘traditional’.The F plot is, once again, much ‘noisier’ than the VG plot, confirming its limited utility in the case of signal intensities as a measure of the level of significance of the effect of a factor. The group standard deviation surface plotted in Fig. 7(d) shows once again that irregularities in the surface are less pronounced than for VG and its evolution is similar to that of the original relaxation surface. The coefficient of variation plot shows a major distortion again when the signal intensity passes through the null point of the I–R signal.An expansion of the plot [Fig. 8(a)] shows that the effect of this factor is not proportional to the intensity of the relaxation surface. The CPMG curves vary more towards the end. This is also the case for the chi-squared plot [Fig. 8(b)]. This greater variability in the final part of the CPMGs implies a change in the proportions or relaxation properties of the slower relaxing T2 components. One would expect the nature of the lipid phase to be responsible for this ‘butter’/‘margarine’ differences between the samples, but as the region of greatest variance is at the end of the CPMG curves, this would imply that the lipid phase has a longer transverse relaxation time than the aqueous phase.The analysis of the CPMG relaxation curves using CONTIN17 and MARQT18,19 showed that the major differences were in the proportion of aqueous phase between ‘light’ and traditional, and in the longer relaxation times of the aqueous phase of the margarine samples.As before, the Durbin–Watson D was calculated for the vectors of SDG, F, VG, VR, c2 G and CVG values in the I–R and CPMG directions. Once again, the c2 G and CVG vectors gave D values close to 2, as can be seen in Fig. 9. However, outside the null-point zone, the VR values are lower than the others, confirming the higher signal-to-noise ratio of this vector. Multivariate analysis Having shown that the relaxation surfaces contain information on the moisture content of the samples, it should be possible to use them to develop a predictive model.PLS regressions were performed using the 36 2000-point vectors as the X-matrix and the corresponding moisture contents, determined by Karl– Fischer titration, as the Y-matrix. The major danger in using multivariate regression techniques on signals, where there are usually many more variables than samples, is over-fitting, in which case noise present in the signal is included in the model to adjust it to the data.This problem may be partly overcome by limiting the number of principal components included in the model. The choice of the cut-off point is sometimes made based on the evolution of the X-matrix eigenvalues, or more usually by examining the root mean squared error of prediction (RMSEP) values calculated using test-set validation or cross-validation. As will be seen below, the Durbin–Watson criterion may be used as a means of detecting which principal components contain most information.Fig. 10 clearly shows the increase in noise intensity and decrease in ‘structure’ with increasing number of the principal component. As the B coefficient vectors are derived from these loadings, if more than the optimal number are used, excessive noise will be included. The evolution of the Durbin–Watson values for both the B coefficient vectors and the X-matrix loadings vectors for the first 10 principal components are plotted in Fig. 11, along with the corresponding eigenvalues. An abrupt increase in the Durbin–Watson values occurs after three principal components, confirming the evolution of the eigenvalues. Although both criteria indicated the same maximum number of components, the Durbin–Watson criterion has the advantage over the eigenvalue of being a measure of the structured variability in Fig. 6 Durbin–Watson D values of the vectors in (a) the I–R direction and (b) the CPMG direction for the factor ‘light’/‘traditional’. 556 Analyst, April 1998, Vol. 123Fig. 7 (a) Group variance, (b) residual variance, (c) Fisher’s F and (d) group standard deviation surface based on the factor ‘butter’/‘margarine’. Fig. 8 Expanded plot of (a) the coefficient of variation and (b) chi-squared for the factor ‘butter’/‘margarine’. Analyst, April 1998, Vol. 123 557the eigenvectors derived from the X-matrix, and not just of the proportion of the total variability extracted from the Xmatrix.Fig. 12 presents the B coefficient surface for the threecomponent PLS regression model between the T1–T2 relaxation surfaces and the measured moisture content. There is a definite similarity between the B coefficient surface and the group variance surface, as both contain information on the variability of the X-matrix, the set of relaxation surfaces. The regression line for the prediction of the moisture content of the butters and margarines based on the T1–T2 relaxation surfaces is plotted in Fig. 13. The statistics for the model are root mean squared error of calibration (RMSEC) = 4.06% and R2 = 0.967. However, as was seen in Fig. 1, the fat and moisture contents of these samples are strongly anti-correlated so it is impossible to distinguish their effects on the relaxation surfaces. The PLS regression model for the moisture content will necessarily Fig. 9 Durbin–Watson D values of the vectors in (a) the I–R direction and (b) the CPMG direction for the factor ‘butter’/‘margarine’.Fig. 10 Loadings of the first 10 PLS principal components (from the bottom up) presented as vectors. The curves have been shifted vertically for clarity. Fig. 11 Durbin–Watson values for the B coefficients and X-matrix loading vectors and the eigenvalues of the first 10 PLS principal components. 558 Analyst, April 1998, Vol. 123include a contribution from the effect of the anti-correlated variation in the fat content.Nevertheless, the ANOVA and chi-squared procedures have demonstrated their usefulness for the detection of information in the signals. The Durbin–Watson criterion has been shown to be of use to quantify the signal-to-noise ratio of the variability estimators and to determine the optimal number of principal components for the PLS regression. These statistical estimators all have the advantage of being very quick to calculate and requiring little computer memory.However, they have the disadvantage of being based on supervised methods, requiring the division of the data set into groups. We thank the French Institut National de la Recherche Agronomique for partial financing of the TD-NMR instrument and Stephen Provencher for the CONTIN source code. References 1 Atta-Ur-Rahman, Nuclear Magnetic Resonance. Basic Principles, Springer, Berlin, 1986. 2 Canet, D., La RMN: Concepts et M`ethodes, InterEditions, Paris, 1992. 3 Farrar, T. C., and Becker, E. D. Pulse and Fourier Transform NMR, Academic Press, New York , 1971. 4 Rutledge, D. N., J. Chim. Phys., 1992, 89, 273. 5 Osborn, B. G., and Fearn, T., Near Infrared Spectroscopy in Food Analysis, Longman, New York, 1986. 6 Vercauteren, J., Forveille, L., and Rutledge, D. N., Food Chem., 1996, 57(3), 441. 7 Rutledge, D. N., Barros, A. S., and Gaudard, F., Magn. Reson. Chem., 1997, 35, 13. 8 Davenel, A., Marchal, P., and Guillement, J. P., in Magnetic Resonance in Food Science, ed. Belton, P. S., Delgadillo, I., Gil, A. M., and Webb, G. A., Royal Society of Chemistry, Cambridge, 1995, pp. 146–155. 9 Gerbanowski, A., Rutledge, D. N., Feinberg, M., and Ducauze, C., Sci. Aliments, 1997, 17, 309. 10 Vackier, M. C., and Rutledge, D. N., J. Magn. Reson. Anal., 1996, 2(4), 321. 11 Vackier, M. C., and Rutledge, D. N., J. Magn. Reson. Anal., 1996, 2(4) 311. 12 Clayden, N. J., Lehnert, R. J., and Turnock, S., Anal. Chim. Acta, 1997, 344, 261. 13 Matlab, Mathworks, S. Natick, MA, 1992. 14 Czerminski, J., Iwasiewicz, A., Paszk, Z., and Sikorski, A., in Statistical Methods in Applied Chemistry, Elsevier, Amsterdam, 1990, pp. 186–207. 15 Durbin, J., and Watson, G. S., Biometrika, 1950, 37, 409. 16 H�oskuldson, A., J. Chemom., 1988, 2, 211. 17 Provencher, S., Comput. Phys. Commun., 1982, 27, 229. 18 Marquardt, D. W., J. Soc. Ind. Appl. Math., 1963, 11, 431. 19 Rutledge, D. N., in Signal Treatment and Signal Analysis in NMR, ed. Rutledge, D. N., Elsevier, Amsterdam, 1996, pp. 191–217. Paper 7/07058F Received September 30, 1997 Accepted November 7, 1997 Fig. 12 B coefficient surface for the PLS regression model between the T1–T2 relaxation surface and the moisture content. Fig. 13 Regression line and 95% confidence limits for the threecomponent PLS regression model for moisture in the butters and margarines. Analyst, April 1998, Vol. 123 5
ISSN:0003-2654
DOI:10.1039/a707058f
出版商:RSC
年代:1998
数据来源: RSC
|
6. |
On-line sample pre-treatment systems interfaced to electrothermal atomic absorption spectrometry† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 561-569
José Luis Burguera,
Preview
|
PDF (91KB)
|
|
摘要:
Solvent extraction Microwave digestion Sorption on columns Vapour generation Precipitation & coprecipitation FI In-situ trapping Transfer capillary Microprocessor controlled valve Thermospray High pressure nebulization INDIRECT (off-line) DIRECT (on-line) Collection in cups ETAAS Sample processing Interfaces Connecting type Liquid chromatography On-line sample pre-treatment systems interfaced to electrothermal atomic absorption spectrometry† Jos�e Luis Burguera and Marcela Burguera* IVAIQUIM (Andean Institute for Chemical Research), Faculty of Sciences, University of Los Andes, P.O.Box 542, M�erida 5101-A, Venezuela Atomic spectrometric techniques working with a continuous supply of sample, such as flame atomic absorption spectrometry or inductively coupled plasma atomic emission spectrometry, are well suited for on-line connection through flow injection arrangements. For the determination of elements at lower concentrations, electrothermal atomization is often needed owing to the requirement for better detection limits.However, connecting a batch technique such as electrothermal atomic absorption spectrometry to a continuous flow or a flow injection system presents some fundamental difficulties, making it a real challenge for the analytical chemist to design suitable interfaces. There are different ways of interfacing the systems to each other; the available versions utilize either on-line sample treatment with off-line measurement or are completely on-line arrangements.Complex samples cannot be directly processed by this technique owing to severe matrix interferences, which have not been minimized despite the development of efficient background correction devices. Successful matrix separation can be achieved on-line through solvent and sorbent extraction, precipitation and coprecipitation, volatile compound generation and liquid chromatography. Moreover, transformation of the sample and/or its fractions into a form that can be analyzed in a graphite furnace often requires unpleasant, tedious and time-consuming digestion procedures.Microwave heating has emerged as a means of dramatically improving leaching, mineralization or digestion processes and has frequently been used with in-batch or on-line systems. This paper outlines the most recent advances in the development of flow injection–graphite furnace interfaces and also describes the on-line sample pre-treatment systems developed so far for electrothermal atomic absorption detection.Keywords: Flow injection; electrothermal atomic absorption spectrometry; interface; on-line sample pre-treatment The specific detectors most commonly used with on-line sample work-up devices for the determination of metallic species in different types of sample are atomic absorption spectrometry (AAS) with flame (FAAS), hydride generation (HGAAS) and electrothermal atomization (ETAAS) as well as inductively coupled plasma atomic emission spectrometry (ICP-AES).These techniques are the traditional workhorses of analytical laboratories.1 Amongst them, FAAS and ICP, which involve working with a continuous supply of sample, are well suited for such on-line connections, although the low nebulization efficiency and dispersion are common drawbacks that degrade the detection limits. ICP coupled to mass spectrometry (ICPMS) most satisfies the need for a technique that is multielement, sensitive, selective, fast and precise.Extensive work has been carried out in the last few decades on this subject, showing that there is great interest in developing automated online systems to improve detection limits,2 remove interfering matrix components3 or perform speciation studies.4,5 Recent developments in ICP-MS instrumentation allow very small samples to be probed and also allow investigation of the spatial distribution of trace elements at the cellular level.6,7 However, such sophisticated analytical tools are not readily available in most routine analytical laboratories because of their high cost of operation and the need for highly qualified personnel. Despite its uni-elemental features, ETAAS is a relatively low cost and robust technique with the appropriate detection capability for monitoring many elements and their common species in biological and environmental samples at sub-ng l21 levels.ETAAS is, therefore, often needed to reach better detection limits in small sample portions.The graphite furnace (GF) has a limited capacity and the aliquot injected has to be dried, pyrolyzed and atomized before any measurement is carried out. Furthermore, complex samples cannot be directly processed by ETAAS owing to severe matrix interferences, which have not been minimized despite the development of efficient background correction devices. Successful matrix separation can be achieved on-line through solvent and sorbent extraction, liquid chromatography (LC), volatile compound generation or digestion, microwave (MW) treatment being the preferred option in this respect (Fig. 1). Coupling the different techniques mentioned above with ETAAS constitutes the most suitable approach to carrying out preconcentration and speciation studies. An examination of the literature reveals that there are a number of books8–14 and reviews6,7,15,16–24 dealing with on-line sample preparation techniques, most of which could be carried out before the introduction of the sample into the GF.The earliest attempts were for off-line sample work-up, although great efforts have been made to design on-line systems; the symbiosis between flow injection (FI) and ETAAS, although troublesome, proved to be an efficient interface for on-line sample pre-treatment and subsequent deposition of the sample into the furnace. † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.Fig. 1 Flow injection interfaces for ETAAS. Analyst, April 1998, Vol. 123 (561–569) 561Continuous operations coupled to ETAAS are carried out in parallel (while a sample is processed in the FI manifold, the previous sample follows the furnace temperature program) or sequentially (the FI manifold is operated in stopped-flow mode during the furnace program).16 These connections brought a great deal of sophistication to the instrumentation and a stronger impulse to the development of new procedures.Since this relatively fast moving area of analytical science deserves attention, the more recent literature dealing with continuous interfaces for sampling in ETAAS as well as the description of on-line sample pre-treatment systems directly coupled to ETAAS will be emphasized in this review. Interfaces for sample introduction into the GF The need for a multistage temperature program prior to atomization in the GF delayed for decades the on-line connection of any continuously flowing manifold directly to the GF, one type of sampling device being needed between the two systems.So far, there are several different ways of connecting the FI and ETAAS systems to each other, the most frequently used being: (1) a collecting vial or a flow-through cell located in the autosampler tray followed by the deposition of a suitable aliquot of sample into the GF, (2) sample introduction in the form of an aerosol through conventional nebulization, thermospray or hydraulic high pressure (HHP) nebulization, (3) automatic injection of small volumes retained in collecting tubes or valve loops, performed by air displacement, and (4) in situ trapping of volatile compounds in a pre-heated GF.The first version refers to on-line sample treatment with off-line measurement by ETAAS, while the others refer to on-line systems. All are meant to increase sensitivity and selectivity (concentrate the analyte of interest and separate it from the interfering matrix) as well as to remove (separation procedures) or destroy (digestion methods) the organic and/or inorganic matter which accompanies the analyte.Direct sample deposition into the GF The sample plug from any on-line sample processing system might be directly collected at a fixed time in autosampler cups25–27 or carried to flow-through cells25,28–30 located in aed position in the carousel of the ETAAS autosampler.The flow cells, also called ‘well-samplers’, with a capacity of 50–100 ml, are continuously filled from the bottom, and excess liquid is withdrawn at the top by gentle suction by one of the channels of the peristaltic pump. The autosampler arm can periodically sample microvolumes of the fractions thus collected and pipette them into the ‘cool’ furnace. Sample introduction as an aerosol Alternatively, the sample could be introduced as an aerosol into a ‘warm’ furnace using conventional pneumatic or ultrasonic nebulization.31 Although a shortening in the overall analytical procedure is obtained, the process proved to be unsuccessful because: (a) there is no possibility of introducing simultaneously several distinct solutions, (b) only a minor fraction of the sample solution is transferred and a separate desolvation unit is needed; sample consumption is therefore large as the extent of nebulization is low, this being a real inconvenience where available sample volumes are necessarily small and (c) the procedure shortens the tube lifetime since the deposition must take place at elevated temperatures.Using a modified nebulizing device, Pinel et al.32 claimed a quasi-quantitative introduction of samples into the furnace. However, the amount of sample deposited depends on the deposition period, a parameter that is very difficult to control in a precise way. Moreover, a certain delay time must be included to produce a steady aerosol and to change samples.Only aqueous samples have been introduced; thus, there was no need for any sample pre-treatment before the GF was reached (Table 1). Thermospray interfaces have proved more efficient for sample introduction into commercial ETAAS systems because they allow periodical deposition in real time.33,34,49 The device consists of a fused silica capillary inserted into an electrically heated stainless-steel tube mounted onto a holder that can be moved backwards and forwards into the furnace housing by means of a pneumatic actuator.When a sample is to be introduced into the GF, six steps follow, controlled by a timer: (a) the furnace is heated to the desired deposition temperature; (b) the water-cooling of the furnace housing is switched off to prevent vapor condensation on the quartz windows; (c) the evacuated chamber is opened; (d) simultaneously, the injection valve is emptying the sample loop; (e) the vaporizer is moved forward so that the tip of the fused-silica capillary enters the injection hole of the graphite tube; (f) the furnace is evacuated by a second vacuum system. When the aerosol formed on heating makes an impact on the pre-heated wall of the graphite tube, a considerable saving in time is achieved, which results in shorter furnace thermal programs.Such an interface was evaluated for the determination of several elements33,49 and for the speciation of arsenic (arsenite, arsenate and dimethylarsinate) using ion-pair chromatography.34 The operation of the interface is controlled by a sequence controller and each action can be programmed independently.The time between consecutive measurements is drastically reduced since the eluate is dried on impact onto the pre-heated platform and hence the pyrolysis step could be omitted. Further shortening of the thermal program is limited by the cool-down step; the cooling, however, is used only until the GF is at the deposition temperature.The experimental arrangement would permit the monitoring of larger eluate fractions, thereby improving the detection limits. A total temperature program time of 40 s, for instance, is feasible for the speciation of arsenic, which is less than that used conventionally (85 s). Thermospray sample introduction into the GF retains the possibility of depositing microlitre sample volumes (10 ml in a few seconds); the signal shapes and sensitivities are similar to those observed in conventional ETAAS and the deposition efficiency and hence the size of the signal depends on the temperature of the furnace wall.Despite its advantages, this approach has not been widely applied; before a commercialized version of the system becomes available, the idea deserves further research effort. The HHP nebulization technique developed by Berndt and Schaldach35 has been successfully applied to sample deposition into a GF for the determination of trace amounts of metals in an aluminum matrix after sorption of their dithiocarbamate complexes on a C18 column mounted in an on-line FI preconcentration system.The aluminum-rich matrix is led to the waste outlet of the nebulization chamber and the trace elements collected on the column are eluted with ethanol. The eluate is forced through a capillary nozzle with an opening of only 10–30 mm at a high pressure propelled by an HPLC pump. The liquid jet is transformed into a fine aerosol at a pressure of 1500–6000 psi, which is 30–50 times higher than for conventional nebulization.The aerosol cloud thus formed is transported by an air carrier gas at about 3 l min21 into the pre-heated GF (at 140 °C) through a flexible tube with a glass tip end introduced into the sampling hole of the GF. The elution, transport and deposition are completed within 30 s for all the elements investigated (Au, Co, Cu, Pb). A sample volume of 400–450 ml reaches the GF, but the sample transport is limited by the nebulization chamber, the supply lines and the transport gas flow rates.The combined effect of efficient nebulization and column preconcentration resulted in as much as a 400-fold 562 Analyst, April 1998, Vol. 123enhancement in the detection power (Table 1). The multiple advantages demonstrated by the experiments carried out with such devices deserve the attention of manufacturers if the great improvement in the nebulization efficiency, low sample and reagent consumption, low memory effects and possibility for on-line sample processing and speciation are taken into account.Automatic injection by air displacement In some experiments, a capillary tubing used as sample collector was manually inserted into the sampling hole of the GF, held in place during injection, and retracted at the end of the sequence.39,46 Later, the capillary was attached to the autosampler arm, which moved its tip automatically into the dosing hole of the graphite tube.38–51 However, the volume retained was not compatible with the capacity of the graphite tube ( �@ 50 ml) and only a fraction of the solution is introduced into the GF, to the detriment of sensitivity.Efforts were made to reduce the injected volume by incorporating valves with low capacity loops or by the use of a pre-heated tube and a very slow flow rate to allow partial evaporation of the solvent, thus avoiding overflow of the GF during sample deposition.38,39 For direct connections and automation of the entire system, microprocessor-controlled valves have been developed.There are two approaches for introducing the flowing sample into the GF: volume-based and time-based sampling. To avoid further dispersion of the sample plug in the FI system, the most concentrated segment should be selected. A multi-port injection valve controlled by a sequencer was one of the first devices used for this purpose.52 While the device injected a chosen aliquot into the furnace, the main stream was diverted to waste through a by-pass during the heating cycle of the atomizer.A similar approach used an eight-port two-position sampling valve, which trapped a portion of the sample in its loop before the sample was blown into the graphite furnace.53 An electronic interface coupled a commercial GF to an HPLC system for arsenic speciation.54 The components of the interface were: two eight-port slider injection valves with pneumatic actuators, eight solenoid valves, an injector for the deposition of samples into the GF and various electronic components.The entire system was under computer control. The paper gives full details of the construction, circuits and the program of the interface. Another interesting interface directly connected an on-line extraction system to the GF.36,37 consists of two six-port injector valves (Rheodyne) connected in series and mounted on a commercial FI analyzer for the introduction of sample and rinsing solutions into the GF.One of the injectors (for sample introduction) was directly connected to the extraction manifold and its other end tube was attached to the sampler arm in order to allow the introduction of the tubing into the GF before injection. A small volume (23 ml) of the extract trapped in the loop of this injector was transported to the graphite tube by a flow of air from a peristaltic pump.A pioneering group on FI-MW digestion systems described a novel interface for digest introduction into the GF.55–59 A commercially available injection valve was actuated to introduce the digest plug into a capillary collector tube of a home- Table 1 On-line preconcentration by nebulization and solvent and sorbent extraction followed by ETAAS detection Element or species Sample type Pre-treatment system Interface Sample deposition type Enrichment factor Ref.Cu, Sn Aqueous None Pneumatic nebulization (aerosol yield: 4–16%) Time-based * 31, 32 Ag, Au, Ru, Co, Mn, Pb, Al, V, Cd, As Aqueous None Thermospray [aerosol deposition into pre-heated (120 °C) GF] Time-based * 33 As species HPLC eluate None Thermospray [deposition into preheated (120 °C) GF] Time-based * 34 Au, Co, Cu, Pb Aluminum-rich matrix Element–APDC complexes retained on C18 column; elution with ethanol; converted into aerosol HHP nebulization [aerosol deposition into pre-heated (140 °C) GF at a yield of 60%] Time-based 50–400 35 Cd, Co, Cu, Fe, Ni, Pb Aqueous Element–DDTC complexes extracted into Freon 113 and back-extracted into HgII solution Vials.Extract trapped in loop of injection valve connected to autosampler arm. Injection by air displacement — Volume-based — 50–100 27 36, 37 Pb Waters Pb–DDTC complex retained on a knotted reactor and eluted with a fixed volume of ethanol Transfer to the pre-heated GF by air displacement Volume-based 142 38 Cd, Pb, Co, Cr species, As species Surface waters, high-purity reagents, CRM Element–DDTC complexes retained on a C18 column (in-tip), eluted with ethanol and concentrate collected in a delivery tube Transfer to the pre-heated GF by air displacement Volume-based (zonesampling) † 39–44 Cr species Aqueous CrVI–APDC complex adsorbed on a knotted reactor, eluted with 55 ml ethanol and collected in a delivery tube Transfer by means of a specially made rocker-arm to the GF by air displacement Volume-based † 45 Fe, Cd, Zn, Cu, Ni, Mn, Pb Sea-water (open ocean and CRM) Metals chelated on a microcolumn of I-8HOQ and eluted with HCl + HNO3 Capillary attached to autosampler arm.Transfer to GF by air displacement Volume-based 250 46 Cd, Pb Sea-water (near shore and CRM) Metals chelated on a microcolumn of Chelex-100 and eluted with HNO3 Capillary attached to autosampler arm. Transfer to GF by air displacement Time-based 34-242 47 Mn Geothermal waters Interferents (sulfur anions) retained on a microcolumn of Dowex-1 and eluted, after analyte measurement, with HNO3 Interferent-free solution containing Mn was trapped in a holding coil and a desired volume transferred to GF by air displacement Time-based — 48 * Depends on deposition time.