摘要:

ANALYST, APRIL 1993, VOL. 118 323 Tutorial Review Data Processing by Neural Networks in Quantitative Chemical Analysis* Martinus Bos, Albert Bos and Willem E. van der Linden Department of Chemical Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands An overview is given of the current usage of artificial neural networks as mathematical models for non-linear calibration procedures. The emphasis is on practical aspects: the choice of the calibration samples, the required network characteristics for a given problem, various training methods and their efficiency and the validation of the network models. Some problems with the application of neural networks in multivariate calibration are considered, together with recent research aimed at solving these problems.Keywords: Neural networks; multivariate calibration; sensor Most quantitative methods of chemical analysis are of a relative nature nowadays and stoichiometric, absolute methods are called upon only i n special cases, i.e., when established standard rnethods have to be applied such as International Organization for Standardization (ISO) methods or when high accuracies are required. Therefore, most modern instrumental methods require calibration. Another trend over the last decade is the rclaxation of the demands with regard to selectivity of analytical methods, compensated for by relying heavily on sophisticated chcmometric software to retain the required level of accuracy. High-resolution techniques and elaborate separation methods have been superseded by simpler equipment that is substan- tially computer based.In general, this approach has been very cost effective owing to the ever increasing performance : price ratio of computer equipment compared with the far more stable price level of measuring equipment. A characteristic example of this trend is the development of the near infrared analyser. Quantitative analysis of complex samples can be performed by using measurements at various wavelengths on unresolved broad bands of the spectra in multivariate calibration techniques. A calibration procedure is called multivariate when the dependent variable (a concentration) is a function of more than one independent variable (measured amount). Some multivariate calibration procedures are more complex than this and produce a mathematical relationship between a number of dependent variables (maybe the total composition of a sample, or at least the concentrations of a number of constituents of the sample) and a large number of independent variables (maybe a complete spectrum taken at fixed wavelength intervals).For the latter the general calibration procedure can be depicted as in Fig. 1. We have rn calibration samples the composition of which is known. In the first calibration sample (cl) component 1 has a concentration cl 1 , component 2 has a concentration cI2, etc. For the second calibration sample (c?) component 1 has a concentration c21 , component 2 has a concentration c22, etc. On each sample we have n measurements. The measure- ments on calibration sample 1 are given by xll,x12, .. ., xln. ~ * Presented at thc Scnsors and Signals Symposium at The Royal Socicty of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18. 1992. For the rn-th sample these measurements are xrnl,xr,12, . . ., xnrn, whereas its composition is given by C , , ~ , C , ~ , . . . . In the calibration procedure a mathematical model is used to represent these data. The calibration process itself then consists of determining the parameters of this model. Once the calibration model and its parameters have been established, it can be used to convert measurements performed on unknown samples into concentration data. Several mathematical difficulties can arise in solving the calibration model equations. Tf rn is much larger than n, the set of model equations will be overdetermined and no exact solution can be found.If n is large, the measurement data will probably contain dependent columns, prohibiting a unique solution. The most critical issue in the use of multivariate calibration is the right choice of the mathematical form of the calibration model. In the ideal situation this choice can be made on physico-chemical grounds. The use of Beer's law in spectro- photometric multi-component analysis is such an instance. However, there are a number of situations in which a theoretical description of the calibration model is not avail- able. One then has to resort to empirical models that should Measurement data model parameter ess 1 ' Concentration data 1 Mathe- matical form of model Fig. 1 General calibration procedure324 ANALYST, APRIL 1993, VOL.118 not only represent the calibration data as accurately as possible, but should also produce the right results for reference samples of known composition that have not been used in the calibration procedure. The various multivariate calibration methods that exist today differ in the mathematical form of the model and the way in which the model equations are solved. Most models rely on the property that the contributions of the measurement variables to the result are additive. Multiple linear regression (MLR), singular value decomposition (SVD), 1 principal component regression (PCR)2 and partial least-squares regression (PLS)3 fall into this category. Non-linear calibration models are required if interactions between the different components in the sample have to be accounted for.Sometimes this is accomplished by extending the measurement vector with quadratic and cross-terms of the measurements4 and followed by the application of a linear method to the extended measurement data. If the measure- ment vector is large and no a priori knowledge about which interaction terms are important is available, this approach drowns in the large number of input data. This paper shows that artificial neural networks (ANNs) are particularly suited as calibration models in this situation. In a recent review of neural networks for solving chemical problems, Zupan and Gasteigers treated the mathematics of various types of neural networks and classified the applica- tions found in the literature.In most applications the neural networks were used as classifiers, e.g., to interpret spectra for determining spectra-structure correlations or to diagnose malfunctions in complex process control from sensory data. Only a few applications concerning multivariate calibration procedures were mentioned. These examples will be covered in more detail in this paper together with some newer examples to show which factors should be taken into account in multivariate calibration with the use of neural networks. Another overview of neural networks6 was of a more general nature and did not mention any specific applications. Artificial Neural Networks Artificial neural networks are empirical input-output models suitable for modelling complex multi-input-multi-output rela- tionships by curve fitting.An important feature is their learning ability. The information that they contain is distri- buted over a large number of model parameters, which accounts for their great flexibility. They mimic human cognitive processes and as such are suited t o processing noisy, incomplete and even, to some extent, inconsistent data. The basic element of a neural network is a simple signal-processing unit, which is called a neuron. Such a neuron can have multiple inputs, but it has only a single output. The transfer function for the input-output relationship can have various forms. All have in common that each input is associated with a weight that determines the extent to which 0 0.4 0.3 1 / / I I the influence of the corresponding input is transmitted to its output.Inputs with weights of the same sign enhance each other and those of opposite signs counteract each other. For multivariate calibration the type of neuron that is used most often is the sigmoid type, which is characterized by the following equations: neti = Cwjioj i (2) 1 output, = 1 + exp (- where subscript i denotes the number of the neuron, wJI is the weight of the connection to neuronj of the preceding layer and oJ denotes the output of the neuron, T, is a factor that determines the slope of the rising part of the output function of the neuron and t), its bias with which the output function can be shifted along the net, axis. Fig. 2 shows the behaviour of the sigmoid neuron. If these neurons are combined in a network, this network can model any desired input-output relationship if it is given the right size and topology.For the topology of a network there are many choices: the neurons can be arranged in onc- or two-dimensional layers, a network can have multiple layers, the neurons can be connected only in the forward direction (from input to output) or they can have feedback connections, etc. In the description of the topology of a neural network, the term ‘hidden’ is used to indicate neurons or layers of neurons that have no connection to the outer world, in contrast to neurons in the input layer, which receive their inputs from outside, and neurons in the output layer, which deliver their outputs to the outer world (see Fig. 3). The behaviour of such a network is completely determined by the topology, the values of the weights associated with the connection between the neurons and the transfer function of the neurons.In some special networks the neurons have a kind of memory in which the current output value is retained and used in the calculation of the output when new inputs are presented to the network. These networks can operate in a ‘context-sensitive’ way. However, they are not used in calibration. With a given topology of the network and a given transfer function for the neurons, the desired behaviour of the network can be approximated by adjustment of the weights of the connections in the network. This adjustment of the weight factors is called the training of the network and is carried out by using the data of a set of calibration standards. Starting with a network the weight factors of which are randomly initialized, the measurements on the standard samples are presented to + 8,) -10 - 5 0 5 10 Net output Input Hidden layer layer layer Fig.2 C, T = 2.0 Sigmoid transfer function. Curve A, T = 0.5; B , T = 1 .0: and Fig. 3 Feedforward fully connected network topologyANALYST, APRIL 1993. VOL. 118 325 the network as inputs and the corresponding outputs of the network are calculated using the equations given above and then compared with the wanted values of these outputs. If there is a difference between these calculated and wanted outputs this error value is used to adjust the weights of the connections between the neurons. Various algorithms are available for this purpose, of which the back-propagation algorithm is the most p ~ p u l a r .~ It is fairly robust, but slow. More efficient algorithms have been published, such as the scaled conjugate gradient method* and the SuperSABg algorithm, but sometimes these speed-up algorithms suffer from strong oscillations. If the right topology is chosen, this training process can reduce the errors in the outputs of the network for the training samples to a negligible value, making the neural network a perfect empirical model for the training data. What one hopes for is that the network now also models the underlying process(es) that generated the training data and that the network can be used to calculate the right output values for input data that were not used in the training. In practice, it has turned out that neural networks trained in the proper way do indeed have this desired generalization capability. Neural Networks as Calibration Models No empirical model is better than the data that were used to derive it, no matter whether the model is based on a neural network or some other empirical description.Hence it is very important that a good representative set of calibration samples is used that covers the whole working range of the analytical method that is to be developed. A second major point in calibration is the dimensionality of the model. The model should contain sufficient parameters to accommodate the systematic variance of the calibration data. If this number of parameters is too small, the calibration model will fit the data poorly, if it is too high, this overfitting will produce model parameters that are strongly influenced by the noise in the calibration data.Both situations will generally cause large errors in the results when the model is used in the analysis of real samples. In neural network calibration models both phenomena also play a role, but overfitting due to the dimensionality of the model is not really a problem, as will be shown later. Far more important in this respect is the way the training of the neural network is performed. The third important issue in developing an empirical calibration model is its validation. The purpose of the validation of a calibration model is to derive estimates of the accuracy of its results. Calibration Samples The development of a calibration model starts with a data set consisting of the measurements and the concentration data for the components to be determined in a series of standard samples.In the acquisition of this data set care should be taken to cover the full range of expected compositions of the samples. A systematic design is necessary to choose the standard samples in such a way that not only is the concentration range of each component covered, but also that there are sufficient examples of possible interactions between them. Choice of the Neural Network for a Given Calibration Problem The number of decisions that have to be made when designing a neural network model for a calibration problem is over- whelmingly great. Generally only two things are fixed a priori: the number of inputs to the net and the number of outputs (and even these are sometimes a matter of discussion).Sometimes the measured variables are redundant and some of the inputs can be left out without impairing the performance Table 1 Issues in the choice of a neural network for calibration Item Questions to be answered Representation of Scaling/normalizing/transformation of sample data input-output data? Data reduction for inputs? Separate networks for each output? Number of neurons? Number of hidden layers'? Which connections between neurons? Feedback connections? Linear, sigmoid, radial base or other? Different transfer functions for Topology of network Transfer function different layers? of the network. If the network is to be used to determine several components simultaneously in one sample, one has the choice to use a network that outputs all wanted information, or to use a separate network for each of the components. Table 1 gives a summary of the more relevant items that have to be dealt with in the choice of a neural network as a calibration model.Pre-processing of input data If the measurement variables are of a different nature they will often show large variations in scale and dynamic range. To facilitate the training of a network having this type of input, it is often advantageous to scale all the measurement variables to have zero mean and a standard deviation of one. Sometimes another mathematical transformation of the inputs will improve the results considerably. Chemical judgement should guide how to express the influence of a certain measurement variable in the best way.If the measurement vector of a sample has a high dimensionality, which often is the case if it consists of a spectrum sampled at a fixed wavelength interval, then data reduction methods may be needed to keep the required training period within reasonable time limits. Principal components analysis, wavelet transformation10 and Legendre polynomials are some of the techniques for representing a large set of measurements on a sample in a few coefficients that can then be used as the actual inputs to the neural network. Scaling of output data The commonly used sigmoid neuron produces an output in the range 0-1. To accommodate output data exceeding this range, obviously rescaling is needed to bring the outputs within the limits that can be reached by the transfer function.For data having a high dynamic range, the use of a linear transfer function for the output neurons is advised. 1 1 Another solution to this problem of a large dynamic range of concentrations measured by potentiometry with ion-selective electrodes was given by Bos et a1.,12 who used the scaled logarithm of the concentrations as the output value to be delivered by the neural network. For samples in which the concentrations have a relatively small dynamic range, scaling around the centre of the sigmoid output function with a small standard deviation will be advantageous. 13 Network topology The configuration of a neural network to be used as a multivariate calibration model is the most difficult step in the calibration procedure.Generally, substantial trial and error steps are involved in establishing the optimum topology of the network and the right transfer functions for its neurons. Part of the difficulty stems from the long time that it takes to evaluate the performance of a chosen configuration. This326 ANALYST, APRIL 1993, VOL. 118 Inputs n n 0- output output layer Input Hidden layer layer Fig. 4 Network with cxtra connections of inputs to output layer needs full training and validation of the network to obtain a reliable measure for its generalization capabilities. In most of the papers cited above these efforts have concentrated on finding the optimum number of hidden neurons in feedforward networks with one or two hidden layers and in which the external inputs are propagated through the network layer by layer.The results in this respect do not always show a clear optimum11 and, if they do, it can be argued that this is caused by overfitting a noisy data set. The conclusion may be that it is best to use the smallest sized network that givcs satisfactory results. This not only reduces the amount of training time that is needed, but also ensures that there is moderate degradation of the performance of the calibration model at the limits of its working range.I3 Although most applications use feedforward networks, a feedback topology can sometimes be used to design a calibration model that replicates the physico-chemical deserip- tion of the process underlying the analytical method, as was shown for the use of ion-selective electrode arrays,l2 where it was used to model the Nicolskii equation.The input information to a network does not necessarily have to be processed strictly layer by layer. Networks can be designed that connect their external inputs not only to the hidden layer but also directly to the output layer," as shown in Fig. 4. If the transfer function of the output layer neurons is linear and the hidden layer neurons are equipped with sigmoid neurons, this type of topology is particularly suited to model calibration relationships that show small deviations from linearity. Transfer functions The most versatile transfer function that can be used to model a variety of relationships is the already mentioned sigmoid type. In its mid-range it has a linear part and at its extremes a convex and a concave part (see Fig.2). By combining enough of these neurons, arbitrary functions can be approximated to the required degree of precision.16 In applications where the calibration function is not too complicated, neurons with a simpler type of transfer function can be used. Generally this will improve the training effi- ciency, as has been shown for linear11 and radial base transfer functions'l that have a bell-shaped form. Training The aim of the training phase is first to adjust the weights of the connections in the network to obtain the best possible behaviour in the validation. In some special eases the network topology is adjusted during the training phase, e.g., to reduce the size of the network for better generalization.'7~lx The cascade-correlation learning architecture developed by Fahl- man and Lebierel9 also changes the network topology dynamically during training.Here, however, the process starts with a network that contains only one layer of output neurons to which the inputs are directly connected. This network is gradually extended during training until it reaches the required performance. In this way a near-minimum multi- layer network topology is built in the course of the training. In most applications, the back-propagation algorithm is used in one form or another to perform the training of the neural networks. The backbone of this algorithm is formed by the following steps: Enter the specific inputs (measurements on a sample) and measure the actual output produced by the chosen network.Compare this actual output with the desired output; calculate a quantitative error based on input-output analysis. Iteratively minimize the error by adjusting the weights of the connections in the network: (a) begin at the output nodes and adjust their weights; (b) propagate backwards to the layer adjacent to the output layer: calculate the errors there and adjust the weights; (c) proceed with this process until the input layer is reached. In the calculation of the quantitative error in the algorithm in step 2, one generally uses the sum of squared errors, either on a per sample basis or summed over all samples of the full calibration set. The latter is called epoch training. There is no clear view on which is to be preferred. The standard back-propagation algorithm as described by Rumelhart and and McClelland2° uses three parameters that have to be provided by the user: the learning rate, the momentum factor and the range in which the initial weights are randomized.The method is fairly robust: it generally leads to useful results, but its efficiency is highly influenced by the tuning of these parameters. Trial and error adjustments of these parameters will be needed to achieve reasonable training times. Convergence speed-up The main problem with the standard back-propagation algorithm is its slowness. Various modifications to this algorithm to speed it up have been suggested. I n some of these the learning rate is adjusted during the training, either globally9 or on a per weight factor basis.21 A different approach was taken by Fahlman22 with his Quickprop algorithm and by M611erx with his scaled conjugate gradient method.In these techniques the second derivative of the error is used in Newton's method to derive updates on the weight factors during training. Recurrent networks A second problem with the back-propagation algorithm and its variants is that they can only be applied to feedforward networks for neurons with a continuous derivative of the transfer function. Some types of networks with feedback connections can be trained by back-propagation by unfolding the recurrent links in time.7 An example of this method can be found in a previous paper." For networks that cannot be trained with the back-propaga- tion algorithm, the genetic algorithm can be an alternative23 if the network size is limited.I3 When to stop the training:? The last step in the training of a neural network is the decision to stop the iterative learning process at the right time.Prolonged training to reduce the errors beyond realistic values often deteriorates the performance on the reference samples,ANALYST, APRIL 1993, VOL. 118 100. I 327 90 A 80 .- 5 70 I B l o LL 0 2000 4000 6000 8000 10 000 Epochs Fig. 5 B, training Characteristics of overfitting. Curve A, reference and curve as shown in Fig. 5 , where the root mean square (r.m.s.) error for the set o f calibration samples and for some reference samples not uscd in the training is plotted against the number of iterations of the algorithm over the complete training set (epochs). It is not sufficient to cut off the training at the moment when the r.m.s.error of the reference set starts to rise, as there is n o guarantee that it is independent of the nature of the reference samples. Generally this behaviour is indicative of the development of an incorrect model in which the noise in the measurements is fitted instead of the phenomena that should explain the data in a chemical sense. Only if the calibration samples and the reference samples behave in the same way can this r.m.s. error be used as a stopping criterion. If they do not, then some modification to the training algorithm is needed to improve the generalization capability of the developing network. Modifications worth mentioning in this respect are applying extra random noise to the calibration set measurements, the descending epsilon method24 and the weight decay method.The application of extra randomly generated noise to the input patterns of the calibration samples prevents the learning process from following the data too closely. However, it slows convergence considerably. In the descending epsilon method, weight adjustments are made only for those calibration samples which produce errors larger than a threshold, epsilon, that is gradually lowered during training. The weight decay method uses an extra term in the update of the weight factors that drives them to zero. Only those weights for which the back-propagation correction terms exceed this decay term will survive in the final network. In an application where this method was compared with the descending epsilon method, it was found to be the more efficient of the two.14 Validation The aim of the validation phase is to derive statistical confidence limits for the results that are to be obtained with the neural network model for unknown samples.The only way to do this is to use the calibrated neural network model on standard samples that have not been used in the calibration process. For these samples the differences between the known concentrations and those calculated with the model are used for the required statistical calculations. Hence not all of the available data on the standard samples can be used to train the neural network. Some standard samples should be kept apart for the validation of the resulting neural network calibration model. With a limited number of standard samples this poses the dilemma of how to divide the data for these two purposes.If most of the samples are used for training, the estimate of the accuracy of the results that can be obtained will be poor. If too many samples are used in the validation step, the model will not be as general as it could have been. Table 2 Applications of neural networks for multivariate calibration Analysis Techni q u e * Ref. Protein in wheat NlK I 1 Active ingredients in pharmaceuticals UV/VIS 11 Kf , Ca2+, NO3-, C1-, Cu2+ ISE potentiometry 12 Fe-Ni thin films XRF 13 Fe-Ni-Cr steels XRF 13 H20 in cheese Production process parameters 14 * NlR = near infrarcd; UV/VIS = ultraviolet/visiblc; ISE =: ion-selective electrode; and XRF = X-ray fluorescence. The information contained in the set of all available standard samples is used maximally in the ‘leave one out’ method (LOOM), in which calibration models are calculated for all available standard samples, except one that is left out of the set and that is used for the validation.By repeating this procedure and changing the sample that is left out of the calculation of the calibration model, one obtains the errors for each of the standard samples. The r.m.s. of these errors can then be used as an estimate of the uncertainty in the results produced by the calibration model for unknown samples. Care should be taken not to use the validation samples implicitly in the training phase, e . g . , to decide when to cut off the training, as this would compromise their independence. Results for unknown samples which lie outside the calibra- tion range cannot be trusted.The extrapolation capabilities of neural network models have been shown to be very poor.13 Practical Applications Table 2 shows the practical applications of neural networks in multivariate calibration found in the recent literature. The last entry in Table 2 needs some clarification. Here the neural network does not model a calibration process of an analytical technique, but represents an empirical model of the manufac- turing process of the samples. The model was trained on process data and uses a quality parameter as its output. Once trained it can be used in the quality control of the end product by using it in inverted form to find adjustments for the process parameters so that the required specifications of the end product are met.Conclusion This work clearly shows that neural networks can be used successfully as empirical non-linear multivariate calibration models. Their self-modelling and self-learning properties greatly reduce the effort that is normally necessary to formulate a calibration model in mathematical terms. However, they are not a cure-all recipe in calibration. Owing to their empirical nature, they require large amounts of training data in the form of standard samples that are costly to prepare o r to analyse by classical means and there is no guarantee that they always deliver optimum results. Further, their training is costly in terms of time and computer resources. Finally, even their reliability cannot be guaranteed Calibration models should not be more complicated than necessary to fit the data to the required precision.There is little sense in using a neural network for a linear calibration problem: long training times will result and, if no special measures are taken, the performance at low and high concentration limits will be poor. 1 1 Linear multivariate calibration techniques such as PCR and PLS are much more suitable in this instance. Despite these serious drawbacks, neural networks have proved their worth in specific applications characterized by a fully.328 ANALYST, APRIL 1993, VOL. 118 complicated non-linear relationship between measurement variables and desired results, for which a formal mathematical description is not available, but where obtaining a large number of calibration samples is not a problem.Some of the drawbacks mentioned will be overcome in the near future. The long training times indicated in the papers citcd above were obtained on computers with the Von Neumann architecture that carry out the iterative training calculations serially with binary representations of the signals. Neurocomputer chips have been developed in which the parallel and analogue nature of the processing of the signals by the neurons is incorporated in the hardware together with their learning capability.7s When these devices become generally available, their application in analytical instruments will provide the latter with rapid autocalibration facilities. Another promising development in this respect is the adaptive logic network.26 Here the network consists of a binary tree of nodes that perform simple operations of binary logic such as AND and O R .The tree takes its input from a binary input vector and produces a single bit as its output. Despite the Boolean nature of its input data, real-valued or integral data can be processed if look-up tables are used for their representation in binary form. For computing non- Boolean outputs, several trees are used in parallel to produce a vector of bits representing the output value. Training and use of these adaptive logic networks are claimed to be fast, because no time-consuming floating-point calculations are necessary. Moreover, the hardware implementation of these structures should present no difficulties as they consist of relatively simple combinatorial logic circuits.Owing to their limited depth, these hardware implementations should deliver very high execution speeds. With regard to the objection that results obtained with neural network calibration models cannot be guaranteed owing to their ‘black box’ nature, it can be pointed out that much mathematical research has already been directed to ascertaining the function approximation capabilities of neural networks of various types.27-29 Nevertheless, the fundamental problem of trying to a model a not fully understood relationship remains: one can only hope that all the required information is contained in the training set so that there will be no surprises in data coming from new samples. Some statistical tools such as principal component analysis, cluster analysis and canonical discriminant analysis can help to ascertain the homogeneity of the input data for the neural network.If these techniques are applied to the output values of the hidden neurons on a per sample basis they can even provide a look inside the neural network ‘black bo~’.30 References 3 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Sjiistrom, M., Wold, S., Lindberg. W., Persson, J . , and Martens, H., Anal. Chim. Acta, 1983, 150, 61. Bos, M., Anal. Chim. Acta, 1984, 166. 261. Zupan, J . , and Gasteiger. J . , A n d . Chim. A c f u , 1991, 248, 1 . Janson, P. A . , Anal. Chem.. 1991. 63, 357A. Rumelhart, D. E.. and McClelland, J . L., Parullel Diytrihuted Processing, MIT Press, Cambridge, MA, 1986.Moller, M. F., Neural Networks, in the prcw. Tollenacre, T., Neurul Networks. 1990, 3, 561. Bos. M., and Hoogendam, E . , A n d . Chim. Actu, 1992,267.73. Long, J . R.. Gregoriou, V. G.. and Gemperline. P. J . , Anal. Chem., 1990, 62, 1791. Bos, M.. Bos, A., and van der Linden, W. E . , Anal. Chim. Acta, 1990, 233. 31. Bos, M.. and Weber, H. T., Anal. Chim. Actu, 1991, 247, 97. Bos, A., Bos, M., and van der Linden, W. E., A n d . Cliirn. Actu, 1992, 256, 133. The AspirinlMIGRAfi\iES Software Tools User’s Manual, ReleaJe V5.0, cd. Leighton, R. R., Mitre, McLean, VA, 1991. Lippman, R. P., IEEE Acouhtiu, Speech Signal Process. Mug., 1987, 4, 4. Karnin, E. D., IEEE Trans. Neurul Networks, 1990, 1, 239. Sankar, A., and Mammone, R . J . , in Proceedings of IJCNN, Seattle, 1991. Fahlman, S . E., and Lebiere. C., in Advance5 in Neural Information Processing Syhtems, ed. Touretzky, S . D., Morgan Kaufmann, San Mateo, CA, 1990. Rumelhart, D. E . , and McClelland, J . L.. Parallel Dictributed Proceshing: Explorations in the Microstructure of Cognition, MIT Press, Cambridge, MA, 1986. Jacobs, R. A , , Neurul Networks. 1988, 1, 295. Fahlman, S. E . , A n Empirical Study o f Learning Speed in Back-propagation Networks, Technical Report CMU-CS-88- 162, Department of Computer Science, Carnegie Mellon University, Pittsburgh, 1988. Goldberg, D. E. ~ Genetic Algorithm5 in Seurch, Optimization und Machine Lcwrning. Addison Wesley. Reading, MA. 1989. Yu. Y . H., and Simmons, R. F., in Proceedings IJCNN, International Joint Conference o n Neurul Networks, Sun Diego, IYYO, IEEE Neural Networks Council, Ann Arbor, MI, 1990, vol. 111, p. 167. Wallinga, H., and Bult. K., IEEE J. Solid-State Circuits, 1989. Armstrong, W. W., Liang. J.-D., Lin, D.-K., and Reynolds, S., Experiments u.5 ing Parsimonious Adaptive Logic, Technical Report TR 90-30, Department ot Computing Science, Univer- sity of Alberta, September 1990. Cybenko, G., Math. Control Signals Syst., 1989, 2, 303. Hornik, K.. Stinchcornbe, M.. and White, H . , Neurul Networks, 1989, 2, 359. Hartman, E . J . , Keeler, J. D., and Kowalski, J. M., Neural Comput., 1990, 2, 210. Dennis. S., and Phillips, S . , in The AApirinlMIGRAINES Software Tools User’s Manual, Release V5.0, cd. Leighton, R. R., Mitre, McLean, VA, 1991, p. 68. SC-24, 672. 1 2 Bos, M., Anal. Cizim. Actu, 1978, 103, 151. van Espen, P.. and Adams, F., Anal. Chim. Acta, 1983, 150, 153. Paper 21042656 Received August 7, 1992 Accepted October 2, 1992

ISSN:0003-2654    

DOI:10.1039/AN9931800323

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 329 Molecular Recognition Using Conducting Polymers: Basis of an Electrochemical Sensing Technology* Plenary Lecture P. R. Teasdale and G. G. Walfacet Intelligent Polymer Research Laboratory, University of Wollongong, Locked Bag 8844, South Coast Mail Centre, New South Wales 252 I , Australia Molecular recognition principles are being increasingly used as the basis for analytical technologies. The combination of a molecular recognition approach with conducting polymer materials has been beneficial, particularly in the field of electrochemical sensing. The electrochemical sensing process usually consists of two steps: analyte recognition and signal generation. Conducting polymers are versatile materials in which molecular/analyte recognition can be achieved in a number of different ways, including the incorporation of counter ions that introduce selective interactions, using the inherent and unusual ion-exchange properties of the conducting polymers; the addition of functional groups to the monomers; and the codeposition of metals within the polymer. Specific examples of these approaches are provided.The molecular recognition properties of conducting polymers can be further refined by the application of appropriate electrochemical potentials, which can induce either large or small changes in the chemical interactions that occur at the polymers. This electroactivity, as well as their conducting properties, also provides the basis for the signal generation steps. A number of electronic signals relating t o some chemical or electrochemical change within the polymer can be measured.These include the faradaic electron transfer typically used for electrochemical sensing, the catalysis of the analytically useful electron transfer by the polymer or the analyte, the change in capacitance signals induced by the analyte species and changes in the polymer resistance which can be measured by a recently developed technique. These features, combined with the molecular recognition properties, make conducting polymers a very promising material for electrochemical sensing technology. Keywords: Molecular recognition; conducting polymer The interesting electronic properties of conducting polymers have attracted the attention of materials scientists for over a decade now.' It seems, however, that this fascination with the electronic properties has overshadowed investigations into the unique molecular recognition capabilities of these materials. Molecular recognition entails a degree of selectivity for a particular species and is the basis of all chemical separations and sensing.The versatile molecular recognition capabilities of conducting polymers, coupled with their ability to commun- icate electronically with input controlling and output monitor- ing systems, provide the basis for a powerful new sensing technology. In particular, the electronic conductivity and electroactivity of these materials provides a means by which their molecular recognition capabilities can be fine-tuned. Furthermore, the changes in electronic properties accompany- ing any molecular interactions can be readily monitored.These features combine especially well to fulfil the require- ments of an electrochemical sensor. Basis of Electrochemical Sensing The active material used in an electrochemical sensor should be capable of analyte recognition and conducting polymers are well endowed with this faculty. This selectivity arises through interactions at the polymerholution interface. The recogni- tion interaction should involve electron transfer or the change of some other electronic property of the active material to enable an analytically useful signal to be monitored. It is worth noting that discrimination between competing molecular interactions is also available at this signal generation stage, because certain species will generate signals at particular potentials (thermodynamics) or at certain frequencies in relation to the perturbation signal (kinetics).In electro- chemistry, the signal generated is usually a d.c. current. The total current can be recorded continuously or the current can be sampled and even differentiated as occurs with various pulsed-potential waveforms. However, other signals such as a.c. current, phase-selective a.c. current, resistance and conductivity can also be used. The modern electrochemical sensing system usually consists of three electrodes and an automated potentiostat. The cell set-up is shown in Fig. 1. The three-electrode arrangement enables the desired potential waveform to be applied and the d.c. current is measured separately. When one of the other types of signals is used for the analytical response, potential waveform generators and current sampling instrumentation can be used.The hardware is usually interfaced with computer software that can both control the input and record the output signals for the system. More recently the advantages to be gained by carrying out electrochemical sensing in a flowing solution have been * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 1618, 1992. t To whom correspondence should be addressed. Fig. 1 Electrochemical cell set-up. W = Working electrode, A = platinum gauze auxiliary electrode (annular), R = reference electrode (Ag-AgCI or Ag-Ag+), and S = salt bridge330 ANALYST, APRIL 1993, VOI,.118 Flow in Ref ere nce e iect rode h' 11 Auxiliary electrode Gasket I I I I Pt disc Fig. 2 Flow-through electrochemical cell 1 ' n 2 Fig. 3 Molccular realized. Extensive 3 structures of selected conducting polymcrs research towards developing appropriate cell designs for flow injection (FT) and high-performance liquid chromatography (HPLC) detection has been conduc- ted. A typical electrochemical detector for HPLC or FIA is shown in Fig. 2. Inherently Conducting Polymers A wide range of polymer materials capable of conducting electricity now exist24 and the use of these materials as sensors has been discussed previously .5,6 The level of conduc- tivity varies depending on the molecular structure of the polymer backbone, the degree of doping and the nature of the counter ion species incorporated.1 At present, the most widely used conducting polymers are based on poly(pyrro1e) (l), poly(thiophene) (2) and poly(ani1ine) (3).The chemical structures of these polymers in their conducting forms are shown in Fig. 3. Conducting polymers can be prepared either chemically or electrochemically; both of these involve electron transfer, the prior term describes a solution process and the latter involves depositing the polymer at an electrode. For the polymers shown in Fig. 3 polymerization is an oxidative process. The electrochemical doping of the polymers also involves oxida- tion, which occurs at a lower potential and the polymer is therefore prepared in its doped conducting form. Poly- (aniline) is doped by protonation as well as oxidation, but as it can only be polymerized from acidic solutions then it too is deposited in its conducting state. The oxidative doping of the conducting polymers produces a positive charge that is delocalized over several polymer units.This necessitates the incorporation of an anionic species as a counter ion in order to maintain charge neutrality. Of these three classes of materials poly(pyrro1es) have been most widely used in electrochemical sensing applications for the following reasons: (i) they can be polymerized from aqueous solution, [poly(aniline) can only be polymerized from Table 1 Molecular interactions Interaction typc Ionic Io n ic-di pol e Dipole-dipole (H-bonding) Ion induction Dipole induction Dispersion Example Ho\ I ,c - H I H t acid and poly(thiophenc) from organic solvents]; and (ii) they can be polymerized at low anodic potentials [poly(thiophene) requires a very high potential, although the substitution of functional groups can lower this, e.g., 3-methylthiophene].Both factors enable incorporation of a wide range of counter ions (A-) into poly(pyrro1e). As will be demonstrated later, the nature of the counter ion is a key element in the development of these molecular recognition systems. Another practical advance in conducting polymer technol- ogy that impacts on their use as sensors is that they can now be prepared in a number of forms. Polymer coatings on a wide range of conducting substrates,7 stand-alone films8 or polymer composites.9 Even particles or colloids,'" which can be mixed with other formulations (e.g., paints), have been produced. Another interesting development is the synthesis of dis- persed metal electrodes. 11,12 By using appropriate poly- merization conditions, metals that enhance the electrocataly- tic or electrochemical properties of the sensor can be dispersed through the polymer.The unique attributes of conducting polymers with respect to analyte recognition and signal generation will now be considered. Analyte Recognition As a molecule approaches an electrode surface then the tendency to interact with that surface will depend on the inducement of one or more of the interactions listed in Table 1. As more of these interactions participate the mol- ecular recognition becomes more selective. The appropriate combination of these interactions results in molecular recognition systems with which we are more familiar, e.g., complexation and chelation, enzyme-substrate interactions and antibody-antigen interactions.As with these familiar modes of molecular selection the ability of the surface to discriminate will depend on the species interacting to different extents (thermodynamics) or at different rates (kinetics). Conducting polymers are particularly versatile molecular recognition systems. This versatility arises because the recog- nition properties can be achieved in several different ways. During assembly of the material, preformed molecular recognition sites can be incorporated as the counter ion (A-).ANALYST, APRIL 1993, VOL. 118 331 This has been used to create surfaces capable of metal recognition by incorporating complexing groups such as dithiocarbamates13 or ethylenediaminetetraacetic acid (EDTA)I4 into poly(pyrro1e.s). Biorecognition sites, such as the enzymes glucose oxidase159'6 or urease,l7 have also been incorporated as counter ions.In some work direct incorpora- tion of the mediator and the enzyme has been reported.18 Incorporation of pol ynucleotidesly has been achieved and used for sensing purposes. Even macromolecules such as antibodies have been incorporated directly;20 in this case the use of additives to function as molecular carriers has proved useful. Surfactants21 and colloidal gold22 have been particu- larly useful for this purpose. Another approach involves direct, covalent binding to the polymer after synthesis.23 Some workers have included whole cells during conducting polymer growth.24 Incorporation of such a wide range of species into poly(ani1ine) or poly(thiophene) is not possible as poly- (aniline) must be grown from acidic solutions and poly- (thiophene) from organic solvents.There are only reports of a few instances where active molecules have been incorporated into poly(aniline).25 An alternative to incorporating preformed molecular recog- nition sites is to manipulate the polymer itself to enable it to perform certain recognition tasks. The conducting polymers are inherent anion-exchange materials in their conducting form due to the positive charges delocalized over their backbones. It has been shown that poly(pyrro1e) is a strong anion exchanger with a capacity of 7.1 x 10-4 mol g-1.26 Poly(ani1ine) has also been shown to have strong anion- exchange properties.27 What is unusual about poly(pyrro1e) (PP) is that the ion-exchange selectivity series is determined by the counter ion (A-) incorporated during synthesis.For example: PP-Cl, Br- > SCN- > S042- > T- > Cr042-; and PP-CI04, SCN- > Br- > 1- > S042- > Cr042-. Selectivity series based on ion-exchange processes are determined primarily by: (i) the charge density of the solute species compared with the density and/or mobility of the functional group (highly charged solutes interact poorly with sparsely situated, monovalent functional groups); (ii) the strength of the acid-base interactions of the participating molecules (stronger acids interact more with bases and vice versa); and (iii) the polarizability of the species involved, hard (weakly polarizable species) acids prefer hard bases and soft (highly polarizable species) acids prefer soft bases.With conducting polymers there is another consideration, the charged sites are able to move but they also interact with one another and are, therefore, spaced at discreet and presumably unchanging (unless by further doping or dedop- ing) intervals. Furthermore, the charge is delocalized to a much greater extent, i.e., over several molecules rather than several atoms as is the case usually. These factors undoubtedly contribute to the unique ion-exchange selectivity of conduct- ing polymers and as these properties are imprinted by the anion originally incorporated during synthesis the selectivity series is easily altered.Ion-exchange capacity can also be modified by changing the counter ion. For example, in extreme cases incorporation of larger more hydrophobic molecules such as dodecyl sulfate28 or poly(viny1 sulfonate)29 will totally eliminate any ion- exchange character from the materials. Addition of such molecules can be used to control the hydrophobicity of the material.3" Although not so convenient the molecular recognition properties can also be modified by adding functional groups to the base monomers (1-3). Some of these additions are shown in Table 2.31-35 It is also possible to produce copolymers3~3Y or interpenetrating network layered structures.40 Previous workers41 $2 have used conductive coatings to prevent potential interferents in the sensing process.One of these approaches41 involved deliberate overoxidation of the polymer to produce a material that was electroncially non- conductive but ionically conductive and excluded anionic species. Another report42 described how the polymerization condi- tions could be modified to enhance the charge and size discrimination capabilities of a polymer containing an enzyme. Other workers have used conducting polymers to prevent fouling by preventing product deposition.43 Electroactivity of Conducting Polymers The conducting polymers described here have a further unique advantage that adds to their molecular recognition capabilities. These polymers are all electroactive and hence their physical, conducting and chemical properties can be varied by changing the potential of the polymer.These changes in potential inducc the following redox reactions. These oxidation-reduction processes are well defined and in most cases reversible as the cyclic voltammograms for poly(ani1ine) show in Fig. 4. It is well known that accompany- ing these oxidation-reduction processes are dramatic changes in the electrical resistance of the material.1 These changes can +e- , rn + A- (2) -e- Fully reduced form (leucoemeraldine) +*e-l ,. First oxidation state Conducting form; Emeraldine salt-poly(ani1ine) membrane -2e- Second oxidation state Fully oxidized form (pernigraniline)ANALYST, APRIL 1993, VOL. 118 332 Table 2 Some functionalized pyrroles Monomer Ref. H 31,32 33 1.50 1 1 1.00 c A -1.00 -1.50 - - I I I I I I -0.2 0.0 0.2 0.4 0.6 0.8 1.0 PotentialN H 34 Fig.4 Cyclic voltammogram of poly(ani1ine) (HCl) on 2 mm platinum disc electrode at 100 mV s-* in 1 rnol 1-1 HCI (pH = 0). The large redox couples correspond to oxidation and reduction of poly(ani1ine) between its three conducting forms. The small peaks in between are due to the overoxidation of the polymer H (CH&--C- Ferrocene I H 35 H Table 3 Selectivity factors (response monochloramine/dichloramine) at different electrodes as a function of potential. Eluent, 0.01 mol 1-l Na2HP04 and 0.01 mol 1-1 NaH2P04; flow rate, 1.2 ml min-1; and sample injection, 20 ml of moll-' NaC104 and 10-3 moll-1 (NH4)$304 in eluent. Polymer deposition, 0.20 mol 1-l pyrrole + 0.10 mol 1-1 counter ion, 0.50 mA cm-2, 30 s Electrode 0.00 -0.40 -0.70 GC* 16.0 19.0 9.0 GC-PP-Cl - - - GC-PP-NO3 6.0 15.0 NR?- GC-PPclodecyl sulfate 5.5 NR NR GC-PP-EDTA 2.0 2.0 3.7 * GC = Glassy carbon.j- NR = No monochloramine response observed. only occur if the bulk molecular structure is dramatically altered. Any such change will also alter the polymer surface structure and hence the molecular recognition capabilities. We have used this phenomenon to advantage. It has been demonstrated using liquid chromatography, with conducting polymers as stationary phases, that the application of an applied potential can affect the molecular The activity of preformed molecular recognition sites such as antibodies can even be influenced in this way.20 Other workers47 have demonstrated enhanced chromatographic resolution of polynucleotides with applied potential to a conducting polymer stationary phase.This phenomenon has also been used to control the recognition step with conducting polymer sensors4x designed for chromium determinations and also for chloramine detec- tion systems.49 The dichloramine detection system was employed as a detector in HPLC and was particularly interesting as it involved the use of a composite sensing system consisting of a conducting polymer with mercury dispersed throughout. The mercury facilitates the signal generation process, which involves electron transfer as the chloramines are reduced, and the polymer provides the chemical selectiv- ity. The effect of applied potential on selectivity is shown in Table 3. The application of electrical potential has also been used by other workers to control the activity of enzymes incorporated into conducting polymers.50 6 4 - N 2 $ 0 g - 2 a -4 I I I I I I -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 E N versus SCE Fig.5 Cyclic voltammogram of poly(pyrro1e) at a microelectrode in mol 1-1 KC1. Polymers were grown for 10 min at 2 mA cm-2 from 0.50 rnol 1-I pyrrole and 0 . 1 mol I-' TBAP in acetonitrile. Scan rate = 20 mV s-1 Anion exchange process Cation exchange process More recent work in our laboratories51 on simple poly- (pyrroles) highlights the complex molecular processes of which these systems are capable. By using microelectrodes and cyclic voltammetry, distinct transitions can be observed as this material moves from being an anion exchanger, to a hydrophobic material and then to a cation exchanger (Fig.5). This ability has been used to advantage in the development of membrane systems for controlled transport of ions .5253 By applying pulsed potentials to a membrane in a diffusion cell, ion incorporation or expulsion and hence transport can be initiated and controlled. Such controlled transport can be used to form the basis of electrochemical controlled recognition. The same dynamic behaviour can be expected of poly- (thiophenes) and poly(ani1ines). However, such transitions occur at different potentials as well as at different rates. Furthermore, with poly(aniline), at least three distinct chem- ical states exist. Signal Generation Following the analyte recognition step an electronic signal related to some chemical or electrochemical change should be generated.The intensity of this signal should be related to the amount of target analyte in the operational environment. This signal is usually generated by perturbing the system electric-ANALYST, APRIL 1993. VOL. 118 333 ally and measuring some electrical output or characteristic. Also, some additional recognition (selectivity) might be available at this signal generation step. If the signal is a current flow the potential at which it is induced or the frequency (kinetics) can be used to differentiate signals. Electron Transfer Direct The most traditional of signals employed by the electroana- lytical chemist is that arising from oxidation-reduction of the target analyte either at or on the electrode surface.The signal is generated by application of a linear (d.c.), pulsed or alternator (a.c.) ramp and by sampling the current on the ramp. Each of these potential waveforms has their advantages and disadvantages as discussed elsewhere.54 This mechanism of signal generation has been employed for determination of metal ions55.56 and organic molecules4~ using conducting polymer based sensors. Indirect Catalysis. In some cases the kinetics of the oxidation- reduction process are slow and this results in broad, drawn out (low sensitivity) responses. In such cases redox mediators can be used. The use of these mediators (catalysts) results in a faster over-all electron transfer rate. Several mediators have been incorporated into conducting polymers for this pur- pose .57-59 In fact some workers have shown that poly(pyrro1e) itself will catalyse the oxidation of some small organic molecules such as ascorbic acid.60 Analyte catalysed electrode oxidation.Another mechanism by which an indirect electron transfer signal can be generated is by oxidation of the electrode material in the presence of the analyte. This approach has been used previously for detection of some anions such as sulfide at mercury [see eqn. (4)].54 S2- A1- e . g . , H g e H g S + 2e- (4) PP'.PP+AI- + e- ( 5 ) With conducting polymers a similar mechanism [eqn. ( 5 ) ] can be envisaged. Upon application of an appropriate potential the conducting polymer can be oxidized provided an appro- priate ion, which can be incorporated into the polymer, is available.Heineman and co-workers61762 first introduced this concept in an FI mode. Ye and Baldwin63 used a similar mechanism for signal generation with poly(ani1ine). More recently this scheme has been examined in more detai164.65 and it has been discovered that it is in fact a multicomponent signal as depicted in Fig. 6, where Aicp is the current arising from a change in polymer conductivity as ion exchange occurs, AioD is due to doping-doping, i.e., polymer oxidationh-eduction and Aics is due to a change in the solution Time - Fig. 6 conducting polymers for detection of electroinactive species Schematic diagram showing breakdown of FI signal using conductivity. We have also shown that the selectivity of this signal generation mechanism can be controlled by the counter ion incorporated into the polymer during synthesis; a not unexpected result given the previous discussion on control of the ion-exchange selectivity series by the counter ion incorpor- ated during synthesis.As would also be expected the selectivity of this signal can be modified by use of different eluents, e . g . , using glycine, I NO3- > CH3COO- > PO4+ > C032- > dodecyl sulfate > 1 CH,COO-; and using dodecyl sulfate, NO3- > CH3COO- > I PO43- > C032- > dodecyl sulfate. Also, the selectivity can be modified by changing the frequency of the applied potential pulse. Changes in Capacitance Several voltammetric techniques can be used to derive information related to changes in capacitance of the sensor materia1."