† Depends on adsorption time. Analyst, April 1998, Vol. 123 563made sampling arm assembly. A time-based solenoid valve was then activated to introduce a defined volume into the GF with the aid of an air-flow.This last sequence was synchronized with the spectrometer computer, which had been pre-programmed to introduce a chemical modifier into each aliquot of the sample and to run the GF temperature program. In situ trapping of volatile compounds The volatile compound generation technique (cold vapor for Hg and hydrides for As, Se, Sb, Sn, etc.) is well established and very well documented in the literature. However, interest in improving knowledge on the biochemical cycle and environmental behavior of such elements led to the development of new separation/preconcentration procedures capable of detecting trace amounts (sub-ng l21 level) of the elements and their species in samples of different nature.Improved sensitivity has been obtained when volatile materials or compounds, such as gas chromatography (GC) effluents,60,61 mercury vapor61–71 or hydrides,72–86 have been introduced directly, on-line, into a pre-heated GF or trapped in situ and subsequently atomized.13,21 These compounds are transferred into the furnace either via the internal gas line of the instrument or through an interface, directly introduced into the sampling port of the tube.The first approach was applied to the determination of tetraalkyllead compounds60 or some organoarsenicals61 previously separated by GC. The end of the GC column was connected via a silica T-piece to both inner gas flow entrances of the GF. The column effluent replaces the normal gas flow inside the graphite tube.Although efficient, in some cases, this procedure has been hampered by the requirement to maintain the GF at the atomization temperature during the entire experiment, thus decreasing the lifetime of the tubes. Additionally, these reactive compounds might decompose by reaction with the internal metallic parts of the GF housing, resulting in loss of sensitivity and possible deterioration of expensive instrumentation.This mode of volatile compound introduction into the GF was therefore replaced by the use of a transfer line (a capillary) made of PTFE, graphite, glass or borosilicate. The capillary is normally narrower than the sampling port and should be inserted as close as possible to the atomizer inner surface, firstly to avoid its heating and consequently the decomposition of the volatile compounds prior to trapping, and secondly to ensure optimum trapping efficiency. The most convenient and almost exclusively used interface is a quartz capillary, which, because of its high thermal resistance, withstands furnace temperatures above 2300 °C if cooled by a gas flow.62 The tip of the quartz capillary could be connected via a suitable length of PTFE tubing to the outlet of the gas– liquid separator.Its other end is then inserted through the sample hole of the GF and held in close contact with the opposite wall. The insertion and withdrawal of the capillary could be either manual or automatic.The need to renew the tube coating after each atomization cycle delayed the complete automation of the technique. An important step towards automation of such systems was the transfer via the capillary mounted on the autosampler arm, accompanied by changes in the spectrometer software. Likewise, the trapping efficiency and the atomization temperature used for desorption strongly depend on the element and on the coating material, although no satisfactory explanation has been given so far21,85 (Table 2).However, in situ trapping on more durable coatings, such as Zr or Ir, appears to have solved the problem, because repeated application of such coatings does not seem necessary during the lifetime of the graphite tube. There are many reports on mercury vapor absorption on gold-,64–66 platinum-,67,68 palladium-69 or iridium-coated63,70,71 graphite tubes and absorption of hydrides on carbide-forming elements (Zr, Nb, Ta and W)72–75 or noble metals (Pt, Pd, Ir or Pd/Ir).73–86 On-line matrix separation/digestion followed by ETAAS detection ETAAS has very low detection limits for many elements in aqueous solutions, but it is very sensitive to matrix interferences even with sophisticated background correction and chemical modification.Therefore, removal (solvent or sorbent extraction) or digestion (MW) of the residual sample matrix, as well as isolation of the analyte-containing compounds (e.g., hydride formation, coprecipitation, LC separation), is preferred.For some samples, the interfering matrix could be eliminated during the ashing step, with the consequent shortening of the tube lifetime.14 Solvent extraction Among the numerous separation techniques developed in FI systems, solvent extraction would be the method of choice for Table 2 On-line volatile compound formation with in situ trapping followed by ETAAS detection Element or species Generation system* Chemical reaction Atomizer type Atomizer coating Tempe Trapping rature/°C Atomization DL†/ pg Ref.Hg PE FIAS-200 0.2% NaBH4 Electrographite tubes Au–Pt 160 750 35 66 Hg PE FIAS-200 0.05% NaBH4 Platform. THGA tube Ir 150 950 90 70 Hg Laboratory-made 0.002% NaBH4 Pyrolytic graphite tubes Au, Ir, Pd 100 ‡ 8–9 71 Alkyllead GC effluent None Graphite tubes None 2000 40–90 60 Alkyllead PE FIAS-200 NaBH4 + 2.5% tartaric acid THGA tube None 350 1600 10.5 77 As, Se, Bi, Sn, Sb Laboratory-made 1% NaBH4 Graphite tubes Pt, Pd 400 2600 8–40 78 Sn PE FIAS-200 0.5% NaBH4 + 0.5% HCl Pyrolytic graphite tubes Pd 300 2300 66 79 Sn PE FIAS-200 0.2% NaBH4 + 0.1 m HCl Pyrolytic graphite tubes Pd 300 2300 66 80 Sb PE FIAS-200 NaBH4 after pre-reduction Electrographite tubes Uncoated 200 2400 15 81 Se Laboratory-made NaBH4 Electrographite tubes Zr 800 2250 120 72 As 92 Pb Varian VGA-76 1.5% NaBH4 + oxalic acid Pyrolytic graphite tubes Zr, W, Pd, Ir 200 2000 250 74 As Laboratory-made Fleitman reaction Pyrolytic graphite tubes Pd 200 2600 10 82 As Varian VGA-76 0.5–1% NaBH4 Graphite tubes Zr 700 2100 15 73 Sb PE FIAS-400 450 2100 10 Bi 300 2100 27 Ge PE FIAS-400 NaBH4+ acetate buffer Pyrolytic graphite tubes Zr 550–800 2500 9–18 75 Ge PE FIAS-400 1.5% NaBH4 + 3 m HCl Pyrolytic graphite tubes Pd 400 2500 18 83 * PE = Perkin-Elmer.† Detection limit. ‡ Detection by furnace atomic non-thermal excitation spectrometry (FANES). 564 Analyst, April 1998, Vol. 123sample pre-treatment for ETAAS, since the analyte concentration is uniform in the entire fraction extracted; injection of a small volume thereof into the furnace would produce the same result, without affecting the reproducibility. However, there are not many papers in the literature dedicated to this subject. B�ackstr�om and co-workers27,36,37 spent some years developing on-line extraction systems specially designed for coupling with ETAAS, all based on the formation of hydrophobic complexes between trace metals (Cd, Co, Cu, Fe, Ni, Pb and Zn) and dithiocarbamates, which are extracted into an organic phase. Owing to their high volatility, the direct injection of these compounds into the furnace is not suitable for ETAAS determinations.Therefore, the extract should be mixed with an acidic solution of mercury to displace the metals into the aqueous phase, mercury forming stronger complexes with dithiocarbamates. A membrane separator ensured quantitative phase transfer in the first stage of the online extraction process while a gravitational T-tube separator was used in the back-extraction.In early studies the aqueous phase containing the back-extracted metals was collected in vials and then analysed by ETAAS,27 but in subsequent experiments a special interface connected the above-described extraction system to the GF.36,37 The delay from starting the GF program to injecting the sample is governed by the time needed by the sampling arm to reach the injection position.The timing of the injectors and the temperature program were synchronized by a microcomputer; the two systems (extraction and measurement) were operated in parallel. For the determination of nickel in blood, serum and other biological samples, Tao and Fang87 optimized an on-line solvent extraction procedure, based on complex formation with ammonium pyrrolidinedithiocarbamate (APDC) followed by extraction into isobutyl methyl ketone (IBMK). A gravity phase separator incorporated in the FI manifold was used to separate the organic phase which was stored in a collector tube, from which a 50 ml aliquot was introduced into the GF.An enrichment factor of 25 and a detection limit of 4 ng l21 were obtained. Sorbent extraction Undoubtedly, the systems based on sorption on mini-columns located in the valve loop,39–44 in the tip of the sampling probe46,47,88–90 or in knotted reactors38,45,50,51,91 are most extensively used for on-line preconcentration of trace amounts of metals.The main differences between the manifolds developed so far include the position of the sorption column in the manifold, the sample loading mode (volume- or timebased), sorbent material [chelating ion exchangers such as Chelex-100, immobilized 8-hydroxyquinoline (I-8HOQ), Muromac A-1, C18 bonded silica gel, anion and cation exchangers, polymer sorbents, etc.] and the eluent delivery mode to the GF.The first successful applications used a conically shaped microcolumn filled with bonded silica with octadecyl functional groups (C18) and employed diethyldithiocarbamates as chelating agents.39–44,88 The metal chelates formed on-line were loaded onto the column, eluted with ethanol and the central portion containing the most concentrated analyte fraction was collected in a tube and introduced on-line into the GF by an air flow. This zone sampling technique in ETAAS permits the introduction of any portion of the concentrate without deterioration of reproducibility as long as the total mass injected remains constant. However, the procedure was not considered rugged enough for long-term unattended operation.39 The presence of water in the eluate dramatically slows down evaporation in the tube; therefore, air segments must be introduced after the rinsing step and also between sample and eluent to avoid mixing of the neighboring phases.Many modifications of the original proposals have been published; all the improvements were designed to minimize dispersion,38,39,44,50 to achieve better enrichment factors38,51 and to automate the process completely.38 A significant reduction of dispersion at the elution and eluate introduction stages was achieved when an air flow was used to drive the eluate into the delivery tube and then into the GF.38 Other improvements are related to the fact that the chelation with dithiocarbamates followed by the sorption onto packed minicolumns has been replaced by direct sorption on chelating resins (Chelex-100,47 Dowex 50W-X892 or Muromac A-190) and on I- 8HOQ.46,88 It appears that solid-phase chelation offers several advantages over solution-phase chelation: the materials are reusable, provide low blank values and exhibit good stability in acidic media; they also have a high exchange capacity and large stability constants, although the non-selectivity of the active sites in some resins led to partial retention of alkaline earth elements present in the matrix.These microcolumns46,47,90 are preferentially located on the autosampler arm tip, thereby forming the interface between the FI system and the GF. The whole process consists of various steps: sample loading, washing, elution, and transfer into the GF followed by signal measurement. With few exceptions, the preconcentration procedures are similar, but the mode of eluate introduction into the furnace varies.For example, the effluent from the I-8HOQ column, containing the trace metals of interest, is directed into the GF by manually turning the sampling arm into the injection position and holding it there for the duration of the elution phase (22 s).46 However, with the Chelex-100 column, the autosampler aspirates the eluent (about 80 ml of nitric acid) through the filled microcolumn and the elements released are eluted directly into the GF.47 The authors claim that the reversal of the flow direction for elution minimizes sorbent compacting and increases the eluent–sorbent interaction.The autosampler program also includes air segmentation between injections; hence, there is no dilution or mixing whatsoever. A minicolumn containing an anion-exchange resin (Dowex- 1) was incorporated in a time-based solenoid injector (TBSI) for the removal of spectral interferences from sulfur anions in the determination of manganese in geothermal waters by ETAAS.48 The interferents were retained on the column while the analyte plug was inserted into the sampling tube of the TBSI for subsequent introduction of the desired volume into the GF.Whatever the chelation type, severalcessful applications in speciation studies by ETAAS have been reported. For instance, AsIII can be quantitatively extracted at pH 2, after complexation with diethyldithiocarbamate (DDTC) and sorption on a reversed-phase C18 column.42 AsV does not form a complex with DDTC under any conditions, a reduction step with a mixture of strong reagents (KI–Na2S2O8–Na2SO3–HCl) being necessary.Similarly, the CrVI species can be separated and concentrated; total chromium is determined after off-line oxidation of CrIII by potassium peroxodisulfate, the CrIII content being obtained by difference.41 Sorption on a polymeric DETATA material was performed for the separate preconcentration of both chromium species at different pH values.93 MW irradiation was used to intensify the processes of complex formation and sorption used for the preconcentration of chromium species under dynamic conditions followed by determination of the metal in the eluate or in the solid phase of the sorbent, using slurry sampling–ETAAS.Aqueous samples containing CrIII and CrV were continuously pumped through a microcolumn packed with DETATA sorbent located in a MW oven operated at 500 W. Quantitative sorption of CrIII was achieved at pH 7 and of CrVI at pH 3.The dynamic process under the MW irradiation takes only 5 min compared with static conditions which require more than 20 min. Also, the preconcentration efficiency is improved 200–300 times. In recent work, a knotted reactor was used to adsorb the CrVI–APDC Analyst, April 1998, Vol. 123 565complex for on-line speciation and preconcentration studies.45 The complex was eluted with a monosegmented discrete zone of ethanol and the analyte was quantified by ETAAS.All these sorption systems allow all the necessary manipulatory operations to be executed on-line: (a) effective entrapment of the analyte in the preconcentration unit; (b) washing of the column to remove possible interfering matrix components; (c) quantitative elution in the smallest possible volume of eluent; (d) reproducible transport of the concentrate to the GF with minimum dispersion; and (e) efficient cleansing of the entire system to avoid carry-over between individual samples.The column material is tolerant to complex matrices although the presence of heavy metals can cause severe interferences owing to competition with the analyte element in complex formation and/or adsorption on the column. Another common feature is related to the concentrated analyte introduction mode into the GF. In all cases the selected volume was transferred by air displacement. Also, all the applications are limited to water samples; sea-water has been the matrix most studied owing to its high salinity and low metal content (Table 1).Liquid chromatographic separations LC, especially high-performance LC (HPLC), is another powerful separation technique with great resolving power, which employs conventional ion-exchange columns for the separation of trace amounts of many non-volatile compounds. These techniques, coupled to ETAAS, are capable of sensitive detection of a wide range of molecular species containing metals or metalloids.The fractions are collected25–30 and periodically sampled in a pulsed-mode operation for offline analysis (Fig. 1). Such systems do not provide real-time separations and the optimization of the experimental parameters is difficult to perform. Computerized, automated interfaces are efficient for on-line connections with real-time sampling of the fractions.35,52–54 The sampling rate in these cases depends on the rate of the GF analysis and subsequent cool-down time; consequently, low flow rates are used but broad peaks are obtained.To increase sample throughput, the sampling procedure was modified to store the chromatographic peaks temporarily and thus be able to analyze multicomponent mixtures of organometallics of the same metal or multi-element, multicomponent mixtures.94 These sampling modes have been successfully demonstrated for the speciation of organocompounds of arsenic,28 lead,94 tin,30,95 selenium,25,96 etc.However, flow control and gradient elution cause background signal fluctuations owing to molecular absorption. This problem is effectively handled by employing suitable background correction systems. Microwave digestion Depending on the sensitivity required, there is not always a need for analyte preconcentration. However, owing to the severe matrix interferences experienced in ETAAS, it is always necessary to isolate and/or destroy the components that accompany the analyte in complex samples. MW-assisted sample treatment is widely used as an alternative to conventional wet digestion techniques as it is applicable to a wide range of sample types with minimal volatilization losses and contamination problems.There are several recent reviews dealing with the principles of MW field interaction with matter,22 the advantages and also the problems related to the use of MW treatment of complex samples for analytical purposes,18 or the multiple applications in geological, biological and environmental fields.19,22 To our knowledge, there is only one literature survey dedicated entirely to ETAAS after ‘in-batch’ MW treatment, in which relevant papers published up to 1994 are presented.The variety of elements determined (Al, As, B, Cd, Co, Cr, Cu, Fe, Mn, Mo, Ni, Pb, Sb, Se, Sn, Sr, Pb, V, Zn, etc.) and of matrices treated (biological, environmental, metallurgical, geological, etc.) are underlined in comprehensive tables. The authors also describe the digestion procedure (reagents used, MW oven power and irradiation time) as well as the experimental conditions (temperature program, chemical modifier and type of background correction) and some analytical figures of merit (detection limit, precision and recovery rate).However, on-line MW sample treatment for ETAAS detection is scarcely mentioned. When a MW digestion system is incorporated in a FI manifold, automatic mixing of the sample with suitable acids, as well as the pumping of the digest through the oven, is permitted.Finally, the digest might be collected in open vessels for subsequent analysis, or the outlet of the MW oven might be directly coupled to an instrument for on-line measurements. The first version was used, for example, for the digestion of food samples for the sequential determination of aluminum by ETAAS.26 Sample and nitric acid, simultaneously injected through an injector conmutator, were mixed by the mergingzones technique before entering a reaction coil wrapped around an Erlenmeyer flask filled with water (dummy load to prevent damage of the magnetron).The digests were collected in the autosampler cups. The performance of the method was evaluated by determining aluminum in shellfish, which typically contain abundant organic matter. The recovery of aluminum was only about 90%, indicating incomplete matrix decomposition, which entailed using a correction factor in order to determine the true amount of analyte in the samples.Total digestion of the sample is not always necessary, provided that reproducible recoveries are always obtained. However, other workers84 centered their interest on the total destruction of the silicate matrix and all mineral species comprising sediment samples in order to ensure the measurement of the total trace element content. Although a strong acidic mixture (HNO3, HCl and HF) was used and clear digests were obtained, a fine residue was observed at the bottom of the autosampler collection tubes on standing.The fact that aluminum was not fully recovered from standard sediments (only 68% of the certified value) suggests that the residue is due to undissolved alumina. As the FI-MW system was not directly coupled to ETAAS, in neither case were there problems associated with pressure changes; no special devices were necessary for de-gassing and the method permits the sequential and economical treatment of different types of sample.To avoid manipulation of the digest prior to the analyte measurement, totally on-line systems are preferred. The design and operation of an on-line automated MW-assisted mineralization and FI-ETAAS system was first described for the determination of lead in biological materials.55 The autosampler provided with the instrument was synchronized with the FI system to introduce the chemical modifier, thus making the entire system automated and on-line.The total analysis time was 3 min including the furnace program time. Some analytical figures of merit are given in Table 3. The accuracy was investigated by determining the lead content of various CRMs. Good agreement was obtained for all the biological samples, but slightly lower values were found for botanical materials, suggesting that more drastic mineralization conditions should be used, e.g., longer residence time of the acid slurries in the MW oven. A similar approach was developed for the determination of titanium dioxide in soaps.56 On-line microwave-assistance of the dissolution process and the time-based introduction of the liquid sample into the graphite tube minimized the pretreatment steps, allowing transfer to the atomizer without exposure to the environment.Less than 30 mg of soap sample were manually placed in a sample introduction device, laboratory-made from a 200 ml plastic pipette tip, placed in the FI manifold, just outside the MW oven.Synchronously with the 566 Analyst, April 1998, Vol. 123introduction of an acidic solution of sulfuric and nitric acid, a solenoid injector introduced two individual zones of an organic solvent in order to confine the aqueous segment containing the dissolved sample. This step was necessary, given the poor reproducibility obtained owing to non-uniform dispersion of the sample plug in the system. In this work, carbon tetrachloride was chosen from among other organic solvents, as being transparent to microwaves, and because of the insolubility of titanium dioxide in this solvent.The sample–acid mixture was mineralized as it passed through the knotted reactor located in the irradiation zone of the oven. A compromise had to be made between the flow rate (0.8 ml min21) and the length of the mineralization tubing (3 m), in order to keep the residence time in the oven at 45 s. The lack of certified standards for such samples necessitated the use of the standard additions method and an alternative technique97 to check the absence of matrix effects and the accuracy of the proposed procedure.The concept of continuous FI-MW-ETAAS has been further developed for the determination of cobalt in whole blood with in vivo sampling.57 The sample was pumped from the forearm vein of the patient and mixed with EDTA as anticoagulant and nitric acid to aid mineralization. A time-based solenoid injector sampled a portion of the mixture and introduced it into the carrier stream.The middle portion of the digest plug was allowed to fill the sampling arm assembly and the solenoid injector was synchronously activated to deposit sequentially 20 ml of the digest into the graphite tube. The software of the instrument was pre-programmed to introduce a chemical modifier, Mg(NO3)2, through the autosampler and to run the furnace program. In these systems a washing procedure