?66 The most common are a.c.techniques where capacitive and faradaic components of the signal can be separated. This approach suffers from a lack of specificity in the signal generation step. However, if coupled with extremely specific chemistries, for example antibody-antigen recognition,67@ then perhaps this approach will prove useful. Changes in Polymer Resistance As stated previously, conducting polymers are unique amongst polymers in that they undergo a redox reaction that switches them from a conductive to a less conductive state. This process can be used analytically in two ways. Firstly, determination of gaseous species, which reduce the polymer with a concomitant increase in resistance.6g770 Secondly, more recently studies involving the use of resistometry , which enables resistance-potential profiles to be determined in situ.71 This, like current measuring techniques, can be used to monitor analyte catalysed changes in potential-resistance profiles.65 In some instances monitoring of these changes in resistance is more sensitive than monitoring the concomitant current flow (Fig.7). Microelectrodes The advantages gained from the use of conventional microelectrodes72 are also available with modified micro- electrodes. That is, they can be used in physically small environments and they can be used in more resistive media, and might be implemented to record signals with faster scan rates (or pulse rates). These are particularly important as the range of operational environments can be extended, and for conducting polymer electrodes low ionic strength can be used so that greater ion-exchange selectivity can be achieved.I I 0.2 100 80 c 6o B 40 20 0.1 N 0 ' E a -0.1 < -0.2 I I I I I I -1.0 -0.8 -0.6 -0.4 -0.2 -0.0 0.2 0.4 EN versus SCE Fig. 7 Cyclic voltammogram and cyclic resistogram recorded in 1.0 mol 1-1 KCl for a thin poly(pyrro1e) film deposited on glassy carbon. Scan rate = 20 mV s-1. Polymer prepared by cycling once to +0.65 V from 0.00 V in 0.10 mol 1-l pyrrole and 1.0 mol I - KCl in water. -, cyclic resistogram; - - - -, cyclic voltammogram. Arrows indicate direction of potential scan334 ANALYST, APRIL 1993, VOL. 118 There are also advantages to be gained in the use of microelectrodes during synthesis: electropolymerization can be carried out in more resistive solutions, a greater range of counter ion species can be incorporated and a wider range of solvents can be employed.73 Conclusions The field of conducting polymers has provided the sensor technologist with a unique series of molecular building blocks.These building blocks can be assembled in such a way that analyte recognition (via chemical interactions and unique signal generation) and quantification can be achieved. The number of applications reported for this sensing technology is growing at an extraordinary rate. Current studies are aimed at producing these sensing materials in an economical form either as reusable (regener- able) or disposable electrodes. G. G. W. acknowledges the support of the Australian Research Council in the form of a QEII Fellowship. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 References Handbook of Conducting Polymers, ed.Skotheim. T. A., Marccl Dekker, New York, 1986, vols. 1 and 2. Diaz, A . F., and Lacroix, J. C., New J. Chem., 1988, 12, 171. Xiachong, L., Jiao, Y., and Li, S . , Eur. Polym. J., 1991, 27, 1345. Bartlett, P. N., Gardner, J. W., and Whitaker, R. G., Sens. Actuators A , 1990, 21,911. Imisidcs, M. D., John, R., Rilcy, P. J., and Wallace, G. G., Electroanalysis, 1991, 3, 879. Bidan, G., Sens. Actuators B. 1992. 6, 45. Gc, H., Ph.D. Thesis, University of Wollongong, 1990. Mirmasheni. A., Price, W. E., Wallace, G. G., and Zhao, H., J. Int. Muter. Syst. Struct., 1993, 4, 43. Orata, D., and Buttry, D. A., J. Electroanal. Chem., 1988,257, 71. Eizadeh, H., Spinks, G., and Wallace, G.G., Muter. Forum, in the press. Yoneyama, H., Shoji, Y., and Kawai, K., Cltem. Lett., 1989, 1067. Ge, H., Zhao, H., and Wallace, G. G., Anal. Chim. Acta, 1990, 238, 345. Yuping, L., and Wallace, G. G., Anal. Lett., 1989, 22, 669. Yuping, L., and Wallace, G. G., J . Electroanal. Chem., 1988, 247, 145. Belanger, D. J., Nadreau, J., and Fortier, G., J . Electroanal. Chem., 1989, 274, 143. Couves, L. P., and Porter, S. J., Synth. Met., 1989, 28, 761. Adelojou, S., Shaw, S . , and Wallace, G. G., J. Electroanal. Chem., submitted for publication. Iwakaura, C., Kajiya, Y., and Yoneyama, H.. J. Chem. SOC. Chem. Commun., 1988, 1019. Shimidzu, T., React. Polym., 1987, 6, 221. Hodgson, A. J., Lewis, T. W., Maxwell, K. M., Spencer, M. J., and Wallace, G. G., J .Liq. Chromatogr., 1990, 13, 309. John, M. J., John, R.. Wallace, G. G., and Zhao, H., in Electrochemistry of Colloids and Dispersions, eds. Mackay, R. A,. and Texter, J., VCH, Wcinhcim, 1992, p. 235. Cardcn, P., Hodgson. A. J., John, R.. Spencer, M. J., and Wallace, G. G., in Electrochemistry of Colloids and Disper- sions. eds. Mackay, R. A.. and Texter, J., VCH, Weinhcim, 1992, p. 245. Schuhmann, N., Lammert, R., Uhe, B., and Schmidt, H. L., Sens. Actuators B, 1990, 1, 137. Dcshpandc, M. V.. and Hall, E. A., Biosens. Bioelectron., 1990, 5, 431. Shinohara, H . , Chiba, T., and Aiazawa, M., Sens. Actuators, 1988, 13, 79. Gc, H., and Wallace, G. G., React. Polym., 1992, 18, 133. Teasdale, P. R., and Wallace, G. G., Polym. Int., 1992, 29. Ge, H., and Wallace, G.G., J. Chromatogr., 1991, 588, 25. Naoi, K., Lien, M., and Smyrl, W. H., J. Electrochem. Soc., 1991, 138,440. Chriswanto, H., and Wallace, G. G., Chromatographia, sub- mitted for publication. 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 Pickup, P. G., J . Electroanal. Chem., 1987, 225, 275. Ashraf, S . , Gilmore, K. G., Ge, H., Too, C. O., and Wallace, G. G.. J . Electroanal. Chem., 1992,340, 41. Havinga, E. E., Hoeve, W., Meijer, E. W., and Wynberg, H., Chem. Muter., 1989, 1, 650. Iyoda, T., Ando. M., Kancko. T., Ohtarsi, A., Shimidzu, T., and Honda, K., Tetrahedron Lett., 1986, 27, 5633. Inagaki, T., Hunter, M., Yang, X. Q., Skotheim, T. A., and Okamoto, Y., J. Chem. Soc., Chem.Commun., 1988, 126. Sato, M., Shimidzu, T., and Yamauchi, A., Mukromol. Chem., 1990, 191, 313. Wei, Y., Hariharan, R., and Patel, S. A., Macromolecules, 1990, 23, 758. Nishizawa, M., Sawaguch, T., Matsue, T., and Uchida, I . , Synth. Met., 1991, 45, 241. Wong, Y., and Rubner, M. F., Macromolecules, 1992,25,3284. Iyoda, T., Toyada, H., Fujitsuka, M., Nakahara, R., Tsuchiya, H., Honda, K., and Shimidzu, T., J . Phys. Chem., 1991, 95, 5215. Freund, M., Bodalbhai, L., and Brajtcr-Toth, A., Talanta. 1991, 38, 95. Hammcre, M., Schuhmann, W., and Schmidt, H. L.. Sens. Actuators B , 1992, 6, 106. Wang, J., and Li, R., Anal. Chem.. 1989, 61, 2809. Ge, H., and Wallace, G. G., Anal. Chem., 1989, 61, 2391. Ge, H., and Wallace, G. G., J . Liq. Chromatogr., 1990, 13, 3261.Ge. H., Teasdale, P. R., and Wallace, G. G., J. Chromatogr., 1991, 544, 305. Dcinhammer, R. S . , Shimazu, K., and Porter, M. D., Anal. Chem., 1991, 63, 1889. Spencer, M. J., Teasdale, P. R., and Wallace, G. G., Anal. Chim. Acta, 1992, 263, 71. Lin, Y., and Wallace, G. G.. Anal. Chim. Acta, 1992,263,71. Aizawa, M., Yabuki, S., and Shinohara, H., Stud. Org. Chem. (Amsterdam), 1987, 30, 353. John, R., and Wallace, G. G., in preparation. Price, W. E., Wallace, G. G., and Zhao, H., J. Electroanal. Chem.. 1992,334, 111. Price, W. E . , Wallace, G. G., and Zhao, H., Polymer, 1993,34, 16. Bond, A. M., Modern Polarographic Methods in Analytical Chemistry, Marcel Dekker, New York, 1980. Imisides, M. D.. and Wallace, G. G., J. Eleciroanal. Chem., 1988, 246, 181. Lin, Y., and Wallace, G. G., J . Electroanal. Chem., 1988, 247, 145. Bull, R. A., Fan, F. R., and Bard, A. J., J . Electrochem. Soc., 1983, 130, 1636. Elzig, A., van der Putten, A., Visscher, W.. and Barendrecht, E., J. Electroanal. Chem., 1987, 233, 11 3. Holdcroft, S., and Funt, B. L., J . Electroanal. Chem., 1988, 240, 89. Saraceno, R., Pach, J. G., and Ewing, A. G., J . Electroanal. Chem.. 1986, 197, 265. Ikariyama. Y., and Hcineman, W. R., Anal. Chem., 1986, 58, 1803. Ikariyama, Y., Galiatsatos, C., Heincman, R., and Yamauchi, S . , Sens. Actuators, 1987, 12. 455. Ye, J.. and Baldwin, R., Anal. Chem., 1988, 60, 1979. Sadik, O., and Wallace, G. G., Electroanulysis., in the press. John, R., Ph.D. Thesis, University of Wollongong, 1992. Bard, A. J., Electrochemical Methods, Wilcy, New York, 1980. John, R.. Smyth, M. R., Spencer, M. J.. and Wallace, G. G., Anal. Chim. Acta, 1991, 249, 381. Gardies, F., Martelet, C., Colin, B., and Madrcnel, B., Sens. Actuators, 3989, 17, 461. Hanawa, T.. and Yoneyama, H., Bull. Chem. Soc. Jpn., 1989, 62, 1710. Hanawa, T., and Yoneyama, H., Synth. Met., 1989, 30, 341. John, R., Talaie, A., Flctcher, S . , and Wallace, G. G., J . Electroanal. Chem. ~ 1991. 319, 365. Stojek, Z., Mikrochim. Acta, 1991, 11, 353. John, R., and Wallacc, G. G., J. Electroanal. Chem., 1991,306. 157. Paper 2105446 I Received October 12, I992 Accepted January 8, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800329

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 335 Present State of Fabrication of Chemically Sensitive Field Effect Transistors* Plenary Lecture Karel Domansky and Jiii Janata Molecular Science Research Center, Pacific North west Laboratory, Richland, WA 99352, USA Mira Josowicz Inst. f. Ph ysic, Universitat der Bundeswehr Miinchen, D 8014, Neubiberg, Germany Danuta Petelenzt HEDCO Microfabrication Facility, University of Utah, Salt Lake City, UT 841 12, USA Fabrication of reliable chemically sensitive field effect transistors still poses significant technical difficulties. Problems related t o integrity of the solid state part of the device, integrity of the final sensor package, definition of the sensitive areas in multisensors and casting of the selective membranes have been addressed. Several strategies, such as employing photosensitive polyimides and electrochemically formed encapsula- tion, were investigated in an attempt t o eliminate procedures at the individual device level.Performing as many fabrication steps as possible at the wafer level leads t o higher yields and reduces fabrication costs by introducing automated processes. Electrochemical testing procedures at different fabrication stages of chemically sensitive field effect transistors are also presented. Keywords: Microsensor fabrication; passivation; chemically sensitive field effect transistor encapsulation; ion selective field effect transistor testing; multisensor Chemically sensitive field effect transistors (CHEMFETs) are now more than two decades old. Many theoretical and practical papers have been published and this topic has been extensively reviewed.'-2 One reason for its popularity is the general trend towards the miniaturization of chemical sensors.Another is that it deals with modern analogues of familiar ion-selective electrodes and other time-tested potentiometric devices .3 Traditionally, there have been four main problem areas in the fabrication of these devices: (i) the integrity of the solid-state part of the device; (ii) the integrity of the final sensor package; (iii) the definition of the sensitive areas in multisensors; and (iv) the adhesion of the selective mem- branes to the solid-state substrate. These topics will be addressed separately in this paper. Some of these problems can be dealt with at the wafer level whereas solutions for the others must be addressed at the individual device level.Obviously, from the point of view of fabrication cost it is important to perform as many fabrication steps as possible at the wafer level. A critical prerequisite of a successful potentiometric sensor is that the transducer part, the amplifier, has sufficiently high input impedance. This requirement has its origin in the fact that the primary sensing signal resulting from the interaction of the chemical species with the sensing layer is weak (of low power). This is an important distinguishing facet of potentio- metric sensors as compared with other electrochemical, particularly amperometric, sensors. As a result, the require- ments on encapsulation integrity of potentiometric microsen- sors are higher than on those of the other types.In the past, different research groups have approached this problem in different ways with various degrees of success. 'I'he purpose of this paper is to summarize our experience in the preparation of viable potentiometric microsensors. It should be understood that views and conclusions presented * Prescnted at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18. 1992. t To whom corrcspondencc should be addressed. here are not the only possible solutions and that different sensing situations might require different fabrication approaches. The paper is divided into four sections that deal with most general problems of CHEMFET fabrication. Operations that can be performed at the wafer level are discussed first followed by the fabrication approaches accomplished at the individual device level.Most of the information presented here has been described in other publications in which the full details can be found. Passivation Integrity of Silicon Nitride Amorphous silicon nitride (Si3N4) films are widely used in integrated circuit technology as passivation layers,4 as masks for selective oxidation of silicon and as gate dielectric materials. For the last two applications, in metal-nitride- oxide-silicon structures, silicon nitride is generally deposited by a high-temperature chemical vapour deposition (CVD) technique; such films have high uniformity and a low processing cost.5,h However, these films are deposited on substrates in a state of stress.As a result, Si3N4 layers are sensitive to subsequent treatment such as annealing or wet etching. The existence of small cracks in the top dielectric Si3N4 film, which simultaneously acts as a passivation layer, might not be particularly important in conventional device applications. However, in solid-state chemical sensor applications, in which a part of the chip is exposed to conducting liquids, integrity of the top layer is of paramount importance. Cracks and/or pinholes invariably downgrade the input characteristics, introduce baseline instability and, ultimately, can lead to a catastrophic failure. 1 The formation of these imperfections severely decreases the yield of the microfabrication process, sometimes by as much as 30%.For an explanation of the root cause of this problem and its solution, the multilayer structure illustrated in Fig. 1 must be considered. It is known that a layer of one material grown or vapour deposited on another material with different mechan- ical properties is often in a state of stress. When only certain336 ANALYST, APRIL 1993, VOL. 118 7 I I I Gate\ I i A __ \ Drain Substrate contact / / Field (b) Field oxide ' Source ' ' Drain ' Gate Silicon substrate Fig. 1 (a) Schematic diagram of the chip investigated with the top Si3N4 layer (drawing to scale). (b) Cross-section (A-A) of the structure. determined by the profilometer, showing the steps in source, drain and gate areas (drawing not to scale but actual vertical dimensions are given) parts of a substrate surface are covered by a film, high stress is known to be concentrated in the substrate near the film edge.7 This stress is induced by the film-edge force originating from the thermal expansion difference between the film and substrate or from the intrinsic stress of the film.* The stress measured at room temperature in thermally grown Si02 is compressive and increases with the temperature of oxidation.9 When the measurement of stress is made at elevated tempera- tures (above 965 "C) the films become almost stress free due to onset of the viscoelastic behaviour of SO2. 10 On the other hand, as the thermal expansion coefficient of Si3N4 (== 2.5 x 10-6-3.85 x 10-6 "C-1)11 is reported to be almost the same as that of Si (= 3.0 x 10-6-4.5 x 10-6 "C-1),12 the interfacial stress is regarded to be the inherent intrinsic stress produced during the deposition.The room- temperature stress for high temperature atmospheric pressure CVD processes was reported to be tensile and decreasing with increasing deposition temperature. By measuring the curva- ture of wafers at elevated temperatures, this inherent intrinsic stress was found to be about 15% greater than the stress at room temperature. 12 Considering this, it is concluded that the Si02 film, after the Si3N4 deposition at 870"C, is in compression because this temperature is not sufficiently high for the viscoelastic behaviour to develop in the SiO2 layer while the Si3N4 is under tension. When cooled to room temperature, the Si02 layer becomes more compressive and the Si3N4 layer less tensile.As a result, a significant stress along the edges of the structure in Fig. 1 develops. This stress is influenced by the cooling rate and is enhanced boy the uneven total thickness of SO2, which ranges from 890 A (gate oxide) to about 4500 8, (field oxide and gate oxide). Chemical etching is synonymous with corrosion. It is known13 that mechanical, particularly tensile, stress acceler- ates the corrosion rate. This phenomenon is known as stress corrosion and is well documented in metals and alloys. It is reasonable to speculate that the increased etching rate of stressed Si3N4 can be attributed to an analogous process. When a wafer is cooled slowly after the CVD step, there is probably more time for the intermediate SiO2 layer to suppress or relieve the stress in the top Si3N4 film.As far as the influence of the annealing is concerned, the annealing temperature of 950°C is too low for the Si02 to become viscoelastic. Consequently, there is no reduction of the interfacial stress between these two layers. Several studiesl4--16 have shown that silicon nitrides deposited by the CVD techniques contain an appreciable amount of hydrogen, which is bonded either to silicon or nitrogen. Annealing at temperatures between those used for deposition and 950 "C causes a significant loss of hydrogen ,I7 which results in changes of both mechanical and electrical properties. In this way, the stress behaviour and consequently the resistance of the Si3N4 to developing pinholes in the buffered hydrofluoric acid can be altered.The density of pinholes in the top passivation (silicon nitride) layer increases significantly with the cumulative time of exposure to BHF. At the same time the greatest occurrence of pinholes appears along the edges of the substrate structure where the stress is concentrated. For this reason, the observed phenomenon is attributed to preferential etching of Si3N4 in locations under stress. The rate of cooling of Si3N4 from the deposition tempera- ture has a major influence on the integrity of the Si3N4 film during etching in BHF. By using a cooling rate of 1-2 "C min-1 the yield is effectively increased to over 90%. Annealing at 950 "C does not appear to increase the density of pinholes but somewhat alters the behaviour of the Si3N4 film in BHF.The cooling rate from the annealing temperature has only a negligible effect on the formation of pinholes in Si3N4. However, it might alter the pH sensitivity of the ion sensitive field effect transistor (ISFET) with a bare silicon nitride gate. Definition of the Sensitive Areas One of the major difficulties encountered with chemical sensors is because the area of the chip housing the electrical circuit and the chemically sensitive area of the sensor must be separated from each other. The former must be protected from the conducting liquids while the latter must be exposed to the very same liquids. Moreover, the performance of the multifunctional CHEMFET is primarily determined by the properties of the chemically sensitive membranes, which must be deposited over separate but closely spaced gates of the FET.Those areas must be geometrically defined but sepa- rated from each other. A substantial part of this process can be accomplished at the wafer levell* by patterning a thick insulating material in such a way that cylindrical openings, into which the chemically sensitive material is deposited, are created above individual gates. Materials used for this area definition have to be chemically inert in order to eliminate possible chemical interactions with the sensor material. Equally important is their good mechanical adhesion to the top passivation layer, usually Si3N4. For most CHEMFETs there is a minimum thickness required for the sensing membrane. For ISFETs it is 40-50 pm. Thus, the minimum depth of the well over the gate must be at least 50 pm.Historically, epoxy (two-component mixture of Epon 825 and Jeffamine D-230) was used in sensor encapsulation. The well was formed on each chip by individual hand deposition under the microscope. The geometry of theANALYST, APRIL 1993, VOL. 118 337 Gross Encapsulation Areas that need to be protected on CHEMFETs are the bonding wires, the electrical contacts on the device and the edges of the silicon chip. In order to ensure the sensor integrity, the selection of the type of encapsulant material, and also its adhesion to the defined regions of the substrate, are important. These factors will dictate the long-term stability of the device. The manual procedures for encapsulation of sensors present many challenges to the common user.Introduction of new encapsulation processes based on electrochemical deposition can be viewed as new solutions to an old problem. This process was developed for encapsulation of a sensor-package that is defined by the CHEMFET die attached to a mechanical support such as a TO5 standard transistor header or a catheter, with all the bonding contacts made of noble metals.19 In this case an electrochemically generated poly- (oxyphenylene) was used. The process starts with an electro- chemical pre-treatment of the area to be encapsulated20 as the electroactive areas that need to be protected are usually not clean. First of all, the regions of the thin noble metals should be free of oxide films of metals such as titanium, tungsten or chromium all of which are used for the deposition of thin films of platinum or gold.The oxide on top of the thin films can result in poor adhesion of the electrochemically formed encapsulation material. For the same reason, the electro- chemical encapsulation process cannot be applied to alumi- nium or silicon directly. If a silicon encapsulation is required, an additional electroless metallization is necessary. Poly- (oxyphenylene) is chemically inert after thermal curing and undergoes almost no oxidation or hydrolysis in either basic or acid media. The generalized electrochemical encapsulation process is carried out by the following steps:19 (i) Electrochemical , pre-treatment of the thin-metal films deposited on the sensor chip, which should be protected by an encapsulant.2O (ii) In situ electrochemical deposition of poly(oxypheny1ene) on all electrically connected areas within the sensor package that play the function of microelectrodes connected in parallel.(iii) Thermal curing of the freshly deposited poly( oxyphenylene) , which converts it to an electrical and chemical insulation barrier. (iv) Electroless co-deposition (coating deposited in a bath without electric current) of palladium-gold alloy on the edges of the silicon chip.2' ( v ) Encapsulation of the edges of the silicon die with poly(oxypheny1ene) by repeating steps 2 and 3. More detailed explanation of each of the steps is given below. Step 1. Electrochemical pulse etching of the surface between 0 and +2 V versus Ag-AgN03-3 mol 1-1 KCI reference electrode in solution containing 0.08 mol 1-1 ethylenediaminetetraacetic acid (EDTA), 5.2% ammonia solution and 2.7 x 10-4 moll-' H202 for 5-10 min.It is then followed by cycling of the applied potential between 0.4 and -0.4 V versus the same reference electrode in 1.0 mol 1-1 KN03 for 10 min. The surfaces prepared by this procedure remain clean for a period of at least 24 h. Step 2. The poly(oxypheny1ene) (POP) is electrodeposited from freshly prepared solution containing 0.23 mol 1-1 of 2-allylphenol (Merck), 0.4 moll-' of allylamine, 0.2 moll-* of butylcellosolve [ethyleneglycolmonobutylether in a water- methanol mixture (1 + 1 by volume)]. [Caution: Allylamine is highly toxic both on inhalation and through skin contact and 2-allylphenol is an irritant; they are both flammable. Extra safety precautions should be followed when handling these materials.] This electropolymerization solution is unstable (it can be used only for a few hours), therefore it is necessary always to prepare it fresh.The electrodeposition is carried out in a one-compartment cell at room temperature by applying 4 V from a constant voltage power supply between the cathode (a platinum coil of Fig. 2 Photograph of a dual gate chip for fabrication of ISFET. The top encapsulation, which defines 'chemically active' areas on the chip, is a 65 pm thick photosensitive polyimide. The dimcnsions of the oval openings are approximately 230 x 600 pm well varied from chip to chip and its preparation was very labour intensive and required good mechanical skills. In an effort to minimize this labour intensive encapsulation and as the first step towards automated fabrication, thick photosensitive materials were investigated.Two materials proved to be particularly suitable. (i) Vacrel (DuPont) polymer, which was developed for the printed circuit board fabrication process and belongs to the acrylic-epoxy family, and (ii) Selectilux (Ciba-Geigy) photosensitive polyimide. Vacrel provides good environmental protection to the under- lying material. However, its operating temperature was limited to approximately 120 "C. The thickness of the commer- cially available material was 100 pm. The layer was laminated to the substrate, exposed and subsequently developed in aqueous solution. The lateral resolution that could be obtained with this material was between 125 and 150 pm.Thus, this material could be used whenever a deep well was required (>lo0 pm, as only a few layers could be laminated sequentially). A better aspect ratio (coating thickness : lateral separation) could be obtained using a photosensitive polyimide. As of today, there are only a few companies producing such polyimides (Toray, DuPont and Ciba-Geigy), but only the last two companies provide material with sufficient viscosity to allow at least a SO pm cured layer to be obtained. Selectilux HTR3-200 (Ciba Geigy) was found superior for sensors applications. The material was spun-on, pre-baked, exposed, developed and cured. Openings as small as 55 pm were patterned in a 60 pm fully cured polyimide. Thus the resulting aspect ratio was better than 1. The practical thickness limit was 70 pm for fully cured polyimide (Fig.2). A thicker layer exhibited adhesion failure. Maintaining a very low tempera- ture gradient during heating to the curing temperature of 400-425"C, and during cooling to room temperature is essential. Fully cured polyimide is not attacked by solvents such as acetic acid, acetone, carbon tetrachloride, ethanol, hexane, methanol, aliphatic hydrocarbons, toluene, xylene, dimethylformamide and cresol, (non-oxidizing acids) but is attacked by strong organic or inorganic bases, e.g., NaOH, KOH, hydrazine and ethylene-diamine, and also by fuming nitric acid, hot concentrated sulfuric acid, antimony tri- chloride and arsenic trichloride. Additionally, polyimide can be used at elevated temperatures of up to 350-400"C without observable deterioration.This is the last step in ISFET fabrication that is performed at the wafer level. All additional steps are carried out on individual devices.338 ANALYST, APRIL 1993, VOL. 118 2 \ 4\ A B Fig. 3 Cross-section of the sensor with gates A and B. 1, Silicon chip; 2, epoxy encapsulation formed according to ref. 23; 3 , blank plasticized PVC membrane; and 4, area of the membrane doped with ionophore. The Si3N4-SiOZ gate insulator, which is approximately 0.2 pm thick. is not shown a large area) and the substrate (anode). The deposition of approximately 1.5-2 pm of the POP layer requires approxi- mately 1-1.5 h. Step 3. The electrodeposited films of POP are cured at 150°C for 30 min. The resistivity of an approximately 2 pm POP layer is 4.5 x 1012 B cm.After two weeks of continuous immersion in 0.1 mol 1-1 NaCl solution it decreased to 7.3 x 1011 B cm. After this period no change in resistivity was observed. Dry films (>2 pm) exhibit dielectric breakdown in excess of 300 VDC. Step 4 . The electroless plating of silicon with Pd or Pd-Au co-deposition follows the electrochemical exchange reaction in an acidic solution. As Pd is a more noble metal than Si, Pd is deposited on the surface due to dissolution of Si. This procedure is based on a chemical activation step that follows the electroless metallization, which was originally designed for autocatalytic deposition of Au and Pa onto n-GaAs in acidic media.21 It can also be applied to the electroless modification of the silicon. The procedure requires preparation of the following solutions.Solution 1, 0.06 mol 1-1 NH,CI, 0.12 mol 1-1 KHF2, 2 mol 1-1 KCI, 0.1 mol I-' citric acid in de-ionized water. Solution 2, 0.02 mol 1-1 KAU(CN)~ dis- solved in de-ionized water and Solution 3, for 900 ml of this solution 0.3 g of PdCI2 is dissolved in 9 ml of concentrated HCI. To this 5 ml of de-ionized water, 864 ml of glacial acetic acid and 22 ml of HF (40%) are added. Shortly before applying the electroless metallization, 1 ml of solution 2, 0.1 ml of solution 3 and 40 ml of solution 1 are mixed. The deposition temperature is kept constant at 50°C for 10 min. The co-deposited Pd-Au alloy is desirable for the subsequent electrochemical deposition of POP and forms a good ohmic contact to Si. Allylamine can be replaced by the same molar amount of ammonia solution.22 Tn this case the resistivity of the POP is lower by one order of magnitude and the uniformity of the POP coating also increases.The major advantage of electrochemical encapsulation is that the de- posited encapsulant is geometrically conformal, i.e., it follows minute details of the contours of the electroactive area. This is particularly important in the fabrication of multisensors. Multiple Sensing ISFETS An improved procedure for preparation of multiple-gate ISFETs with polymeric membranes has been developed? It is based on the simple idea that most ion-selective membranes consist of a common matrix made of the polymer and plasticizer and that the selectivity is imparted on the mem- brane by the presence of a low relative molecular mass ionophore 24 Preparation of the selective material for an TSFET then proceeds in two steps.First, a blank membrane containing only the polymer and the plasticizer is applied to the encapsulated FET chip. It envelops the entire probe. In the second step, the low relative molecular mass electroactive components of the membrane (ionophores andor other additives) are introduced by locally doping this blank mem- brane (Fig. 3). The main benefit of this simple procedure is a continuous and therefore mechanically stable membrane without electrical shunts between the membrane and the Fig. 4 Photograph of the chip, taken through the drop of water under the microscopc, during the pinhole test. Dark circles arc the bubbles of hydrogen cvolving indicating the locations of thc pinholes encapsulant.This approach for the most part alleviates the problem of adhesion of the membrane. However, the overall performance of the ISFET will benefit from specially formu- lated, well adhering membranes that have been developed recently.25 The idea of creating ion-selective membranes by doping a standard plasticized poly(viny1 chloride) (PVC) tubing has been used previously.26 Despite the fact that the composition of the membranes and of the swelling solutions have not been optimized, the membranes have already shown a respectable sensitivity. The blank membrane solution was made by dissolving PVC and plasticizer at a 30:70 mass ratio in doubly distilled cyclohexanonc. Solutions of the electroactive compounds are prepared by dissolving the ionophore in a 1 + 1 mixture of toluene and light petroleum.This mixture swells but does not dissolve previously cast membranes. After the doping has been completed the excess solvent is removed in a mild vacuum. Membranes prepared by this procedure show com- parable behaviour to corresponding macroscopic ion-selectivc electrodes. The average lifetime of these devices is 2 months.23 Testing Electrochemical testing is the most direct procedure for evaluating the integrity of the device; it can be performed at any stage during fabrication of the CHEMFET as well as for the final device. The procedure was described in the late 1970s' but it is obvious that it is often missing from studies of these devices. The test consists of applying a sufficiently negative potential to the device so that the electrolysis of water and the evolution of hydrogen would result and these effects would be observable under a microscope and/or recorded as a current-voltage relationship.Solutions used for these tests are electrolytes, usually aqueous NaCl. The potential at which the formation of the hydrogen bubbles occurs, indicating an electrical failure, is not critical. It is usually > -1.5 V against any common metal. The most important information is the location of the bubble, which can best be observed in the early moments of its formation. The test depends on the geometry of the object under study. If it is carried out at the wafer level it is typically performed as follows. The wafer is allowed to hydrate for several hours.After the hydration period it is placed under a microscope in a special holder containing NaCl solution and a voltage of -6 VANALYST. APRIL 1993, VOL. 118 339 is applied between the aluminium electrode on the back side (cathode) of the wafer and a metal probe tip that touches the surface of the test solution. On establishing contact with the solution, hydrogen bubble formation is observed at the pinhole locations. The leakage test performed on the final device is identical in principle, only the arrangement of the immersion of the device in the test solution varied with the shape and size of the final package. A visual leakage test is usually sufficient to locate the source of the electrical problem (Fig. 4). However, it is also possible to perform a more rigorous electrochemical test in which the device is connected as the working electrode in a normal three-electrode poten- tiostatic experiment, using a proper reference electrode.Additional information can be deduced from the value of the breakdown voltage and from the shape of the current-voltage curve. In general, a fault at the noble metal (e.g., Pt or Au) will yield a current-voltage curve, the breakdown voltage of which is more negative than that of Si. A shallow current- voltage curve is usually indicative of a leak obstructed by a long and narrow insulating crack in the encapsulation. These types of fault are, however, clearly distinguishable during the visual test. Conclusions In addition to the major technical issues discussed in this paper there are several other practical points that need to be addressed.First is the preconditioning of the completed device. Like some other solid-state devices CHEMFETs should undergo a so-called ‘burn-in’ period during which time some of the physical parameters stabilize as they reach their equilibrium values. In CHEMFETs the operational parameter that changes most is the threshold voltage. The primary reason is the stabilization of the charge structure within the gate region of the transistor. Passage of a small (tl mA) drain-to-source current for a period of 24-48 h usually accomplishes this task. Some reports of large baseline drifts of ISFETs of the order of several (2-5) mV h-1 can be attributed to the failure to burn-in the devices. The second type of chemical preconditioning consists of exposure of the sensitive layer to a solution similar to that in which it is expected to operate, e .g . , a calibrating solution for an ISFET. Both preconditionings can usually be accomplished at the same time. Obviously, the lifetime of the sensitive layer must be longer than the necessary preconditioning period of the device. Because CHEMFETs are usually operated in a constant current (feedback) mode the actual value of the threshold voltage is not important as long as it lies within the limits of the gate voltage applicable, typically from 0 to +3 V for n-channel transistors. For the same reason, the actual value of the transconductance is also not critical; a typical value of 1 FA mV-1 is sufficient. A properly preconditioned mem- brane ISFET will have a stability of k 1 mV per week,23 and a usable lifetime of over 1 month.Bare gate silicon nitride ISFETs have been continuously operated over 6 months. Under those conditions any problems are usually caused by the failure of the reference electrode. There is no practical limit on the lifetime of bare CHEM- FET chips. This is important from the point of view of fabrication and storage of the unfinished devices. The in-use lifetime of the CHEMFET depends on the type of chemically sensitive layer. It can be years for bare silicon nitride pH ISFETs to months for membrane ISFETs and days for enzyme FETs (ENFETs). Lifetime data for various gas FETs are not yet available. There are some reports of severe light sensitivity of these devices.The magnitude of this problem depends on the processing conditions during the fabrication of the basic devices. Although this problem has not yet been universally solved, in our experience it can be avoided by the choice of the proper calibration and operating conditions. Temperature sensitivity is another issue often quoted against ISFETs. The majority of potentiometric measurements of any type are carried out under isothermal conditions. In a potentiometric measurement it is difficult to separate the source of the error because of the effect of variable temperature on the reference electrode and on the indicator device. It is fair to say that some types of CHEMFETs, such as ISFETs, have reached the stage of a possible commercial application. Others are still in a research and development phase but already show every promise to play an important role in the future sensing domain.In ours and other laboratories ISFETs that consistently perform as well or better than conventional ISEs in a variety of measuring applications have been prepared and tested. The remaining developments in the area of ISFETs are expected mainly in the development of multigate ISFET probes with the associated electronics on the same chip. A logical step would be the application of advanced data processing techniques, such as chemometrics and neural networks. Automated packaging and membrane deposition techniques based on the approaches described in this paper should drastically reduce the cost of manual operations in their preparation. The obvious question that is frequently asked is ‘Why are CHEMFETs not commercially available?’ In our opinion the most likely reasons are of a marketing rather than of a technological nature.Manufacturers of conventional sensors have not yet identified a commercially attractive application that would warrant the introduction of a radically new sensor despite the fact that it would result in a superior performance. Results described in this paper have been obtained over a period of sevcral years under the sponsorship of National Institute of Health (USA), Bundesministerium fur Forschung und Technologie and Institut fur Medizintechnik Munster, both Germany. References 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20 Janata, J., in Solid State Chemical Sensors, eds.Janata, J., and Huber, R. J., Academic Press, New York, 1985. Sibbald, A., J . Mol. Electron., 1986, 2, 51. Janata, J., Chem. Rev., 1990, 90, 691. Domansky, K., Petelenz, D., and Janata, J., Appl. Phys. Lett., 1992,60, 2074. Rosler, K. S . , Solid State Technol., 1977, 20, 63. Kern, W., and Rosler, R. S . , J . Vac. Sci. Technol., 1977, 14, 1082. Isomae. S . , J . Appl. Phys., 1985, 57. 216. Isomar, S . , J . Appl. Phys., 1981, 52, 2782. Jaccodine, R. J . , and Schlegel, W. A., J. Appl. Phys., 1966,37, 2429. EerNisse, E. P., Appl. Phys. Lett., 1979, 35, 8. Tokuyama, T., Fujii. V., Sugita, V., and Kishino, S . , Jpn. J. Appl. Plzys., 1967, 6. 1252. Tamura. M., and Sunami, H., Jpn. J. Appl. Phyy., 1972, 11, 1097. Uhlig, H. H., and Revie, R. W., Corrosion and Corrosion Control, Wiley, New York, 3rd edn., 1985. Stein, H. J., and Wegener, H A . , Appl. Phys. Lett., 1977, 124, 908. Stein, H. J., Picraux, 5 . T . , and Holloway, P. H . , IEEE rlruns. Electron Devices, 1978, ED-25. 1008. Peercy, P. C., Stein, H. J . , Doyle, B. L.. and Picraux, S. T., J . Electron. Muter., 1979, 8, 11. Stein, H. J., Peercy, P. S . , and Sokel, J . , Thin Solid Films, 1983, 101, 291. Ho, N. J., Kratochvil, J . , Rlackburn, G. F., and Janata, J., Sens. Actuators, 1983, 4, 413. Potje-Kamloth, K., Janata, P., Janata, J . , and Josowicz, M., Sens. Actuators, 1988, 18, 415. Josowicz, M., Janata, J., and Levy, M., J. Electrochem. Soc., 1988, 135, 112.340 ANALYST, APRIL 1993, VOL. 118 21 Stremsdoerfer, G., Perrot, H . , Martin, J . R., and Clkchet, P., J . Electrochem. SOC., 1988, 135, 2881. 22 Potje-Karnloth, K . , Janata, J . , and Josowicz, M., Ber. Bun- senges. Phys. Chem., 1989, 93, 1480. 23 Bezegh, K . , Bczegh, A., Janata, J . , Oesch, U . , Xu, A., and Simon, W., Anal. Chem., 1987,59,2846. 24 Morf, W. E., The Principles of Ion-Selective Electrodes and of Membrane Transport, Elsevier, Amsterdam, 1981. 25 26 Harrison, J . D., Cunningham, L. L., Li. X., Teclemariam, A . , and Permann, D., J. Electrochem. SOC., 1988, 135,2473. Fogt, E. J . , Calahan, P. T., Jevne, A., and Schwinghammer, M. A . , Anal. Chem., 1985, 57, 1155. Paper 21059300 Received November 6, I992 Accepted December 4, 1992

ISSN:0003-2654    

DOI:10.1039/AN9931800335

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 34 1 Sodium-selective Membrane Electrode Based on pfert-Butylcalix[4]arene Methoxyethylester" K. Cunningham and G. Svehla Department of Chemistry, University College Cork, Cork, Ireland S. J. Harris School of Chemistry, Dublin City University, Glasnevin, Dublin 9, Ireland M. A. McKervey School of Chemistry, The Queen's University of Belfast, Belfast, UK Twelve calixarenes have been studied as prospective ionophores for sodium. Five of these have shown promising results in the liquid membrane form. All five have been successfully incorporated into poly(viny1 chloride) membranes, and applied as ion-selective electrodes for sodium. Electrode responses, selectivities towards alkali and alkaline earth metals as well as towards hydrogen and ammonium ions, temperature coefficients and response times have been studied.Based on these results p-tert-butylcalix[4]arene methoxyethylester has been chosen as the best ionophore for the preparation of sodium-selective electrodes. Details of electrode construction are presented. Keywords: Sensor; potentiometry; ion-selective electrode; calixarene The introduction of new ion-selective electrodes has been an important development in potentiometry over the last 20 years. Much of this research has been directed towards developing neutral molecular carriers which possess the ability to complex selectively certain species such as cations. The development of poly(viny1 chloride) (PVC)-membrane elec- trodes based on neutral carriers has been an actively researched field since the discovery of the valinomycin-based potassium-selective electrode.1 Advantages of these mem- branes over the established glass membrane electrodes include a lower membrane resistance, easy construction, robustness and, in some cases, a working range which is almost free from the effect of pH. Cyclic and acyclic ionophores with defined cavities, which have ion-complexing properties, have been well documented in the literature2 and in recent years there has been consider- able interest in a new series of ionophores generally known as calixarenes.3-5 Derivatives of these ligands show receptor ionophoric activity when certain functional groups, such as an ester, ketone, amide or thioamide are attached to the phenolic oxygen atom. As a result, these ionophores have the ability to complex selectively alkali metal cations into the cavity present in the cone conformation6.7 and are therefore suitable as neutral carriers in ion-selective membrane electrodes.The ability of the tetrameric calix[4]arenes to act as sodium sensors has been demonstrated,8,9 while the larger hexameric calix- [6]arene derivatives have also been studied as caesium-selec- tive PVC-membrane electrodes.I0 The performance of sodium-selective electrodes based on 12 new calix[4]arene derivatives is now described. A preliminary account of one of these ionophores, (calixarene 1 in Fig. l), has already been published,ll but not in the context of the 12 ionophores described here. Electrodes based on ion-selective membranes respond, theoretically, in a Nernstian fashion to changes in the activity of only the primary ion (i).However, in reality, the presence of other cations also contributes to the response and this effect is summarized in the Nicolski-Eisenman equation E = cj + Slog[aj + ZkYf ( a j ) Z ~ ' z i ] * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18, 1992. Here E is the measured potential, Cj is a constant, having the role of the standard potential in the Nernst equation, S is the prelogarithmic factor, a is the activity, k$Ot is the selectivity coefficient which is a weighting factor that takes account of the contribution to the over-all potential by the interfering species, and z is the number of electrons involved in the electrode reaction.The subscript i refers to the analyte, and j to the interfering ions. When plotting E against logai, the slope of the calibration graph equals S. The temperature dependence of ion-selective electrodes can be defined through the Gibbs-Helmholtz equation; for practical purposes this can be converted into the form ET = E298 + ( T - 298) (aE/aT), where ET is the potential at temperature T , is the potential at 25 "C and (3E/aT), is the temperature coefficient of the electrode at constant pressure. The equation holds for the wide temperature ranges encountered in laboratory and clinical measurements. The temperature coefficient has to be determined experimentally. In these experiments the more extensive treatment, originating from Vasconcelos and Calixarene 1 2 3 4 5 6 7 r R3 - OR1 R' -CH2C02CH2CHz0Me -CH2COZPh -CH,COPh -CHzCON Etz -CHZCSNEtz -CHZCOZ-BU' (3X )-CH,C07Et n R2 R3 -C H 2- -But -CH2- -But - C H r -Bu' -CH2- -But -CH2- -But -CH2- -But -But (1 x )-CH;CO;CHzCH20COC(Me)=CH~ -CH2- 4 -CH2C02CHzCH20COC(Me)=CH2 -CH2- 8 A polymer of ( 7 ) 9 10 11 12 -Bu' 4 -CH,CO,Et (2 X )-CHZ- 4 ( 3 X )-CH&OzEt 6 -CHZCO,Et (2 X )-CH20- -Bu' - C H r -But -CH*- ( ? x )-CH2C02Me -CH=CH2 Fig.1 General structures of calixarenes studied342 ANALYST, APRIL 1993, VOL. 118 Machado,l~~~3 was not applied, in which the effect on standard potential, prelogarithmic factor and logarithmic term can be separately assessed; their hysteresis methods were also not applied. 14 The response time of the electrode can be characterized by the time constant t; this is defined through the equation Tf the time constant is less than 140 ms, the electrode attains its equilibrium potential (E,J within 1 s.Experimental Ionophore The structures of the ligands studied in this work are shown in Fig. 1. The calixarenes were all prepared according to the pro- cedures published in separate papers: for calixarenes 1 , 3 and 11 see ref. 15; for 2 and 4 see ref. 4; for 5 and 6 see ref. 16; for 7 and 12 see ref. 17; for 8 see ref. 18; for 9 see ref. 19; and for 10 see ref. 20. All the ligands were purified before being incorporated into the electrodes. Other Reagents The other membrane components were: potassium tetrakis(4- chloropheny1)borate (KTpCIPB); 2-nitrophenyl octyl ether (NPOE), (99%); tetrahydrofuran (THF), (HPLC grade); and poly(viny1 chloride) (PVC, ISE grade).Solutions of the highest available purity of the alkali and alkaline earth metal chlorides were prepared in doubly distilled water, except for beryllium where the sulfate was studied. Metal picrate solutions were made up from the metal chloride and picric acid, mixing these in stoichiometric ratios. The solutions were neutralized by adding small amounts of dilute hydrochloric acid against indicator paper. In the phase-transfer studies the calixarenes were dissolved in AnalaR grade dichloromethane. Electrode Fabrication The calixarene indicator electrodes were constructed accord- ing to the methodology described by Moody and Thomas2l with appropriate modifications. Each electrode contained a chloride-coated silver wire as the internal reference electrode, immersed in an internal filling solution of 0.1 mol dm-3 NaCl.An interchangeable PVC tip was placed at the bottom of the electrode body and the PVC membrane was glued to the PVC tip which was in contact with the analyte. The PVC mem- brane was prepared by dissolving appropriate amounts of the ionophore (0.66% m/m), KTpClPB (0.17% m/m) and plasti- 100 mV I t lu 1 I I I I -6 -5 - 4 -3 -2 - 1 0 LodM “1 Fig. 2 Response of PVC-membrane electrodes, containing calix- arenes 1-5, to sodium. Membrane composition; calixarene, 0.66% ; KTpClPB, 0.17%; NPOE, 65.84%; and PVC, 33.33% cizer [NPOE (65.84% d m ) ] . (These amounts were optimized in preliminary experiments, not to be detailed here.) Then the PVC (33.33% mlm), dissolved in a suitable amount of THF, was added.The mixture was stirred until a ‘syrup’ consistency was obtained and this was cast onto a flat glass plane with a glass ring on it. O n evaporation of the solvent a membrane was obtained, which was peeled off, cut and then pasted onto the PVC electrode tips. Instrumental Arrangement The potentiometric cell contained the calixarene indicator electrode, a Metrohm 60702.100 saturated calomel electrode and the analyte solution. A thermostated glass vessel (kept normally at 25°C) and a magnetic stirrer was applied. Electrode potentials were measured by a Metrohm 654 mV/pH meter and recorded as a function of time with a Linseis LM 24 chart recorder. Methodology In preliminary experiments it was ascertained that each of the 12 calixarenes had ionophoric characteristics and that they all responded to sodium ions.Then membranes were cast, and the linear response ranges, slope of the potential versus log concentration curves, limits of detection and selectivity coefficients were determined for each of them. The response of each electrode was measured for the cations over the range 1 x 10-6-1 mol dm-3 and the slope determined from the linear region of the response as well as the limit of detection. The selectivity coefficients were, initially, obtained by using the separate solution method,22 for the p-tert-butylcalix[4]arene methoxyethylester, which was singled out as the best calix- arene, then the mixed solution method.23.24 From these experiments five ionophores were selected (1-5 in Fig.l ) , that showed the most promising characteristics; all further studies were restricted to these. The temperature dependence of the electrode assembly was studied by measuring the changes in potential to a solution of the primary ion as the temperature was varied over a small temperature range (283-313 K). Phase-transfer extractions were based on the method described by Pederson2’ and used as an indication of the relative complexing power of the calixarenes for different metal cations. This procedure involved shaking a dilute solution of the calixarene in dichloromethane (2.5 x 10-4 mol dm-3) with an equal volume of the aqueous neutral metal picrate solution (2.5 x 10-4 mol dm-3) for 3 min. After separation, ultraviolet (UV) spectrophotometric analysis of the organic layer at 378 nm enabled the determination of the amount of picrate present in this phase, this in turn allowed the percentage of metal extracted by the calixarene to be determined.No extraction was observed in the absence of the calixarene. Results and Discussion Elimination of Ionophores 6-11 The potential versus log concentration plots were promising for all of the 12 calixarenes, but there were marked differences between selectivities. Table 1 lists the logarithms of selectivity coefficients for all the calixarenes. On the basis of these results it is clear that ionophore 10 is selective first of all to rubidium, 11 to potassium, and 12 to caesium, and therefore unsuitable for the measurement of sodium. Ionophore 6 shows a poor selectivity in the presence of Rb, while 7, 8 and 9 show poor selectivity in the presence of K.As the listed ionophores offered no advantage over others in the set, they were eliminated from further studies. Tonophores 4 and 5 are the poorest as far as selectivity towards sodium is concerned; however, further studies wereANALYST, APRIL 1993. VOL. 118 343 Table 1 Logarithm of thc selectivity coefficicnts for 12 calixarenes (all treated as sodium-selective ionophores) Log ( k r ’ ) values Calixarene (cf. Fig. 1) 1 1 2 Ion ms* ss-t ss Li + -2.7 -3.8 -2.5 Na+ 0 0 0 K+ -2.4 -2.5 -1.5 Rb + -2.2 -2.6 -1.4 Cs+ -2.0 -3.4 -1.2 Bc’+ -3.7 -3.2 -2.6 Mg’+ -3.5 -4.3 -3.3 Ca* + -3.5 -4.3 -3.0 Sr’+ -3.2 -3.1 -2.8 Ba?+ -3.1 -4.1 -3.3 H+ -2.3 -2.8 -1.2 NHlf -2.0 -2.8 -2.4 ms = mixed solution method. t ss = separate solution method.3 4 -2.7 - 0 0 -2.3 -0.4 -3.7 -1.0 -3.9 -0.5 -3.9 - -4.2 -0.6 -4.3 -0.4 -3.3 -0.6 -4.4 -1.1 -3.1 -0.5 -3.5 - ss ss 5 -0.6 0 -0.8 -1.1 -1.5 -0.8 -1.4 -0.4 -0.5 -0.9 -0.3 +0.2 ss 6 -2.1 0 -1.4 -0.2 -2.9 -2.8 -5.4 -3.4 -5.9 -3.0 -2.3 -2.7 ss - . 7 s - 9 ss ss ss -3.1 -2.4 -2.8 0 0 0 -1.3 -0.9 -1.5 -3.4 -1.4 -2.4 -3.1 -1.2 -2.3 -4.9 -3.1 -3.3 -5.3 -2.8 -3.2 -4.9 -3.0 -3.1 -4.7 -2.6 -2.5 -4.6 -2.1 -3.1 -4.1 -2.2 -2.7 -2.9 -1.8 -3.1 10 -0.8 0 +0.7 +1.0 +0.9 -1.3 -1.4 -1.8 -1.4 -1.2 -0.5 +0.5 ss 11 - 1.8 0 +0.5 -1.8 -1.7 -2.9 -3.5 -3.4 -3.4 -3.4 -3.0 -1.8 ss 12 -1.3 0 +1.2 + I S +2.1 -1.6 -1.5 -1.5 -1.8 -0.8 -0.3 +0.4 ss carried out on these, because of the interesting extraction data (see below). We also studied ionophore 2 further, although it is obviously less interesting than either I or 3.There is not much to choose between these last two ionophores and both could be adapted for the determination of sodium; finally we chose 1 as the best as it showed slightly better selectivity in the presence of potassium, which is the most frequently occurring interferent in real samples. The interference caused by Rb+ was of lesser concern. The electrode based on calixarene 2 is limited by a smaller degree of selectivity particularly to K+, Rb+, Cs+ and H+. The ketone derivative, 3, showed excellent selectivity to the cations studied which were more favourable than 1 in some instances but the short lifetime of the membrane was the main limitation. The amide 4 and thioamide 5 derivatives were not as selective as the esters (1 and 2) and ketone (3).Calixarene 4 only showed a small preference for sodium over the other cations and had a limited linear range. A liquid membrane electrode, which contained the same amount of calixarene and KTpClPB dissolved in trichlorobenzene, performed in the expected manner, but experiments with PVC-membrane electrodes containing plas- ticizers of different polarity failed to reproduce this result. Calixarene 5 showed a slightly greater preference for sodium than 4 but the electrode suffered from a large interference from ammonium ions. Another limitation to 4 and 5 is the small selectivity for sodium over K+ and H+. Stability Constants and Phase-transfer Results Stability constants for certain calixarene-metal complexes have been determined for calixarenes 1 and 4.26 Logarithms of stability constants obtained for the complex in methanol (4.7 for Na+, 2.3 for K+ and <1 for Li+, Rb+, Cs+ and C$+) also indicate that calixarene I has a preference for sodium and forms a medium-strength complex which suggests that the binding is reversible.It is interesting to mention that (logarithms of) stability constants for calixarene 4 are signifi- cantly higher than those for 1 [Li+(3.9), Na+(7.9), K+(5.8), Rb+(3.8), Cs+(2.4), Mg2+(1.2), Ca’+(>9) and Ba2+(7.2)]. The phase-transfer results for the five calixarenes, pre- sented in Table 2, show that calixarenes 1-4 show a preference for sodium while calixarene 5 has a slight preference for complexing K+. The Na+ extraction values for calixarenes 1-3 are greater than the other values by a factor of 2-10 which indicates favourable selectivity for sodium ions.In general, calixarene 4 has very high extraction values for all the cations studied which would suggest that all of the cations can be complexed by this calixarene with only a small preference for sodium, except for Mg’+. Finally calixarene 5 has a relatively Table 2 Summary of the phase-transfer results as a percentage of the metal picrate extracted by the calixarene into dichloromcthane from an aqueous solution of neutral pH Calixarene Cation studied Li + Na+ K+ Rb+ cs+ Mg2f Ca?+ Sr*+ Ba2+ 1 7 23 7 5 7 3 5 6 - 5 5 8 9 5 6 3 2 3 - low percentage extraction for all the metals with a slight preference for K+. Performance of the Sodium Ion-selective Electrodes The PVC-membrane electrodes were studied as sodium-selec- tive electrodes, as described previously.Table 3 shows the response of these electrodes to sodium ions for calixarenes 1-5. The table should be studied with reference to Fig. 2, which shows the different calibration graphs obtained with these electrodes. The potential scale on Fig. 2 is not ‘absolute’; the diagram was constructed so that the response of each ionophore could be clearly seen and compared. However, from the data in Table 2 (row 2) it is possible to reconstruct ‘absolute’ potential values. The limit of detection was deter- mined from the intersection of the two best lines drawn from the two portions of the response of the electrode. The slopc, S, and correlation coefficient , Y, were determined for the points of the linear region above the limit of detection.All of the responses were sub-Nernstian, with calixarene 3 being the closest to the theoretical value of 59 mV decade-’ while calixarenes 1 and 2 gave an acceptable sensitivity. Calixarene 4 has a limited linear range (10-4.4-10-1.9 mol dm-3) and a reduced sensitivity compared with calixarenes 1-3. Finally, calixarene 5 also shows less sensitivity but it does n o t have a maximum on the calibration graph. The limits of detection were satisfactory and of the same order of magnitude as values quoted by other researchers.27 The correlation is good, in most instances, for the data on the linear part of the response particularly for calixarene 1. The lower coefficients for calixarenes 3 and 5 were caused by difficulties in reproducing data on the points just above the limit of detection.344 ANALYST, APRlL 1993, VOL.118 -50u- Table 3 Summary of the response of the PVC-membrane-selective electrodes to sodium ions Calixarene n*l I 1 1 2 3 4 5 Slope/mV decade- 54.2 55.6 53.0 46.1 43.6 Potential at 1 x 10-2 mol dm-YmV 15 70 102 - 10 12 Limit of detection/ mol dm-3 1 x 10-4 3 1 x 10-3 8 1 x 10-3 9 1 x 10-4 4 1 x 10-3 5 Correlation coefficient ( r) 0.999 0.994 0.989 0.997 0.969 Lifetimdd > 100 7 3 30 60 Table 4 Response of the PVC-membrane sodium-selective electrodes to sodium ions over the temperature range 10-40 “C Calixarene and “a] present 1 5 Calixarene [Na+]/mol dm-3 0.001 0.01 0.001 0.01 Temperature coefficient, correlation coefficient ( r ) 0.940 0.955 0.876 0.893 6E/6T(mV K-1) -0.55 -0.53 +0.12 +0.05 Lifetimes were estimated by carrying out regular checks on the sodium response of the electrode, which was kept in a 0.1 mol dm-3 sodium chloride storage solution.The electrode was considered to be at the end of its working life when the sensitivity started to fall and this was often accompanied by loss of selectivity. The lifetime varies considerably, with calixarene 1 showing a satisfactory performance after three months and calixarene 3 only lasting 3 d. Table 4 summarizes the temperature response of two of the ion-selective electrodes. Calixarene 1 was selected because it was clear that this compound is the best ligand, and compared with calixarene 5 which could be regarded as the worst.Values for the temperature coefficient (relating to the whole cell, including the external reference electrode) indi- cate, that in order to obtain reproducible results, it is necessary to control the temperature when making such measurements. It has to be emphasized that very few workers have in fact studied the temperature dependence of their electrodes; hence it is not easy to make comparative state- ments in this respect. The response time ( i e . , the time to attain final, equilibrium potentials) was always less than 2 s , indicating that the time constant was less than 300 ms. We did not attempt to determine these constants more accurately, as they were only of academic interest. The electrodes were all suitable for batch measurements and for flow injection purposes.With a calixarene-based silver electrode, that had similar time- response characteristics, successful automated potentiometric titrations of halides were carried out.28 The performance of all five PVC-membrane Na-selective electrodes was also studied, in more detail, in the presence of various ions at different concentration levels. For the sake of comparison, the diagrams for calixarenes 1 and 5 ( i . e . , the best and worst) are presented in Figs. 3-5. It is perhaps worthwhile pointing out, that the lines on these figures were drawn by a simple least-squares computer program; if they were drawn by hand, the sodium response for calixarene 5 could be better represented, i.e., by an almost horizontal line up to a concentration of 1 X 10-3 mol dm-3, and a steeper line for higher concentrations.However, even a treatment such as this would not alter the fact that this calixarene is the least suitable of the five compounds studied. More detailed figures for calixarene 1 have been published elsewhere.” 