is mandatory between samples.It is interesting that the mixing of blood samples with Triton X-100 improved the drying characteristics of the sample and reduced the accumulation of carbonaceous residue inside the graphite tube. Some of the analytical characteristics of such a system are given in Table 3. The accuracy of the procedure was investigated by determining the cobalt content in whole blood certified samples; the results obtained were in agreement with the certified values. A focused MW oven proved more efficient for the mineralization of adipose tissue samples followed by on-line determination of iron and copper by ETAAS.58 A sampling unit device, containing about 20 mg of the lyophilized subcutaneous adipose tissue sample, was introduced into the radiation zone with the aid of a mechanical arm.An acidic mixture (2 mol l21 sulfuric and 1.0 mol l21 nitric acid in a 4 + 1 proportion) was allowed to flow through the sample unit in a closed flow coil of 0.1 m, with the oven activated at a power of 40 W for 10 s.The carrier solution propelled the sample plug which was mixed downstream with Triton X-100 and chemical modifier (Pd–Mg) solutions and the resulting solution filled the sample arm assembly. The first 50 ml were discarded and the next 300 ml were retained in the PTFE collector tube of the sampling arm assembly until the graphite tube was ready to perform its temperature program. A solenoid valve synchronized with the computer of the spectrometer allowed the deposition of aliquots containing the sample onto the graphite tube platform. The accumulation of greasy solids on the wall of the exit tubing was avoided by the intercalation of a stream of Triton X-100 (0.02% v/v) in the system. The accuracy of the procedure was tested by determining iron and zinc in 20 different adipose tissue samples by flame AAS after liquid–liquid extraction of the metal chelates of sodium diethyldithiocarbamate with IBMK.98 The system described allowed the analysis of adipose tissues containing 2.3–8.6 and 1.5–4.4 mg g21 of iron and zinc, respectively, with good precision and accuracy (Table 3), in a totally closed system with minimal sample manipulation and exposure to the environment.Another interesting FI-ETAAS approach was recently used to determine iron in geothermal fluids containing large amounts of dissolved sulfate and sulfide anions.59 It is well known that these anions have a depressing effect on the iron ETAAS response owing to background absorption.Use of lutetium as a chemical modifier minimized the sulfate interference. The presence of sulfide ions still deteriorated the precision; therefore, sulfide was precipitated with hydrogen peroxide in a knotted reactor exposed to MW radiation to aid the precipitation process. The knotted reactor promoted radial mixing of sample and reagent providing reproducible conditions for the precipitation and also acted as a collector of the precipitate.A portion of the sulfate and sample containing iron was sequestered downstream in a sampling arm for the sequential deposition of fixed aliquots onto the GF platform. Meanwhile, the sulfur precipitated was dissolved on-line with carbon tetrachloride and diverted to waste. This sequence was synchronized with the computer of the spectrometer, which was pre-programmed to introduce aliquots of Lu-modifier prior to each sample deposition.For the less sensitive iron line (296.7 nm) the linear range was from 12 to 280 mg l21, with a characteristic mass of 104.8 pg and a detection limit of 72 pg (3.6 mg l21). The precision for ten consecutive measurements was around 3%; the accuracy was checked by performing recovery tests and by comparison with an independent analytical method. Table 3 On-line MW sample treatment with ETAAS detection Element or species Sample type Digestion mixture Chemical modifier Digestion coil length/m MW power/W Irradiation time/s DL*/ mg l21 (pg) RSD (%)† Ref.As, Cr Biological tissues; marine sediments; CRM HNO3–H2O2; HNO3–HCl– HF; (digest collected) NR 30 275 for tissues; 360 for sediments 240 NR 1 84 Al Shellfish HNO3 (digest collected) Mg(NO3)2 2 800 120 10 (NR) 4.3 26 CrIII, CrVI River water Aided sorption with HCl on DETATA column NR 0.01 500 300 0.03 (NR) 3–4 93 Pb CRM HCl–HNO3 Pd 2 700 25 0.1 (0.8) 2.5 55 Co Whole blood (in vivo uptake) HNO3 Mg(NO3)2 5 300 50 0.3 (6) 2.5–3.0 57 Fe, Zn Adipose tissue HNO3–H2SO4 Pd–Mg 0.1 40 10 NR (20) NR (30) 3.2 2.3 58 Ti Soaps HNO3–H2SO4 Mg–NH4 + 3 700 45 4 (80) 1.8–3.5 56 Fe Geothermal waters Precipitation of S with H2O2 Lu 2 400 NR 3.6 (72) < 3 59 * DL = Detection limit.† n = 10. ‡ NR = Not reported. Analyst, April 1998, Vol. 123 567The characteristics of the on-line MW systems given in Table 3 show their great versatility in processing samples of different nature with minimal analyst attention, in a very short time.Obviously, the irradiation time depends on the coil length inside the oven and on the flow rate. Such systems are not only used to digest samples, but also to aid some analytical processes such as sorption on columns93 or precipitation reactions.59 Conclusion The literature survey presented in this paper clearly shows that the main obstacles for efficiently and harmoniously combining FI with ETAAS have been overcome and that the approach is promising for the totally automated determination of many elements at ng l21 levels. Coupling a micro-scale FI sample processing system on-line with ETAAS leads to a more powerful integrated system (Fig. 1). Its most significant advantages are the improved detection limits achieved by preconcentration and matrix separation and the low degree of contamination, owing to the totally closed systems used. The weakness of such approaches is the lack of commercially available connections for eluate or digest sequestration and automatic injection into the GF.The systems developed so far have been strongly dependent on the creativity of researchers and on laboratory facilities. Additionally, direct coupling might require the furnace to be maintained at the atomization temperature for the period necessary to carry out the whole operation, producing overheating of the furnace housing and shortening the lifetime of the tube. Specially designed atomizers have to be used to avoid these difficulties.It can be expected, however, that owing to the potentiality and versatility of these systems to effect complex operations, future research activities will be focused on solving instrumental difficulties and on widening the range of applications to include more work on speciation. With commercialized dedicated hardware and software becoming available, FI-ETAAS will experience a faster development in the future. Such integrated systems permit fully automated operations, avoiding time-consuming manual work and enhancing accuracy and precision.References 1 Alcock, N. W., Anal. Chem., 1995, 67, 503R. 2 Crain, J. S., and Kiely, J. T., J. Anal. At. Spectrom., 1996, 11, 5251. 3 Bloxham, M. J., Hill, S. J., and Worsfold, P. J., J. Anal. At. Spectrom., 1994, 9, 935. 4 Sanz-Medel, A., Analyst, 1995, 120, 799. 5 Tangen, A., Trones, R., Greibrokk, T., and Lund, W., J. Anal. At. Spectrom., 1997, 12, 667. 6 Taylor, A., Branch, S., Crews, H.M., Halls, D. J., Owen, L. M., and White, M., J. Anal. At. Spectrom., 1997, 12, 119R. 7 Burguera, J. L., and Burguera, M., J. Anal. At. Spectrom., 1997, 12, 643. 8 Fang, Z., Flow Injection Atomic Absorption Spectrometry, Wiley, Chichester, 1995. 9 Valc�arcel, M., and Luque de Castro, M. D., Automatic Methods of Analysis, Elsevier, Amsterdam, 1988. 10 Automation in the Laboratory, ed. Hurst, W. J., VCH, New York, 1995. 11 Stockwell, P. B., and Corns, W. T., Automatic Chemical Analysis, Taylor & Francis, London, 1996. 12 Kingston, H. M., and Haswell, S. J., Microwave Enhanced Chemistry, ACS, London, 1997. 13 D�edina, J., and Tsalev, D., in Hydride Generation Atomic Absorption Spectrometry, Chemical Analysis, a Series of Monographs on Analytical Chemistry and its Application, ed. Winefordner, J. D., and Kolthoff, I. M., Wiley, New York, 1995, vol. 130. 14 Slavin, W., Trace Element Analysis in Biological Specimens, ed. Herber, R. F. M., and Stoeppler, M., Elsevier, Amsterdam, 1994, ch. 3. 15 Fang, Z., Xu, S.-k., and Tao, G., J. Anal. At. Spectrom., 1996, 11, 1. 16 Zhi, Z.-l., R�ýos, A., and Valc�arcel, M., Crit. Rev. Anal. Chem., 1996, 26, 239. 17 Matusiewicz, H., and Sturgeon, R. E., Prog. Anal. Spectrosc., 1989, 12, 21. 18 Burguera, M., and Burguera, J. L., Quim. Anal., 1995, 15, 112. 19 Smith, F. E., and Arsenault, E. A., Talanta, 1996, 43, 1207. 20 Burguera, M., and Burguera, J. L., Anal. Chim. Acta, in the press. 21 Matusiewicz, H., and Sturgeon, R. E., Spectrochim. Acta, Part B, 1996, 51, 377. 22 Zlotorzynski, A., Crit. Rev. Anal. Chem., 1995, 25, 43. 23 Subramanian, K. S., Spectrochim. Acta, Part B, 1996, 51, 291. 24 Chakraborti, R., Das, A. K., Cervera, M. L., and de la Guardia, M., Fresenius’ J. Anal. Chem., 1996, 355, 99. 25 Laborda, F., Vicente, M. V., Mir, J. M., and Castillo, J. R., Fresenius’ J. Anal. Chem., 1997, 357, 837. 26 Arruda, M. A. Z., Gallego, M., and Valc�arcel, M., J.Anal. At. Spectrom., 1995, 10, 501. 27 B�ackstr�om, K., Danielsson, L. G., and Nord, L., Analyst, 1984, 109, 323. 28 Brinckman, F. E., Blair, W. R., Jewett, K. L., and Iverson, W. P., J. Chromatogr. Sci., 1977, 15, 493. 29 Brinckman, F. E., Jewett, K. L., Iverson, W. P., Irgolik, K. J., Ehrhardt, K. C., and Stockton, R. A., J. Chromatogr., 1980, 191, 31. 30 Parks, E. J., Brinckman, F. E., and Blair, W. R., J. Chromatogr., 1979, 185, 563. 31 Sotera, J. J., Christiano, L.C., Conley, M. K., and Kahn, L. H., Anal. Chem., 1983, 55, 208. 32 Pinel, R., Benabdallah, M. Z., and Guihal, C., Anal. Instrum., 1987, 16, 275. 33 Bank, P. C., de Loos-Vollebregt, M. T. C., and de Galan, L., Spectrochim. Acta, Part B, 1989, 44, 571, 34 Bendicho, C., Anal. Chem., 1994, 66, 4375. 35 Berndt, H., and Schaldach, G., J. Anal. At. Spectrom., 1994, 9, 39. 36 B�ackstr�om, K., and Danielsson, L. G., Anal. Chem., 1988, 60, 1354. 37 B�ackstr�om, K., and Danielsson, L. G., Anal.Chim. Acta, 1990, 232, 301. 38 Sperling, M., Yan, X.-p., and Welz, B., Spectrochim. Acta, Part B, 1996, 51, 1891. 39 Welz, B., Sperling, M., and Sun, X.-j., Fresenius’ J. Anal. Chem., 1993, 346, 550. 40 Fang, Z., Sperling, M., and Welz, B., J. Anal. At. Spectrom., 1990, 5, 639. 41 Sperling, M., Yin, X., and Welz, B., Analyst, 1992, 117, 629. 42 Sperling, M., Yin, X., and Welz, B., Spectrochim. Acta, Part B, 1991, 46, 1789. 43 Berglund, M., Frech, W., Baxter, D. C., and Radziuk, B., Spectrochim.Acta, Part B, 1993, 48, 1381. 44 Xu, S.-k., Sperling, M., and Welz, B., Fresenius’ J. Anal. Chem., 1992, 344, 535. 45 Nielsen, S., and Hansen. E. H., Anal. Chim. Acta, in the press. 46 Azeredo, L. C., Sturgeon, R. E., and Curtius, A. J., Spectrochim. Acta, Part B, 1993, 48, 91. 47 Fern�andez, F. M., Stripeikis, J. D., Tudino, M. B., and Troccoli, O., Analyst, 1997, 122, 679. 48 Burguera, J. L., Burguera, M., Rivas, C., Carrero, P., Gallignani, M., and Brunetto, M.R., J. Anal. At. Spectrom., 1995, 10, 479. 49 Bank, P. C., de Loos-Vollebregt, M. T. C., and de Galan, L., Spectrochim. Acta, Part B, 1988, 43, 983, 50 Chen, H., Xu, S.-k., and Fang, Z., Anal. Chim. Acta, 1994, 298, 167. 51 Fang, Z., Xu, S.-k., Dong, L., and Li, W., Talanta, 1994, 41, 2165. 52 Castillo, A. I., and Segar, D. A., Proceedings of the International Conference on Heavy Metals in the Environment, Toronto, Canada, CEP Consultants, Edinburgh, 1975, vol. 1, 1-183-204. 53 Vickrey, T. M., Buren, M. S., and Howell, H. E., Anal. Lett., 1978, A11, 1075. 54 Haswell, S. J., Stockton, R. A., Bankroft, K. C. C., O’Neill, P., Rahman, A., and Irgolic, K. J., J. Autom. Chem., 1987, 9, 6. 568 Analyst, April 1998, Vol. 12355 Burguera, J. L., and Burguera, M., J. Anal. At. Spectrom., 1993, 8, 235. 56 Burguera, M., and Burguera, J. L., Lab. Robot. Autom., 1993, 5, 1. 57 Burguera, M., Burguera, J. L., Rond�on, C., Rivas, C., Carrero, P., Gallignani, M., and Brunetto, M.R., J. Anal. At. Spectrom., 1995, 10, 343. 58 Burguera, J. L., Burguera, M., Carrero, P., Rivas, C., Gallignani, M., and Brunetto, M. R., Anal. Chim. Acta, 1995, 308, 349. 59 Burguera, J. L., Burguera, M., and Rond�on, C., Anal. Chim. Acta, in the press. 60 De Jonghe, W., Chakraborty, D., and Adams, F., Anal. Chim. Acta, 1980, 115, 89. 61 Blair, W. R., and Brinckman, F. E., Anal. Chem., 1977, 49, 378. 62 D�edina, J., Frech, W., Lundberg, E., and Cedergren, A., J.Anal. At. Spectrom., 1989, 4, 143. 63 Bermejo-Barrera, P., Moreda-Pi�neiro, J., Moreda-Pi�neiro, A., and Bermejo-Barrera, A., J. Anal. At. Spectrom., 1997, 12, 317. 64 Lee, S. H., Jung, K., and Lee, D. S., Talanta, 1989, 36, 999. 65 Hladk�y, Z., Ri�sov�a, J., and Fi�sera, M., J. Anal. At. Spectrom., 1990, 5, 691. 66 Sinemus, H. W., Stabel, H. H., Radziuk, B., and Kleiner, J., Spectrochim. Acta, Part B, 1993, 48, 643. 67 Baxter, D. C., and Frech, W., Anal. Chim. Acta, Part B, 1989, 225, 175. 68 Baxter, D. C., Nichol, R., and Littlejohn, D., SpecYan, X., Ni, Z., and Guo, Q., Anal. Chim. Acta, Part B, 1993, 272, 105. 70 Sinemus, H. W., Stabel, H. H., Radziuk, B., and Kleiner, J., Spectrochim. Acta, Part B, 1993, 48, 1719. 71 Dittrich, K., Franz, T., and Wennrich, R., Spectrochim. Acta, Part B, 1994, 49, 1695. 72 Garbos, S., Walcerz, M., Bulska, E., and Hulanicki, A., Spectrochim. Acta, Part B, 1995, 50, 1669. 73 Haug, H. O., and Liao, Y.-p., Fresenius’ J. Anal. Chem., 1996, 356, 435. 74 Haug, H. O., Spectrochim. Acta, Part B, 1996, 51, 1425. 75 Haug, H. O., and Yiping, L., J. Anal. At. Spectrom., 1995, 10, 1069 76 Shuttler, I. L., Feuerstein, M., and Schlemmer, G., J. Anal. At. Spectrom., 1992, 7, 1299. 77 Erber, D., Bettmer, J., and Cammann, K., Fresenius’ J. Anal. Chem., 1994, 349, 738. 78 Sturgeon, R. E., Willie, S. N., Sproule, G. Y., Robinson, P. T., and Berman, S. S., Spectrochim. Acta, Part B, 1989, 44, 667. 79 Li, Z., McIntosh, S., Carnrick, G. R., and Slavin, W., Spectrochim. Acta, Part B, 1992, 47, 701. 80 Tao, G., and Fang, Z., Talanta, 1995, 42, 375. 81 Sinemus, H. W., Kleiner, J., Stabel, H. H., and Radziuk, B., J. Anal. At. Spectrom., 1992, 7, 433. 82 Burguera, M., and Burguera, J. L., J. Anal. At. Spectrom., 1993, 8, 229. 83 Tao, G., and Fang, Z., J. Anal. At. Spectrom., 1993, 8, 577. 84 Sturgeon, R. E., Willie, S. N., Methven, B. A., Lam, J. W. H., and Matusiewicz, H., J. Anal. At. Spectrom., 1995, 10, 981. 85 Hilligsoe, B., Andersen, J. E. T., and Hansen, E., J. Anal. At. Spectrom., 1997, 12, 585. 86 Tsalev. D., D’Ulivo, A., Lampugnani, L., Di Marco, M., and Zamboni, R., J. Anal. At. Spectrom., 1996, 11, 979. 87 Tao, G., and Fang, Z., Spectrochim. Acta, Part B, 1995, 50, 1747. 88 Beinrohr, E., Cakrt, M., Rapta, M., and Tarapci, P., Fresenius’ Z. Anal. Chem., 1989, 335, 1005. 89 Porta, V., Abollino, O., Mentasti, E., and Sarzanini, C., J. Anal. At. Spectrom., 1991, 6, 119. 90 Sung, Y. H., Liu, Z. S., and Huang, S. D., J. Anal. At. Spectrom., 1997, 12, 841. 91 Yan, X. P., and Adams, F., J. Anal. At. Spectrom., 1997, 12, 459. 92 Burguera, J. L., Burguera, M., Carrero, P., Marcano, J., Rivas, C., and Brunetto, M. R., J. Autom. Chem., 1995, 17, 25. 93 Kubrakova, I., Kudinova, T., Formanovsky, A., Kuz’min, N., Tsysin, G., and Zolotov, Y., Analyst, 1994, 119, 2477. 94 Vickrey, T. M., Howell, H. E., and Paradise, M. T., Anal. Chem., 1979, 51, 1880. 95 Jewett, K. L., and Brinckman, F. E., J. Chromatogr. Sci., 1981, 19, 583. 96 Chakraborty, D., Hillman, D. C. J., Irgolic, K. J., and Zingaro, R. A., J. Chromatogr., 1982, 249, 81. 97 L�opez, I., Vi�nas, G. P., and Hern�andez, M., J. Anal. At. Spectrom., 1992, 7, 529. 98 Nord, L., and Karlberg, B., Anal. Chim. Acta, 1983, 145, 151. Paper 7/08112J Received November 11, 1997 Accepted February 10, 1998 Analyst, April 1998, Vol. 1
ISSN:0003-2654
DOI:10.1039/a708112j
出版商:RSC
年代:1998
数据来源: RSC
|
7. |
Fourier transform infrared spectrometry: a versatile technique for real world samples† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 571-577
Llewellyn Rintoul,
Preview
|
PDF (313KB)
|
|
摘要:
Fourier transform infrared spectrometry: a versatile technique for real world samples† Llewellyn Rintoul, Helen Panayiotou, Serge Kokot, Graeme George, Gregory Cash, Ray Frost, Thuy Bui and Peter Fredericks* Centre for Instrumental and Developmental Chemistry, Queensland University of Technology, 2 George Street, Brisbane, Qld 4000, Australia The versatility of FTIR spectrometry was explored by considering a variety of samples drawn from industrial applications, materials science and biomedical research.These samples included polymeric insulators, bauxite ore, clay, human hair and human skin. A range of sampling techniques suitable for these samples is discussed, in particular FTIR microscopy, FTIR emission spectroscopy, attenuated total reflectance and photoacoustic FTIR spectrometry. The power of modern data processing techniques, particularly multivariate analysis, to extract useful information from spectral data is also illustrated. Keywords: Fourier transform infrared spectrometry; infrared microscopy; emission; attenuated total reflectance; photoacoustic; chemometrics; hair; skin; bauxite; clay; polymer Commercial FTIR spectrometers have now been available for about 25 years.The extraordinary power of these instruments in terms of their speed and sensitivity has revolutionised infrared spectrometry so that dispersive instruments have now been almost completely replaced. The benefits of the interferometric approach are well known and derive from the multiplex advantage, the throughput advantage and the accuracy conferred by the use of laser referencing.1–5 Hence the speed and accuracy of data collection are much better than for the previous generation of dispersive instruments.However, although certain techniques benefit greatly from the improved speed of data collection, of equal importance to the analyst is the flexibility and versatility derived from the high sensitivity of the instruments.This has allowed the development, over two decades or more, of a range of techniques used in conjunction with an FTIR spectrometer, which allows accurate and reproducible IR spectra to be collected for a wide variety of sample types. Prior to the development of FTIR, IR spectrometry was predominantly carried out in the transmittance mode on relatively simple samples, although the technique of attenuated total reflectance (ATR) also found some use. Now, techniques such as diffuse reflectance, specular reflectance, attenuated total reflectance, photoacoustic spectrometry, infrared microscopy, long path spectrometry, and remote fibre optic spectrometry are routinely used.Furthermore, the nature of the sample has changed and become considerably more complex. Previously, IR spectrometry found most use in the characterisation and analysis of organic compounds. Although IR is still used for this purpose, it now also finds much application for diverse samples such as minerals, polymers, thin films, living cells and biological tissues.Many of the IR techniques used to characterise these samples are energy wasteful and in some cases less than 1% of the available energy finally reaches the detector. Because of the sensitivity of FTIR, these sampling techniques can be fairly routine despite the low available energy. Another aspect of the FTIR revolution derives from the inherent stability of the instruments, which gives highly reproducible spectra, provided that care is taken over sample preparation and presentation.The reproducibility of FTIR spectra, combined with the high signal-to-noise ratio, has allowed the development of computerised data processing techniques of direct benefit to analysts. Whereas height or area measurement of a single band was the typical analytical approach prior to FTIR, it is now commonplace to apply multivariate techniques such as principal components regression (PCR) or partial least squares (PLS) to solve analytical problems.This area has been given the name chemometrics and is now a major field of study in its own right. This paper will demonstrate the versatility of modern FTIR spectrometry by describing some of the range of techniques available. In particular, the techniques of infrared microscopy and emission FTIR will be discussed in some detail. Examples are drawn from current work being carried out in the Vibrational Spectroscopy Laboratory at the Queensland University of Technology.The breadth of current FTIR applications means that the paper cannot be comprehensive, but rather is meant to give some insight into the possibilities of the FTIR technique. Most of the applications described have an analytical basis. Experimental FTIR spectra were collected using a Perkin-Elmer (Norwalk, CT, USA) System 2000 FTIR spectrometer equipped with a germanium on KBr beamsplitter and both room temperature deuterated triglycine sulfate (DTGS), and liquid nitrogencooled mercury–cadmium–telluride (MCT) detectors. This spectrometer is attached to a Perkin-Elmer infrared microscope which has its own small element MCT detector.The microscope is equipped with a television camera to display images, and also with a computerised stage, programmable in the x and y directions. Photoacoustic spectrometry was carried out with an MTEC (Ames, IA, USA) Model 200 photoacoustic cell which was flushed with ultra-pure helium before use.Background spectra were obtained by the use of finely powdered carbon black. ATR spectra were obtained with a Graseby–Specac (Orpington, Kent, UK) horizontal ATR accessory equipped with a 45° ZnSe internal reflection element (IRE) or with a Perkin-Elmer vertical ATR accessory equipped with a 45° KRS-5 IRE. Emission FTIR spectra were obtained using a modified Bio- Rad (Richmond, CA, USA) FTS7 spectrometer which has been described previously.6 A schematic diagram of the sampling arrangement of the emission system is shown in Fig. 1. The sample was placed on a platinum plate (diameter 6 mm), the temperature of which could be controlled between ambient and about 1100 °C. Background spectra were obtained by the use of a graphite plate in place of the platinum plate. It was necessary to collect a background spectrum at every temperature of † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.Analyst, April 1998, Vol. 123 (571–577) 571interest to allow for the variation in emission intensity at different temperatures. Spectral data processing was carried out with the GRAMS32 software package (Galactic Industries, Salem, NH, USA). Chemometric operations were carried out using the SIRIUS suite of programs. Infrared microscopy The infrared microscope has become one of the most significant accessories for an FTIR spectrometer.7 It finds application in a wide variety of analytical situations such as forensic science and materials science.Typically, the throughput of an infrared microscope is around 5% compared with the open beam of the spectrometer. It is the sensitivity of the FTIR spectrometer which allows the microscope to be a routine tool despite the large loss of energy. Dedicated small element MCT detectors are typically used with microscopes to achieve the highest sensitivity. It is becoming common to map surfaces by FTIR microscopy.Generally, this is achieved automatically by moving the sample with a computer-controlled stage so that a grid pattern of measurements is made of the sample area. The sample surface should be flat and horizontal or problems with focus will arise. Recently, FTIR microscopes have become available which autofocus before each measurement, which obviates the necessity for a perfectly flat and horizontal sample surface. An alternative approach to FTIR mapping uses an MCT focal-plane array detector to obtain spectroscopic images in a single measurement.8,9 Currently, such array detectors have relatively few pixels (for example, 64364 is common) and are very expensive, but there will no doubt be significant developments in the near future.Studies of bauxitic pisoliths Bauxite ore occurs at Weipa in Queensland, Australia, as pisoliths, which are small, approximately spherical, pebbles with diameters in the range 0.5–2 cm.The pisoliths are composed of a number of minerals including gibbsite {Al(OH)3}, boehmite {AlO(OH)}, haematite {Fe2O3}, goethite {FeO(OH)}, quartz and the clay kaolinite. The distribution of the different minerals within the pisoliths is of interest as it throws some light on the genesis of the ore body. FTIR microscopy offers an approach to solve this problem because the minerals involved have significantly different IR spectra. 10–12 Pisolith samples were sectioned with a diamond saw, and the exposed surface was polished with silicon carbide paper of grit P400 at first, followed by 3 mm diamond paste on a soft polishing cloth.FTIR microscopic reflectance spectra of the four common minerals are shown in Fig. 2. These spectra were taken of the polished mineral surface and are predominantly specular in nature. The distortion of the hydroxyl stretch region, for example in gibbsite, can be clearly seen. Application of the Kramers–Kronig transformation allows the calculation of an absorbance spectrum from the specular reflectance data.The exact nature of the reflectance spectra of solid samples depends on the nature of the sample surface. Polished surfaces give specular spectra, whereas matt surfaces gives diffuse reflectance spectra. Both of these types of spectra are readily understood by the application of the Kramers–Kr�onig transformation or the Kubelka–Munk equation, respectively. A difficulty arises when the surface is such that spectra are mixed specular and diffuse.These spectra cannot be corrected. It is possible that different parts of the same sample, such as a pisolith, may provide specular, diffuse and mixed spectra. This can make data analysis extremely complicated. In the case of the pisoliths it was found that best results were obtained by using the spectra as they were collected, without applying any correction routines.12 A pisolith was sectioned through its centre, polished and mounted on the computerised stage of the FTIR microscope.It was then mapped in a grid pattern at a spatial resolution of 200 mm with a spectral resolution of 8 cm21. Each measurement took 16 s. A 1 cm pisolith required 250 spectra with a total measurement time of just over 1 h. Of course, a much higher spatial resolution is possible up to the diffraction limit of around 10 mm, but this would lead to an enormous database of 106 spectra and a measurement time of 4444 h for a single pisolith sample! The data can be manipulated to provide false colour maps of the distribution of the various minerals within the pisolith.Fig. 3(A)–(D) show the distributions (the lighter the colour, the higher the concentration), for this particular pisolith, for the four minerals gibbsite, boehmite, quartz and kaolinite. Gibbsite and boehmite are the two major constituents and both occur in rings with boehmite occupying the centre of the pisolith. Kaolinite is less common and occurs as a single ring.Quartz occurs as a random distribution of individual grains. Although it would not be possible to map pisoliths at the highest spatial resolution, the approximately circular distribu- Fig. 1 Schematic diagram of sampling arrangement for FTIR emission spectrometry. Fig. 2 FTIR microscopy reflectance spectra of some of the different minerals found in a bauxitic pisolith: (A) quartz; (B) gibbsite; (C) boehmite; and (D) kaolinite. 572 Analyst, April 1998, Vol. 123tion of most of the minerals means that useful information may be obtained from a line of spectra taken across the diameter of the pisolith section. In this way some high spatial resolution information was obtained. A similar approach to that used for the mapping of bauxitic pisoliths was also used to map human gallstones which are of similar size and often show a circular distribution of their component compounds.13 In the case of gallstones the components are both organic (e.g., cholesterol, bilirubin) and inorganic (e.g., calcium carbonate) and the FTIR microscope is sensitive to both types of compound.FTIR–chemometric study of human hair The FTIR microscope is becoming an indispensable tool in forensic science for the characterisation of physical evidence such as fibres, illicit drugs and paint chips. In this study, the power of FTIR microscopy combined with chemometrics was demonstrated. FTIR spectra of a set of six black, untreated, female hair samples (three Asian and three Caucasian) were collected by FTIR microscopy.The samples were flattened by rolling, to reduce lensing effects which distort the spectrum, fastened to a gold mirror and 10 spectra of each hair were collected from different points on the sample. A reference data set was established of the spectra for two samples from each of the two racial groups. The remaining spectra were used as validation data. The reference spectra were subjected to principal components analysis (PCA)14 on a double-centred (column and row mean-scaled), variance scaled matrix.Fuzzy clustering15 and SIMCA (soft independent modelling of class analogies)16 were used to identify atypical spectra which were removed from the set. Typical spectra are shown in Figure 4. To the eye these spectra appear essentially identical. The results of exploratory PCA are presented in Fig. 5. PC1 explains 61.6% of the variance and PC2 explains a further 32.1%.The plot reveals six clusters. PC1 discriminates between the FTIR spectra on the basis of the different races. The Asian black hair (CR1 and CR2) has negative scores on PC1, whereas the Caucasian black hair (CR1 and CR2) has positive scores. The two sets of validation spectra (AV and CV) fall within their respective groups. This preliminary study demonstrates the power of FTIR microscopy to measure reproducible low noise spectra, together with the extraordinary capability of chemometrics to distinguish between almost identical spectral data sets.The result is that it appears possible to distinguish between black Asian hair and black Caucasian hair by FTIR spectrometry. This may be of great significance in forensic science. FTIR emission spectroscopy The high sensitivity of the FTIR spectrometer, particularly when equipped with a photocurrent device such as an MCT detector, facilitates the direct measurement of the emission spectrum of a sample at elevated temperature.An emission infrared FTIR spectrometer may be constructed by replacing the infrared source of a conventional FTIR spectrometer by a mirror designed to direct a parallel beam of light, emitted from the sample placed on a hot stage external to the instrument, on to the beamsplitter of the interferometer.6 Two advantages of emission techniques will be illustrated by examples below. They are the ease with which spectra of samples at elevated temperature are obtained in situ and the high sensitivity.Emission Fig. 3 Mineral distribution, in a bauxitic pisolith by FTIR microscopy mapping. (A) gibbsite; (B) boehmite; (C) kaolinite; and (D) quartz. The lighter the pixel, the higher is the concentration of the particular mineral in that area. Analyst, April 1998, Vol. 123 573techniques are inherently more sensitive than other infrared techniques because the small signal is measured directly rather than requiring the detection of small differences between large signals as in transmittance or reflectance measurements. The same vibrational modes are probed in either emission or absorbance measurements, hence the structural information content is the same in both cases.Because the emittance spectrum emanates from downward vibrational energy level transitions, there must be a significant proportion of molecules in the excited vibrational state. Governed by Boltzmann's distribution, the signal strength is dependent on the frequency of the transition and the temperature of the sample. To obtain a normalised spectrum in a recognisable form from an emission measurement it is necessary to find the ratio of the sample emission to that of a black body at the same temperature as the sample; in these and most cases graphite was used to approximate a black body.At 40 °C, bands with frequencies below 1000 cm21 y be readily observed in the ratioed emittance spectrum and by 160 °C the signal-to-noise ratio of the high frequency region has improved so that the CH stretching bands are clearly manifest.For some applications, such as following reaction kinetics, it is also useful to linearise the emission signal with respect to concentration of the emitting species by relating the emittance to the absorbance of the sample. The platinum stub on which the sample is placed also contributes to the emission signal and absorption of this signal and also reabsorption of the sample emittance by the sample and instrumental background emission must be considered.The relationship A = 0.5log[(Ibb 2 Ie)/(Ibb 2 IPt)] (where Ibb, Ie and IPt are the intensities emitted by the graphite reference, the sample and the platinum stub, respectively) accounts for most of these effects in thermally thin samples and has proved useful in the study of samples applied as a thin film, but is not expected to hold true in the study of minerals, for example, where sample thickness and reflectance effects may dominate.17,18 Degradation of polymer insulators Many of the insulators used in high tension electricity transmission are now made of polymeric materials such as EPDM (ethylene–propylene–diene monomer) copolymer.The degradation mechanisms of these insulators are of significance as their dielectric properties can change markedly with time of outdoor exposure. In a study of this problem, the surface degradation of a polymeric insulator was investigated by absorption and emission FTIR in order to develop a simple nondestructive analytical procedure.The surface was swabbed several times with a xylene-soaked cotton bud. Each time the xylene solution containing small amounts of low molecular mass material and surface contaminants was collected. The solution was concentrated by evaporation and FTIR spectra were obtained by two methods. An absorption measurement was obtained by allowing a small amount of the xylene solution to evaporate on a KBr plate.The spectrum was then obtained as a transmission measurement. In another approach, a drop of the solution was placed on the platinum plate of an emission FTIR spectrometer which was maintained at 120 °C in a nitrogen atmosphere, to prevent further oxidation. Both absorption and emission spectra of the surface degradation products and contaminants are shown in Fig. 6. The quality of the emission spectrum in terms of signal-to-noise ratio is much better than that of the absorption spectrum.The presence of carbonyl groups formed by oxidation processes can be clearly seen and these form a ready index of the degradation of the polymer.19 Furthermore, absorptions in the region 1500–1700 cm21 indicate that there is residual xylene in the absorption sample, but not in the emission sample. This example demonstrates the sensitivity of the emission technique to very small amounts of sample.At the relatively low sample temperature of 120 °C, spectra can only be obtained in the reduced spectral range of about 2000–500 cm21, but this is usually sufficient to characterise the sample.19 The temperature of 120 °C was Fig. 4 Typical FTIR microscopy absorbance spectra of female hair in the region 1700–700 cm21. CR1 and CR2 are Caucasian subjects; AR1 and AR2 are Asian subjects; and CV and AV are validation data from additional Caucasian and Asian subjects, respectively. Fig. 5 PCA scores plot for FTIR microscopy absorbance spectra of female hair samples. Fig. 6 Spectra of surface material, obtained by a xylene swab, from an EPDM insulator. (A) Absorbance spectrum of evaporated solution on a KBr disk; and (B) emission spectrum of evaporated solution obtained at 120 °C under N2. 574 Analyst, April 1998, Vol. 123chosen to be high enough to obtain useful spectra, but sufficiently low to inhibit further oxidation. Thermal decomposition of the clay montmorillonite Because of the capability to access elevated temperatures, and to control the temperature accurately, an FTIR emission system is well suited to follow the thermal decomposition of mineral samples.20 This is illustrated in Fig. 7, which shows a set of emission FTIR spectra of a calcium montmorillonite clay (Clay Mineral Society Standard STx-2), collected at 100 °C intervals between 300 and 1100 °C. The loss of the hydroxy groups is clearly seen by the reduction in intensity of the O–H stretching emission bands near 3606 cm21, which begins around 500 °C and is complete by 600 °C.Because the hydroxyl emission bands are at the high frequency end of the spectrum, their measurement only becomes viable at temperatures above about 150 °C. Other changes can be seen in the Si–O stretching emission band near 1000 cm21 and the Al–O band near 600 cm21 as the layer structure of the montmorillonite changes into the three-dimensional network structure of wollastonite.The emission approach allows us to see clearly the temperature ranges in which these structural changes occur and, importantly, allows the thermal decomposition processes of dehydration, dehydroxylation, phase change and oxidation to be studied in situ at the elevated temperatures. Attenuated total reflectance (ATR) spectrometry ATR has been a common sampling technique for infrared spectrometry since it was developed for dispersive instruments in the early 1960s.21 The increased sensitivity of FTIR spectrometers makes ATR a simple and routine technique.In recent years, accessories in which the internal reflection element (IRE) is horizontal, rather than vertical, have become popular, especially for quality control purposes. The fixed, and rather small, pathlength of the ATR experiment also makes it suitable for aqueous solutions and several specialist accessories are available for such samples. The fact that the penetration depth of an ATR experiment depends on the angle of incidence and the refractive index of the IRE, in addition to the refractive index of the sample, means that depth profiling is possible by the use of different IREs and angles of incidence. However, the penetration depth always remains small, in the range 0.05–0.12l for a typical sample.1 Good contact between the sample and the IRE is required for ATR spectrometry, hence the preferred samples are viscous fluids or flexible solids.The work described below illustrates the simplicity and effectiveness of ATR as a sampling technique. Studies of sunscreens on human skin Skin has been studied by the ATR sampling technique almost since its development.22,23 The increased sensitivity of FTIR means that good quality spectra may now be collected with a measurement time of around 1 min, which is a considerable reduction compared with the 10 min required for a dispersive instrument.22 The rate of absorption of topical sunscreens by the skin was studied by ATR-FTIR spectroscopy.Fig. 8 compares an ATR spectrum of clean skin with a typical ATR spectrum of skin which has been treated with a commercial topical sunscreen. A horizontal ATR accessory equipped with a 45° incidence angle ZnSe IRE was used and studies were performed on the ball of the thumb. The sunscreen preparation to be tested was applied at a level of about 2 mg cm22 and spectra were collected hourly until the presence of sunscreen absorptions in the spectra could no longer be detected.Fig. 9 shows some of the spectra obtained from this experiment. Initially, bands due to the sunscreen are prominent, but after 12 h the spectrum is almost that of normal skin. The presence of two active components, octyl p-methoxycinnamate (OPM) and butyl-pmethoxydibenzoylmethane (BPM), can be detected by absorptions at 1167 and 1375 cm21, respectively. A plot of the relative band areas against time (Fig. 10) shows that the OPM is absorbed by the skin more quickly than the second active component BPM. The OPM has essentially disappeared from the skin surface after 6 h, whereas traces of BPM remain until nearly 10 h. Fig. 7 Set of FTIR emission spectra obtained at 100 °C intervals from 300 to 1100 °C of a calcium montmorillonite clay. Fig. 8 ATR spectra in the range 4000–700 cm21 of (A) clean human skin and (B) human skin 4 h after treatment with a commercial topical sunscreen. Fig. 9 Set of ATR spectra in the range 1800–1100 cm21 showing the absorption by skin of a topical sunscreen at (A) 0; (B) 2; (C) 4; (D) 6; (E) 8; (F) 10; and (G) 12 h. Spectrum H is of normal clean skin. Bands used to quantify the presence of the active components octyl p-methoxycinnamate (OPM) and butyl-p-methoxydibenzoylmethane (BPM) are indicated at 1167 and 1375 cm21, respectively. Analyst, April 1998, Vol. 123 575The use of a standard horizontal ATR accessory is not optimum for the study of skin since it is possible for only a few easily accessible areas of skin to be measured.This problem can be addressed by the use of a flexible fibre optic cable composed of an IR transmitting material such as chalcogenide, and equipped with an ATR probe head.24 Such probes are now commercially available from a number of sources. There is a significant loss of energy where the light is launched into the fibre, and also along the length of the fibre. Consequently, it is recommended that the more sensitive MCT detector is used in conjunction with the fibre optic cable.When this is done measurements of skin are routine and can be completed in 1–2 min. Photoacoustic spectrometry Although the technique of photoacoustic spectrometry (PAS) has been known for many years, it was not used in the midinfrared region until the advent of FTIR. The combination of PAS and FTIR delivers a potent tool to the analyst for the study of solid samples.The major benefit of PAS is that, to a large extent, the general features of the spectrum are independent of the sample morphology.25,26 This is in contrast to other IR sampling techniques for solids such as diffuse reflectance, for which the particle size and shape of the sample are critical. PAS can therefore be used to obtain spectra of a wide range of difficult solid samples including tough polymers, coal, minerals and fibres. Minimal sample preparation is required to obtain an acceptable spectrum, but more care is required for serious quantitative work.Depth profiling with PAS is possible, to a certain extent, by varying the velocity of the moving mirror.27 The slower the mirror moves, the greater is the penetration depth (thermal diffusion length). The actual penetration depth also depends on the sample properties of thermal conductivity, density and specific heat capacity, and typically can only be varied within the approximate range 1–10 mm.However, the PAS method has been used successfully, for example, to investigate surface treatment of wool fibres.28 When a rapid scan interferometer is used, each wavelength in the spectrum has a different modulation frequency and hence a different thermal diffusion length. This problem can be overcome by the use of a step-scan interferometer, which allows all wavelengths to be modulated at the same frequency, and also allows the advantages of phase-modulated/phase-resolved depth profiling methods.29–31 Photoacoustic FTIR of polymer chips Polymers are often very difficult samples to characterise by infrared spectrometry.They can be extremely tough and are often found as pellets or chips, or as manufactured articles. The following work illustrates how PAS proved much superior to FTIR microscopy for a particular polymer sample. Samples of acrylonitrile–butadiene–styrene (ABS) polymer were received in chip form. The surface of some of the chips was contaminated with an unknown black contaminant.Initial work in the reflectance mode with an FTIR microscope yielded spectra A and B in Fig. 11. Both spectra show the distortion common to specular reflection spectra, which is not surprising since many polymers have a smooth surface. Furthermore, both spectra are truncated to 700 cm21 because the FTIR microscope is equipped with the most sensitive narrow-range MCT detector. Kramers–Kr�onig analysis resulted in weak and inconclusive spectra.In contrast, PAS of the ABS component, and of the contaminant, gave excellent spectra over the full range of 4000–400 cm21, with little sample preparation other than cutting chunks small enough to fit in the PAS cell. The spectra are shown as C and D in Fig. 11. Spectrum C shows the features of an ABS spectrum including the moderately intense C·N stretching vibration near 2200 cm21. This band is extremely weak in the reflectance spectrum of the polymer chips (spectrum A).The contaminant is clearly not related to the ABS, and its simple spectrum indicates that it is polyethylene. The benefit of PAS in this case is that clear, unambiguous spectra were obtained rapidly and with a minimum of sample preparation. Conclusions The versatility of FTIR spectrometry has been demonstrated by considering a variety of samples of mineral, polymer and biological origin. The samples considered were not the samples of classical chemical research, but rather were samples more typical of industrial applications, materials science and biomedical research.This versatility of FTIR can be seen to arise from the high sensitivity of the technique, which allows the use of a range of sampling methods suitable for different sample types. Such sampling methods include FTIR microscopy, FTIR Fig. 10 Plot showing relative rate of absorption by the skin of the two active components (A) OPM and (B) BPM. Fig. 11 FTIR spectra of contaminated polymer chips. FTIR microscopy reflectance spectra of (A) grey bulk polymer material and (B) black contaminating material.FTIR photoacoustic spectra of (C) grey bulk polymer material and (D) black contaminating material. 576 Analyst, April 1998, Vol. 123emission, attenuated total reflectance, photoacoustic spectrometry, diffuse reflectance and specular reflectance. In order to utilise the full power of the FTIR spectrometer, the infrared laboratory should be equipped with as many of these sampling methods as possible.Chemometric methods, such as multivariate analysis, have been shown to be extremely powerful for extracting information from spectra. It was possible to discriminate between the FTIR microscopy spectra of human hair from people of different races by the application of chemometric methods, despite the fact that, to the eye, the spectra could not be distinguished. References 1 Griffiths, P. R., and de Haseth, J., Fourier Transform Infrared Spectrometry, Wiley, New York, 1986. 2 Smith, B. C., Fundamentals of Fourier Transform Infrared Spectroscopy, CRC Press, Boca Raton, FL, 1996. 3 Nishikida, K., Nishio, E., and Hannah, W., Selected Applications of Modern FT-IR Techniques, Gordon and Breach, New York, 1996. 4 Johnston, S., Fourier Transform Infrared: a Constantly Evolving Technology, Ellis Horwood, Chichester, 1991. 5 Practical Fourier Transform Infrared Spectroscopy: Industrial and Laboratory Chemical Analysis, eds.Ferraro, J. R., and Krishnan, K., Academic Press, New York, 1990. 6 Vassallo, A. M., Cole-Clarke, P. A., Pang, L. S. K., and Palmisana, A. J., Appl. Spectrosc., 1992, 46, 73. 7 Messerschmidt, R. G., and Harthcock, M. A., Infrared Microspectroscopy. Theory and Applications, Marcel Dekker, New York, 1988. 8 Lewis, E. N., Kidder, L. H., Levin, I. W., Kleiman, V. D., and Hellwell, E. J., Opt. Lett., 1997, 22, 742. 9 Marcott, C., Story, G. M., Dowrey, A. E., Reeder, R.C., and Noda, I., Mikrochim. Acta (Suppl.), 1996, 14, 157. 10 Farmer, V. C., The Infrared Spectra of Minerals, Mineralogical Society, London, 1974. 11 Jonas, K., and Solymar, K., Acta Chim. Acad. Sci. Hung., 1970, 66, 1. 12 Rintoul, L., and Fredericks, P. M., Appl. Spectrosc., 1995, 49, 1608. 13 Wentrup-Byrne, E., Rintoul, L., Smith, J. L., and Fredericks, P. M., Appl. Spectrosc., 1995, 49, 1028. 14 Massart, D. L., Vandeginste, M. G. M., Deming, S. N., Michotte, Y., and Kaufman, L., Chemometrics: A Textbook, Elsevier, Amsterdam, 1988. 15 Otto, M., Chemom. Intell. Lab. Syst., 1988, 4, 101. 16 Meglen, R. R., Chemom. Intell. Lab. Syst., 1988, 3, 17. 17 Wilmhurst, J. K., J. Chem. Phys., 1963, 39, 2545. 18 Kozlowski, T. R., Appl. Opt., 1968, 7, 795. 19 George, G. A., Celina, M., VassalloA. M., and Cole-Clark, P. A., Polym. Degrad. Stab., 1995, 48, 199. 20 Frost, R. L., and Vassallo, A. M., Clays Clay Miner., 1996, 44, 635. 21 Harrick, N. J., Internal Reflection Spectroscopy, Wiley-Interscience, New York, 1967. 22 Puttnam, N. A., J. Soc. Cosmet. Chem., 1972, 23, 209. 23 Baier, R. E., J. Soc. Cosmet. Chem., 1978, 29, 283. 24 Compton, D. A. C., Hill, S. L., Wright, N. A., Druy, M. A., Piche, J., Stevenson, W. A., and Vidrine, D. W., Appl. Spectrosc., 1988, 42, 972. 25 Graham, J. A., Grim, W. M., and Fateley, W. G., in Fourier Transform Infrared Spectroscopy, ed. Ferraro, J. R., and Basile, L. J., Academic Press, New York, 1985, vol. 4, pp. 345–392. 26 Vidrine, D. W., in Fourier Transform Infrared Spectroscopy, ed., Ferraro, J. R., and Basile, L. J. Academic Press, New York, 1982, vol. 3, pp. 125–148. 27 Yang, C. Q., Bresee, R. R., and Fateley, W. G., Appl. Spectrosc., 1987, 41, 889. 28 Carter, E. A., Fredericks, P. M., and Church, J. S., Text. Res. J., 1996, 66, 787. 29 Dittmar, R. M., Chao, J. L., and Palmer, R. A., Appl. Spectrosc., 1991, 45, 1104. 30 Sowa, M. G., and Mantsch, H. H., Appl. Spectrosc., 1994, 48, 316. 31 Bertrand, L., Appl. Spectrosc., 1988, 42, 134. Paper 7/07111F Received October 1, 1997 Accepted December 9, 1997 Analyst, April 1998, Vol. 123 577
ISSN:0003-2654
DOI:10.1039/a707111f
出版商:RSC
年代:1998
数据来源: RSC
|
8. |
Fourier transform infrared microscopy: some advances in techniques for characterisation and structure–property elucidations of industrial material† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 579-586
John M. Chalmers,
Preview
|
PDF (407KB)
|
|
摘要:
Fourier transform infrared microscopy: some advances in techniques for characterisation and structure–property elucidations of industrial material† John M. Chalmers*a, Neil J. Everalla, Karen Hewitson‡a, Michael A. Chestersb, Martin Pearsonb, Andrew Gradyc and Barbara Ruzickad a ICI Technology, Wilton Research Centre, P.O. Box 90, Wilton, Middlesbrough, Cleveland, UK TS90 8JE b Department of Chemistry, University Park, University of Nottingham, Nottingham, UK NG7 2RD c Bio-Rad Laboratories Ltd., Bio-Rad House, Marylands Avenue, Hemel Hempstead, Hertfordshire, UK HP2 7TD d CCLRC Daresbury Laboratory, Daresbury, Warrington, Cheshire, UK WA4 4AD FTIR-microscopy has become one of the foremost vibrational spectroscopy techniques for problem-solving and analysing and mapping the chemical structure and physical characteristics associated with industrial materials and their fabricated products.Many recent advances have utilised the attributes of reflection techniques, such as specular reflection spectroscopy approaches, while emerging capabilities becoming available to the industrial spectroscopist include both spectral imaging and use of synchrotron radiation as a source.This paper seeks to illustrate each of these recent advances and developments through applications of FTIR-microscopy to industrial problem-solving case studies. Keywords: Fourier transform infrared-microscopy; specular reflectance; infrared imaging; synchrotron radiation; polymers Transmission and reflection FTIR-microsopy techniques are well established tools within many industrial laboratories.1 High spatial resolution (ca.!10 mm) measurements of infrared spectra have found particular favour in studies and analyses related to polymers, their fabricated products and commerical articles made from them.2,3 They are used routinely in fingerprinting contaminants, characterising the components of multilayer or multi-phase systems, and point-by-point profiling chemical property gradients or mapping physical property anisotropy.Consequently, advances or new approaches, which extend the range of applicability, enhance the information content of an analysis, and/or make more efficient the measurement process are important to the continued success and problem-solving developments of FTIR-microscopy. In this paper we aim, through three case/evaluation studies involving polymeric products, to demonstrate the use of specular reflectance measurements for a quantitative compositional determination, illustrate potential advantages of using radiation from a synchrotron as the infrared source, and show the potential of infrared imaging.Specular reflectance FTIR-microscopy front surface specular reflectance measurements have been shown to be particularly convenient for polymer characterisations.4 They do however require that the sample is ‘optically thick’, the sample surface is clean from contamination, and that the area examined is essentially optically flat. Also, since it is the absorption index spectrum which is usually analysed, generally if high signal-to-noise ratio (SNR), reasonable information content spectra are required, then the technique is best suited to more highly absorbing materials, such as aromatic polymers.4 However, the measurements have the advantages that they are non contact and require no sample preparation, and hence sample property integrity is maintained, which is imperative if physical property characteristics, such as crystallinity, are to be determined.A pure specular reflectance spectrum will exhibit the overlay of dispersion in the refractive index with the absorption index profile; these can be separated by applying the Kramers–Kronig (KK) transform to the recorded specular reflectance spectrum. 5. Reported specular reflectance studies have included work on uniaxially drawn poly(ethylene terephthalate),6,7 drawn liquid crystalline polymers,8–10 polyurethane cure,5,11 carbon-filled polymers,2,4,5 and methyl methacrylate copolymers.4 Here, we report the quantitative determination of the percentage crystallinity, to a reasonable precision, of injection-moulded poly(aryl ether ether ketone) (PEEK) samples.(Some of these data have been previously reported briefly by us elsewhere.4) Synchrotron infrared radiation FTIR-microscopy In a synchrotron, electrons move at relativistic velocities in a circular path, emitting electro-magnetic radiation.The properties of infrared (and other) radiation emanating from electron storage rings are well published.12–16 Synchrotron infrared radiation is highly collimated. It has very high brilliance and a much smaller source size, compared with a conventional source, and has very low thermal noise. It is also strongly linearly polarised in the plane of the ring, is pulsed at very high frequency, and has a very wide spectral distribution. The combination of low divergence, small source size and brightness have been utilised to good effect in recent years to extend the capabilities of FTIR-microscopy.13,17–22 These features couple to improve the spatial resolution and increase the SNR of low �etendue measurements.In conventional infrared spectroscopy applications, where �etendue is not limited to a particularly low value, the cross-over point below which the synchrotron source delivers more light than a conventional source is in the far-infrared (ca.l ~ 100 mm, where l is the infrared wavelength); for reflection–absorption infrared spectroscopy (RAIRS) experiments, where �etendue is of the order of 1022 mm2 sr, it occurs at ca. l = 10 mm (1000 cm21); for an infrared microscope system at high spatial resolution, with an �etendue of ca. 1024 mm2 sr, the cross-over point lies toward the † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.‡ Present address: Department of Chemistry, Sheffield-Hallam University, City Campus, Pond Street, Sheffield, UK S1 1WB. Analyst, April 1998, Vol. 123 (579–586) 579high wavenumber end of the mid-infrared, at ca. 1 mm, i.e., the synchrotron source should deliver higher SNR across the whole mid-infrared region for infrared-microscopy studies at the highest spatial resolutions.12 The spatial resolution cross-over at which a synchrotron source offers significant advantage over a conventional source, such as a Globar, for mid-infrared FTIRmicroscopy applications occurs at ca.a 40 mm apertured sample size. Reported synchrotron mid-infrared FTIR-microscopy studies have included examples of polymer laminate layer characterisations,17,18,21 single 2–12 mm polymer particles,19 and single cell interrogations.22 Here, we report some preliminary FTIR-microscopy spectra recorded from multi-layer food packaging films which were recorded on the recently commissioned system at the synchrotron facility at Daresbury, UK, for which the beam size is 40 mm 3 12 mm.The FTIR-microscope system, although added to an already existing beamline optimised for RAIRS studies in the far-infrared,12 is still shown to enhance significantly the spatial resolution for mid-infrared FTIR-microscopy work. The spectra recorded are compared with examinations made in a similar manner on a FTIR-microscope fitted with a conventional Globar source. FTIR-microscopy imaging Mid-infrared micro-spectroscopic imaging is a rapidly emerging and developing technology, which has recently been commercialised.24 In current systems, the conventional single channel multiplex detector (e.g., MCT) of a FTIR-microscope is essentially replaced by a focal plane array (FPA) detector, thereby affording the dual instrumental advantage of both multichannel and multiplex to a measurement.The former clearly brings large time-saving benefits over functional group or chemical species mapping, which have until now been more conventionally carried out in the mid-infrared region by rasterscanning or point-by-point mapping a sample.25–27 Infrared FPAs, or staring arrays, were developed originally for military and surveillance purposes, and have only recently become commercially accessible.25–27 The firstcroscopy publication in 1995 used a liquid-nitrogen cooled InSb detector (128 3128 pixels), limited to 5000–2500 cm21,25,26 in 1996 use of a liquid-helium cooled Si:As camera (64 3 16 pixels were used), with an operating spectral range out to 400 cm21, was reported;27 while, in 1997 work with both narrowband ( > 900 cm21; 64 3 64 pixels), and wide-band ( > 400 cm21) liquid-nitrogen cooled MCT detectors have been presented. 28,29 Presently, the commercial system offers both the InSb (10 000–2000 cm21) and narrow-band MCT (5000–1000 cm21) FPA cameras as options.24 In FTIR-microscopy imaging, the spectrometer is operated in step-scan mode.SNR may be improved by, either or both, the integration time at each step or the number of co-added and averaged frames taken at each step. Typical integration times for a single image plane are of the order of a few milliseconds or less, and typically one might expect to co-add up to 100 images per step, which for, say, a 64 3 64 array would produce 4096 8 cm21 spectra and an image plane at every 4 cm21 in ca. 17 min. This time-efficient collected wealth of information should be compared to that from a conventional point-by-point mapping measurement, which typically may take several hours for a much smaller map. Application areas reported for imaging studies include work on chemical imaging of a surfactant dispersion in water,25 monkey brain tissue,26 silicone leakage from artificial implant in female breast tissue,30 a packaging film laminate,28 and a biomineralised tissue (bone).28 Here, we report an attempt to discriminate between two polyethylenes, which differ essentially only in their methyl group concentration, and which form the components of a co-extruded tube.Experimental Specular reflectance FTIR-microscopy measurements The FTIR specular reflectance spectra were recorded using a Nicolet (Warwick, UK) 850 FTIR spectrometer interfaced to a Nic-Plan IR microscope. The microscope, which was fitted with a Nicolet Instruments narrow-band liquid nitrogen cooled MCT detector, was operated in the reflectance mode.The single-beam reflectance background was recorded under identical conditions from a gold coated disk. For all specular reflectance measurements the incident infrared beam was focused at the sample surface, using a typical aperture size of 360 mm 3 360 mm. The specular reflectance spectra were recorded at 4 cm21 resolution, 100 scans per spectrum. Replicate (triplicate) measurements were made from different positions on the surface of each sample.The absorption index spectra were then derived from the recorded specular reflectance spectrum by application of the Kramers–Kronig transform algorithm, available as the ‘Dispersion Correction’ option within the Nicolet OMNIC software supplied with the FTIR spectrometer. The KK transform assumes normal incidence. In practice, in FTIR-microscopy reflectance mesurements, radiation is incident and collected at near normal incidence (ca. 30°). The microscope objective favours collection of specularly reflected radiation over any diffusely scattered light, which is inherently spread over a wider angular range, and clearly demands less of an optically flat surface area for measurement than that required for a macromeasurement. The PEEK samples were plaques ca. 0.6 mm thick. Their preparation and the determination of their crystallinity by wide angle X-ray scattering (WAXS) has been reported elsewhere.31 The method is designed to work for pure specular reflectance spectra and therefore the sample must have essentially a surface which is optically flat over the area interrogated, i.e., no (miminal) diffuse scattered radiation should be present; the sample must be ‘optically thick’, i.e., no back surface reflected radiation should be detectable, and the sample should be ‘optically homogeneous’ through its depth, see Fig. 1. In practice, for PEEK, this means that samples need to be ca. 100 mm thick or greater, although thinner films filled with carbon black are often acceptable, since the carbon effectively absorbs all radiation which penetrates the bulk of the sample.Synchrotron FTIR-microscopy system The FTIR-microscope set-up has been fitted as an adjunct to a synchrotron radiation source already utilised for far-infrared studies, in particular RAIRS work, for which the system was optimised.12 The beamline at the Daresbury (SRS) in the UK Fig. 1 Schematic showing the generation of a specular reflectance spectrum from a sample surface.Only the ticked (]) radiation should be collected, rays marked with a cross (X) should not be generated or minimised, i.e., the sample should be ‘optically thick’, such that no ‘transflectance’ radiation is generated or collected, and the surface smooth and flat such that no or minimal diffuse reflectance radiation is produced. 580 Analyst, April 1998, Vol. 123O O C O n emerges from a storage ring of current 250 mA at 2 GeV.A schematic of the installation is shown in Fig. 2. The microscope is a Nicolet Nic-Plan IR microscope fitted with a 250 mm 3 250 mm MCT-A liquid nitrogen cooled detector. This is interfaced to a Nicolet 730 mid-IR FTIR spectrometer. For all the infrared measurements reported here, a 153 Reflechromat objective was used. Samples were contained within a Spectra-Tech diamond window compression cell. All the spectra were recorded at 4 cm21 resolution, for which 128 scans were co-added, except the cheese packaging film layer sandwiched between the poly(propylene) (PP) and ethylene–vinyl acetate (EVA) layers, for which 1024 scans were co-added.Single-beam background spectra were recorded under similar conditions through a KBr particle compressed in the diamond compression cell, alongside the contained sample being interrogated. For the Globar comparative studies, measurements were made on a Nicolet 850 FTIR spectrometer interfaced to a Nicolet Nic-Plan FTIR-microscope fitted with a 250 mm 3 250 mm MCT-A detector.All spectra were recorded at 4 cm21 resolution, 500 scans, from samples contained within a Spectra- Tech diamond window compression cell. Again a 153 Reflechromat objective having a 0.58 numerical aperture (NA) was used. Single-beam background spectra were recorded under similar conditions through a KBr particle compressed in the diamond compression cell, alongside the contained sample being interrogated.The samples examined were approximately 10 mm thick cross-sections microtomed from the packaging films, using a Leitz Model 1400 microtome (Germany) fitted with a Leica ‘c’ profile steel blade. FTIR-microscopy imaging The infrared imaging measurements were made on a Bio-Rad FTS Stingray 6000 (Bio-Rad, Digilab Division, Cambridge, MA, USA). The Stingray system comprised the Bio-Rad FTS 6000 step-scan FTIR spectrometer, the Bio-Rad UMA 500 infrared microscope which was equipped with an FPA detector and specialised imaging software.The detector was a 64 3 64 MCT FPA detector, i.e., 4096 active pixels, with a size of approximately 7 mm 3 7 mm per pixel, with an operating range from ca. 5000–1000 cm21. A 50 mm ZnSe formation lens directed the infrared radiation onto the FPA, see Fig. 3. The image area displayed was about 450 mm 3 450 mm. Doublesided step-scan interferograms equivalent to 4 cm21 resolution were recorded, from the sample supported on a ZnSe window.One hundred frames were co-added at each step of the interferometer, at a step-rate of 1 Hz (1 step s21); the total time for a scan was approximately 35 min. The sample examined was an approximately 20 mm thick cross-section microtomed from a bi-layer co-extruded polyethylene tube of external diameter 3 mm and internal diameter about 2.4 mm. The section was prepared with a Leitz Model 1400 microtome equipped with Leica ‘c’ profile steel blade.Results and discussion Specular reflectance FTIR-microscopy, crystallinity of PEEK PEEK is a high temperature engineering thermoplastic. It is a semicrystalline polymer with the structure, In a previous study,32 the crystallinities determined by WAXS from a series of specially prepared PEEK plaques were correlated with absorbance ratio measurements from multiple internal reflectance (MIR) spectra made on the samples. This macro surface sampling method was developed about the time that modern FTIR-microscope systems were just beginning to be commercialised, and certainly several years before deriving absorption index spectra from specular reflectance measurements by FTIR-microscopy became routinely practicable. Recently, we had a requirement to develop a simple, reasonably precise method by which the surface crystallinity of injection-moulded tensile bars could be assessed and mapped.Although FTIR–attenuated total reflectance (ATR) microscopy is now a common-place technique, it was not appropriate, nor practicable for the required measurements.(Like MIR, ATR is an internal reflection spectroscopy method.) The FTIR–ATR microscopy technique is a single reflection method, hence does not provide the absorption intensity and SNR achievable with a multiple internal reflection approach. Secondly, reproducible contact sampling between the ATR objective internal reflection element and the hard surface of the PEEK tensile bars proved to be impractical to achieve; also, since localised high contact pressures are involved, there existed the possibility of pressureinduced crystallisation occurring, a phenomenon we have noted to happen while sampling certain PEEK samples using a diamond compression cell.33 Consequently, we derived a FTIRmicroscopy specular reflectance approach, preliminary results of which have been briefly reported elsewhere.4 Fig. 2 Schematic showing the optical layout and path of the synchrotron infrared radiation beamline to the FTIR-microscopy system.Fig. 3 Schematic showing Bio-Rad FTS Stingray 6000 mid-infrared imaging FTIR-microscopy system. Analyst, April 1998, Vol. 123 581Fig. 4 illustrates application of a KK algorithm to a recorded specular reflectance spectrum. Pure specular reflectance spectra of organic polymers, because of dispersion in the refractive index associated with absorption features, yield ‘first-derivative- like’ infrared spectra, see Fig. 4. When a specular reflectance spectrum is subjected to the KK transform it may be separated into the refractive index dispersion and, the more analytically useful, absorption index components,5 as shown schematically in Fig. 4. Fig. 5 compares a set of derived absorption index spectra generated from applying the KK transform to a set of FTIR-microscopy specular reflectance recorded from the series of standard PEEK plaque samples. Fig. 6 shows the linear least-squares correlation plot between the Fig. 4 Schematic illustrating the generation of the absorption index spectrum by applying the Kramers–Kronig transform algorithm to the recorded specular reflectance spectrum from a PEEK moulding. Fig. 5 Absorption index spectra over the range 1400–900 cm21 for a series of PEEK plaques of differing crystallinities. The spectra were generated from a Kramers–Kronig transform of the specular reflectance spectra. Spectra have been offset for clarity. The peak heights were quantified relative to the dashed baseline, as illustrated on the bottom spectrum.Fig. 6 Plot of absorption index band ratio 1305 cm21/1280 cm21 against WAXS determined percentage crystallinity for a series of PEEK plaques. Fig. 7 Synchrotron infrared radiation FTIR-microscopy spectra recorded from the three consecutive layers of a sweet packaging film laminate section. A photograph of the section is displayed as an inset. Fig. 8 Comparison of the best ‘middle layer’ FTIR-microscopy spectra recorded from the sweet packaging film laminate section using both synchrotron and Globar source FTIR systems.Fig. 9 Synchrotron infrared radiation FTIR-microscopy spectra recorded from a four layer cheese packaging film laminate section. A photograph of the section is shown above the spectra, alongside a scale index, for which each small division represents 10 mm. Fig. 10 Comparison of attempts to obtain spectra of the cheese packaging film laminate section layer sandwiched between the PP and EVA layers using both synchrotron and Globar source FTIR systems. 582 Analyst, April 1998, Vol. 123Fig. 11 FTIR-microscopy spectra recorded from a cross-section microtomed slice, ~ 50 mm, (shown in schematic form) from a two layer coextruded polyethylene tube. Fig. 12 Stingray FTIR system broad band infrared image of the polyethylene tube section. Fig. 15 Infrared image at 1378 cm21 of the polyethylene tube section, and the infrared spectrum of the inner polyethylene layer corresponding to the pixel at cursor position. Fig. 13 Infrared image at 1304 cm21 of the polyethylene tube section. Fig. 14 Infrared image at 1378 cm21 of the polyethylene tube section. Analyst, April 1998, Vol. 123 583absorption index band intensity ratio 1305 cm21/1280 cm21 and the WAXS determined sample percentage crystallinity. From ten replicate measurements of the ca. 22% crystalline sample, a measurement error of ca. ±1.5% absolute crystallinity was determined; this is significantly better than that quoted for univariate and multivariate calibration of Raman data,34 for which the lowest error quoted for the isotropic plaque samples was ca. 3.2%. Synchrotron FTIR-microscopy, polymer film laminates For part of our preliminary evaluation of the Daresbury synchrotron source FTIR-microscope system, we chose to examine sections microtomed from two multi-layered packaging materials. As with the polyethylene tube sections prepared for our image studies, see below, no special care was taken to microtome an optimal section.The sections were examined, for convenience, contained within a diamond window compression Fig. 16 Infrared image at 1378 cm21 of the polyethylene tube section, and the infrared spectrum of the outer polyethylene layer corresponding to the pixel at cursor position. Fig. 17 Infrared image at 1622 cm21 of the polyethylene tube section, and the infrared spectrum from a pixel at contaminant position (marked by the cursor), and the infrared spectrum from a pixel away from contaminant position. 584 Analyst, April 1998, Vol. 123cell; although no or minimal compression was applied, to avoid layer spreading. Similar sections were also examined in a similar manner on a similar FTIR-microscope system equipped with a conventional mid-infrared Globar source, which was sited in the industrial laboratory of the ICI authors. For both the synchrotron and Globar source FTIR-microscope examinations only a 153 Cassegranian objective was used.Fig. 7 shows the infrared absorbance spectra recorded from the three layers evident in the sweet packaging film laminate section, shown in the inset to Fig. 7. The aperture width used for examining each layer was the maximum which could be imaged onto the layer; for the middle layer, this was of the order of 8–10 mm, for the outer layers this was of the order of 30–40 mm. For the synchrotron data, shown in Fig. 7, the middle layer absorption features are clearly separated from those of the two surrounding PP layers. The middle layer infrared spectrum is characteristic of a vinyl acetate co-ethylene polymer. Fig. 8 compares, under the conditions used, the best ‘middle layer’ spectra that could be achieved with the synchrotron and Globar source systems. In the case of the Globar derived spectrum, the spectrum of the vinyl acetate polymer based layer could not be cleanly separated from the PP.[Internal reflection spectra (MIR, Ge prism, 45° incidence angle) of the two laminate air contact surfaces showed that each of the PP layers was coated with a vinylidene chloride–acrylonitrile–ester polymer, which were however much too thin for examination by FTIR-microscopy.] In the second multi-layer laminate examined, which was a cheese packaging film, four layers were readily observable in the microtomed section. The two widest layers, see Fig. 9, were readily separated in the synchrotron FTIR-microscope measurement and identified as a PP layer and an EVA copolymer layer.Sandwiched between these two layers was a thinner layer, the spectrum of which could not (under the conditions used) be separated completely from absorption features from its surrounding layers. Yet there is sufficient detail in the spectrum to suggest the layer is based on a nitrogen containing resin, e.g., the absorption features near 1640, 1547, and 1508 cm21.Internal reflection spectra (MIR, Ge prism, 45° incidence angle) of the two laminate air contact surfaces showed the inside surface to be an EVA copolymer, while the outside surface was a vinylidene chloride–acrylonitrile–ester polymer. The top spectrum of Fig. 9, clearly shows overlaid on the PP spectrum a doublet at 1066 and 1042 cm21, characteristic of vinylidene chloride (VdCl) based polymers. The apertures used in recording the synchrotron FTIR-microscopy measurements were: 15 mm 3 240 mm, VdCl + PP ‘spectrum’; 27 mm 3 300 mm, PP layer; 9 mm 3 30 mm, ‘probable nitrogen containing resin layer’; 36 mm 3 300 mm, EVA layer.Fig. 10 compares the attempt to obtain a synchrotron FTIR-microscope spectrum of the layer sandwiched between the PP and EVA layers, with the Globar source FTIR microscope best attempt equivalent. While the layer spectrum discrimination in the fingerprint region below 2000 cm21 is much poorer for the Globar case, demonstrating the superior spatial resolution achievable with the synchrotron, it is evident, by comparison, that in our suboptimal set-up the synchrotron loses intensity rapidly towards the high wavenumber region of the mid-infrared spectrum, and is well down on the theoretical SNR gain expected.As stated previously, the optical interface system is optimised at 100 cm21 (for the RAIRS work), and it is known that at least onethird of the radiation is being lost due to a beam stop in place to remove X-rays.An improved interim optical interface system is being planned, prior to the eventual commissioning of a dedicated mid-infrared FTIR-microscopy beam-line. FTIR-microscopy imaging, co-extruded polyethylene tube In this particular study, we explored the discriminating potential of infrared imaging to separate similar materials, which differed to a limited extent essentially only in their concentration of a particular chemical moiety. The example selected was a coextruded tube comprising two polyethylenes, which, within the spectral operating range of the focal plane array detector, could essentially only be distinguished by the relative intensity of their methyl deformation mode at 1378 cm21, see Fig. 11. The co-extruded tube had an external diameter of 3 mm, with a wall thickness of 0.6 mm, made up of equal thicknesses of the two polyethylene layers. The inner layer material was a low methyl concentration, estimated at ca. 6 –CH3 groups per 1000 C atoms, polyethylene type, whereas, the outer layer was made from a linear low density polyethylene (LLDPE) type, in which the olefine co-monomer modifier was identified as butene, to give an estimated methyl group concentration of ca. 25 –CH3 groups per 1000 C atoms. The (user-definable, colour-scale) broad band infrared image of a microtomed cross-section of this co-extruded tube recorded using the Stringray FTIR-microscope system is shown in Fig. 12. The edges of the tube are clearly visible, due to their poor optical contrast, as a consequence of significant edge flow/ distortion occurring during the microtoming process.No special attention was given to the preparation of the microtomed crosssection; it was prepared by a limited experienced user on a routinely used laboratory microtome, which was equipped only with a steel blade. The ‘knife-edge’ flow lines are readily discernible in the infrared (false colour-scale flat view) image taken at 1304 cm21, see Fig. 13. Fig. 14 shows the infrared image based on the infrared absorbance intensity at 1378 cm21. There is a distinct boundary discernible (in the middle of the image) between the two coextruded polyethylene layers, even though this is only based on a difference in methyl group concentration of about 20 per 1000 C atoms, which in the section examined represents an absorbance difference of significantly less than 0.1. Within each layer image the colour is not uniform, because of the less than optimal microtoming of the section. When displayed as colour coded images all the intensity to the right of the boundary line ranged from yellow to red, which was in sharp contrast to the lower methyl material to the left of the boundary, where the colour ranged from dark green to light blue.