100 50 0 1 u -200 I I I I I I -6 -5 -4 -3 -2 -1 0 LodM “1 Fig. 3 Response of the electrode with calixarene 1 to: A , sodium; B, potassium; C, hydrogen; and D, strontium ions. Membrane compo- sition; calixarene 1, 0.66%; KTpCIPB. 0.17%: NPOE. 65.84”/0: and PVC, 33.33% 100 G 50 > , E o w -6 -5 -4 -3 -2 -1 0 Log [M m+ ] Fig. 4 Response of the sodium-selective electrode to the alkali metal, hydrogen and ammonium cations: A , sodium; B , lithium; C, potassium; D, rubidium; E, caesium; F, hydrogen; and G, ammo- nium.Membrane composition: calixarene 5 . 0.66%; KTpClPB, 0.17% ; NPOE, 65.84% ; and PVC, 33.33% 100 I -6 -5 -4 -3 - 2 -1 0 LodM “I Fig. 5 Response of the sodium-selective electrode to sodium and the alkaline earth metal cations: A, sodium; B, beryllium; C, magnesium; D, calcium; E, strontium; and F, barium. Membrane composition: calixarene 5, 0.66%; KTpCIPB, 0.17%; NPOE, 65.84% and PVC. 33.33%ANALYST, APRIL 1993, VOL. 118 In conclusion, calixarene 1 is the best electrode produced in our research to date and an application for a patent is being processed. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 References Pioda, L. A. R., Stankova, V., and Simon, W., Anal. Lett., 1969, 2, 665. Morf, W. E . , The Principles of Ion-selective Electrodes and of Membrane Transport, Elsevier, Amsterdam, 1981, pp.266-267. Gutsche, C. D.. Top. Curr. Chem., 1984,123, 1. Harris, S. J., Arnaud-Neu, F., Collins, E. M., Deasy, M., Ferguson, G., Kaitner, B., Lough, A. J . , McKervey, M. A., Marques, E., Ruhl, B. L., Schwing-Wcill, M. J., and Seward, E. M., J. Am. Chem. SOC., 1989, 111. 8681. McKervey, A., and Bohmer, V., Chem. Br., 1992, 28, 724. McKervey, M. A., Seward, E . M., Ferguson, G., Ruhl, B. L.. and Harris, S. J., J. Chem. SOC., Chem. Commun., 1985, 388. Ferguson, G., Kaitner, B . , McKervey, M. A., and Seward, E. M., J. Chem. SOC., Chem. Commun., 1987, 584. Diamond, D., Svehla, G., Seward, E. M., and McKervey, M. A., Anal. Chim. Acta, 1988,204, 223. Telting-Diaz, M., Smyth, M. R., Diamond, D., Seward, E. M., Svehla, G., and McKervey. M.A., Anal. Proc., 1989, 26, 29. Cadogan, A., Diamond. D., Smyth, M. R., Svchla, G., McKervey, M. A., Seward, E. M., and Harris. S . J., Analyst, 1990, 115, 1207. Cunningham, K . , Svehla, G., Harris, S. J . , and McKervey, M. A., Anal. Proc., 1991, 28, 294. Vasconcelos, M. T. S. D., and Machado, A. A. S. C., Analyst, 1988. 113, 49. Vasconcclos, M. T. S. D., and Machado, A. A. S. C., Analyat, 1990, 115, 195. Vasconcelos, M. T. S. D.. and Machado, A. A. S. C., Anal. Lett., 1988, 21, 1987. 15 16 17 18 19 20 21 22 23 24 25 26 27 28 345 Arnaud-Neu, F., Barrett, G., Cremin, S . , Deasy, M., Fer- guson, G . , Harris, S. J., Lough, A. J., Guerra, L., McKervey, M. A., Schwing-Weill, M. J., and Schwinte, P., J. Chem. Soc., Perkin Trans. 2, 1992, 1119. Harris, S. J., Barrett, G., and McKervey, M. A., J . Chem. Soc., Chem. Commun., 1991, 1224. Harris, S. J . , McKcrvcy, M. A., Svehla, G., and Diamond, D., US Pat., 5 132345, July, 1992. Arnaud-Ncu, F., Schwing-Weill, M. J . , Ziat, E., Crcmin, S . , Harris, S. J . , and McKcrvcy, M. A., New J. Chem., 1991, 15, 33. McKervey, M. A., Harris, S. J., Collins, E. M., Guthrie, J . , and McArdlc, C., Eur. Pat. Appl., 90313382.5, December, 1989. Harris, S. J., McKervey, M. A . , Arnaud-Neu, F., Cremin, S . , Cunningham, D., McArdle, P., McManus, M., Schwing-Weill. M. J . , and Ziat. K . , J. Incl. Phenomena, 1991, 10, 329. Moody, G. J., and Thomas, J. D. R., in Chemical Sensors, ed. Edmonds, T. E., Blackie, London, 1988, p. 76. Moody, G. J . , and Thomas, J . D . R., Selective Ion-sensitive Electrodes, Merrow, Watford, 1971, pp. 10-14. Van der Linden, W. E.. in Comprehensive Analytical Chemistry, ed. Svehla, G., Elsevier, Amsterdam, 1981, vol. XI, Craggs, A., Moody, G. J . , and Thomas, J . D . R., J. Chem. Educ., 1974, 54, 541. Pederson, C. J . , Fed. Proc. Fed. Am. SOC. Exp. Bid., 1968,27, 1305. McKervey, M. A., Arnaud-Neu, F., Cremin, S., and Harris, S. J . , unpublished work. Cadogan, E., Diamond, D . , Smyth, M., Deasy, M., McKervey, M. A., and Harris, S. J., Analyst, 1989, 114, 1551. O’Connor, K. M., Svehla, G . , Harris, S. J.. and McKervey, M. A., Talanta. 1992, 39, 1549. p. 374. Paper 210621 2 G Received November 23, 1992 Accepted January I , 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800341

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993. VOL. 118 347 Neural Network Based Recognition of Flow Injection Patterns* Margaret Hartnett and Dermot Diamondt School of Chemical Sciences, Dublin City University, Glasnevin, Dublin 9, Ireland Philip G. Barker Interactive Systems Research Group, School of Computing and Mathematics, The University of Teesside, Middles broug h, Cleveland, UK Response patterns produced from a flow injection (FI) system employing sodium, potassium and calcium ion-selective electrodes as detectors were used for training and testing back-propagation neural networks. A variety of different network parameters were investigated including a study of the effects of variation of the learning rate coefficient and the momentum on the rate of training. The networks studied demonstrated the ability to recognize the identity of sample components even in FI patterns severely distorted by noise addition, peak height variation and baseline shift.Keywords: Neural network; pattern recognition; flow injection; sensor array; ion-selective electrode Artificial neural nctworks are a form of artificial intelligence that mathematically simulate biological nervous systems. They have been employed in a diverse range of applications in recent years including robotic control,l forecasting2 and pattern classification.3 Artificial neural networks have many advantages over statistical techniques including the ability to adapt continuously to new data through the use of less rigid assumptions about the underlying data distributions. There are a number of different types of network models including Hopfield networks, bidirectional associative memory models, the neocognitron, adaptive resonance theory models and counter-propagation network models.The choice of a particular type of network depends to a large extent on the nature of the problem for which it is being used4 but the most commonly used network at the moment is the multi-layer feed-forward network. A more complete description of the different network models and their attributes can be found elsewhere .5-9 In this paper we consider the use of neural network technology in order to solve some of the problems associated with the use of ion-selective electrodes (ISEs) as a means of detection in flow injection (FI) systems. These problems include noise pick-up due to the high impedance nature of the sensors, baseline shift due to the floating earth nature of the measurement and cross-selectivity of the ISEs for other ions present in a sample.Although there have been many applications of neural networks to problems relevant to analytical chemistry, very little has been published on their use for interpreting signals from ion-selective electrode arrays in the primary literature.") A back-propagation training algorithm was applied to a feed-forward network, which was used for the recognition and identification of patterns produced by the response profiles of the transient signals of three ISEs (for sodium, potassium and calcium) used in an FT system. The response profiles of the electrodes were distorted by a number of mechanisms including the addition of noise and the shift in baseline potentials of the peaks, and the ability of the networks to recognize previously unseen distortions of the patterns investigated subsequently .* Presented at the Sensors and Signals Symposium at The Royal Society o f Chemistry Autumn Meeting. Dublin, Ireland, September 16-18, 1992. t To whom correspondence should be addressed. Theory Multi-layer feed-forward networks are composed of three or more layers of interconnected nodes or neurons. These nodes (see Fig. 1) behave in a similar fashion to biological neurons. Each node receives a series of inputs (xl-x,) which are each weighted by the values of the connection weights (W,-W,) between the nodes. These weighted inputs are then summed and operated on by a linear or non-linear transfer function, which yields the output signal of the neuron.A non-linear, continuously differentiable, monotonically non-decreasing function with an output value defined within 0 and 1 known as the sigmoid function is used in this work and is the most commonly used transfer function with the back-propagation training algorithm. The continuous output signal of a neuron can be interpreted in a number of ways such as an indication of the degree of confidence that an associated feature is present or absent, rather than providing a binary answer describing the presence or absence of the feature. Alternatively, the output signal of a neuron may suggest the amount of a feature that is present." For the purpose of this work a simple binary classification of the input patterns was required, i.e., an output value of 0 from an output neuron indicated that the associated component was definitely absent from the sample and an output value of 1 indicated that the component was definitely present in the sample. The difference between the continuously valued output of the network and the required discrete binary pattern classification was considered to be the error of the network classification procedure.Studies of the classification abilities of different networks were performed by considering the magnitudes of the mean squared error (MSE) to the different training and testing patterns. /I Fig. 1 A schematic representation of an artificial neuron348 Hidden layer ANALYST, APRlL 1993, VOL. 118 Desired binary - classification - Calculate error Here, n is the number of patterns presented in the training or testing set; p is the pattern index; j is the component index of output vector; tpj is the desired value of component j of the output vector to pattern p ; op, is the network value of output neuron j to pattern p .The first layer of the network (Fig. 2) is called the input layer and the inputs to the neurons in this layer are the input values of the training or testing set. The last layer of the network is called the output layer and the output signals from the neurons in this layer correspond to the outputs associated with the input pattern. Between these two layers there exists one or more hidden layers of neurons, which receive the weighted outputs from the input layer and produce output signals that act as the input to the output layer.Supervised training of the network involves changing the values of the connection weights between the neurons in order to minimize the error described earlier as the difference between the desired output of the network and the actual output from the neurons in the output layer for a given training pattern. This training process, which involves the repetitive presentation of different training patterns and comparison of desired and actual network output values, continues until the error converges or reduces to some pre-defined level. The back-propagation algorithm is a method by which the sum squared error ( E ) is reduced by propagating the error backwards from the output to the hidden and input layers and Sodium Potassium Calcium present/ present/ present/ absent absent absent Response Response Response profile from profile from profile from sodium potassium calcium ISE ISE ISE Fig. 2 A diagrammatic representation of a feed-forward neural network architecture Fig.3 A A Continuous network output alteration of the connection weights between the layers in a gradient descent fashion. Two main parameters are used for controlling the weight modifications and these are the learning rate and momentum as shown in Fig. 3. The learning rate determines the size of the increments by which the connection weights are varied during training and the momentum term refers to the addition of a fraction of the previous weight change to the current weight change in order to reduce oscillation or instability in the training procedure.12 Mathematically, these relationships are described by the expression Awji(n + 1) = r ( 8 p j o p j ) + aAWjqnf (3) where n = index of the presentation number; q = learning rate; a = momentum; W,i = weight on connection between neuron i in previous layer and neuron j in present layer; 6 = tpj - op,, or the difference between the desired classification and the actual network output; and opi = output state of neuron i in the previous layer. The performance of the trained network can then be tested by examining the response of the network to a test set of patterns on which it was not trained. P! Experimental Flow Injection System The FT-sensor array system used for transient ion detection has been described previously by Forster and Diamond,l3 and for the purposes of this study, signal traces produced from electrodes selective for sodium, potassium and calcium were used.A schematic representation of the FI system is depicted in Fig. 4. The experimental data were captured via an Analog Devices RTI-815 data acquisition card fitted inside an IBM 286 compatible PC. Electrode Fabrication The electrodes used for the detection of sodium, potassium and calcium were based on three ionophores [p-tert-butyl- calix[4]methyl acetate,14 valinomycin (Fluka) and Calcium Ionophore I1 (ETH 129; Fluka), respectively], each of which schemat :ic diagram of Direction of data flow L (lnputdata) the information flow in the feed-forward neural network using a back-propagation ti -aining algorithmANALYST, APRIL 1993, VOL.118 349 Sample injection Carrier phase Carrier trs I 3 ton selective electrode outputs Waste 2 Channel peristaltic Reference electrode output Saturated Saturated KCI KCI Fig. 4 A schematic diagram of the FI system used for data acquisition were immobilized within a plasticized poly(viny1 chloride) (PVC) matrix. Membranes were cast from solutions of the ionophore, ion excluder, plasticizer and PVC in tetrahydro- furan as described previously.15 Software Routines for the control of the FT system, data acquisition from the ISEs and data processing were written in Microsoft QuickBASIC. Data preparation for the neural networks and results analyses from the neural nets were performed using software written in BORLAND TURBO C and SUN C version 1.1.Two types of neural network software were investigated, namely, Neural Technologies NT5000 running on an 80486 PC and Neuralworks Professional I1 running on a SUN SPARC workstation. Results Studies Performed Using Neural Technologies NT5000 There were seven possible combinations of the sodium, potassium and calcium ions on their own or with the other ions in the group. Data files composed of 80 data points per electrode response profile (240 data points in total) were prepared and pre-scaled for the neural net. The initial training set was composed of the electrode response profiles to the 7 possible combinations of the cations and also contained a synthetic profile corresponding to the absence of any of the cations. The number of iterations required for the error to converge to the pre-defined value of 0.03 was investigated with respect to the learning rate coefficient and momentum for different numbers of hidden layer neurons.Fig. 5 shows how the number of presentations of the training set required for convergence varied with the learning rate coefficient. It can be seen from this figure that an apparent minimum in the number of presentations of the training set required for convergence exists for a learning rate coefficient in the region of 0.25 and 0.4 for the different network topologies studied. On either side of this minimum we can see that at very low learning rates the nets converge slowly to a solution because of the relatively small changes in the connection weights.At very high learning rates the nets also converge slowly to a solution because of instability in the learning process produced by the large changes in the connection weights. The stabilization in the network learning process arising from increasing the momentum (ie., feedback error) is demonstrated strikingly in Fig. 6, which shows how the number of presentations of the training set required for convergence varies with momentum for neural networks with 7, 8 and 9 neurons in the hidden layer. 170 160 150 5 140 0 ' E 130 2 0,120 5 .- U r ; 110 % 90 C .- 100 +, 2 a 80 70 60 I 0.001 0.01 0.1 1 Learning rate coefficient Fig. 5 9 neurons in the hidden layer Convcrgence dependency on learning rate. A, 7; B , 8; and C , 600 4- 3 500 0, C C .- .- 400 0, 5 c 300 0 .- U 4- c 3 200 2 LL 100 0 I I I I I l l 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Momentum Fig.6 Fig. 5 Convergence dependency on momentum. A, B and C as for In order to investigate the effect of various distortions on the ability of the networks to, predict correctly and quali- tatively, the composition of an unknown solution, test sets of distortions of response profiles in the training set were composed. It was found that networks trained with FI peaks of a particular height had the most difficulty recognizing and identifying patterns in which the heights of the FI peaks had been altered.16 The networks also had difficulty with patterns produced by shifting the baselines of the FI peaks although the magnitude of the MSE was not as large as the FI patterns in which the peak heights had been changed.The test patterns synthesized by adding noise to the training patterns provided the least difficulty to the networks. These results suggested that the trained networks were sensitive to the position of the peak baseline in the pre-defined voltage range and the relative height of any feature above this baseline rather than to the underlying shape of the peak features. A training set composed of 70 random patterns representing the FI peaks distorted by different combinations of baseline350 ANALYST, APRIL 1993, VOL. 118 shift, peak variation and noise addition, at levels much worse than normally encountered under experimental conditions, was designed and a testing set of 56 patterns was composed in a similar fashion. Tables 1 to 7, list the outputs of the network with 50 neurons in its hidden layer to the different patterns in the test set, it also lists the ratios of the network outputs to the ions that are supposed to be present in the sample to the ions that are supposed to be absent from the sample. They also contain a listing of the different distortions applied to the original experimental trace required for the generation of the par- ticular pattern. Table 1 lists the outputs from the network to the distorted pattern generated from FI traces to a sample that contained sodium only.The nature of the distortions employed and the resultant network responses suggest that as the height of the sodium ISE peak became smaller and hence closer to the level of the noisy signals from the potassium and calcium ISEs, that it became more difficult to distinguish between the presence of sodium or calcium in the sample, because of the sensitivity of the network to low amplitude calcium ISE signals relative to the sodium and potassium ISE peak amplitudes.The worst response from the network for this series of patterns was obtained for pattern 7 in which the height of the sodium ISE peak had been reduced by 70% and noise at 40% of the original peak height had been added. The network response to this pattern was 0.999 (Na), 0.032 (K), 0.546 (Ca), however, even with this level of distortion it is evident that the network output indicates a degree of confidence in the presence of sodium at almost twice that of the presence of calcium. Table 2 lists the outputs of the network to the distorted patterns generated from FI traces to samples which contained only potassium.Interference signals from the sodium ISE to the potassium present in the sample produces the relatively low output ratio of K : Na seen in pattern 13 (6.566) in which the baseline had been shifted in the negative direction by approximately 8% of the original peak height, and the height of the potassium ISE peak had been reduced by 80%. Noise at Table 1 Network outputs to patterns in the testing set generated from samples containing sodium only Distortions of original pattern employed Network outputs Output ratios Pattern number 1 2 3 4 5 6 7 8 Output for sodium 1,000 1 .ooo 1.000 1 .ooo 0.986 1 .000 0.999 1 .ooo Output for potassium 0.067 0.01 1 0.038 0.245 0.017 0.021 0.032 0.251 Outputfor - calcium 0.017 0.010 0.02 0.035 0.021 0.104 0.546 0.023 Na:K 14.926 90.909 26.316 4.082 58.000 47.619 31.219 3.984 Na : Ca 58.823 100.000 50.000 28.571 46.952 9.615 1.830 43.478 Noise addition as percentage of peak height(%) - - 30 60 50 40 70 - Baseline shift as percentage of peak height (YO) -62 -38 -42 = 10 - - - =30 Peak height reduction as percentage of sodium peak height (YO) 60 40 50 80 70 - - - Table 2 Network outputs to patterns in the testing set generated from samples containing potassium only Distortions of original pattern employed Network outputs Pattern number 9 10 11 12 13 14 15 16 Output for sodium 0.020 0.008 0.042 0.029 0.152 0.023 0.013 0.164 Output for potassium 1 .ooo 1.000 1.000 0.999 0.998 1 .Ooo 1 .000 1 .Ooo Outputfor - calcium 0.005 0.007 0.023 0.006 0.006 0.028 0.095 0.348 Output ratios K:Na K : C a 50.000 200.000 125.000 142.857 23.809 43.478 34.448 166.500 6.566 166.333 43.478 35.714 76.923 10.526 6.098 2.874 Noise addition as percentage of peak height (YO) 40 30 50 - - - 70 60 Baseline shift as percentage of peak height (YO) -23 -30 = 10 -20 - - =-8 - Peak height reduction as percentage of potassium peak height (%) 60 60 30 70 80 - - - ~~~~ Table 3 Network outputs to patterns in the testing set generated from samples containing calcium only Distortions of original pattern employed Pattern number 17 18 19 20 21 22 23 24 Network outputs Output for sodium 0.062 0.017 0.080 0.037 0.080 0.038 0.049 0.037 Output for potassium 0.020 0.008 0.012 0.006 0.002 0.010 0.010 0.020 Output for - calcium 0.910 0.995 0.989 0.998 1 .Ooo 1 .om 1.000 1 .ooo Output ratios Ca : Na 14.677 58.529 12.362 26.973 12.500 26.316 20.408 27.027 C a : K 45.500 124.375 82.417 166.333 500.000 100.000 100.000 50.000 Noise addition as percentage of peak height ( Y ) 60 40 30 75 80 - - - Baseline shift as percentage of peak height (YO) =213 -22 ==75 - - - =85 =43 Peak height reduction as percentage of calcium peak height (%) 60 40 65 20 40 - - -ANALYST, APRIL 1993, VOL.118 351 Table 4 Network outputs to patterns in the testing set generated from samples containing sodium and potassium only Distortions of original pattern employed Noise addition as percentage of peak height (YO) 30 60 - Baseline shift as percentage of peak height (%) -40 =-20 i= 12 -28 =8 - - - Peak height reduction as percentage of peak height (%) 40 (Na), 60 (K) 60 (Na), 40 (K) 20 (Na), 20 (K) 70 (Na), 60 (K) 60 (Na), 30 (K) - Network outputs Output ratios Pattern number 25 26 27 28 29 30 31 32 Output for sodium 1 .ooo 0.997 1.000 0.986 1.000 0.993 1 .ooo 1 .ooo Output for potassium 1 .om 1.000 1 .Ooo 1 .ooo 1.000 1 .000 1 .00o 1 .om Outputfor - calcium 0.136 0.585 0.023 0.024 0.007 0.021 0.000 0.032 Na : Ca 7.353 1.704 43.478 41.083 142.857 47.286 31.250 - K:Ca 7.353 1.709 43.478 41.667 142.857 47.619 31.250 - - 40 20 Table 5 Network outputs to patterns in the testing set generated from samples containing sodium and calcium only Distortions of original pattern cmployed Noise addition as percentage of peak height (%) Baseline shift as percent age of peak height (YO) -43 (Na), - ~ 1 2 7 (Ca) Peak height reduction as percentage of peak height (YO) 40 (Na), 40 (Ca) 60 (Na) ,60 (Ca) Network outputs Output ratios Output for calcium Na: K C a : K 0.971 40.000 38.840 0.546 19.920 10.920 ~~ Pattern number 33 34 Output for Output for sodium potassium 1 .ooo 0.025 0.996 0.050 70 (Na), 20 (Ca) 35 36 0.999 0.003 1 .ooo 0.020 0.982 333.000 327.333 0.988 50.000 49.400 40 - -30 (Na), -90 (Ca) 1 .om 0.005 1 .ooo 0.152 0.988 200.000 197.600 0.993 6.579 6.118 40 60 37 38 -13 (Na), =38 (Ca) -26 (Na), -76 (Ca) =13 (Na), =138 (Ca) - 85 (Na), 30 (Ca) 45 (Na), 65 (Ca) 39 0.7% 0.020 0.995 39.900 49.750 20 40 1 .ooo 0.093 0.798 10.753 8.581 50 Table 6 Network outputs to patterns in the testing set generated from samples containing potassium and calcium only Distortions of original pattern employed Noise addition as percentage of peak height (YO) 40 60 Baseline Peak height shift as reduction as percentage percentage of peak of peak height (YO) height (YO) - - -10 (K), 60 (K), 40 (Ca) - 40 (K), 20 (Ca) -25 (Ca) =38 (K) , - - 103 (Ca) --40 (Ca) -23 (K), =61 (Ca) --lS(K), 70(K),60(Ca) 45 (K), 45 (Ca) - 25 (K), 50 (Ca) 4 3 (K), - ==36 (Ca) Network outputs ~ Output ratios Output for calcium K:Na Ca:Na 0.999 1000.000 999.000 0.998 998.000 998.000 Pattern number 41 42 Output for Output €or sodium potassium 0.001 1 .ooo 0.001 0.998 43 44 0.005 1 .000 0.003 1 .ooo 1 .om 200.000 200.000 1 .000 333.333 333.333 30 - 0.002 1.000 0.997 500.000 498.501 45 45 46 O.OO0 1 .ooo - - 0.989 47 48 0.003 1 .ooo 0.047 1 .OO0 0.987 333.333 329.000 1 .ooo 21.276 21.276 - 70 60% of the original peak height also produced low output ratios in pattern 16 (6.098 for K : Na and 2.874 for K : Ca).Table 3 lists the outputs of the network to the distorted patterns generated from FI traces to samples which contained only calcium. The high sensitivity of the network to low amplitude signals from the calcium ISE and the high selectivity of the other ISEs to calcium meant that all the patterns were classified correctly with a very high degree of confidence (worst output ratio for Ca: Na is 12.500 for pattern 21). Interestingly, the output ratios for Ca:K were around an order of magnitude larger than for Ca : Na, suggesting that under the conditions of this study, the system is able to discriminate spurious responses from t h e potassium ISE much better than those obtained from the sodium ISE following injection of pure calcium standards.Table 4 lists the network outputs to distorted FI traces generated from a sample containing sodium and potassium. In these examples the FI traces were distorted by shifting the baseline and adding noise as before, but the heights of the sodium and potassium ISE peaks were changed independentlyANALYST, APRIL 1993, VOL. 118 352 0.7 Table 7 Network outputs to patterns in the testing set generated from samples containing sodium, potassium and calcium (a) - Network outputs Distortions of orginal pattern employed Noise addition Baseline shift Peak height reduction Pattern Output for Output for Output for as percentage of as percentage of as percentage of number sodium potassium calcium peak height (YO) peak height (Yo) peak height (YO) 49 so 51 52 53 54 55 56 1 .000 1 .000 1 .000 0.991 1.000 0.999 0.999 1.000 1 .ooo 1 .ooo 1 .000 1.000 0.917 1 .000 0.999 1 .000 0.990 0.992 0.991 0.965 0.984 0.999 0.981 0.760 - -47 (Na), -40 (K).