(In the colour scale coding used, red represented more intense absorbance). Figs. 15 and 16 each show the same image as Fig. 14, but overlaid on each is a crossed dotted lines cursor.Accompanying each of the images in Figs. 15 and 16 is an absorbance spectrum taken from the pixel at which the cursor intersection was placed. While, superimposed on both of these spectra there is clearly an interference pattern, the spectra are readily distinguished by their relative absorbance intensities at 1378 cm21. In the course of the experimental study it was noted that in some images a high intensity (red colour centred) very localised area was discernible, see the colour-scale image of Fig. 17. This image is based on the absorbance at 1622 cm21, and the spectrum associated with a pixel at the centre of this area is shown beneath the image. The broad absorption feature near 1622 cm21 and that extending from about 1150 cm21 to lower wavenumbers suggest an inorganic surface contaminant. In the colour image the localised red area is surrounded by a narrow yellow ring, then a wider band of green, which was all contained well within an area of 10 3 10 pixels; the remainder of the image is blue in colour.A spectrum associated with a pixel that was in this blue region showed only the presence of polyethylene, see Fig. 17. Conclusions The three application examples reported here all demonstrate the high potential for a continuing expanding and developing field of industrial applications for FTIR-microscopy techniques. Analyst, April 1998, Vol. 123 585Properly conducted specular reflectance measurements are clearly capable, in the right circumstances, of providing precise quantitative methods.In a comparable study with a conventional source system, the spatial resolution advantages of using synchrotron derived infrared radiation as the source for FTIRmicroscopy measurements have been demonstrated for the Daresbury installation. Good infrared image discrimination has been demonstrated for two generically similar polymers; an important attribute for future property gradient and material anisotropy studies on polymeric articles.For both the synchrotron and imaging work reported here, higher quality results would have undoubtedly been achieved if a 323 objective (NA 0.65) had been used, and if more attention had been paid to the microtoming process, by using a glass or diamond blade. Good sectioning will be imperative for high contrast imaging of chemically similar materials. The Daresbury FTIR-microscopy set-up would benefit significantly from being on a dedicated beam-line.Notwithstanding, use of synchrotron infrared radiation as a source for low �etendue FTIR-microscopy measurements for industrial problem-solving and research will expand, and specific uses of the inherently polarised characteristic of the beam have yet to be explored. (The initial Brookhaven dedicated beam-line is now interfaced to a Spectra-Tech IRms scanning FTIR microprobe spectrometer. The first microscope facility at Brookhaven was built in 1993,20 two more are already under construction.) Industrial applications of imaging are set to grow rapidly, particularly as wide spectral range FPAs have now become available commercially at a reasonable cost.Studies of multi-phase and multilayer commercial products will all benefit from these two important new instrumental advances. Stephen Ellison, Sheffield-Hallam University, UK, was responsible for recording much of the data relating to the specular reflectance measurements on PEEK.We also wish to acknowledge the advice and help given by Mark Surman, Daresbury Laboratories, towardshe synchrotron studies. We would like to thank both the CCLRC Daresbury Laboratory and the EPSRC for providing support for the construction of the infrared beamline and this work, and for providing a research assistantship (B.R.) and CASE studentship (M.P.). References 1 Katon, J. E., Vib. Spectrosc., 1994, 7(3), 201. 2 Chalmers, J. M., Croot, L., Eaves, J.G., Everall, N., Gaskin, W. F., Lumsdon, J., and Moore, N., Spectrosc. Int. J., 1990, 8, 13. 3 Chalmers, J. M., and Everall, N. J., Macromol. Symp., 1995, 94, 33. 4 Chalmers, J. M., Everall, N. J., and Ellison, S., Micron, 1996, 27, 315. 5 Clayborn, M., Colombel, P., and Chalmers, J., Appl. Spectrosc., 1991, 452, 279. 6 Cole, K. C., Guevremont, J., Ajji, A., and Dumoulin, M. M., Appl. Spectrosc., 1994, 48, 12, 1513. 7 Everall, N. J., Chalmers, J. M., Local, A., and Allen, S., Vib.Spectrosc., 1996, 10, 253. 8 Kaito, A., and Nakayama, K., Macromol., 1992, 25, 4882. 9 Bensaad, S., Jasse, B., and Noel, C., Polymer, 1993, 348, 1601. 10 Jansen, J. A., Paridaans, F. N., and Heynderickx, I. E. J., Polymer, 1994, 3514, 2970. 11 Claybourn, M., in Proceedings of the 9th International Conference on Fourier Transform Spectroscopy, SPIE Vol. 2089, 1993, p. 180. 12 Slater, D. A., Hollins, P., Chesters, M. A., Pritchard, J., Martin, D. A., Surman, M., Shaw, D.A., and Munro, I. H., Rev. Sci. Instrum., 1992, 63, 1, 1547. 13 Ugawa, A., Ishii, H., Yakushi, K., Okamoto, H., Mitani, T., Watanabe, M. Sakai, K., Suzui, K., and Kato, S., Rev. Sci. Instrum., 1992, 63(1), 1551. 14 Yarwood, J., Shuttleworth, T., Hasted, J. B., and Nanba, T., Nature, 1984, 312(20/27), 742. 15 Williams, G., Rev. Sci. Instrum., 1992, 63, 1535. 16 Duncan, W. D., and Williams, G. P., Appl. Optics, 1983, 22(18), 2914. 17 Reffner, J. A., Martoglio, P.A., and Williams, G. P., in Microbeam Analysis, Proceedings of the 28th Annual MAS Meeting, New Orleans, July 31–August 5, 1994, p. 79. 18 Ellis, L. E., Platek, S. F., and Satzger, R. D., in Microbeam Analysis, Proceedings of the 28th Annual MAS Meeting, New Orleans, July 31–August 5, 1994, p. 81. 19 Ward, K. J., in Microbeam Analysis, Proceedings of the 28th Annual MAS Meeting, New Orleans, July 31–August 5, 1994, p. 83. 20 Reffner, J. A., Carr, G. L., and Williams, G. P., in Microbeam Analysis, Proceedings of the 29th Annual MAS Meeting, Breckenridge, August 6–11, 1995, p. 113. 21 Reffer, J. A., Carr, G. L., and Williams, G. P., in Progress in Fourier Transform Spectroscopy, Proceedings of the 10th International Conference, Budapest, August 27–September 1, 1995; Mikrochimica Acta [Suppl. 14], 1997, 339. 22 Wetzel, D. L., Reffner, J. A., and Williams, G. P., in Progress in Fourier Transform Spectroscopy, Proceedings of the 10th International Conference, Budapest, August 27–September 1, 1995; Mikrochimica Acta [Suppl. 14], 1997, 353. 23 Reffner, J. A., private communication. 24 Bio-Rad Stingray 6000 FTIR-microscope system, Bio-Rad Laboratories, Digilab Division, Cambridge, MA, USA, 1997. 25 Lewis, E. N., Treado, P. J., Reeder, R. C., Story, G. M., Dowrey, A. E., Marcott, C., and Levin, I. W., Anal. Chem., 1995, 67, 3377. 26 Lewis, E. N., Gorbach, A. M., Marcott, C., and Levin, I. W., Appl. Spectrosc., 1996, 50, 263. 27 Lewis, E. N., Kidder, L. H., Arens, J. F., Peck, M. C., and Levin, I. W., Appl. Spectrosc., 1997, 51, 563. 28 Marcott, C., and Reeder, R. C., Poster M44 at 11th International Conference on Fourier Transform, August 10–15, 1997, Athens, Georgia, USA. 29 Lewis, E. N., Kidder, L. H., and Levin, I. W., Plenary Lecture PL13 at 11th International Conference on Fourier Transform, August 10–15, 1997, Athens, Georgia, USA. 30 Kidder, L. H., Kalasinsky, V. F., Luke, J. L., Levin, I. W., and Lewis, E. N., Nature Medicine, 1997, 3, 235. 31 Blundell, D. J., and Osborn, B. N., Polymer, 1983, 24, 953. 32 Chalmers, J. M., Gaskin, W. F., and Mackenzie, M. W., Polym. Bull., 1984, 11, 433. 33 Chalmers, J. M., and Dent, G., Industrial Analysis with Vibrational Spectroscopy, The Royal Society of Chemistry, Cambridge, UK, 1997. 34 Everall, N. J., Chalmers, J. M., Ferwerda, R., van der Maas, J. H., and Hendra, P. J., J. Raman Spectrosc., 1994, 25, 43. Paper 7/07070E Received September 30, 1997 Accepted December 9, 1997 586 Analyst, April 1998, Vol. 123
ISSN:0003-2654
DOI:10.1039/a707070e
出版商:RSC
年代:1998
数据来源: RSC
|
9. |
Photothermal spectrometry for membrane and interfacial region studies† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 587-593
R. D. Snook,
Preview
|
PDF (102KB)
|
|
摘要:
Photothermal spectrometry for membrane and interfacial region studies† R. D. Snook*, R. D. Lowe and M. L. Baesso‡ Department of Instrumentation and Analytical Science, University of Manchester Institute of Science and Technology, PO Box 88, Manchester UK M60 1QD In this paper the application of photothermal techniques to membrane and interfacial studies is described. Thus the techniques of photothermal and concentration gradient induced beam deflection, photoacoustic spectrometry and thermal lens spectrometry are described as applied to the study of thin films, skin and diffusion through membranes.Techniques for the measurement of thermal diffusivity in thin polymer films and skin are also described. Their theoretical basis for application is discussed together with the limitations of each technique. Finally the future development of photothermal techniques for these applications is assessed. Keywords: Photothermal spectrometry; thermal lens spectrometry; beam deflection techniques; thin films; skin and membrane studies The importance of studying the properties of thin films, membranes and diffusion processes cannot be underestimated.Thus thin films and membranes are of crucial importance in many manufactured items in everyday life, e.g., food packaging and laminated structures as well as in scientific and technological applications, e.g., polymer and separation science, clinical science and of course medical science and its applications.Inherent in many of these scientific studies is the need to determine rates of diffusion into or through such films which in turn again requires measurements of physico- chemical properties of the film or membrane. For diffusion measurements it may also be necessary to make measurements in the interfacial region between the film or membrane and the contacting medium which may be in any of the gaseous, liquid or solid states. Although desirable it is often not possible to make interfacial measurements rather than in the bulk medium because of the requirement to probe, non-destructively and noninvasively, very small volumes.Confocal microscopy is useful for probing a few micrometres of the sample surface and has provided an extremely valuable tool for studying biological and clinical samples especially when combined with fluorescence or Raman spectrometry but for depths and film thicknesses of more than a few microns there are very few techniques other than the photothermal ones described here.It is for this reason that such techniques are going through somewhat of a renaissance following the success of photoacoustic spectrometry in the 1980s. A glance at any of the conference proceedings of the topical Meeting on Photoacoustic and Photothermal Phenomena Series1 shows the breadth of interest in electronic materials, non-destructive testing and spectroscopy of thin films in chemistry and biology and the use of photothermal spectrometries for their investigation.It is the aim of this paper to discuss some of the advances made in selected areas of photothermal spectrometry with special reference to the measurement of optothermal parameters in different materials and diffusion through thin films and membranes. In particular beam deflection techniques and thermal lens spectrometry (TLS) will be described although some reference will be made to step-scan Fourier transform photoacoustic spectrometry for depth profiling.Theoretical basis of photothermal spectrometries All photothermal techniques have a common origin in the use of electromagnetic radiation in the UV to infra-red region of the spectrum as probing radiation and in this respect are similar to conventional UV/VIS spectroscopies. Unlike these, however, the sample is not quantified by transmission or luminescence but rather by the non-radiative de-excitation of excited states which results in a change in temperature of the substrate which can be measured by its effect on the substrate or contacting medium.Often these techniques employ a laser to provide a modulated or pulsed excitation beam which is tuned to the absorption maximum of a chromophore and focussed into the sample. Subsequent non-radiative relaxation generates a thermal wave which can be detected in the sample or in the contacting medium using any of several different detection schemes summarised in Table 1.The use of laser excitation brings several advantages and indeed for techniques such as TLS the Gaussian intensity profile provided with a TMoo single mode laser is an essential † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997. ‡ Present Address: Laborat�orio de Foto�ermica-DFI, Universidade Estadual de Maring�a, Avenida Colombo, 5790-87020-900-Maring�a, PR-Brazil. Table 1 Detection techniques for photothermal spectrometries Technique Detection principle Photoacoustic spectrometry (PAS) using— (a) Microphone Pressure wave in a closed cell containing the sample, contacting gas and microphone.(b) Piezo or pyro-electric Attached sensor detects the sensor thermal wave in the sample by deformation or temperature rise. Thermal lens spectrometry (TLS) using— (a) Beam deflection Off-axis thermal gradient induced deflection of a probe beam parallel to pumping beam. (b) Co-axial probe beam Symmetric lens detected by a detection probe coaxial with the pump beam.Photothermal beam deflection (PBD)— (a) Thermally induced Beam deflection caused by a thermally induced refractive index gradient (dn/dT). (b) Concentration induced Beam deflection caused by a concentration induced refractive index gradient (dn/dC). Photothermal radiometry— (a) Laser induced optothermal IR thermal emission spectrometry Analyst, April 1998, Vol. 123 (587–593) 587requirement for the technique to work.Other general advantages which the use of laser excitation sources bring to photothermal spectrometry are: high power; coherence; directional and spatial stability; and the ability to probe small volumes of sample. These properties are especially important in enabling the techniques to be applied to membrane and interfacial studies. The generation of the photothermal signal can be considered in two parts: first the absorption of light to generate heat and second the diffusion of heat through the sample and adjacent or contacting medium to generate a thermal wave or thermal gradient which can be detected.Absorption of light and heat generation The simplest case is absorption by an homogeneous sample which is uniformly illuminated by a harmonically modulated beam as described by McDonald et al.2 Then the beam intensity at any depth will be given by: I x I x t ( ) exp( )( cos ) = + 0 2 1 b w (1) where Io is the intensity at the sample surface, i.e., x = 0; b is the optical absorption coefficient; and w = 2pf where f is the modulation frequency.The value of b may have to be modified in certain cases to account for scattering and other factors. The optical absorption length 1b is approximately equal to 1/b is and represents the depth in a highly absorbing sample at which all the incident radiation is absorbed. Ignoring the static term the heat produced (H) per unit volume will be: H I x i t = hb b w 0 2 exp( )exp( ) (2) where h is the fraction of absorbed power converted to heat.However the assumption of uniform irradiation is not always held when the pump beam size is smaller than the sample. Normally the beam is Gaussian in profile and Io can be replaced with I r I r R r ( ) exp = - Ê Ë Á � � � 0 2 2 2 (3) where r is the radial distance from the axis of a Gaussian beam of radius R. Diffusion of heat The diffusion of generated heat is described by the equations of Carlslaw and Jaeger:3 - Ñ + = k t r dt d 2 C t H h (4) where k is the thermal conductivity, r is the density and Ch is the specific heat.If the periodic time dependency of eqn. 2 is introduced then eqn. 4 becomes: -— + = 2r i D H w t k th (5) where Dth is the thermal diffusivity. Omitting the term H and assuming uniform illumination then a solution to this equation can be written as: t = Aexp(±qxere q i D = w th (7) Thus a thermal wave can be visualised whose amplitude decays by a factor of 1/e within a thermal diffusion length given by: m w = 2 Dth (8) This expression of the thermal diffusion length (m) is of special importance in the studies described in this paper as it defines the maximum depth from which a photothermal signal can be detected in a sample.It also shows that by changing the modulation frequency w = 2pf the thermal diffusion length can be changed which forms the theoretical basis of a number of photothermal methods for depth profiling including the recent innovation of step scan Fourier transform IR photoacoustic spectrometry as reported by Jiang and Palmer4 in which entire spectra in the IR region can be obtained at different fixed depths. The thermal wave, if not completely damped by the surface of the sample heat will flow into the contacting medium which is usually a gas.The resulting time-dependent temperature fluctuation creates a fluctuating refractive index gradient which if probed with a laser beam parallel with and close to the surface yields quantitative information about the sample via deflection of the probe beam. This is the basis of photothermal beam deflection techniques in which the relative probe beam deflection is measured after passing over the sample using either a position sensitive detector or CCD array.The other key parameter in photothermal spectrometry is the thermal diffusivity (Dth) and it is essential that accurate values of Dth are known in order to accurately determine thermal diffusion lengths.The relationship between the optical absorption length (1/b) and the thermal diffusion length (m) is also important in determining whether or not the sample is photothermally saturated. Ideally it should not be and to achieve this state where the signal depends upon the absorption coefficient of the sample the thermal diffusion length should be less than the optical absorption length. It is these principles which underpin the use of photothermal spectrometry for thin film and interfacial studies and in the rest of the paper a few examples of interest are discussed with reference to the measurement of thermal diffusivity and diffusion of chemicals through and into skin and membranes.Skin is of particular interest because it is the primary barrier to absorption into the body of many topically applied cosmetics and sun screens. The study of diffusion through skin is also important to determine penetration rates of topically applied drug treatments and prophylactics.Other chosen examples are diffusion through physical membranes and the properties of polymer films. Photothermal studies of skin Several papers have shown the potential of PAS for studying the penetration of topically applied substances through skin.5–7 The latter paper by Giese et al.7 for example described such a study of the penetration of sun screens into human skin in the visible region of the spectrum with some success.A major problem however is the need for an accurate value of Dth to be able to define the thermal diffusion length and relate this to the penetration depth. This problem becomes exacerbated with different regions of the skin, i.e., stratum corneum compared with the whole epidermis and there are few reliable values of Dth in this context. Brown et al. developed two simple photothermal methods for the accurate determination of Dth in stratum corneum.8 In the first of these two methods the technique of photothermal beam deflection was used to measure D using the method of Kuo9 in which the transverse beam deflection profile at different pump beam modulation frequencies was used to derive Dth.These 588 Analyst, April 1998, Vol. 123authors used an argon ion (515 nm) pump beam and a He–Ne probe beam in the optical configuration shown in Fig. 1 which employed a quadrant cell position sensitive detector to determine both normal and transverse components of beam deflection resulting from thermal wave propagation into the contacting air.For a Gaussian pump beam the normal and transverse components of beam deflection can be described theoretically in terms of the Hankel spectrum of the surface temperature of the sample, Ts(l) and dn/dT of the contacting medium. Using the derivation of Aamodt and Murphy10,11 the normal and transverse deflections can be written as: f l l n s f f n n T T b b x y = - - ¥ ò 2 d d ( ) exp( )cos( ) (9) f l l l l t s f n n T T b x y = - - ¥ ò 2 d d d ( ) exp( )sin( ) (10) respectively. In these equations bf is the principal root of the radical (l2 + 2j/mi 2)1 2, where mi is the thermal diffusion length of the medium (i = q for a gas, s for a solid and S for the surface).The practical manifestation of the normal and transverse beam deflections is shown in Fig. 2 for a sample of pig stratum corneum. The pump–probe beam separation refers to the lateral displacement of the probe beam from the pump beam.It is essential that the probe beam is also as close as practically possible to the surface of the sample so that its deflection is determined largely by the thermal properties of the sample and not the surrounding gas, i.e., it must be within a thermal wavelength of the sample surface. At zero separation the pump beam and probe beam are therefore orthogonal. For the purposes of determining Dth it is the transverse deflection which is of importance.Kuo et al.9 showed that the separation of the zero crossing points, xo [Fig. 3(b)] can be changed by changing the modulation frequency (f). In effect this changes the thermal wavelength in the sample and furthermore they showed that if xo is plotted against 1/f 1 2 a straight line is derived with a gradient of g given by g = (1.4pD)1 2 (11) Thus a value of Dth can be obtained and indeed Kuo et al.,9 used this method to determine Dth for semiconductor materials.In the work of Brown et al., Dth for stratum corneum was found to be 2.9 ± 0.5 3 1024 cm2 s21 at 50% relative humidity (RH) which is in good agreement with the values of Warner et al.12 Using the same method Brown determined D for stratum corneum under different relative humidities by carrying out the experiments over saturated salt solutions (Table 2). Water only has a value of 14.2 3 1024 cm2 s21 so it is clear that as the skin becomes more hydrated the thermal diffusivity increases towards that of water which is entirely consistent with complete hydration of the stratum corneum.In the same paper a thermal lens method was also used to determine Dth which resulted in the same value for stratum corneum at 50% RH. These values of Dth now allow accurate definitions of thermal diffusion lengths in other photothermal studies. For example Imhof and co-workers13,14 used the technique of optothermal radiometry to determine hydration of skin and for depth studies of topically applied substances in which Dth is an important parameter.In optothermal radiometry an IR laser pulse is used to briefly illuminate a sample of skin. Heat generated by absorption and relaxation is then recorded as IR emission either at a fixed wavelength or using an FTIR spectrometer if spectral information is required. Stratum corneum hydration measurements were made using this technique by exploiting the spectral properties of water.A Q-switched Er : YAG laser (l = 2.94 mm) was used to excite near to the water absorption maximum and the wavelength of the resulting emitted radiation was recorded at 13 mm. The Fig. 1 Optical configuration for photothermal beam deflection experiments. Pump beam, Ar ion, l = all lines; Probe beam, He : Ne, 632 nm. Fig. 2 (a) Normal and (b) transverse components of probe beam deflection for stratum corneum. Ar ion pump power, 2 W. Reprinted from Brown, S. M., Baesso, M.L., Shen, J., and Snook, R. D., Anal. Chim. Acta, 1993, 272, 717, with kind permission of Elsevier, Amsterdam. Table 2 Thermal diffusivity of stratum corneum measured under atmospheres of different relative humidities Saturated Thermal salt diffusivity/ solution RH (%) cm2 s21 MgCl2 33 1.7 ± 0.331024 NH4NO3 66 6.6 ± 0.831024 NaCl 76 9.1 ± 1.031024 K2CrO4 87 11.4 ± 0.631024 Analyst, April 1998, Vol. 123 589optothermal response of skin under these conditions takes the form of that seen in a semi-infinite homogenous slab in which the signal can be described as: S(t) = Aet/t erfc(At/t) (12) where t = 1/b2Dth, A is a constant and t is the optothermal decay time.The optothermal decay time for fully hydrated skin is of the order t = 300 ± 20 ms. Using values of Dth obtained by Brown et al.8 these authors were able to study hydration recovery and wound healing through measurements of the change in hydration. In a separate series of experiments these authors were able to measure epidermal thickness using an empirical model in which characteristic time constants t2 and t3 of delayed thermal waves which originate in the epidermis are expressed as the difference between two exponentials whose peak delay is given by Dt A A = - é ë ê ù û ú t t t t t t 2 3 2 3 3 2 2 3 ln (13) The thickness Dz can then be determined, again provided that an accurate value of Dth is known, using the following expression: Dz = AB(2DthDt) (14) Using this method Imhof et al.also determined the reduction of skin thickness by tape stripping and the thickness recovery of such damaged skin. Of course optothermal radiometry can provide information from a greater depth than the thickness of stratum corneum and the thermal diffusivity will vary in different layers of skin. To study whole epidermis and obtain Dth for skin Baesso, Shen and Snook15 used a laser-induced photoacoustic method in which a sample of stratum corneum and epidermis (40 mm thick) was placed in a photoacoustic cell and illuminated with a argon ion laser beam at 514.5 nm.In this work phase resolved measurements were made in which the analytical expression for the photoacoustic signal phase can be found by solving the thermal diffusion equation and calculating the averaged temperature fluctuation in the sample. The resulting expression is: f p wt wtb = - - - é ë êê ù û úú - - 2 1 1 1 1 1 2 tan ( ) tan ( ) / (15) where again tb = (b2Dth)21.Thus the signal phase angle was measured over values of w = 100–400 Hz corresponding to thermal diffusion lengths of 27 mm to 15 mm (calculated from m = ABB (Dth/pf), assuming a value of b = 81 cm21 as measured by spectrophotometry. The values of t and tb were obtained by fitting a curve to the experimental data using eqn. 15 and Dth was evaluated to be 4.1 ± 0.8 3 1024 cm2 s21, i.e., higher than stratum corneum owing to the greater extent of hydration of epidermis at the same RH of 50%.Phase-resolved measurements can also be used for depth profiling because the signal phase lag between the surface and subsurface layers or absorbers permits discrimination of these components and indeed Palmer and Dittmar16 have shown that this method can be more accurate than the frequency modulation technique. Palmer and co-workers16,17 have described and modelled step-scan FTIR–PAS to allow phase-resolved measurements to be made in the IR, applying this method to study depth profiles in polymer coatings.In the step-scan technique the PA signal is obtained at a constant frequency over the whole spectral range which leaves the thermal diffusion length constant at all wavelengths. Depth information is then obtained from the signal phase lag at each wavelength. This technique is in marked contrast to dynamic scanning interferometry where each wavelength is modulated at a frequency dependent upon that wavelength thus yielding the undesirable situation of different thermal diffusion length for each wavelength. The step-scan FTIR–PAS technique shows great promise for depth profiling because of the feature of a fixed thermal diffusion length and the fact that full spectral identification is obtained.Indeed in the context of skin research Baesso et al.18 have applied the technique described by Palmer to study the penetration of substances through stratum corneum. In that preliminary work the authors described the rapid diffusion of dimethylsulphoxide (DMSO) through skin using the phase resolved method at a probed depth of 20 mm.The characteristic DMSO absorption bands at 2900, 1350 and 1000 cm21 were clearly visible and still weakly detectable after 24 h. Clearly the technique could be developed for full IR spectral depth studies of topically applied substances although some problems can be envisaged which are also common to the conventional PAS studies of skin diffusion processes (Giese et al.7) These problems stem from the fact that the pump beam is used to probe through the topically applied substance. Thus as the substance diffuses into the sample both the optical absorption coefficient will change as will the thermal diffusivity unless the latter is fortuitously matched to the substrate.It is for this reason that a different approach has been investigated in our laboratories which relies upon direct measurement of the diffusion through membranes using a beam deflection technique.Although not yet applied to skin the technique has been evaluated to measure diffusion through membranes. Beam deflection techniques for membrane diffusion studies In the photothermal beam deflection work described so far it is assumed that probe beam deflection occurs as a result of a thermally induced refractive index gradient as the sample induced thermal wave travels into the contacting medium. In the absence of photothermal pumping however there is also the possibility of beam deflection occurring in the presence of a concentration gradient through the parameter dn/dC, i.e., the change in refractive index caused by a change in concentration.Thus if a Gaussian laser beam passes through a concentration gradient normal to the beam propagation axis it will be deflected. The refractive index change across the beam becomes: d d d d d d n r n c c r = æè öø æè öø (16) where n is the refractive index and r is the radial position across the beam measured from the propagation axis.This phenomenon is the basis of a very simple measurement technique for determining diffusion coefficients and concentration gradients in the interfacial region of membranes or surfaces. Fig. 3 illustrates the principle. A single mode Gaussian probe beam passes through the interfacial region of the sample and onto a linear CCD array. When a concentration gradient is imposed Fig. 3 Probe beam deflection apparatus for concentration gradient studies in membrane/solution interfaces Probe beam, He : Ne, 632 nm. 590 Analyst, April 1998, Vol. 123across the beam by a species diffusing through the membrane the beam is deflected and moves along the CCD array in the angle of deflection being proportional to the concentration gradient. This type of system was first reported by Mandelis and Royce19 who used a simple apparatus to study the diffusion of electroactive species in the interfacial region of an electrochemical cell.In their theoretical treatment Mandelis and Royce derived an expression for the deflection of the beam, f: f( , ) exp z t mz mx z Dt x Dt = - + æ è çç ö ø ÷÷ - æ è ç ö ø ÷ 1 12 4 0 2 0 2 (17) where: m A Dt = æ è ç ö ø ÷ 2 p (18) and: A C n n C = æ è ç ö ø ÷ æè öø 0 0 d d (19) where t is time; z is the distance from the sample to the detector; x0 is the offset distance of the beam from the membrane; D is the diffusion coefficient (cm2 s21); C is the concentration (Co being the initial concentration); and n is the refractive index.It is possible using these equations to plot theoretical beam deflection profiles for diffusing species of different diffusion coefficients as shown in Fig. 4. The steep curve towards the minimum indicates the build up of a concentration gradient across the beam until the maximum gradient is established (the minimum, t0, shown on the curve). After that point the gradient decays as a result of establishing concentration equilibrium with the bulk solution.For times greater than t0 the diffusion coefficient D can be evaluated using the expression t0 ! x20 4D as predicted in eqn. (17). Although first developed for studying diffusion in the interfacial region it is possible to apply the model to diffusion through membranes. In the apparatus shown the laser is a single mode TMoo helium neon laser; the membrane cell accommodates a circular membrane glued to the centre tube which can be adjusted in height to maintain the same hydrostatic level either side of the membrane. The CCD array detector is a 3648 pixel array (8 mm pitch) interfaced to a transputer board and PC.The beam axis is positioned 0.25 mm from the membrane and the beam is brought to a focus at the centre of the membrane. For a concentration gradient which decays normally to the plane of the membrane the beam is deflected towards the membrane which for large gradients can be inconvenient as the beam strikes the membrane holder.Using this apparatus, however, it has been possible to monitor refractive index gradients for different solutes and membranes and to monitor the release of nitroglycerin from transdermal drug delivery patches.20 To illustrate the technique Fig. 5 shows the beam deflection expressed as pixel numbers versus time for a 0.05% aqueous polyethylene glycol (RMM 600) injected into the upper chamber at to = 0 s.The lower chamber contained distilled water and the membrane was of the mixed nitrocellulose and nitroacetate type (0.45 mm pore size). The deflection clearly follows the theoretical profile although after 60 s oscillations are observed. The return to the original equilibrium conditions as predicted by the theoretical plots takes about 10 min in this case, although the return time depends upon the cell volume as equilibrium conditions are achieved when there is a uniform concentration in the lower compartment and across the membrane.From the experimental plot, to can be deduced as (40 ± 4) s which yields a diffusion coefficient of (3.8 ± 0.25) 3 1026 cm2 s21 which is comparable to the calculated value of (3.75) 3 1026 cm2 s21 from the relationship D = KD M2aD where KD and aD are constants for the polymer solvent system at given values of temperature and pressure.21 We have also studied beam deflection variation with both molecular weight of dissolved PEG and concentration and shown that clear linear relationships between deflection and these parameters exist using both renal dialysis membranes and Whatman filter papers.In these experiments however the polymer solution was pumped slowly into the membrane compartment at 6 ml h21 using a syringe pump.20 Thus it is entirely feasible to use this method for concentration and molecular weight determinations as well as studying diffusion coefficients. One drawback of the technique, however, is the lack of specificity in a complex diffusing solute i.e., when more than one chemical species is present, the deflection cannot be attributed to one or more of these species.To overcome this problem the next stage of research is to combine thermal lens spectrometry with concentration beam deflection as the former yields, like UV/VIS spectrometry, spectral information. In the last part of the paper therefore the potential for thermal lens spectrometry for thin film studies is explored before returning to the problem of combined TLS and beam deflection measurements for interfacial studies.Thermal lens studies of thin films Thermal lens spectrometry is normally associated with chemical analysis because of its excellent sensitivity compared with conventional UV/VIS absorption spectrometry and its use to determine quantum efficiencies and lifetimes of excited states. Fig. 4 Theoretical probe beam deflection with time for chemical species diffusing through the probe beam.Fig. 5 Replicate experimentally determined probe beam deflection with time for diffusing polyethylene glycol (RMM 600). Analyst, April 1998, Vol. 123 591Many of these aspects are described in the review of Lowe and Snook22 but few of these refer to its use for determination of species in, or parameters of thin films. The reason for this paucity of data of thermal lens aspects of thin films is the lack of appropriate models. In conventional thermal lens spectrometry an excitation laser is used to excite chromophores in a sample which generates heat via their non-radiative relaxation.Because a TMoo beam is used with Gaussian intensity profile more heat is generated at the centre of the beam than the wings which results in a symmetric refractive index gradient. As the coefficient of refractive index with temperature (dn/dT) for liquids is generally negative (positive for solids) a probing beam passing through the refractive index gradient becomes divergent and the degree of divergence can be related to the thermal gradient and hence the absorbance of the sample solution.In effect the region around the pump beam focus behaves as a thermal lens, the magnitude of which can be determined through the thermal lens parameter q q l = P n T A k ( / ) d d (20) where P is the power of the pump beam, A is the absorbance, l the wavelength and k the thermal conductivity of the solution.The term (dn/dT)/l k is often referred to as the enhancement factor and the significantly improved detection limits over conventional UV/VIS spectrometry (2¡©3 orders of magnitude) are in part attributable to this factor and in part due to the power dependency of the signal. A complete description of the TLS signal was developed by Shen et al.23 in which the effect was considered as a phase retardation imposed across the probe beam caused by the refractive index gradient.The simplest way of measuring the effect of the beam divergence is to use a photodetector and pinhole in the far field to measure a change in intensity as the beam spreads. Alternatively a CCD detector can be used to capture the entire beam profile. The time-resolved intensity in the far-field is then given as: I t mV m V t t m V ( ) tan [ ) ] /( / ) = - E I I ¢� ¡Æ ¢« + + + + + E E A ¢¦ - 1 2 2 1 2 2 1 2 1 2 2 2 q c (21) where the degree of mode mismatching m is given by: m e = E E A ¢¦ w w r 1 2 (22) and V Z Z = 2 c (23) where w1r is the radius of the probe beam at the sample and we is the radius of the pump beam focus.Z2 is the distance from cell to detector and Zc is the confocal distance. Eqn. 21 effectively describes the intensity of the probe beam as the thermal lens builds up after a brief exposure of the sample to the pump beam and an important parameter in the equation is the characteristic time which can be defined as t = (we)2 4Dth .Using a curve fitting routine eqn. 21 can be fitted to the experimental rise-time with very good agreement to derive both the thermal lens parameter, tc and hence Dth. Indeed it was this procedure which allowed TLS to be used for the determination of Dth for skin in the paper of Brown et al.8 One of the problems with the model which yielded this equation, and most other models, is that assumptions are made in which the sample is of infinite axial and radial dimensions. Therefore heat losses to the surroundings and sample container walls are not considered, i.e., the thermal gradient has decayed to zero in the sample.For a thin film which we are considering here this is not the case and either the thermal properties of the cell walls and surroundings must be considered or the minimum thickness conditions for compliance of experimental results with the theory must be determined. Accordingly Shen and coworkers24,25 determined the minimum radius and thickness for which their model holds for different pump¡©probe beam ratios.The derived minima were determined and the thin film model proved experimentally using 100 mm and 200 mm optical cells and Cu (EDTA) solutions. Providing that measurements are made within a time equal to 10 tc then the equation holds (tc values in solution are typically 1¡©10 ms). In effect this permits measurements to be made with a minimum radius of 25 mm and minimum thickness of 20 mm before t cell walls or surroundings have to be considered in the model.This approach has enabled thermal diffusivities of polymer films to be determined using the two beam thermal lens spectrometer previously reported.26 For these solids, however, the induced optical path length change and associated induced phase shift involves dl/dT, the sample thickness change as well as dn/dT. dl/dT is largely dominated in a solid sample by the thermal expansion coefficient, expansion being induced by the heating effect of the laser.Thus in eqn. 24 dn/dT is replaced by ds/dT: d d s T n l l T n T T T = - ©¡e c o©ª ¡À+ ©¡e c o©ª ¡À 0 0 1 0 0 ¢Ò ¢Ò ¢Ò ¢Ò (24) where the first term represents the sample thickness change and the second term the refractive index change with temperature. Thus using an Ar+ laser operated at 488 nm the TLS spectrometer and the time-resolved model thermal diffusivities of polyvinyl acetate polycarbonate and low and high density polyethylene were determined as shown in Table 3.These results are in good agreement with literature values obtained with photoacoustic and other methods.27 Thermal lens spectrometry is also widely established for chemical analysis and brings superior detection limits than say UV/VIS spectrometry. Although this factor has driven the development of the technique there are many other uses including lifetime studies, fluorescence quantum yield studies and photokinetic studies and for a complete review the reader is referred to the work of Lowe and Snook22 or the book of Bialkowski.28 One often overlooked feature of TLS, however, is the ability to probe very small volumes, in effect the interaction region between the pump and probe beam which can be as little as a few hundred pico litres.Thus it is entirely feasible that TLS can be used to study the interfacial region and indeed this is one of the purposes of developing the time-resolved thin film model to determine how close to a membrane TLS spectrometry could be performed, without requiring consideration of the effects of heat loss through the membrane.Table 3 Thermal diffusivities of polymer films measured by time-resolved thermal lens spectrometry Thermal Thickness/ diffusivity/ Sample mm 3 103 cm2 s21 Low density polyethylene 190 1.6 ¡¾ 0.05 High density polyethylene 100 2.2 ¡¾ 0.1 Polycarbonate 500 1.3 ¡¾ 0.1 Polyvinylacetate 50 0.8 ¡¾ 0.05 592 Analyst, April 1998, Vol. 123Future work The ultimate goal is to study the interfacial region using a combination of TLS and beam deflection to obtain species identification along with their concentration gradients.This will facilitate study of, for example, electrochemical redox species and discrimination between diffusing species in electrochemical and membrane studies. At first glance this seems an impossibly difficult task as the probe beam passing close to the membrane or electrode would be lensed and deflected at the same time.Also the concentration gradient would not be linear across the beam, rather falling as an exponential according to diffusion theory. Thus the crosssectional shape and deflection of the beam in the far-field would be determined by concentration and thermal effects. To deconvolute these effects requires first a model which is adaptable to probe beam profile imaging to record the crosssectional profile and deflection. This has been achieved by Soroka29 in which he demonstrated the effects of a concentration gradient of the beam profile.Secondly, the deconvolution of concentration effects and thermal effects is required. Fortunately nature helps in this as the chemical diffusion coefficients are typically two orders of magnitude different into solution from thermal diffusivities (1026 cm2 s21 compared with 1024 cm2 s21 for Dth). Thus many time-resolved thermal lens profiles can be obtained during the more slowly evolving concentration gradient.When realised over the next few years the sensitivity of TLS could therefore be exploited for sensitive species dependent measurements in interfacial regions. For diffusion into membranes or substrates such as skin the future looks good for techniques like optothermal radiometry in which in vivo measurements can be made. For in vitro measurements or for inanimate materials studies the use of stepscan FTIR with photoacoustic spectrometry is very promising and finally for diffusion studies through membranes and diffusion coefficients the technique of optical beam deflection has much to offer in the future.We thank Unilever Research (Port Sunlight), UK and the Engineering and Physical Research Council (UK) for their valuable support for some of the projects mentioned in this paper. We also acknowledge Dr. S. M. Brown’s contribution to the skin humidity work. References 1 Eighth International Topical Meeting on Photoacoustic and Photothermal Phenomena. Guadeloupe, ed.Fournier, D., and Roger, J. P., J. Phys. IV, 1994, Colloque C7. 2 McDonald, F. A., Wetsel, Jnr., G. C., and Jamieson, G. E., Can. J. Phys., 1986, 64, 1265. 3 Carlslaw, H. S., and Jaeger, J. C., Conduction of Heat in Solids, Appendix II, Oxford University Press, London, 1959. 4 Jiang E. Y., and Palmer, R. A., Anal. Chem., 1997, 69, 1931. 5 Anjo, D. M., and Moore, T. A., Photochem. Photobiol., 1984, 39, 635. 6 Moore, T. A., Photochem, Photobiol. Rev., 1983, 7, 187. 7 Giese, K., Nicolaus, A., Sennhenn, B., and Kolmel, K., Can. J. Phys., 1986, 64, 1139. 8 Brown, S. M., Baesso, M. L., Shen, J., and Snook, R. D., Anal. Chim. Acta, 1993, 282, 711. 9 Kuo, P. K., Favro, L. D., and Thomas, R. A., in Photothermal Investigations of Solids and Fluids, ed. Sell, J. A., Academic Press, London, 1989, ch. 6, p. 198. 10 Murphy, J. C., and Aamodt, L. C., J. Appl. Phys., 1980, 51, 4580. 11 Aamodt, L. C., and Murphy, J. C., J. Appl. Phys., 1981, 52, 4903. 12 Warner, V., Giese, K., Sennhenn, B., Plamann, K., and Kolmer, K., Phys. Med. Biol., 1992, 37, 21. 13 Bindra, R. M. S., Imhof, R. E., and Mochan, A., J. Phys. Colloq., 1994, 4, C7-445. 14 Bindra, R. M. S., Imhof, R. E., Mochan, A., and Eccleston, G. M., J. Phys. Colloq., 1994, 4, C7-465. 15 Baesso, M. L., Shen, J., and Snook, R. D., Analyst, 1994, 119, 361. 16 Palmer, R. A., and Dittmar, R. M., Thin Solid Films, 1993, 223, 31. 17 Jiang, E. Y., Palmer, R. A., and Chao, J. L., J. Appl. Phys., 1995, 78, 460. 18 Baesso, M. L., Snook, R. D., and Andrews, J. J., J. Phys. Colloq., 1994, 4, C7-449. 19 Mandelis, A., and Royce, B. S. H., Appl. Optics., 1984, 23, 2892. 20 Lowe, R. D., and Snook, R. D., 30th Colloquium Spectroscopicum Internationale, Melbourne, Australia, September 1997, paper no. C96. 21 Polymer Handbook, Wiley, New York, 3rd edn., 1989, vol. V11, p. 62. 22 Lowe, R. D., and Snook, R. D., Analyst, 1995, 120, 2051. 23 Shen, J., Lowe, R. D., and Snook, R. D., Chem. Phys., 1993, 165, 385. 24 Shen, J., and Snook, R. D., J. Appl. Phys., 1993, 73, 5286. 25 Baesso, M. L., Shen, J., and Snook, R. D., J. Appl. Phys., 1995, 75, 3732. 26 Baesso, M. L., Shen, J., and Snook, R. D., 7th International Conference on Organised Molecular Films, Numana, Italy, 1995, paper no. 7, p. 23. 27 Leite, N. F., Cella, N., Vargas, H., and Miranda, L. C. M., J. Appl. Phys., 1987, 61, 3025. 28 Bialkowski, S. E., Photothermal Methods for Chemical Analysis, Chemical Analysis Monographs, Wiley-Interscience (New York) 1996, vol. 134. 29 Soroka, A. J.,PhD Thesis, UMIST, 1995. Paper 7/06757G Received September 17, 1997 Accepted January 19, 1998 Analyst, April 1998, Vol. 123 593
ISSN:0003-2654
DOI:10.1039/a706757g
出版商:RSC
年代:1998
数据来源: RSC
|
10. |
Determination of the crystallinity of calcined and graphitic cokes by X-ray diffraction† |
|
Analyst,
Volume 123,
Issue 4,
1998,
Page 595-600
Frank R. Feret,
Preview
|
PDF (88KB)
|
|
摘要:
Determination of the crystallinity of calcined and graphitic cokes by X-ray diffraction† Frank R. Feret Alcan International Ltd., Arvida Research and Development Center, Jonquière, Québec, Canada G7S 4K8 Although the Scherrer equation has been the basis of the XRD method for the determination of the crystallinity of calcined coke, the most accurate interpretation of coke crystallinity involves profile analysis. In this paper, the general background related to coke crystallinity (Lc) determination is described, Alcan and ASTM standard methods are compared and line breadth as a function of crystallite size is characterized. At present, it is generally understood that the coke graphitization process occurs during calcination of coke and is related not only to the calcination temperature, but also to other parameters (composition, process conditions). As coke graphitization affects electrolytic cell performance, it should be monitored and quantified.The methods used for the determination of the degree of graphitization (DOG) and described in the literature are useful for cases involving calcination temperatures above 1800 °C.These methods are critically examined. Because known methods do not apply to the low-temperature graphitization processes (1200–1500 °C), a new approach was conceived. It is based on profile line analysis and an estimate of the graphitic contribution in coke specimen can be made. The new approach involves a commercial X-ray diffractometer and a corresponding software package and relies on profile fitting of the (002) coke peak.A calcined coke specimen or the presence of a graphitic portion in calcined coke can then be clearly recognized. During the profile fitting operation it is possible to resolve the graphitic portion from the calcined peak profile mathematically, and then to express it quantitatively. Among several profile functions that were tested, the split Pearson VII was found to give the best fit to data corresponding to calcined coke.A novel equation by which the DOG can be estimated in calcined coke is given. Selected examples of the DOG determination are described. Keywords: Calcined coke; graphitic coke; crystallinity; degree of graphitization; profile line analysis; X-ray diffraction Over the past several decades, X-ray diffraction (XRD) has generally been used for the determination of crystallinity (known as the parameter Lc) in green and calcined petroleum coke.Most industrial laboratories involved in Lc determination apply their own or the ASTM method (D 5187) which appeared in 1991. The methods cover the determination of the mean crystallite thickness of a representative, pulverized sample of calcined coke by interpretation of an XRD pattern produced using conventional X-ray scanning techniques. The XRD pattern is obtained in a scan of the carbon (002) reflection covering the ranges 14–34 °2q (Cu tube) or 19–39 °2q (Co tube).In the far past, the data were recorded in the form of a strip chart. Initially, the interpretation of the scan was carried out manually; it was based on a graphical procedure. Later, computer software designed to read and store the angular and intensity measurements became involved in the task. Whether manual or computer aided, the determination of a baseline is done first, followed by peak height, half peak height and the half peak height angles 2q1 and 2q2.Differences between Alcan and ASTM methods Sample preparation The ASTM method allows the use of any of the following techniques for packing the coke sample into the diffractometer specimen holder: back filling, front filling, side loading and even briquetting. Probably the document needed to include the practices of various users. On the other hand, the Alcan method recommends that a cavity slide sample (a packed powder mount) be prepared. The point is that the way in which the coke specimen is ‘packed’ affects the results.The packing influences X-ray penetration of the specimen and this clearly affects the peak broadening mechanism. In the case of a typical matrix the X-ray beam penetrates only approximately 20 mm, so that the specimen appears thin. However, for carbon (which is a light matrix) the X-rays penetrate the whole sample. Consequently, diffracted beams from underlying layers cause line broadening. This means that the resulting