- -100 (Ca) - - 40 (Na), 40 (K), 30 (Ca) 60 - 70 (Na), 40 (K), 55 (Ca) 40 30 (Na), 70 (K), 35 (Ca) 60 -12 (Na), -30 (K), 50 (Na), 30 (K), 20 (Ca) 30 -28 (Na), =24 (K), 55 (Na), 55 (K), 25 (Ca) 70 -70 (Na), -60 (K), =35 (Na), -30 (K), -75 (Ca) =25 (Ca) =60 (Ca) -150 (Ca) - of each other. The worst network output to any of the patterns in the test set was produced to pattern 26. Although the sodium and potassium were positively identified as being present, the high levels of noise on the calcium ISE signal made it difficult to determine whether there was also a calcium peak present.However, even in this difficult example, the output ratios of 1.704 (Na : Ca) and 1.709 (K : Ca) the network is almost twice as confident of the presence of the correct ions in the test solution. Table 5 lists the network outputs to distorted FI traces generated from a sample containing sodium and calcium. The network had the most difficulty with patterns 38 and 40. For pattern 38, the noise addition at 60% of the original peak height and the interference signals from the potassium ISE to the sodium and calcium present reduced the output ratios for Na : K and Ca : K to just over six. Similarly for pattern 40 in which noise at 50% of the maximum peak height had been added and the heights of the sodium and calcium TSE peaks had been reduced, the same output ratios were reduced to 10.753 and 8.581, respectively.Table 6 details the network outputs to distorted FT traces generated from a sample containing potassium and calcium. In all cases described in this table the patterns were identified correctly with a high degree of confidence, the worst example being pattern 48 where increasing the noise added to 70% of the original peak heights caused the output ratios for K : Na and Ca:Na to fall to around 20 from values of over 300 (pattern 47), even though other distortions were not as severe. Table 7 details the network outputs to distorted FI traces generated from a sample containing all three cations.The network was able to cope confidently with all distortions investigated. The least confident result was a 0.760 prediction for calcium (pattern 56), which may have been caused by a larger shift in the baseline (150% of peak height compared with 70% for sodium and 60% for potassium). It can be seen from these tables that for 44 of the 56 ion combinations investigated, the patterns in the testing set were classified correctly, based on the assumption that an output of greater than 0.9 indicated the presence of the species and an output of less than 0.1 indicated its absence. However, for those patterns which were not classified correctly according to these criteria there was still a greater degree of confidence in the presence of the correct ion(s) rather than the other ions which may have been present in the sample.If we consider the outputs of the neural network to ions which are present relative to ions which are absent, it can be seen that in around 38 of the 48 possible patterns studied, the classification of the pattern indicated a degree of confidence of at least 10 times greater for the presence of the correct ion relative to the other possible ion permutations. In fact, in every case, the network 0.6 * "A* 0.5 n "A 0.4 0.3 0.2 0.1 1 7 13 19 25 31 37 43 49 55 61 67 73 79 L c, 0-0.3' I I ' I I I I ' ' ' 1 - 6 12 18 24 30 36 42 48 54 60 66 72 78 w Q 0.5 4 0.4 0.3 g 0.2 0.1 0 ' I I I I I I I I I 1 ' I ' 6 12 18 24 30 36 42 48 54 60 66 72 78 0.6 (d) 0.4 0.2 0 8 16 24 32 40 48 56 64 72 80 Data point index Fig.7 ( a ) Undistorted FI trace of sample containing potassium and calcium. A, potassium ISE; B, calcium ISE and C, Na ISE. (6)-(d) FI traces of a sample containing potassium and calcium distorted by reducing the heights of the potassium and calcium peaks by 60 and 40%, respectively. shifting the baseline by 10% of the maximum peak height and adding noise at 60% of the maximum peak height. The desired pattern classification was 0, 1, 1, the network classification of the pattern was 0.001, 0.998, 0.998. (b) Sodium ISE; (c) potassium ISE; and (d) calcium ISE favoured the correct composition, with the worst case being the output ratio of 1.7 for the correct prediction of the presence of sodium and potassium compared with the presence of calcium as discussed above (see Table 4, pattern n um ber 26).The effects of baseline shift, noise addition and peak height reduction on the array response is shown graphically in Fig. 7.ANALY 353 'ST, APRIL 1993, VOL. 118 0.08 the interpretation of the MSE of the test set as a function of the number of hidden layer neurons.16 However, it was observed that there were differences between the variations and values of the MSEs from the NT5000 software and the Neuralworks Professional 11 soft- ware. Having determined that these differences did not arise from differences in the algorithms used for training, it is postulated that they may be due to variations in the random values assigned to the connection weights at the start of training. This suggests that differences in the starting positions on the error surface can determine the nature of the minimum found by the gradient descent process, which would be expected on a degenerate surface roughened by local minima.Further investigations of this effect could be performed using a simulated annealing approach, which tends to avoid the problems of local minima. Alternatively, the differences in the results obtained by the two types of software may be due to differences in the way in which the PC and the SUN SPARC performed mathematical functions and manipulated numbers. 0 50 100 150 200 250 300 350 Presentations of the training set Fig. 8 training set Overtraining in the later stages of training. A, test set and B, Fig. 7(a) shows the test pattern produced from an FI trace of a sample containing potassium and calcium.Fig. 7(b)-(d) shows the test patterns produced when the original trace was distorted by three methods: (i) by shifting the baselines of the ISE peaks by approximately 10% of the original potassium ISE peak height; (ii) by reducing the peak heights of the potassium and calcium ISE peaks by 60% and 40% of their original values, respectively; and (iii) by adding 60% maxi- mum peak height noise to the FI peaks. The neural net with 50 neurons in its hidden layer produced a classification of the pattern as 0.001 (sodium absent), 0.998 (potassium present) and 0.998 (calcium present). On careful visual inspection of the signals in Fig. 7(b)-(d) no definite conclusion with respect to the presence of the calcium ion in the sample was reached, yet the neural network interprets the underlying trend correctly, which would be very difficult using a simple threshold level or integration based pattern recognition algorithm.Following the apparent success of the pattern classification demonstrated by the neural networks trained using the NT5000 software, it was decided to investigate the learning process more closely using a different type of neural network simulation software namely Neuralworks Professional I1 which would provide more information concerning processes occurring within the network. Studies Performed Using Neuralworks Professional JI Fig. 8 highlights the later stages of training for a net containing 20 neurons in the hidden layer and shows the variation of MSE values from the training and testing sets with the number of presentations of the training set.This figure demonstrates how the MSE of the training set can continue to decrease while the MSE of the test set can begin to increase. This effect is a feature of overtraining and indicates how a network can learn the training set very well to the detriment of the general model. It was observed that overtraining or the potential for overtraining was a problem for all the neural nets studied. For some of the nets the overtraining was much more obvious and occurred before the convergence point defined in the NT5000 software, indicating firstly the problems posed by using the MSE measurement on the training set during training as the sole means of determining how well a network is learning a particular model, and secondly the difficulties associated with Conclusion Back-propagation networks have been investigated for the detection and identification of metal ions in solution based on the response profiles of the transient signals of ISEs to these ions in an FI system.The effects of distorting the patterns in fashions similar to those found in practice on the ability of the networks to perform their identification task was studied, and training and testing sets were devised to consider these deleterious effects. The networks performed well on the test sets even at extremely high degrees of distortion. ' m e same testing and training sets were used to investigate the training process further using a second type of neural network software. The second type of neural network I software allowed the user to study different network par- ameters during training.In particular the MSE on both the training and test sets was investigated as training progressed and demonstrated the effects of overtraining. This indicated the difficulties associated with the use of the MSE attribute as a sole means of determining whether a network had learned a particular model. Although there are a wide variety of parametric and non-parametric techniques used for similar pattern classifica- tion tasks, simple algorithms used for this purpose, such as those which look for a feature above a baseline would have great difficulties with some of the patterns we have presented to the neural networks for classification, especially in ex- amples where the baseline itself shifts, or where the baseline is difficult to define.Similarly some of the more advanced statistical pattern classification procedures, which use distance as a discriminant (such as the k nearest neighbour method) or those in which the discriminant is a hypersurface dividing the pattern space (such as linear discriminant analysis), are dependent on assumptions concerning the probability density functions of the different classes and of their variances. The neural network approach taken in this contribution makes fewer assumptions of this type and as such may be more applicable in those cases where patterns are produced by non-linear processes or in examples in which the class probability density functions are non-Gaussian. If the two types of neural network software used in this study are to be compared, several issues must be considered, one of the most important being the nature of the user requirements.While both types of software perform the standard back-propagation algorithm and have well devel- oped graphics interfaces depicting the topology of the networks they differ markedly in the amount of information they make available to the network designer. Neuralworks Professional I1 provides a large number of different tools for the designer to investigate the processes occurring within the net during training and which are unavailable with the NT5000354 ANALYST, APRIL 1993, VOL. 118 software, Neuralworks Professional TI also supports many different types of networks including Hopfield and counter- propagation networks, which are not available with the NT5000 software.The price to be paid for this extra versatility, however, is greater complexity, meaning that it is more difficult for the user to learn how to use Neuralworks Professional I1 than the NT5000 software. The NT5000 software also includes a hardware analogue and digital processing system, which allows for the real time acquisition and processing of signals from sensors and which is not available with Neuralworks Professional 11. It would be of interest to expand the use of neural network technology to FI pattern recognition using other types of neural networks, such as the counter-propagation net, to investigate the features of different network architectures relevant to similar pattern recognition problems, and also to investigate real time pattern recognition using spatio-temporal neural networks, such as recurrent back-propagation nets, to classify dynamically generated patterns derived from the real time responses of sensors.The FI patterns used for this study were provided courtesy of F. J. S. De Viteri Alonso, School of Chemical Sciences, Dublin City University. 1 2 References Pomerleau, D. A., Gowdy, J . . and Thorpe, C. E., Eng. Appl. Artif. Intell., 1991, 4, 279. Shada, R., and Patil, R. B., Proceedings of the International Joint Conference on Neural Networks, Washington, DC, January 1990, Lawrence Erlbaum, NJ, 1990, vol. 11, p. 491. 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Barto, A. G., and Anandan, P., IEEE Trans. Syst. Man Cybernet., 1985, SMC-15, 3 , 360. Caudill, M., A 1 Expert., 1991, 6, 28. Aleksander, I., A n Introduction to Neural Computing, TAB books, Blue Ridge Summit, PA 1988. Hecht-Nielsen, R., Neurocomputing , Addison-Wesle y . London, 1990. Pao, Y. H . , Adaptive Pattern Recognition and Neural Networks, Addison-Wesley, London. 1989. Lippmann, R., IEEE ASSP Mag., 1987,4, 4. Barker, P. G., Anal. Proc., 1991, 28, 110. van der Linden, W. E., Bos, M., and Bos, A., Anal. Proc., 1989. 26, 329. Williams, R. J., The Logic Of Activation Functions, Parallel Distributed Processing, eds. Rumelhart, D. E., and McClelland, J. L., MIT Press, Cambridge, MA, 1986, vol. I, ch. 10. Rumelhart, D . E., Hinton, G. E . , and Williams, R. J . , Learning Internal Representations By Error Propagation, Parallel Distri- buted Processing, eds. Rumelhart, D. E., and McClelland, J. L., MTT Press, Cambridge, MA, 1986, vol. I, ch. 8. Forster, R. J.. and Diamond. D., Anal. Chem., 1992,64, 1721. Cadogan, A. M., Diamond, D., Smyth, M. R., Deasy, M., McKervey, M. A., and Harris, S. J., Analyst, 1989, 114, 1.551. Moody, G. J . , and Thomas, J . D. R., Chemical Sensors, ed. Edmonds, T. E.. Chapman and Hall, New York, 1988, ch. 3. Barker, P. G., Hartnett, M. K . , and Diamond, D., Detection And Identification Of Potentiometric Flow Injection Analysis Peaks, Proceedings of the Neural Networks Workshop: Tech- niques And Applications, Ellis Horwood, in the press. Paper 2106090F Received November 16, 1992 Accepted January 29, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800347

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 355 Incorporation of Hydroxamic Acid Ligands Into Nafion Film Electrodes* Damien W. M. Arrigan, Brian Deasy, Jeremy D. Glennon, Brian Johnston and Gyula Svehla Department of Chemistry, University College, Cork, Ire land This paper describes studies on Nafion-coated glassy carbon electrodes incorporating hydroxamic acids. Desferrioxamine, a trihydroxamic acid, and glycine hydroxamate, a monohydroxamic acid, can be incorporated by ion exchange into the Nafion film. These ligands were detected in the film visually, after reaction with iron(iii), and electrochemically, by voltammetric oxidation. The complex of iron(iii) with desferrioxamine was electrochemically active a t the iron(iii) centre in the Nafion film. The electrochemistry of this complex in the film agrees well with the solution electrochemistry reported by other workers: irreversible at low pH and reversible at high pH.The ability of the modified electrode to act as a sensor for iron(li1) was assessed. Complexation of iron(ii1) from dilute solution into the polymer film was possible, and 5 x 10-7 mol 1-1 iron(iii) could be detected after a 10 min preconcentration, with differential-pulse voltammetric detection in pH 10 ammoniacal buffer. Keywords: Modified electrode; h ydroxamic acid; Na fion; voltammetry; iron(ii1) complex The development and application of chemically modified electrodes (CMEs) for metal ion determinations has received considerable attention in recent years. Immobilization of metal ion binding reagents at or on an electrode surface can lead to selective accumulation of the analyte of interest with subsequent determination by a voltammetric technique.1 Examples of ligands that have been immobilized include ethylenediamine,2 phenanthroline,3 dimethylglyoxime,4 thiacrown,s dithiopyridine6 and dithiocarbamate7 com- pounds. Ion-exchanger CMEs can also be employed for metal ion determinations.s.9 The analytical utility of such CMEs is a result of the ability of the electrode to accumulate the analyte and so serve to concentrate the analyte from dilute solution onto the electrode surface. Hydroxamic acids play an important role in microbial iron metabolismlO.11 and are, along with the catechol functionality, the most common chelating group found in siderophores. Stability constants of the order of 1030 are possible for hydroxamic acid containing siderophore complexes with iron(n1). 10 The formation of coloured complexes with iron(ii1) and other metal ions has allowed hydroxamic acids to be used in colorimetric determinations.12 Applications in flow injec- tion13 and solid-phase extraction14 have been reported recently .Ambersonls--17 has carried out a voltammetric study of synthetic dihydroxamic acids. These reagents are irreversibly oxidized at glassy carbon electrodes (GCEs) via a one proton, two electron mechanism, followed by hydrolysis to give the carboxylic acid and various nitrogen species.'7 This electro- activity has allowed amperometric detection of the synthetic as well as natural ligands and their metal ion complexes after liquid chromatographic separations. 18-19 We report here on the incorporation of hydroxamic acid ligands into Nafion-coated GCEs.Studies on the covalent attachment of hydroxamic acid functional groups to graphite electrodes and the formation of hydroxamic acid carbon paste electrodes have been reported elsewhere.20 The approach to electrode modification with hydroxamic acids reported here allows the presence of the ligand to be detected in the Nafion film both visually [by its reaction with iron(111)I and electrochemically (by oxidative voltammetry). Addition- ally, the electrodes so prepared show utility for preconcentra- * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18, 1992. tion and voltammetry of iron(m) and thus might have a role in the development of voltammetric sensors for this species. Experimental Desferrioxamine, a natural trihydroxamic acid, was obtained from Ciba Laboratories as the methanesulfonate salt (Des- feral, DFA+).Glycine hydroxamic acid (GHA+) was obtained from Sigma. These reagents were used as received. Nafion 5% solution in lower aliphatic alcohols and 10% water was obtained from Aldrich. Iron(iir) nitrate nonahydrate (BDH, general purpose reagent) was used for preparation of the iron(II1) solutions, prepared in 5 X 10-3 mol 1-1 nitric acid (analytical-reagent grade). All other reagents were of ana- lytical-reagent grade. Doubly distilled water was used throughout the experiments. The GCEs (Metrohm) were 3 mm diameter discs.These were polished on A1203 slurry and sonicated in doubly distilled water for 5 min and either air dried or dried in a glassware oven, before use. Cyclic voltammetry (CV) was carried out with a Princeton Applied Research (PAR) 174A polarographic analyser, a Metrohm E612 VA scanner and a PAR RE0074 x-y recorder. Linear-sweep (LSV) and differential-pulse voltammetry (DPV) were performed with either the PAR 174A or a Metrohm 626 Polarecord. For DPV, the pulse amplitude was always 50 mV, and the scan rate was either 5 or 10 mV s-1. A conventional three-electrode cell employed the modified electrode, a platinum wire and a saturated calomel electrode as the working, counter and reference electrodes, respec- tively. After the GCEs had been polished and dried as described above, they were coated with 20 p1 of 1% Nafion solution in methanol.This volume covered both the glassy carbon disc and the white poly(tetrafluoroethy1ene) (PTFE) insulating surround (a total area of 0.38 cm2). The solvent was allowed to evaporate to give the Nafion-coated GCE. (The thickness of this film can be estimated to be 2.7 pm, assuming that the density of dried Nafion solution is equivalent to that of bulk Nafion membranes ."'> The Nafion-coated GCE was then placed in an aqueous solution of the desired ligand (10-3 mol 1-l) for a given time period (18 h, unless stated otherwise). The hydroxamic acid ligand partitioned into the Nafion film during this time, to give the glassy carbon-Nafion- hydroxamic acid modified electrode. Iron(ii1) measurements employed the medium exchange technique, whereby the modified electrode was placed in the iron(ii1) solution for a given time [this gave the glassy356 ANALYST, APRIL 1993, VOL.118 carbon-Nafion-iron(I1i)hydroxamate modified electrode] and then removed, rinsed, carefully dried and placed in the voltammetric cell for the measurement step. The electrode was then regenerated by placing in 0.08 mol 1-1 ethylene- diaminetetraacetic acid disodium salt (EDTA). This served to remove any iron bound within the modifying film. The absence of analyte from the electrode was verified by a voltammetric scan before any further iron(rI1) accumulation steps. Results and Discussion Nafion, a perfluorosulfonate cation-exchange polymer, has been used extensively for electrode modification .22 Electrode modification with Nafion is a simple procedure and because it is a cation exchanger, various redox-active cations can be incorporated and detected electrochemically.In its commer- cially available form, DFA+ contains a protonated amino group. By placing the Nafion-coated GCE in a solution of this protonated ligand, the following ion-exchange reaction occurs: DFA+CH3S03- (sln) + NafionS03-Na+ (film) -+ NafionS03-DFA+ (film) + Na+CH3S03- (sln) so that DFA+ partitions into the Nafion film, giving the glassy carbon-Nafion-DFA+ modified electrode. Similarly, GHA+ will be protonated at the amino group in slightly acidic solution, and this ligand was also found to partition into the Nafion film. This gave the glassy carbon-Nafion-GHA+ modified electrode.Detection of Ligands in the Film These ligands were detected visually in the films by placing the modified electrode in a solution of iron(m) (1 x 10-2 rnol 1-1 for 30 s). Removal of the electrode and examination of the film for colour allowed the deep red colour characteristic of octahedral iron(ii1) hydroxamate complexes to be observed. Because both the carbon disc and its white PTFE surround were coated with Nafion, the colour formation in the Nafion coating over the PTFE is easily visible. The deep red colour in the film indicates formation of 1 : 1 DFA+-iron(ii1) (log Plol = 30.623) and 3: 1 GHA+-iron(ir1) (log PIo3 = 26.524) com- plexes; both complexes retain a net positive charge and are retained within the film. The electroactivity of hydroxamic acidsls-19 makes their detection in the film by oxidative voltammetry attractive.Their oxidation results in cleavage of the C-N bond, yielding a carboxylic acid and various nitrogen species. Differential- pulse voltammetry of glassy carbon-Nafion-DFA+ and glassy carbon-Nafion-GHA+ modified electrodes in acetate buffer (pH 4) revealed two oxidations for each ligand, as shown in Fig. 1. The DFA+ modified electrode exhibited peak poten- tials at 0.22 and 0.52 V, whereas the GHA+ modified electrode exhibited peaks at 0.15 and 0.51 V. A second scan immediately after the first revealed little oxidation behaviour, indicating that the electroactive component at the glassy carbodfilm interface had been depleted during the first scan. Amberson17 found hydroxamic acids to be irreversibly oxi- dized; in the Nafion films used here, if the ligand is irreversibly oxidized then the fragments of the compound will occupy the sites near the Nafionkarbon interface thus preventing fresh ligand diffusing to the interface to be oxidized on the second scan.Two oxidations are observed for each ligand, one a trihydroxamic acid and the other a monohydroxamic acid. 'Nafion has three phases:25>26 (i) a hydrated ionic cluster phase; (ii) a highly hydrophobic fluorocarbon backbone phase; and (iii) an interfacial region between these two. It is possible that the ligands are distributed between two different regions of the Nafion, and that oxidation is more difficult in one than the 0.0 0.5 E N 1 .o Fig. 1 Oxidative differential-pulse voltammograms for glassy car- bon-Nafion-hydroxamic acid electrodes.A, DFA+; and B, GHA+. 1 and 2 refer to the first and second scans on the electrode. Electrolyte, 0.1 mol I-' acetate (pH 4); and scan rate, 5 mV s-1 t 4- C 2 3 0 0.0 0.5 1 .o 0.0 0.5 1 .o E N versus SCE Fig. 2 Oxidative voltammograms for glassy carbon-Nafion-hydrox- amic acid electrodes in 0.1 mol 1-1 ammonia-ammonium nitrate buffer (pH lo), scan rate 5 mV s-1. ( a ) DFA+; and (b) GHA+. Times of exposure of electrode to ligand solution: A, 5 ; B, 10; C, 15; and D, 20 min other. This type of behaviour has been reported for inorganic complexes in Nafion films.27 Better DPV peak shapes were obtained when a pH 10 ammoniacal buffer was employed in the voltammetric oxida- tion of the ligands in the film.At pH values above the pK, values of the ligands, de-protonation of the ligand occurs as a separate step to the electron transfer, whereas at pH values below the pK, de-protonation and electron transfer occur simultaneously. More facile electron transfer will result in better DPV peak shapes. (Hydroxamic acids usually have pK,ANALYST. APRIL 1993, VOL. 118 357 values of about 9.) Both the DFA+ and GHA+ electrodes underwent only one oxidation in this electrolyte, at peak potentials of 0.23 and 0.37 V, respectively (Fig. 2). Second scans immediately after the first again indicated the depletion of the electroactive component at the glassy carboaafion film interface, due to irreversible oxidation. The increase in the peak current with increasing time of exposure of the electrode to the ligand solution indicates that more ligand is being incorporated into the Nafion film.Voltammetry of the Iron(I1r) Complex The visual detection of the hydroxamic acid ligands in the film is due to their reaction with iron(nr) and transportation of this ion into the Nafion film. Iron(1rr) hydroxamate complexes are electroactive at the metal centre28--30 and we have demon- strated that the complex can also be detected in the film by voltammetry. Complexation of iron(II1) by DFA+ gives ferrioxamine, FA+. A study by Helman and Lawrence30 has shown that FA+ is irreversibly reduced at low pH electrolytes, whereas at pH values above 6.5 oxidation responses can be observed on the reverse half-cycle in CV. At pH 7.5 reversible electrochemistry of the FA+ complex is observed at a mercury electrode .3O Raymond and co-workers2*~29 have also demon- strated reversible electrochemistry of iron(II1) hydroxamate complexes at neutral and basic electrolytes.The reduction of FA+ is a one electron reduction at the iron centre. The electrochemistry of the glassy carbon-Nafion-FA+ modified electrode [i.e., the DFA+ modified electrode after reaction with iron(111)I was examined in various electrolytes. Fig. 3 shows voltammograms of the glassy carbon-Nafion- DFA+ electrode after 30 s in mol 1-1 iron(rII), in 0.1 mol 1-1 nitric acid electrolyte. The characteristic red colour was present in the portion of the film over the PTFE before and after voltammetry. The two reduction waves might be due to reduction of ion exchanged (-0.52 V) and com- plexed (-1.13 V) iron(n1).A small oxidation response at -0.34 V indicates a reversible couple, Ell2 = -0.43 V and AEp = 0.18 V. This is consistent with reduction and re- oxidation of non-complexed iron(1II) in the Nafion film. The reduction at - 1.13 V has no corresponding oxidation; if it is assumed that this is the complexed iron(nr) reduction, an irreversible response would be expected at the low pH of the electrolyte employed, as complexation of the iron(rIr) product by DFA+ is not favourable below the pK, of the ligand groups.30 On the second CV cycle, only the more negative reduction is observed. This is due to formation of fresh FA+ complex in the Nafion film from the non-complexed iron(rI1). This peak decreases on repetitive cycling of the potential and eventually no response is observed.1.5 0.0 E N Fig. 3 Cyclic voltammograms for the glassy carbon-Nafion-DFA+ electrode after 30 s in 10-2 mol 1-I iron(w). Electrolyte, 0.1 moll-' nitric acid; and scan rate, SO mV s-'. 