Lc values are artificially deflated, which can only be tolerated for cokes.When analyzing anthracite and graphite a very thin smear should be used. This practice does not add broadening due to X-ray penetration, but it does lower measured intensity. In general, there is an inverse correlation between the thickness of the specimen used and the Lc value obtained. Therefore, although results obtained with different preparation methods are acceptable for process control, they should not be compared with those from other places.In other words, any inter-laboratory comparison should involve results obtained using the same sample preparation technique. Interpretation of measured data There is a difference in the calculation procedure in the ASTM method and Alcan standard method. Both methods are based on Scherrer’s equation.1 In the Alcan method the Scherrer equation employed is Lc cos = 0 89 . l b q (1) where 0.89 = Scherrer’s constant, l = radiation wavelength, q = angular position of the peak of interest and b = 2q1 2 2q2 (in °2q); pure diffraction broadening represented in † Presented at the XXX Colloquium Spectroscopicum Internationale (CSI), Melbourne, Australia, September 21–26, 1997.Analyst, April 1998, Vol. 123 (595–600) 595the case of coke by FWHM (full width at half maximum). In the ASTM method, the Scherrer equation has been replaced by Lc = - 0 89 2 1 2 . (sin sin ) l q q (2) This approximation is valid only when q = (q1 + q2)/2 and b are both small.Most of the calcined coke peaks are asymmetric and some are very asymmetric. Therefore, q (q1 + q2)/2 and 2q1 2 2q2 is not small. Moreover, it seems that because the original Scherrer equation itself is simple, there is no need for the approximation. The Scherrer constant depends largely upon the crystallite shape, the (hkl) indices and the definitions taken for b and Lc. Various investigators have assumed values from 0.70 to 1.70 for this constant.For cokes it is set equal to 0.89 for the sake of uniformity in published results. Line breadth of coke as a function of its crystallite size With powder samples, a peak observed at some particular diffraction angle, 2q, is generally due to diffraction from several symmetry-equivalent planes. A pure diffraction maximum produced by a crystalline powder has a natural profile which is determined largely by the crystallite-size distribution.2 Crystallite size is defined as the size of a microdomain that causes Xray diffraction.Diffraction is more sensitive to the microdomains and less to the particle size. A particle, even if it looks like a perfect crystal, typically is composed of many crystallites. They feature numerous lattice imperfections and small mosaic blocks. An (hkl) reflection is caused by crystallites with (hkl) planes parallel to the specimen surface. A calcined coke is considered as a two-dimensional, random-layered structure.On a diffraction pattern such a structure is revealed by the presence of only (hk0) and (00l) reflections. In the case of graphite-type materials, crystallites are stacks of graphitic carbon platelets located parallel to one another. The shape of a diffraction peak is important in the measurement of lattice distortions, whereas its breadth is significant in the determination of crystallite size. The geometrical properties of the diffractometer introduce aberrations into the pure diffraction profile which cause it to be more or less asymmetric, broadened and displaced from its theoretical 2q angle.As a result, the profile shapes obtained with a conventional powder diffractometer are not easily described. In order to estimate the magnitude of the peak broadening the Scherrer equation was used first. In the Scherrer equation the b parameter represents the pure diffraction broadening by the sample contribution alone. In reality, what is measured is the ‘observed’ peak breadth B = FWHM.This is because the true sample contribution b is superimposed by broadening (b) caused by the instrument itself. Determination of the pure diffraction breadth b constitutes a major effort associated with crystallite-size analysis. Scherrer’s original postulate was that the peak breadths are strictly additive2 so that B = b + b. This has since been found not to be generally applicable. Warren3 derived the relationship between integral breadths that B2 = b2 + b2 where the pure diffraction and instrumental broadening profiles are both assumed to have a Gaussian shape.However, it has been shown by various other investigators that the instrumental profile follows other functions more closely. Depending on the class of diffractometer and corresponding resolution, the b contribution is in the range 0.07–0.15 °2q. For example, using a Philips PW 1700 diffractometer the measurement of the 3.35 Å graphite line gives a value of about 0.13 °2q.A similar result (0.12 °2q) can be obtained from a (100) quartz line. By contrast, more recent diffractometers contribute broadening of about 0.05–0.1 °2q, depending on the diameter of the focal circle and applied slits. For the calculations carried out below, a value of 0.10 °2q was assumed. Since values of k, l, b and 2q are given, a calibration curve of Lc versus B can be constructed. For a Co tube, assuming after Warren3 that b2 = B2 2 b2 and substituting the given values in eqn. (1), we have L B c = � � - � 0 89 1 789 57 3 0 10 0 966 2 2 .. . ( . ) . (3) where the factor 57.3 is used to convert the value of b from degrees to radians. Hence the line width is given by B Lc 2 2 2 2 94 44 0 10 = + . . (4) This relationship is presented in Table 1 for crystal dimension of various materials. The contribution of the instrumental broadening to the sample profile is stable. However, its significance increases rapidly with crystal dimension.In calcined coke measurements, the instrumental contribution (parameter b) is negligible and almost never considered. The measured breadth (FWHM) is assumed to represent the sample contribution. In the case of anthracite the contribution due to the instrument must be taken into account. For crystals as large as a few mm (alumina, gibbsite), the instrumental broadening is dominant. The sample contribution to the measured peak width is too small to be measured.Modern determination of the FWHM is based on a procedure, called profile fitting. In this procedure, measured data are stored in the form of a digital scan. Next, a computer simulation produces a mathematical representation of the entire line profile ƒ(e). Background is compensated for automatically. Smoothing (by means of a third-degree polynomial method) is optional, but improves the fit quality remarkably when activated. Computation leads to the following results: (002) peak angular position; peak intensity (at the top, in counts s21); peak net area (normalized); FWHM; integral width; and centroid (centre of gravity).The normalized peak area is peak intensity (in counts s21) multiplied by the measuring step in degrees. Such areas are not dependent upon the measuring conditions. The FWHM is the overall width of the line profile at half-maximum intensity measured above the background. The FWHM used in the past was measured manually (with a ruler) based on peak scan.The integral width is defined as the integrated intensity of the line profile above background (peak net area), divided by the peak intensity:4 B I I IW p d(2 ) = Ú 1 2 ( ) q q (5) To calculate the integral breadth one needs digital data corresponding to the peak profile. The centroid is a measure of peak location. Using the Scherrer equation the Lc parameter was calculated next for each FWHM value. Table 2 gives an example of data which were generated using the DIFFRAC-AT software and a Siemens D5000 diffractometer.The profile fitting provides not only the Lc parameter but also valuable additional information. Moreover, a graphical repre- Table 1 Variation of line width (°2q) with crystal dimension Material Crystal dimension Lc/Å Linewidth, B (°2q) Coke 30 3.15 50 1.89 100 0.95 Anthracite 500 0.214 Boehmite 1 000 0.138 Alumina, gibbsite 10 000 0.100 100 000 0.100 596 Analyst, April 1998, Vol. 123sentation (XRD scan) of the coke peak is also available. In contrast to methods based on intensity measurement in a few predetermined places on the coke peak, profile fitting requires a much longer measuring time (now 14 min) which is dictated by the need to complete the (002) peak scan.Fortunately, most of the time the measurement is carried out automatically for a series of samples. As far as calcined coke is concerned, the Scherrer equation (based on the determination of the FWHM) will remain the basis of the XRD method for crystallinity determination.However, for graphitic cokes this approach is largely inefficient and cannot be applied. Determination of degree of coke graphitization (DOG) Graphitic cokes Calcined coke specimens are clearly defined by their profile shape and angular range. Graphitic cokes mark a gradual transformation from random structure to crystalline graphite. They tend to be marked by an unusual, asymmetric shape and specific range of the integral width and centroid.Even when a graphitic coke displays a symmetric profile, such a profile appears narrower than for a calcined coke, and it is shifted towards a larger diffraction angle (smaller d value). The possibility of finding graphite in an ordinary calcined coke has been recognized for some time. However, if the Lc determination procedure has not been programmed to provide coke diffractograms, then they will not be available for inspection. In such a case, coke analyses are carried out automatically, and numbers corresponding to Lc are computer generated for each sample.The danger is that an automatic procedure may miss a graphitic coke. Fig. 1 shows a graph which correlates coke calcination temperature with Lc for a series of samples. One sample calcined at 1300 °C indicates a much higher Lc value than expected. In a subsequent XRD study this sample resulted in a diffractogram which did not resemble the rest (see Fig. 2). Fig. 2 is a superposition of three diffractograms covering the 20–40 °2q range that were acquired using identical recording conditions. The diffractograms correspond to calcination temperatures of 1200, 1250 and 1300 °C. The most intense, highly asymmetric and simultaneously the narrower peak is related to the sample calcined at 1300 °C. For this peak profile, the Scherrer equation simply does not apply. Clearly, it is a composite peak which is made of two distinct superimposed peaks.One peak corresponds to ordinary calcined coke and it shapes the left part of the peak shoulder, at a smaller 2q angle. The other peak is shifted to the right (larger 2q angle). It corresponds to the graphitic part of the coke material. The degree of coke graphitization (three-dimensional crystalline ordering) influences the amount of this graphitic portion in the sample, and consequently affects the angular position of the second peak and its intensity. Ultimately, under favorable conditions (temperature), the entire sample might turn into graphite.Fig. 3 illustrates XRD scans of two peaks measured using identical scanning conditions. The broad, weak peak is from a calcined coke specimen; the sharp and intense peak is from pure graphite. The distinction is clear. If the initial analysis using profile analysis software, such as DIFFRAC-AT, indicates that the investigated coke peak is composed of two contributions, then a line profile analysis is required.Further analysis is carried out in two separate stages: (i) profile fitting (separation) of the two contributions and (ii) deconvolution of the instrument contribution from the resolved profiles and determination of their crystallinity and crystallite Table 2 Parameters obtained for coke samples Sample Parameter 4 Alcan standard Peak position (°2q) 29.92 30.02 29.76 29.0 29.96 Peak intensity/ counts s21 1611 1592 1424 1370 1256 Peak net area/ counts s21 °2q 5813 5677 5262 5578 5181 FWHM (°2q) 2.970 2.958 3.069 3.510 3.560 Integral width 3.72 3.67 3.82 4.23 4.31 Centroid (°2q) 29.58 29.57 29.52 29.47 29.42 Lc/Å 31.78 31.91 30.76 26.52 26.9 Fig. 1 Calibration curve of Lc versus temperature. Fig. 2 Superposition of three diffractograms. Fig. 3 Comparison of diffractograms for calcined coke and graphitic coke. Analyst, April 1998, Vol. 123 597size distribution. The second subject will not be discussed in this paper. Theoretical considerations Mathematically, each profile of the observed maximum h(e) is the convolution, or fold, of the pure diffraction profile f(e) and the weight function of the apparatus g(e):5 h g f ( ) ( ) ( ) e h e h h = - -• •Ú d (6) where the variables e and h have the same units as 2q.The process for obtaining f from h and g is called deconvolution. In order to adjust a mathematical profile function to the measured data, least-squares refinement techniques are mostly used.Computer simulation produces a mathematical representation of the diffraction curve. To represent the expected profile shapes many mathematical models have been tried with varying degrees of success. Most models employ 12 intrinsic parameters to describe the instrument aberration and wavelength-dependent contributions to the profile, and three parameters to describe the sample-dependent variables of line position, height and line broadening.6,7 The number of parameters needed to define a line varies with the selected profile shape function and whether it is assumed to be symmetrical or not.Among the most important functions one can select are Lorentz (Cauchy), Gauss, Voigt (convolution of Cauchy and Gauss functions), pseudo-Voigt, Pearson VII. The first three of these functions are symmetrical. Since the actual profiles are never pure Lorentz or pure Gauss, the corresponding equations have only limited practical value. The remaining functions are more complex as they require four parameters. When a line is not assumed to be symmetrical, it is treated as two half lines sharing the same location and height, but not the same shape—it is said that the function is split.The problem with graphitic cokes is that the pure diffraction profile f(e) is composed of two contributions, coke and graphite. Consequently, the two contributions must first be separated to obtain two separate line profiles. Only then can the instrumental contribution be eliminated by deconvolution.In practice, any of the above functions could be used to separate the superposed coke peak. The selection of an appropriate model function for the deconvolution is important, because errors incurred in this operation are transmitted to the next stage. In order to confirm unequivocally the presence of a graphitic portion in the specimen and to resolve the (002) diffraction profile, a PROFILE program is run first. PROFILE is the new profile fitting program of the Siemens DIFFRAC-AT package which helps obtain accurate line locations, intensities and widths from resolved and non-resolved X-ray diffraction lines.The number of peaks to be resolved could be either identified automatically by PROFILE or set by the operator. Experimental results Data collection was achieved using a Siemens D5000 automated powder diffractometer. The instrumental details are given in Table 3. The Siemens D5000 diffractometer was not equipped to work with monochromatic radiation (Ka1), and therefore all measurements were carried out with a Ka1 plus Ka2 doublet.The presence of the a2 component enhances the line broadening and introduces asymmetry into the profile. As a consequence, profile fitting has to be performed using Ka1 plus Ka2 profiles. The a2 elimination is achieved automatically by computation during profile fitting. The initial experimental effort was aimed at selecting the best mathematical model to represent the (002) coke peak.A calcined coke characterized by a highly asymmetric profile shape was selected for the test. Fig. 4 illustrates examples of four different fits obtained with Gauss, Lorenz, Voigt and split Pearson VII mathematical models. Various individual parameters of the model were tried. Clearly, the split Pearson VII model provides the best fit between theoretical and measured line profiles. Consequently, the split Pearson VII function with undefined parameters was selected for profile fitting runs. If employed for a calcined coke, it needs eight adjustable parameters.Next, a graphitic coke was used. Fig. 5 shows the outcome of the profile fitting in a graphical form. Two separate peaks were resolved from the original line profile. Selected numerical data corresponding to this profile fitting is given in Table 4. Significant new parameters are: angular (2q) and d position of both peaks (Å), peak height (counts s21) and integrated intensity (area) and FWHM (°2q).The goodness of profile fitting is estimated by reliability (Rel) and theoretical reliability (TR) parameters (%). The reliability is related to the difference between observed and calculated profiles. Obviously, the lower the number the better. The profile fitting results are saved as a DPF file. Using the data in Table 4, one can estimate the proportions of both contributions in the sample or the degree of graphitization.Degree of graphitization (DOG) In the past, it was generally assumed that the graphitization process depends on the calcination temperature, which is Table 3 The Siemens D5000 measuring conditions used at Alcan Radiation Co-Ka Detector Kevex, Peltier cooled Monochromator No Generator 40 kV, 40 mA Divergence slit Fixed, 2 mm (coke), 1 mm (graphite) Step size 0.01 °2q Scattering slit 2 mm (coke), 1 mm (graphite) Receiving slit 0.6 mm (coke), 0.1 mm (graphite) Scanning range 20–40 °2q (coke), 25–35 °2q (graphite and graphitic coke) Measuring time 1 s per step (coke), 5 s per step (graphite and graphitic cokes) Wavelength 1.7889 Å PHD window 20–80% Sample changer 40 position Sample spinner On Fig. 4 Example of four different fits. 598 Analyst, April 1998, Vol. 123assumed to be high. At present we know that the graphitization process is related not only to the calcination temperature, but also to other parameters (composition, process conditions). Sørlie and Gran8 proved that the cell operating temperature is sufficiently high for catalytic graphitization of the carbon lining to take place.With increasing graphitization, the size of the graphitic layers and the total number of such layers (contributing to the XRD process) increase continuously, whereas the apparent interlayer spacing decreases. In the open literature, the degree of graphitization (DOG) has been measured using one DSC and two different XRD methods. The DSC method involves the specific heat capacity and is based on the two-dimensional model of phonon transport.By this method a single parameter, the in-plane Debye temperature, qD, characterizes the quality of graphene layers.9 The Debye temperature is the temperature at which all phonon modes are excited, and the specific heat capacity reaches a nearly constant value. The degree of graphitization, g, was defined as a function of the Debye temperature: g T = q q D, graphite D, ( ) (7) where g approaches 1.0 at high temperatures and is much less than 1.0 at low temperatures.The authors claim that the numerical values of g obtained by this technique are very close to those obtained by the XRD method involving the (002) interplanar spacing. However, the specific heat capacity is an indicator of bulk lattice quality and not of the planar stacking in one direction. For typical non-graphitic carbons the interlayer spacing represented by the (002) peak is constant at 3.44 Å and in graphite it is 3.354 Å.Maire and Mering10 defined a degree of graphitization (g) as g d d = - - = - 3 44 3 44 3 354 3 44 0 086 002 002 . . . . . (8) where d002 is the average interlayer distance (Å) measured by XRD. Statistically, the degree of graphitization is the probability of parallel orientation for two consecutive graphite layers. Another XRD method is based on the intensity ratio of the (002) diffraction peak of the carbon specimen relative to a defined graphite standard:11 g I I = s r (9) where Is and Ir are the X-ray intensity/mass ratios for the specimen and reference, respectively.Eqns. (7)–(9) are considered useful for cases involving calcination temperatures above 1800 °C. This is because above this temperature the graphitic portion of the peak doublet becomes dominant and the error of the peak position corresponding to the graphitic portion becomes tolerable. None of the above methods provides an accurate estimate of DOG for the low-temperature calcination ranges.In contrast, using the data obtained from the PROFILE software one can now attempt to calculate DOG for samples calcined at low and high temperatures. This can be done in two ways. First, knowing the d positions of two resolved peak profiles, left (L) and right (R) (Fig. 5), eqn. (8) can now be employed for profile on the right. However, we do not feel that eqn. (8) provides an accurate or adequate description of DOG. This is because various amounts of the graphitic material in the sample may feature the same d002 interlayer distance.In contrast, a ratio of the normalized surface area A of the graphite and carbon portions represents a better choice: g A A = ¥ graphite coke + graphite 100% (10) We feel that the last definition represents reality most appropriately. For example, using the data in Table 4 (the area intensity), one can estimate the proportions of both contributions in the sample.Consequently, g = + � = 4768 4768 27125 100 15% (11) The degree of graphitization defined by eqn. (10) can be used to study differences among samples in a group. However, it has to be realized that the two resolved peak profiles have not yet been corrected for instrumental broadening. This needs to be done next in order to obtain results on an absolute scale. From the data in Table 4, it is apparent that the FWHM values for the two profiles are 1.60 and 0.569 °2q, respectively.If the profiles are considered individually, the instrument contribution in each is represented by approximately 0.1 °2q. Then, especially for the second profile, the specimen and instrument contributions are of the same order of magnitude. It has been suggested that the h profile should be a factor of 1.2 broader12 than the g profile for successful deconvolution. Based on the limited initial data, this seems to be the case for graphitic cokes. Discussion Modern XRD software allowing profile analysis is a necessary tool in the analysis of coke crystallinity.When compared with the Scherrer equation the profile analysis constitutes a new benchmark in coke crystallinity determination. Additional new parameters such as integral width, peak position and centroid are important and should be collected and compared over a long run. Using profile analysis software, coke (002) graphical representation becomes available for every sample so that a simple visual examination allows first a quick evaluation of a potential specimen graphitization.If either from a visual observation of the (002) peak profile or from the software analysis of the numerical data there is no confirmation of calcined coke, then the Scherrer equation simply does not apply. In such a case the Scherrer equation cannot be used to calculate Fig. 5 Outcome of the profile fitting: new method. Table 4 Example of profile fitting results Position Peak Height Area FWHM No.Function °2q d/Å (counts s21) (normalized) (°2q) 1A Split P VII 25.88 3.439 12 827 27 125 1.602 2A Split P VII 26.50 3.359 7 374 4 768 0.569 Analyst, April 1998, Vol. 123 599crystallinity from the FWHM parameter measured for the (002) coke peak. Consequently, a different analytical procedure must be used. The new procedure involves profile fitting and provides a tool for clear recognition of a calcined coke specimen, or the presence of a graphitic portion in calcined coke.Using software such as DIFFRAC-AT (PROFILE), it is not only possible to resolve the graphitic portion from the calcined coke peak profile mathematically, but also to express it quantitatively. The numerical data generated by PROFILE can be used in an additional stage to deconvolute the instrumental contribution from both profiles, and to evaluate the respective crystallite size and crystallite size distribution. The details will be covered in a technical communication. Development of analytical methodology related to the characterization of graphitic cokes could be continued even further. Using the Rietveld XRD approach and a full diffractogram analysis rather than a single peak, it should be possible to obtain additional information such as the probability P of finding a 3R stacking fault, the strain parameters, the preferred orientation and the Lc/La anisotropy ratio. The author thanks Dr. F. M. Kimmerle, Alcan Chief Analytical Chemist, for reviewing the manuscript and his valuable comments. Mrs. Jean Bell’s corrections of the manuscript were also greatly appreciated. The author further thanks Dr. G�otz Menges of Siemens, Karlsruhe, and Mr. Daniel Roy of Alcan International for their help with the experiments. References 1 Scherrer, P., Göttinger Nachrich., 1918, 2, 98. 2 Alexander, L. E., J. Appl. Phys., 1959, 21, 126. 3 Warren, B. E., J. Appl. Phys., 1941, 12, 375. 4 Alexander, L. E., J. Appl. Phys., 1954, 25, 155. 5 Klug, H. P., and Alexander, L. E., X-Ray Diffraction Procedures, Wiley-Interscience, New York, 1974. 6 Alexander, L. E., J. Appl. Phys. A, 1959, 21, 126. 7 Schreiner, W. N., and Jenkins R., Adv. X-Ray Anal., 1983, 26, 141. 8 Sørlie, M., and Gran, H., Light Met., 1995, 497. 9 Rogers, D. K., Jones, S. P., Fain, C. C., and Edie, D. D., Carbon, 1993, 31, 303. 10 Maire, J., and Mering J., Chemistry and Physics of Carbon, Marcel Dekker, New York, 1970, vol. 6, p. 125. 11 Aune, F., Brockner, W., and Øye, H. A., Carbon, 1992, 30, 1001. 12 Schwartz, L. H., and Cohen, J. B., Diffraction from Materials, Academic Press, New York, 1977. Paper 7/07845E Received October 31, 1997 Accepted January 8, 1998 600 Analyst, April 1998, Vol.
ISSN:0003-2654
DOI:10.1039/a707845e
出版商:RSC
年代:1998
数据来源: RSC
|
|