1, First cycle; 2, second cycle; and 3, blank scan In 0.1 moll-1 sodium nitrate, only one reduction wave is observed, at -1.17 V, equivalent to the more negative reduction observed in nitric acid electrolyte. The Nafion film retains its red colour after CV for both electrolytes, indicating the presence of FA+ in the regions of the Nafion over the PTFE surround of the electrode. Given that only one reduction is observed in the sodium nitrate, the lower pH of the acid electrolyte might induce dissociation of the complex in the film, to result in the two waves observed. Fig.4 shows a set of voltammograms for the glassy carbon-Nafion-DFA+ electrode after exposure to iron(1rr) solution, where the electrolyte was 0.1 mol 1-1 ammonia- ammonium nitrate (pH 10). As usual the Nafion film was red after removal of the electrode from the iron(r1r) solution. The first CV cycle shows a reduction at -0.58 V and an oxidation at -0.72 V. Visual examination of the Nafion film at this stage reveals a yellow-orange colour instead of the previous red. The second CV cycle shows a reduction at -1.33 V and an oxidation at -0.77 V. Subsequent cycling decreases the magnitude of both the cathodic and anodic peaks; placing the electrode in EDTA solution removes all colour as well as the electrochemical responses , and the complete procedure of iron(m) accumulation, CV cycling and observation of peak potential shift can be repeated.The shift in reduction potential from -0.58 to -1.33 V is attributed to a pH change within the film: iron(II1) is accumulated from a solution 5 x 10-3 mol 1-1 in nitric acid and subsequent voltammetry is at pH 10, so that once the electrode is placed in the voltammetric cell after iron(II1) accumulation the pH within the Nafion film begins to change. The reduction potential of FA+ in solution is known to shift negatively with increasing pH2830 as do those for other iron(II1) hydroxamates.29 These complexes exhibit reversible electrochemistry in solutions at neutral to basic pH. The FA+ complex has been shown here to exhibit reversible electro- chemistry (Fig.4, curve 2) in a Nafion film when pH 10 buffer is the electrolyte, with E1/2 = -1.05 V and AEp = 0.56 V. The large AE, for the process indicates complications in the electron-transfer kinetics and/or analyte diffusion in the film.31 The colour change observed in the film, from red to yellow-orange, is perhaps due to partial dissociation of the 1:l FA+ hexadentate complex when the pH of the film increases. Although the FA+ complex is known to be hexadentate over a wide range of solution pH values, in the Nafion film, the close proximity of sulfonate sites can induce dissociation of some of the ligand groups, causing the observed colour change. Acetate buffers at different pH values were examined. The highest current responses were obtained for the glassy carbon-Nafion-DFA+ electrode after accumulation of iron(1Ir) in these.A buffer of pH 4 gave the largest current. This can be attributed to swelling of the polymer film while in this electrolyte, hence allowing increased diffusivity of the FA+ complex and so larger currents. Fig. 5 shows a CV for the t Y c 2 3 V I 1 I -1.5 0.0 E N versus SCE Fig. 4 Cyclic voltammograms in 0.1 mol 1-I ammonia-ammonium nitrate electrolyte (pH lo). Other conditions as in Fig. 3. 1, First cycle; 2, second cycle; and 3, blank scan358 ANALYST, APRIL 1993, VOL. 118 t c 2 3 u -1.5 0.0 E N versus SCE Fig. 5 Cyclic voltammogram in 0.1 mol 1-1 acetate buffer, pH 4. Other conditions as in Fig. 3. A, After 30 s in 10-2 mol I-* iron(ii1); and B, blank scan 60 40 f . 0 d = 20 I I t 0 2 4 6 8 1 0 Electrolyte pH Fig.6 Peak currents for the glassy carbon-Nafion-DFA+ electrode after 30 s in 10-2 moll-' iron(1ii) obtained from cyclic voltammetry in different pH electrolytes modified electrode in this electrolyte. The electrode reaction is irreversible at this pH due to competition of protons with iron(ii1) for the available complexing sites.30 Our results for the electrochemistry of FA+ in Nafion films are in good agreement with those in the literature for solution electro- chemistry of iron(iii) hydroxamates:28-30 irreversible at low pH and reversible at high pH. In acetate buffers above pH 4, the peak currents decreased dramatically, perhaps due to shrinkage of the polymer, to values the same order of magnitude as those recorded in the other electrolytes.This is illustrated in Fig. 6, indicating that acetate buffer of pH 4 is the optimum electrolyte for cyclic voltammetric detection of complexed iron( 111) with the glassy carbon-Nafion-DFA+ electrode. Utility of the Electrode as a Sensor for Iron(m) As an examination of the use of the glassy carbon-Nafion- DFA+ electrode as a sensor for iron(iir), the electrode was tested for the ability to detect trace concentrations of this analyte. By using LSV, detection limits for iron(ii1) of 5 x 10-5 mol 1-1 were obtained for a 10 min preconcentration period. Preconcentration took place at open circuit and the electrode was then removed from the sample vessel and placed in the voltammetric cell. The linear range was short, up to 2 x 10-4 mol 1 - 1 , owing to equilibration of the accumulation t Y S 2 3 u 0.5 1.3 E N -0.5 1.3 Fig.7 Differential-pulse voltammograms for iron(iii) at the glassy carbon-Nafion-DFA+ electrode after 10 min prcconcentration. Iron(rii) concentrations: A, 0.5 X 10-6; B, 1 x C, 3 X 10-6; D, 5 x 10-6; E, 7 x 10-6; and F, 10 x lO-6mol I-'. Electrolyte as in Fig. 4; and scan rate, 10 mV s-1 reaction [complexation of iron(ri1) by DFA+]. The peak current was linear with a preconcentration time of up to 20 min, so that lower detection limits should be possible using LSV. However, at the low iron(iii) concentration of interest the background current was large relative to the analytical signal. The use of a more sensitive voltammetric technique that discriminates against the background current should allow lowering of the detection limit.Very broad peaks were obtained using DPV in acetate buffer at pH 4. The use of a higher pH buffer (ammonia-ammonium nitrate) resulted in better DPV responses. The electrode reaction is more reversible in the higher pH electrolyte than the lower (see Figs. 4 and 5 , respectively). With DPV as the detection technique and the ammoniacal buffer as electrolyte , iron(ir1) could be determined down to 5 X 10-7mol1-1 using a preconcentration period of 10 min. The current response was linear with concentration up to 10-5 moll-' ( r = 0.9988, n = 6). Voltammograms for this range are shown in Fig. 7. This represents a decrease in the detection limit of two orders of magnitude over the LSV technique. There is a large back- ground current present at the negative potentials required for reduction of the complex at pH 10.This limits the applicability of the electrode to concentrations higher than 5 x 10-7 moll 1 of iron(ii1). The apparent presence of two peaks for the higher concentrations of iron(m) in Fig. 7 is due to reduction of the complex in different regions of the Nafion.25--27 Open circuit accumulation of iron(i1i) is an improvement over the covalently modified hydroxamic acid electrode, and the detection limits with both LSV and DPV are one and three orders of magnitude better, respectively, than the DFA+-based carbon-paste electrode.20 The ability of hydroxamic acid containing Nafion film coated electrodes to uptake iron(1ii) from dilute solution makes thcm attractive as sensor devices for this species.Additionally, because the complex is coloured, it is possible that spectroelectrochemical techniques can be applied to give further information about the Nafion-hydroxamic acid com- posite film. The formation of complexes between hydroxamic acids and other transition metal ions means that the selectivity of glassy carbon-Nafion-hydroxamic acid electrodes must bc carefully evaluated. However, the ability of the sulfonate sites of the Nafion to accumulate cationic species by ion exchange provides an additional source of interference. PreliminaryANALYST, APRIL 1993, VOL. 118 359 investigations into possible interferents with the glassy car- bon-Nafion-DFA+ electrode have shown that cadmium, zinc and copper(n) ions can be transported into the film and detected electrochemically. Cadmium and copper(I1) had a detrimental effect on the iron(n1) response. The availability of an alternative preconcentration scheme within the electrode modifier might lead to a device capable of speciation measurements: iron(rr) was accumulated from dilute acid solution and was reduced at -0.53 V, in 0.1 mol 1-1 ammonia -ammonium nitrate buffer.This response is well separated from that for iron(rr1). Conclusions The work presented in this paper demonstrates that hydrox- amic acid reagents can be successfully incorporated into Nafion-coated electrodes. This approach to preparation of hydroxamic acid modified electrodes has been more successful than either covalent attachment to graphite electrodes, or preparation of DFA+-containing carbon-paste electrodes.20 By using the preparation technique described here it is possible to detect the ligand on the electrode surface, employ open circuit accumulation of iron(m) and detect sub-pmol concentrations of this analyte.The electrochemistry of the DFA+ complex of iron(1n) in the Nafion film is similar to the solution behaviour reported by other workers . 2 8 3 Further work is required on the evaluation of this class of electrodes as sensors for iron(m), as only a preliminary investigation is reported here, particularly in the area of interferences. Species such as vanadium(v) and aluminium(nr), which form complexes with DFA+, must be examined for their effect on the electrode response. References 1 Wang, J . , in Electroanalytical Chemistry, ed.Bard, A. J . , Marcel Dekker, New York, 1989, vol. 16, p. 1. 2 Cheek, G. T., and Nelson, K. F., Anal. Lett., 1978, A l l , 393. 3 Kasem, K. K., and Abruna, H. D., J. Electroanal. Chem., 1988, 242, 87. 4 Baldwin, R. P., Christensen, J. K., and Kryger, L., Anal. Chem., 1986,58, 1790. 5 T&aka, S . , and Yoshida. H., Talanta, 1989, 36, 1044. 6 Sugawara, K., Tanaka, S . . andTaga. M., J. Electroanal. Chem., 1991, 304, 249. 7 O’Riordan, D. M. T., and Wallace, G. G., Anal. Chem., 1986, 58, 128. 8 Hernandez, L., Hernandez, P . , Blanco, M. H . , and Sanchez, M., Analyst, 1988, 113, 41. 9 Cassidy, J. F., and Tokuda, K., J. Electroanal. Chem., 1990, 10 Nielands, J . B., Ann. Rev. Biochem., 1981, 50, 715. 11 Raymond, K. N., and Carrano, C. J., Ace. Chem. Res., 1979, 12, 183.12 Agrawal, Y. K.. and Roshiana, R. D., Rev. Anal. Chern., 1980, 4, 159. 13 Senior, A. T., and Glennon, J. D.,Anal. Chim. Acta, 1987,196, 333. 14 Glennon, J. D., and Srijaranai, S . , Analyst, 1990, 115, 627. 15 Amberson, J . A., in Electrochemistry, Sensors and Analysis, eds. Smyth, M. R., and Vos, J . G., Elsevier, Amsterdam, 1986, 16 Amberson, J. A., and Svehla, G., Anal. Proc., 1986, 23,443. 17 Amberson, J . A., Ph.D. Thesis, Queen’s University, Belfast. 1987. 18 Glennon, J. D., Woulfe, M. R., Senior, A. T., and Ni Choileain, N., Anal. Chem., 1989, 61, 1474. 19 Glennon, J . D., and Senior, A. T., J. Chromatogr., 1990, 527, 481. 20 Arrigan, D. W. M., Glennon, J. D., and Svehla, G., Anal. Proc., 1993, 30, 141. 21 Hoyer, B . , Florence, T. M., and Batley, G., Anal. Chem., 1987, 59, 1608. 22 Garcia, O., and Kaifer, A. E., J. Electroanal. Chem., 1990,279, 79, and references cited therein. 23 Schwarzenbach, G., Anderegg, G., and L’Eplattenier, F., Helv. Chim. Acta, 1963, 46, 1400. 24 Brown, D. A., Chidambaram, M. V., and Glennon, J. D., Inorg. Chem., 1980, 19, 3260. 25 Buttry, D. A., and Anson, F. C., J. Am. Chem. Soc., 1983,105, 685. 26 Vining, W. J . , and Meyer, T. J., J. Electroanal. Chem., 1987, 237, 191. 27 Rubenstein, I . , J . Electroanal. Chem., 1984, 176, 359. 28 Cooper, S. R., McArdle, J . V., and Raymond, K. N., Proc. Natl. Acad. Sci. USA, 1978, 75, 3551. 29 Abu-Dari, K . , Cooper, S. R., and Raymond, K. N., Inorg. Chem., 1978, 17, 187. 30 Helman, R., and Lawrence, G. D., Bioelectrochem. Bioenerg., 1989, 22, 3394. 31 Murray, R. W., in Electroanalytical Chemistry, ed. Bard, A. J . , Marcel Dekker, New York, 1984, vol. 13, p. 191. 285,287. p. 10s. Paper 2lO5368C Received October 7, 1992 Accepted January 4, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800355

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 36 1 Poly(pyrro1e) Based Amperometric Sensors: Theory and Characterization * Michael E. G. Lyons, Cormac H. Lyons, Catherine Fitzgerald and Thomas Bannon Physical Chemistry Laboratory, University of Dublin, Trinity College, Dublin 2, Ireland The utilization of electropolymerized electronically conducting polymers as amperometric chemical sensors and electrocatalysts is described, with specific emphasis on poly(pyrro1e) based materials. A theoretical model describing the operational principles of the polymer sensor is also described. The processes of substrate transport t o the sensor surface, and subsequent substrate reaction at the latter, were analysed in the context of rotating disc v.oltammetry. Two distinct situations were considered. In the first, the substrate does not partition into the layer, but simply reacts, via Butler-Volmer kinetics, at the polymer/solution interface.The second situation arises when the substrate partitions into the polymer layer. In this instance, substrate diffusion and reaction within the film was analysed. The porous nature of the polymer film is specifically taken into account. The redox chemistry of poly(pyrro1e) films doped with CI- and DBS- ions was examined using cyclic voltammetry and complex impedance spectroscopy and the mechanism of redox switching in these materials was investigated. The electrode kinetics of the quinone-hydroquinone redox couple at the doped poly(pyrro1e) films was also examined using cyclic voltammetry and rotating disc electrode voltammetry, and the mechanistic pathway was elucidated.Keywords: Amperometric sensor; poly(pyrro1e) electrochemistry; electronically conducting polymer; theory The design, fabrication and application of novel ampero- metric chemical and biological sensors has been the subject of considerable research interest in recent years. The subject of chemically modified electrodes has also been a very fruitful area of research activity. With respect to the latter field of research, considerable emphasis has been placed on the fabrication and characterization of electroactive polymer (redox polymer and electronically conducting polymer) modi- fied electrodes. General surveys of the electrochemistry of electroactive polymers have been provided by Hillman,l Lyons' and Evans.3 The application of polymer modified electrodes in analytical chemistry is of increasing interest. The recent review by Wring and Hart4 provides a good summary of this topic. Some specific examples can be mentioned.For instance, osmium-containing redox active metallopolymers have been used to determine nitrite amperometrically using a flow injection system.' Enzymes can be immobilized in electronically conductive polymer films to form novel am- perometric biosensor devices.6.7 It should be noted, however, that much of the analytical work reported to date has been mainly qualitative in char- acter. With some exceptions,Gg little effort has been made to provide a quantitative description of the systems examined. This is surprising given the fact that such an analysis is necessary in order to ensure that the sensor design is conducted on a rational basis.Further, the mechanism by which electronically conducting polymers enhance or catalyse the rate of interfacial electron transfer processes has not been examined to any great extent.9-13 This is also surprising, given that this is the fundamental process that governs the operation of a chemical sensor operating in the amperometric mode. With respect to the latter comment the work of Jakobs et al.11 Haimerl and Merz,9 Miller and Bockris,l2 Mao and Pickup13 and Lyons and co-workers14,I' is of note. The sensitivity and limit of detection of a polymer based sensor will also depend mainly on the background current exhibited by the material. The latter will, to a great extent, be a function of the morphology of the polymer, which in turn will depend on the * Prescnted at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn meeting, Dublin. Ireland, September 16-18, 1992.electropolymerization conditions employed, 16 and on the nature of the dopant counter ion incorporated in the polymer matrix to ensure electroneutrality . Consequently, a thorough characterization of a polymer based amperometric sensor device should also include an examination of the redox switching characteristics of the polymer, although the latter topic has been the focus of considerable attention in recent years. 17 A number of these themes will be addressed in the present paper. In particular, a quantitative analysis of the operational characteristics of conducting polymer amperometric sensors is outlined.Further, the characterization of the redox chemistry of these materials using cyclic voltammetry and complex impedance spectroscopy is presented. Finally, the mechanism and kinetics of mediation of outer sphere electron transfer processes at poly(pyrrole) (PPy) based sensor materials is discussed. Amperometric Detection at Conducting Polymer Surfaces: A Theoretical Appraisal The following analysis will be concerned with amperometric detection at a conducting polymer coated rotating disc electrode (RDE). This specific electrode configuration was chosen for the following reason. The transport of substrate to an RDE surface is well defined hydrodynamically (the diffusion layer thickness can be evaluated quantitatively) and steady-state conditions pertain.The latter condition simplifies the mathematical analysis considerably. Further, ampero- metric detection is usually conducted under steady-state con- ditions. It is assumed that the sensor is operated at a potential such that the polymer is electronically conductive, although it will be shown later that a more physically correct description involves the invocation of a semiconducting band structure for the material. The problem is to obtain an analytical expression for the faradaic current under steady-state conditions. The situation is outlined schematically in Fig. 1. The general reaction sequence at a conducting polymer modified electrode can be represented as follows: k'D k'ME sm - sL - products where kID and represent the mass transfer and heterogeneous modified electrode rate constants, respectively362 Conducting polymer , 1 electrode Support I X = O x = L Fig.1 Schematic representation of operating as an amperometric chemical Solution a. conducting polymer film sensor (units: cm s-1). The former quantity is given by the following expression : where D is the substrate diffusion coefficient (.=lo-5 cmL s-1 for aqueous solutions), X, is the diffusion layer thickness, v is the kinematic viscosity, w is the rotation speed and B is the Levich factor, i.e., 1.550: v-i. Typical values for k'ME are not very numerous, but a value of 1 x 10-3 cm s-1 can be quoted from work reported by Haimerl and Merz' for the quinone- hydroquinone redox couple at PPy coated Pt electrodes in buffered aqueous media of low pH.This result should be compared with a value of about 1 x 10-6 cm s-1 obtained at uncoated Pt electrodes under similar experimental conditions. Values of klME = 1 X cm s-l have been reported8 for the oxidation of catechol at conductive composite electrodes consisting of Ru02 microparticles dispersed in Nafion matrices. Hence a significant acceleration in rate is observed as a result of surface modification using a Conductive matrix. Following an approach suggested by Albery and Hillman,l8 application of the steady-state approximation to the reaction sequence outlined above results in the following expression for the reciprocal of the steady-state current (i): klD = D/X, = 1.55D2/3~-'/6~'/2 = Bw1/2 (1) This is the Koutecky-Levich equation for the amperometric sensor where n = number of electrons transferred, F = Faraday constant, and A = geometric area of the electrode.Hence we note from eqn. (2) that a plot of reciprocal current versus w-t is linear, with slope yielding B-1 and intercept (corresponding to the situation of infinite rotation speed) giving the kinetically significant quantity k'ME-'. The Koutecky-Levich type of analysis outlined above is to be preferred to that analysing the shape of the cyclic voltammet- ric response9 in that the processes of solution phase material transport (k' term) and chemical reaction/diffusion within the layer ( k f M E term) are clearly separated. The essential theoretical problem, therefore, is to evaluate k f M E . Details of the mathematical solution to the differential equation describing the transport and reaction kinetics within the layer have recently been reported from this laboratory.15 In the present paper, discussion will be focused on the current response obtained using this analysis.Two possible modes of reaction can be considered. The first case corresponds to the situation where the polymer film is impermeable to substrate. Hence the substrate will simply react at the polymer/solution interface. We do not consider the case where the reaction may be limited by the rate of charge percolation through the layer, because the polymer is electronically conductive. This form of rate limitation must be considered if redox polymer films are used as amperometric sensors because the conductivity in these materials occurs via an electron hopping mechanism between localized redox sites attached to the polymer backbone, a process which will ANALYST, APRIL 1993, VOL.118 intrinsically be less efficient than the rather facile charge carrier transport along conductive organic polymer chains. Consequently, the situation at an impermeable conductive polymer electrode is exactly analogous to reaction at a conventional metallic electrode and the heterogeneous modi- fied electrode rate constant is given by the simple Butler- Volmer equation k ' M E = k'E = k"~xp[a@] (3) where klF is the heterogeneous electrochemical rate constant, cx is the transfer coefficient, k" is the standard rate constant and 0 is a normalized potential given by 0 = F(E - E")/RT. The second case represents the fundamentally more com- plex situation where the polymer layer is permeable to substrate.In this situation substrate transport and reaction kinetics within the layer must be evaluated. In the latter instance the total current is obtained by solving the bounded diffusion problem for substrate transport and chemical reac- tion within the layer, and can be shown to have the following form: i = nFADsDFks"{tanh[L/XK](~sxK -k kDFX[i t a n h [ L / X ~ ] ) ~ ' } (4) where XK represents the reaction layer thickness given by: X K = (DF/k'E)' ( 5 ) and k is the partition coefficient ( L is the layer thickness and XD is the diffusion layer thickness). In these latter expressions we have differentiated between substrate diffusion in solution (Ds) and in the layer (DF).The reaction layer thickness quantifies the distance into the layer the substrate travels before it is consumed via chemical reaction at the polymer strands. Diffusional transport in solution and diffusiodreaction processes in the polymer film can be readily separated by inversion of the current expression outlined in eqn. (4) to obtain nFAs"/i = X[,/Ds + ( x ~ / k D ~ ) c o t h [ L / X ~ ] (6) k'ME = (kDF/XK)tanh[L/XK] (7) We compare eqns. (2) and (6) to obtain and the current due to diffusiodreaction processes within the layer, iF, is given by iF = nFAkSWDFXKp'tanh[L/XK] ( 8 ) At this stage attention must be focused on the ratio L / X K , where L is the layer thickness. When LIXK >> 1 then tanh[L/XK] = 1 and the expression for iF outlined in eqn.(8) reduces to iF = nFAks"DFXK-L (9) Alternatively, if L/xK << 1 then tanh[LIXK] = L/xK, and eqn. (8) admits the form iF = nFAkDFswL/XK2 = nFAkfEksWL ( 10) We now note the following important point. The reaction layer thickness XK will depend on potential and we can write that X K = (Dt./k'~); = (DF/k")l exp[ - &@/2] (11) Therefore, as the applied potential 0 increases, the factor exp[ - a@] will decrease and consequently the reaction layer thickness XK will also decrease. Froin eqns. (9) and (11) we obtained that for L >> XK, i.e., thick layers, the current- potential response will admit the form iF = n FAks" D,ik"kxp[ a@/2] ( 12) and the experimentally determined transfer coefficient aexpl = d 2 . For a simple single electron transfer reaction LY == 0.5 usually. Hence for a similar reaction taking place at a porousANALYST, APRIL 1993, VOL.118 /' 363 ----- \------/-7 conducting polymer matrix one would expect that aexpl = 0.25, i . e . , for thick films the Tafel slope 6 = 2.303RT/01e,plF, will be twice that expected. This result has a physical basis. It is well established that thick electrodeposited layers of electron- ically conducting polymers are fairly porous. Hence, for large values of applied potential, XK becomes very small and only a small fraction of the inner surface of the pore will be utilized. The perceived sensitivity of rate to applied potential (expressed as a Tafel slope) will be less than that expected for a planar electrode because only a small proportion of the active surface is being utilized.The porous matrix will also be electronically conductive at high potentials and so the reactant will only have to diffuse a short way into the layer before it is consumed via chemical reaction. Hence, as noted in eqn. (13), the reaction rate iF will be independent of layer thickness. In contrast, for thin layers when L << XK, the current-potential expression obtained from eqns. (10) and (11) becomes 3 0 iF = nFAks" Lk"exp[ 0101 L In this instance normal Tafel behaviour is observed with O1expl = 01 and the entire layer is utilized. Further, the reaction rate will depend in a linear manner on the layer thickness. Hence the reaction kinetics of a solution phase substrate at a conducting polymer surface will only depend on the layer thickness for small values of the latter. If thick layers are used then the rate will be independent of layer thickness. This may account for the varied and inconsistent observations regarding rate/layer thickness dependencies reported in the liter- The prediction that the observed transfer coefficient aexpl can depend on the LIX, ratio has not been appreciated in previous studies reported in the literature.v.12 ature.'~.ll,l~ Experimental The PPy-DBS- (DBS- = dodecylbenzenesulfonate ion) polymer was formed via potential step electropolymerization (deposition potential, 800 mV versus Ag-AgCI) onto a Pt disc electrode immersed in a solution containing 50 mmol dm-3 pyrrole and 0.1 mol dm-3 C12H2sC6H4S03Na.A similar procedure was used to generate PPy-Cl-, except that in this instance the supporting electrolyte was NaCl (0.1 mol dm-3).Full experimental details have been published elsewhere. 14 The mechanism of nucleation and growth under controlled potential potentiostatic conditions has also been addressed in a recent paper,l6 and so will not be discussed here. All solutions were prepared using ultra-pure Milli-Q water and AnalaR-grade reagents. A conventional three-electrode electrochemical cell was used. Potentials were measured, and are quoted, with respect to an Ag-AgC1 (sat. KCl) reference electrode. A large surface area Pt foil served as counter electrode. Solutions were de-gassed with oxygen-free nitrogen prior to electrochemical measurements. Cyclic voltammetry and complex impedance spectroscopy measurements were conducted under microcomputer control using an EG&G PAR Model 378 complex impedance software package.The complex impedance of pre-grown polymers was determined in supporting electrolyte solution as a function of electrode potential in the frequency range from 100 kHz to 0.5 mHz. An EG&G PAR Model 5208 lock-in amplifier was used to measure the in phase and quadrature components of the impedance in the frequency range from 100 kHz to 5 Hz. For measurements conducted at lower frequencies, a fast Fourier transform technique was used. Redox Chemistry of Surfactant Doped PPy Films A specific example of a conductive polymer electrode that has been shown to be a very effective sensor material for the amperometric determination of ascorbate14JO will now be considered.The redox chemistry of a PPy film doped with the surfactant counter ion DBS- , C12H2sC6H4S03-, is described the redox characteristics of the latter material (PPy-DBS-) are compared with those exhibited by PPy doped with chloride ion (PPy-C1-). The voltammetric responses obtained for PPy-Cl- and PPy-DBS- in 0.1 rnol dm-3 NaCl are presented in Figs. 2 and 3, respectively. The layer thickness was about 2 pm in each instance. It is clear from these voltammograms that the dopant counter ion exhibits a marked effect on the shape of the voltammetric response. A broad ill-defined response is observed for the PPy-CI- layer, in which the charge distribu- tion is considerably broad along the potential axis, signifying the presence of considerable electrostatic repulsive interac- tions between the oxidized sites in the polymer layer.21-22 The observed broadness may also be attributed, in part, to counter ion transport effects within the polymer matrix.23 The standard redox potential lies in the range 80-90 mV.The latter range is quoted because the voltammetric peaks shift slightly on repetitive potential cycling around the redox switching region. The latter observation is due to polymer restructuring effects. 1 4 - 2 3 ~ 4 The material PPy-CI- is electron- ically insulating at -400 mV and electronically conducting at 500 mV. The voltammetric profile illustrated in Fig. 2 is also characterized by the pseudo-capacitative current plateau region located at potentials anodic to the oxidation peak where the material behaves as a plastic metal.Redox switching in PPy-CI - is accompanied by anion transport in the polymer film and can be described in terms of the following reaction: PPy + X-(as) % PPy+X-(pol) + e- (14) t .+ f 2 3 0 -0.4 0 0.5 E N Fig. 2 Cyclic voltammogram of PPy-CI- (35 mC deposition charge) in 0.1 mol dm-3 NaCl. Sweep rate, 25 mV s-1. First sweep (solid line), third sweep (broken line)364 ANALYST, APRIL 1Y93, VOL. 118 Hence polymer chain oxidation results in the generation of positive charges (polarons) on the backbone that are delocal- ized over four monomer units. In order that electroneutrality is preserved, ingress of anions from the electrolyte solution into the polymer matrix occurs. The reverse occurs on reduction: anions are ejected from the polymer matrix into the solution.This mechanism has been confirmed using the electrochemical quartz crystal microbalance (EQCM) tech- nique.25-27 This hypothesis can also be confirmed using cyclic voltammetry. The material PPy-CI- acts as a permselective membrane so that cations are excluded from the coating. The Nernst equation describing the redox switching reaction involving anion injection/ejection can be written as follows: E, = E,O + (RT/F) ln([P+]/[P"][X-I} (15) where [P+] and [PO] represent the concentrations of the oxidized and reduced units in the polymer, respectively, and E, denotes the apparent formal potential of the polymer redox couple. The quantity E i ) is the formal potential of the polymer redox couple. If we assume that E, is evaluated by taking the mean of the anodic and cathodic voltammetric peak potentials then under these conditions [P+] = [P"] and eqn.(15) reduces to E, = EpO -(RT/F)ln[X-] (16) We note, therefore, that a plot of E, versus log[X-] should be linear with a slope given by 2.303RT/F, i.e., the potential of the PPyO-PPy+ redox couple shifts in a negative direction by 59 mV decade-' increase in anion concentration [X-1. This prediction is valid for an ideal anion-selective permselective ion-exchange membrane. The shift in apparent formal poten- tial is simply caused by the existence of a Donnan potential difference between the anion-exchange polymer film and the supporting electrolyte solution in contact with it. This effect is illustrated in Fig. 4. The apparent formal potential does indeed decrease with increasing anion concentration in the solution.However, the slope has a super-Nernstian value of -80 mV decade-'. Super-Nernstian shifts have also been observed by other workers.28.29 The origin of this super- Nernstian shift is not clear at present. Work is currently being carried out in this laboratory to examine this effect further. At this stage it must be stated that the analysis of voltammetric peak shifts does not convey structural informa- tion. What it does convey is information on the stoichiometry of the redox process. The determination of redox reaction stoichiometry via analysis of shifts in voltammetric peak potentials is a well established procedure in electrochemistry. 100 50 > E G 0 - 50 -2.0 -1.0 0 Log(c/mol dmp3) Fig. 4 Plot of standard voltammetric peak potential versus NaCl concentration for a PPy-CI- coated electrode.The potential values were obtained by averaging the anodic and cathodic peak potentials recorded at a slow sweep rate, 5 mV s- 1 . T = 298 K A point of caution must also be noted here. Voltammetric profiles recorded for electronically conducting polymers are usually fairly broad. This can give rise to appreciable errors in the exact determination of peak potentials. This will not negate the observed trends, however, especially the sign of the dE, versus dlogc plot. The voltammetric peaks obtained for ppy-CI- are fairly broad. Hence the uncertainty in the assignment of peak potential may lead to an uncertainty in the assignment of the magnitude of the slope of the dE, versus dlogc plot for this material.A markedly different voltammetric response is obtained for the PPy-DBS- layer (Fig. 3). On the initial potential sweep a very marked well defined reduction peak is observed, which may be attributed to the ordering of the polymer matrix by the rather large surfactant DBS- counter ion. In contrast, the anodic part of the voltammogram is rather broad and ill defined. An approximate value for the apparent formal potential in 0.1 mol dm-3 NaCl solution is -525 mV. A more well defined response is observed if the voltammetry is carried out in phosphate buffer, pH 7 (Fig. 5 ) . In this instance the redox peaks are well defined and the standard potential is -445 mV. A further feature of interest can be noted from Fig. 3. Considerable polymer restructuring occurs on repeti- tive potential cycling, if the latter perturbation involves continuous redox switching. This restructuring effect results in a shifting of the reduction peak to more positive potentials together with a significant diminution of the peak current. The current response during the anodic sweep also changes, the charge being distributed over a wide potential window.This results in an increase in background current response at potentials more positive than 0 mV. Our experiments have indicated that the PPy-DBS- layer can be subjected to a repetitive potential cycling programme in the region from -300 to +600mV without any noticeable increase in back- ground current. This indicates that little restructuring and morphology change occurs under these conditions.Hence, restructuring effects occur only if the potential limits are extended to include the region of redox switching and will always be accompanied by layer reduction. The 'bottom line' from the viewpoint of electroanalytjcal application is that PPy-DBS- layers exhibit extremely stable electrochemical responses and minimal background currents provided that the layer is not operated in a potential region that involves redox switching. As the latter process occurs at fairly negative potentials, the material is electronically conductive over a fairly large potential window (about 1100 mV). This is a very desirable property from the viewpoint of amperometric sensor applications. 1 I I - 800 - 400 0 E/mV Fig. 5 Voltammetric profile recorded for PPy-DBS- in the region of redox switching in phosphate buffer solution, Ph 7.Sweep rate, 20 mV s-lANALYST, APRIL 1993. VOL. 118 365 The PPy-DBS- material exhibits a number of unusual features. Firstly, analysis by scanning electron microscopy has indicated30 that the electrodeposited layer exhibits a rather compact morphology. Secondly, the fact that a long chain alkylsulfonate ion is used as dopant, results in the imposition of a rather ordered polymer film. This is evident in the well resolved reduction peak in the voltammogram. Thirdly, we have previously noted14 that the material exhibits a very low pseudo-capacitative current in the oxidized state, compared with most of the other electrochemically deposited conducting polymers. The subject of pseudo-capacitative current has been a vexing problem in the area of electronically conducting polymers and has been examined by a number of workers in recent years.17,31,32 Some presume that the current is capacita- tive and others assign to it a faradaic origin.If the first assignment is valid then the low current response can be attributed to the rather dense polymer morphology: the layer is so compact that there is not great exposure of surface area to the pools of electrolyte within the polymer matrix. On the other hand, if the current in the plateau region is assigned a faradaic origin, then the very low current observed may be due to efficient layer oxidation caused by the ordering of the polymer imparted by the large surfactant DBS- ion. There are no unoxidized regions remaining in the polymer at potentials more positive than 0 mV.This is clearly not the situation for PPy-CI- (Fig. 2). In the latter material there is a certain amount of 'over oxidation' in the region of the current plateau and consequently the current response is significant. Finally, the DBS- ion is large and thus should not be readily transported through the polymer matrix. This implies that redox switching should be accompanied by cation rather than anion transport and the material should be cation permselec- tive. This is in contrast to the situation previously encountered with PPy-CI-. Consequently, polymer oxidation will be accompanied by cation ejection and reduction by cation injection as follows: PP~-C+RSO~-(POI) S PPy+RS03-(pol) + C+(aq) + e- (17) One can again apply a nernstian analysis to this reaction to obtain: E, = EpO + (RT/F)ln{ [PPy+RS03-]/[PPyRS03-C+]} + (RT/F)ln[C+] (18) In this instance we note that the apparent formal potential should vary in a linear manner with the log of the cation concentration and that dE,/dlog[C+] = 2.303RT/F, i .e . , +59 mV decade-'. In simple terms this means that an increase in cation concentration will result in a positive shift in the formal redox potential. This feature is clearly illustrated in Fig. 6. In this instance a slope of 56 mV decade-' is obtained, 460 I d I - 560 -2 - 1 0 Log(c/mol dm 3, Fig. 6 Plot of standard voltammetric peak potential versus NaCl concentration for a PPy-DBS- coated electrode. Experimental conditions are similar to those outlined in Fig.4 which indicates that PPy-DBS- behaves almost ideally as a permselective cation-exchange membrane. As redox switching in PPy-DBS- involves cation ejection/ injection, it is interesting to examine the effect of cation size on the efficiency of cation transport in the polymer matrix. The results of such a study are outlined in Fig. 7, in which the numerical value of the voltammetric charge corresponding to layer reduction is plotted as a function of the dehydrated cation radius. One can assume that this charge Q, represents an approximate measure of the ease of cation injection into the film. The significant feature exhibited from these data is that the reduction charge Q, decreases as the dehydrated radius of the cation increases. Hence the phenomenon is kinetic and reflects a simple size effect: large cations are transported less readily than small cations.Hence Cs+ is transported less efficiently than either Li+ or Na+. This observation can be explained in the following manner. The surfactant doped polymer can exhibit considerable hydrophobicity. This is due to the conditions employed during electropolymerization, where 50 mmol dm-3 pyrrole and 0.1 mol dm-3 DBS- were used. It should be noted that the concentration of surfactant employed was considerably greater than the critical micellization concentration, which is 1.2 x 10-3 mol dm-3.33 Consequently, it is probable that micellar aggregates form in the solution and that the deposited film is ordered such that the pyrrole chains orient themselves around the hydrophilic surface of the micelle.The polymer environment will, therefore, exhibit a marked hydrophobic character. Consequently, before the cation can enter the matrix it must lose a significant fraction of its hydration sheath: ion-solvent interactions (Born, ion-dipole, ion-quad- rupole, ion-induced dipole) must be disrupted. This concept of partial dehydration remains speculative at present. Studies utilizing the EQCM technique should enable one to evaluate the amount of water associated with the cations as they are transported through the polymer matrix during redox switch- ing. Complex Impedance Spectroscopy of PPy-Cl- and PPy-DBS- Films Complex impedance spectroscopy has proved to be a very useful method for the quantification of electronic and ionic resistivities in electronically conducting polymer films.In principle, if the range of frequency over which the impedance response is examined is very large then both the electronic resistance RE and the ionic resistance RI can be determined. The mathematics of the impedance response of an electronic- ally conducting polymer film has recently been developed largely by Albery and co-workers34-3h and Fletcher.37 The analysis is based on the porous nature of electronically 2'5 r--- cs + 0 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 rlnm Fig. 7 Plot of the voltammetric charge for layer reduction corres- ponding to the facility of cation transport within the PPy-DBS- matrix as a function of cation radius. The voltammograms were recorded at a slow sweep rate, in a number of metal chloride solutions366 ANALYST, APRlL 1993, VOL.1113 conducting polymers. A full quantitative theory which can be used for data analysis is lacking at present. As a consequence, the impedance response obtained for the doped polymers will only be discussed in a qualitative manner. A full quantitative analysis of the data will be presented in a subsequent paper. The mathematical model employed is based on a dual rail transmission line (Fig. 8). One rail of the transmission line represents the electronic conductivity, the other rail, the ionic conductivity in the pores. The two rails are connected via a circuit element associated with interfacial redox reactions involving polaron and bipolaron states at the pore wall. Typical complex impedance spectra obtained for PPY-CI- and PPy-DBS- layers in contact with 0.1 mol dm-3 NaCl and 0.1 mol dm--7 NaDBS solutions, respectively, are illustrated in Figs.9 and 10. These spectra were obtained when the polymers were in the oxidized, conductive state. For the C1- doped material (Fig. 9), the applied d.c. potential was 400 mV and the deposition charge was 100 mC. A number of features can be noted from the impedance spectrum such as the presence of two semicircles at high frequencies. This feature was predicted in a recent paper by Fletcher.37 The small semicircle at very high frequencies is related to conduction processes associated with polaron motion along the polymer backbone (intrachain conduction) and between chains (inter- chain conduction). The former is associated with regions of high structural order and is, therefore, very fast, while the latter is associated with regions of low structural order and is, therefore, slower.The second semicircle at lower frequencies is attributed to an RC (resistancekapacitance) circuit element describing charge transfer reactions involving the localized polaronic and bipolaronic states at the pore wall. Clearly this feature dominates the over-all response in the high-frequency region. At lower frequencies a Warburg-like feature can be noted, while in the very low frequency region a vertical capacitative response is observed. The response at inter- Fig. 8 A dual rail transmission line model for electronically conducting polymers. RE denotes the electronic resistance, RI is the ionic resistance and Cis the distributed capacitance of the polymer. Rs represents the uncompensated solution resistance 80 ,G G 0 100 140 Z'IR 0.67 Hz .mediate and low frequencies is governed totally by reactions at the pore walls and not by processes occurring within the bulk polymer. A similar situation is found for the PPy-DBS- material (Fig. lo). In this instance the applied d.c. potential was again 400 mV and the deposition charge 75 mC. Only one semicircle is observed at high frequencies, and a quasi-vertical response is observed in the lower frequency domain. The impedance response totally reflects interfacial redox processes at the pore walls. The impedance spectra recorded for the doped PPy materials in the reduced, insulating states are illustrated in Figs. 11 and 12.The applied potential in both instances was -900 mV. Tn Fig. 11, for the C1- doped polymer, two semicircles are noted. The impedances are much greater in this instance, as expected. The response obtained for the DBS- doped material (Fig. 12) is less well defined, but exhibits the shape predicted by Fletcher,37 who has noted that the shape of the response corresponding to the interphasial electrochemistry at the pore walls will depend on the ratio of the double layer capacitance of the pore wall to the pseudo- 0 2.5 Z'IkR Fig. 10 Complex impedancc spcctrum for PPy-DBS- in 0.1 rnol dm-3 NaDBS. The applied potcntial was 400 mV. Deposition charge, 75 mC. The layer is in the oxidized, conductive state . . . . 40 ....-._. I I , 0 40 80 120 Z'IkO Fig. 11 Complex impedance spcctrum for PPy-CI- in 0.1 mol dm-3 NaCl.The layer is fully reduced and insulating. Applied potential = -900 mV. Same deposition charge as in Fig. 9 30 Fig. 9 Complex impedance spectrum for PPy-Cl- in 0.1 rnol dm--l NaCl. The layer is fully oxidized and conductive. Upper frequency limit, 100 kHz; lower limit, 0.67 Hz. Applied d.c. potential = 400 mV. Deposition charge = 100 mC 180 0 37 Z'IkQ Fig. 12 Complex impedance spectrum recorded for PPy-DBS- in 0.1 mol dm--l NaDBS. The layer is fully reduced and insulating. Applied potential, -900 mV. Same deposition charge as in Fig. 10ANALYST, APRIL 1993, VOL. 118 367 capacitance describing the charge/discharge reactions of the polaronic and bipolaronic states. A defined semicircle will only be observed if this ratio is very small.If this is not true, and the ratio is appreciable, then a response similar to that illustrated in Fig. 12 is predicted. The major point to note from this qualitative analysis is that over most of the available frequency range, the impedance response is dominated by interfacial redox chemistry involving localized polaronic and bipolaronic states, and not by the intrinsic conductivity of the polymer chains. This point has not been fully realized by other workers. Mediation of Outer Sphere Electron Transfer Reactions at Doped PPy Films The redox chemistry of the quinone-hydroquinone redox couple at doped PPy films will now be discussed. It is well established3840 that the quinone-hydroquinone redox transformation in solutions of low pH follows a 2e-, 2H+ process at unmodified electrode surfaces.The redox stoichiometry can be readily confirmed by cyclic voltammetry, and determining the way in which the voltammetric peak potentials vary with changes in solution pH. We have shown41 that a similar mechanism operates at a PPy coated electrode. As illustrated in Fig. 13, the voltammetric peak potentials for the Q-QH2 reaction at PPy films vary in a regular manner with changes in solution pH. The voltammetric response was examined as a function of pH for both a slow (2 mV s-1) and a fast (100 mV s-1) sweep rate. In all instances it can be seen that the slope, dE,/dpH, is close to the value of -0.059 V decade-’, when the voltammograms are recorded - at 298 K. In general we can write that dE,/dpH = -(2.303RT/F)(rn/n) 0.6 0.5 0.4 0.3 3 0.2 0.1 0 where rn denotes the number of protons transferred and n represents the number of electrons transferred in the redox process. Hence if rn = n , the predicted slope at 298 K is -0.059VpH-1.Hence, as n = 2 for the Q-QH2 reaction, then the stoichiometry of the reaction in low pH solution is Q + 2H+ + 2e- + QH2 (20) A typical voltammogram for the quinone-hydroquinone redox couple (1 mmol dm-3 in each component) at a PPy- DBS- coated glassy carbon electrode in McIlvine buffer, pH 2.2, is illustrated in Fig. 14. In this instance the sweep rate employed was 5 mV s- 1 . A corresponding voltammogram recorded at a PPy-Cl- coated electrode at the same pH is outlined in Fig. 15. The scan rate employed was 10 mV s-1. In both instances the peak separations (190 mV for the DBS- doped material and 210 mV for the CI- doped material) indicate that the redox process is quasi-reversible.One can utilize the peak separation values to determine the hetero- geneous electrochemical rate constant using a protocol originally proposed by Nicholson.42 One can relate the observed peak separation WE, to a parameter c. The latter is related to the standard electrochemical rate constant via c = (Do/D,).’*ko/[Donv(n~/R~]~ (21) where Do and DR denote the diffusion coefficients of the oxidized and reduced forms of the redox couple, and v is the scan rate. The other symbols have their usual meanings. Clearly n = 2. The required value of c was obtained from the published working curve relating nWE, values to c. Further, Do = DR = 3.46 x 10-5 cm2 s-1 for the quinone-hydroqui- none couple at PPy-CI- and Do = 1.92 X 10-5 cm2 s-1, DR = 1.67 x 10-5 cm2 s-1 for the reaction at PPy-DBS-.If these 0 I I I I I I 1 2 3 4 5 6 PH Fig. 13 Variation of voltammctric peak potentials with solution pH for the quinone-hydroquinonc rcdox couple at a PPy-DBS- electrode. The data were recorded at 2 and 100 mV s-l. Note that d F /dpH = -59mVdecade-1 (Y = 2mVs-1) and -60 mrdecade-1 (Y = 100 mV s-1). The corresponding anodic slopes wcrc slightly higher with dE,,/dpH = -64.3 mV decade-’ (Y = 100 mV s-1). Hence the ratio of protons to electrons is unity. 0, Anodic; and 0, cathodic. A, 100; and B, 2 mV s-l 0.3 E N 0.6 Fig. 14 Voltammetric response for quinone-hydroquinone at a PPy-DBS- coated clectrode. Conditions are similar to those outlincd in Fig.14, but thc sweep rate in this instance is 10 mV s-1 0 0.7 E N Fig. 15 Voltammetric response for the quinone-hydroquinone (1 mmol dm-3 in each Component. McIlvinc buffer, pH 2.2) redox couple at a PPy-Cl- coated electrode. Sweep rate, 5 mV s--l368 ANALYST, APRIL 1993, VOL. 118 t Y c i?! 13 0 t Y t 2 3 0 -0.1 0.3 0.8 EN I I I -0.1 0.4 0.8 EIV Fig. 16 Typical RDE voltammograms recorded for the quinone-hydroquinone redox couple at (a) PPy-CI- and (b) PPy-DBS- electrodes. Sweep rate, 2 mV s--l. Rotation speed, 500 rev min-1. Experimental conditions are similar to those outlined in Fig. 14 - values are substituted into eqn. (21), then the heterogeneous rate constant for the quinone-hydroquinone reaction at PPy-Cl- is 9.2 x 10-dcms-1, and at PPy-DBS-, 4.52 x 10-4 cm s-1 (oxidation) and 4.84 x 10-4 cm s-1 (reduction).The mechanism of the Q-QH2 reaction can be determined by a detailed analysis of the current-potential response curves. A typical rotating disc voltammogram recorded for the Q-QH2 couple at a PPy-DBS- coated electrode at a rotation speed of 500 rev min-1 is illustrated in Fig. 16. Again, the shape of the RDE voltammogram indicates quasi-reversible electrode kinetics. This voltammogram was recorded at the slow sweep rate of 5 mV s-1. The current-potential data are illustrated in a Tafel format in Fig. 17. A dual Tafel slope behaviour is observed for reduction, whereas only a single Tafel region is observed for oxidation. The numerical values of the Tafel slopes are b, = 0.053 V decade-', b,, = -0.39 V decade-' (low potentials) and bc2 = -0.291 V decade-' (high potentials).As the overall Q-QH2 reaction sequence involves the transfer of both protons and electrons, the reaction pathway can be described in terms of a square scheme.39-JO.43 As a buffer solution is being used, the proton transfer processes can be assumed to be at equilibrium. Hence, formally, the kinetics can be analysed in the context of a simple consecutive EE sequence, in which the heterogeneous electrochemical rate constants for the individual electron transfer steps depend on pH. The RDE voltammograms were recorded in a low pH buffer (pH 2.2). Under such conditions the general nine- member square scheme reduces to the following CECE (chemical, electron transfer chemical, electron transfer) reaction sequence: Q + H+-+QH+ QH+ + e - -+= QH' QH* + H+ + QH2+* QH2+* + C- + QH2 This sequence can be used to explain the observed kinetic data (Fig.17). We firstly consider the Tafel plot for reduction. The Tafel slope at low potentials is 40 mV decade-1, correspond- ing to a = 3/2. This corresponds to the second electron -10 --I1 3 - I= -I -12 -13 0 0 cr 0.1 0.3 E N 0.5 Fig. 17 Tafel plot analysis for quinone-hydroquinone reaction at PPy-DBS-. Voltammetric conditions as outlined in Fig. 16. General experimental conditions as in Fig. 14. 0, Cathodic; and 0, anodic transfer in the reaction sequence being rate determining. At higher potentials there is a change in the nature of the rate-determining step as implied by the change in observed Tafel slope.If the first electron transfer becomes rate determining under these conditions then the expected slope would be 120mV decade-' or a = 1/2. However, the observed slope is about 290 mV decade-' (corresponding to a = 0.2), slightly over twice the expected value. This observa- tion is in good accord with the theory outlined earlier in this paper. The apparent doubling in the Tafel slope value is due to the porous nature of the polymer layer. The first electron transfer step is rate limiting, but only a small portion of the polymer layer is reactive and eqn. (12) pertains. Hence the data reported support the theoretical analysis presented here. The dual slope Tafel behaviour and the changeover in the identity of the rate-determining step is in exact accord with the Hammond postulate.Hence, the harder a reaction is driven, the earlier will be the transition state. The observed Tafel slope of about 50 mV decade-' observed for the correspond- ing oxidation reaction again points to an assignment that the second electron transfer is rate determining in this instance. A full kinetic analysis of this system will be presented else- where.41 We are currently examining the redox chemistry of a number of redox couples at conducting polymer coated electrodes in order to test fully the applicability of the theoretical analysis presented here. The results of this work will be published in a subsequent paper. This is a contribution from the Electroactive Polymer Research Unit, Trinity College. The work has received financial support from EOLAS, the Irish Science and Tech- nology Agency (Strategic Programme), the British Council, and from the Commission of the European Communities (CEC Science Programme).M. E. G. L. is grateful for this support. As always, M. E. G. L. is also grateful to Professor P. Bartlett for useful discussions. References 1 Hillman. A. R., in Electrochemical Science and Technology of Polymers, ed. Linford, R. G., Elsevier Applied Science, Amsterdam, 1987, pp. 103-291. 2 Lyons, M. E. G., Ann. Rep. CR. Soc. Chem., 1990, 87, 119. 3 Evans, G. P., in Advances in Electrochemicul Scirncr und Engineering, eds. Gerisher, H.. and Tobias, C. W., VCH, Weinheim, 1990, vol. 1, pp. 1-74. 4 Wring, S. A., and Hart, J. P., Analyst, 1992, 117, 1215. 5 Malone, M. M., Dohcrty.A. P., Smyth, M. R., and Vos, J . G., Analyst, 1992, 117, 1259. 6 Bartlett, P. N., and Whittaker, R . G., J. Efectroanal. Cliem., 1987, 224,27 and 37. 7 Bartlett, P. N., Tebbutt, P., and Whittaker, R. G., Prog. React. Kiner., 1991, 16, 55.4NALYST, APRIL 1993, VOL. 118 369 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Lyons, M. E. G., Lyons, C. H., Michas, A., and Bartlett, P. N., Analyst, 1992, 117, 1271. Haimerl, A., and Merz, A., J. Electroanal. Chem., 1987, 220, 55. Saraceno, R. A., Pack, J. G., and Ewing, A. G., J. Electroanal. Chem., 1986, 197, 265. Jakobs, R. C. M., Janssen, L. J. J., and Barendrecht, E., Electrochim. Acta, 1985, 30, 1313 and 1433. Miller, D. L., and Bockris, J. O’M., J. Electrochem. SOC., 1992, 139, 967. Mao. H., and Pickup, P.G., J. Phys. Chem., 1992, 96, 5604. Lyons, M. E. G., Breen, W., and Cassidy, J . F., J. Chem. SOC., Faraday Trans.,, 1991, 87, 115. Lyons, M. E. G., Bartlett, P. N., Lyons, C. H., Breen, W., and Cassidy, J. F., J. Electroanal. Chem., 1991, 304, 1. Lyons, M. E. G., Lyons, C. H., McCabe, T.. and Corish, J., J. Appl. Electrochem., in the press. Faraday Discuss. Chem. Soc., 1989, 88. Albery, W. J., and Hillman, A. R., J. Electroanal. Chem., 1985, 170, 27. Mao, H., and Pickup, P. G., J. Electroanal. Chem., 1989, 265, 127. Lyons, M. E. G., Lyons, C. H., McCormack, D. E., McCabe, T. J., Breen, W., and Cassidy, J. F., Anal. Proc., 1991,28,104. Lyons, M. E. G., Fay, H. G.. McCabe, T., Corish, J., Vos, J. G., and Kelly, A. J., J. Chem. SOC., Faraday Trans., 1990,86, 2905. Lyons, M. E. G., Fay, H. G., and McCabe, T., Key Eng. Muter., 1992, 72/74, 381. Warren, L. F.. and Anderson, D. P., J. Electrochem. SOC., 1987, 134, 101. Breen, W., McGee, A., Cassidy, J. F., and Lyons, M. E. G., J. Electroanal. Chem., in the press. Reynolds, J. R., Sundarison, N. S., Pomcrantz, M., Basak, S., and Baker, C. K., J. Electroanal. Chem., 1988, 255, 355. 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 Naorik, K., Lien, M. M., and Smyrl, W. H., J. Electrochem. SOC., 1991, 138, 440. Dusemond, C., and Schwitzyebel, G., Ber. Bunsen-Ges. Phys. Chem., 1991,95, 1543. Wallace, G. G., personal communication. Naegeli, R., Redepenning, J., and Anson, F. C., J. Phys. Chem., 1986,90, 6227. Breen, W., Ph.D. Thesis, University of Dublin, Trinity College, 1991. Cai, Z., and Martin, C. R., J. Electroanal. Chem., 1991, 300, 35. Tanguy, J., Mermilliod, N., and Hoclet, M., J. Electrochem. SOC., 1987,134, 795. Hunter, R. J., The Zeta Potential in Colloid Science, Academic Press, London, 1981. Albery, W. J., Chen, Z., Horrocks, B. R., Mount, A. R., Wilson, P. J., Bloor, D., Monkman, A. T., and Elliott, C. M., Faraday Discuss. Chem. Soc., 1989, 88,247. Albery, W. J., Elliott, C. M.. and Mount, A. R., J. Electroanal. Chem., 1990,288, 15. Albery, W. J., and Mount, A. R., J. Electroanal. Chem., 1991, 305, 3. Fletcher, S., J. Electroanal. Chem., 1992, 307, 127. Vetter, K. J., Z. Elektrochem., 1952,56, 797. Laviron, E., J. Electroanal. Chem., 1983, 146, 15. Laviron, E., J. Electroanal. Chem., 1984, 164, 213. Lyons, M. E. G., Bannon, T., and Lyons, C. H., J. Electroanal. Chem., submitted for publication. Nicholson, R. S., Anal. Chem., 1965, 37, 1351. Albery, W. J., Electrode Kinetics, Clarendon Press, Oxford, Paper 2105719K Received October 27, 1992 Accepted January 4, 1993 1975, pp. 142-146.

ISSN:0003-2654    

DOI:10.1039/AN9931800361

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST. APRIL 1993, VOL. 118 371 Electronic Nose for Monitoring the Flavour of Beers* Timothy C. Pearce, Julian W. Gardnert and Sharon Friel Department of Engineering, University of Warwick, Coventry, UK CV4 7AL Philip N. BartlettS and Neil Blair School of Chemistry, University of Bath, Bath, UK 5A2 7AY The flavour of a beer is determined mainly by its taste and smell, which is generated by about 700 key volatile and non-volatile compounds. Beer flavour is traditionally measured through the use of a combination of conventional analytical tools (e.g., gas chromatography) and organoleptic profiling panels. These methods are not only expensive and time-consuming but also inexact due t o a lack of either sensitivity or quantitative information. In this paper an electronic instrument is described that has been designed to measure the odour of beers and supplement or even replace existing analytical methods.The instrument consists of an array of up to 12 conducting polymers, each of which has an electrical resistance that has partial sensitivity to the headspace of beer. The signals from the sensor array are then conditioned by suitable interface circuitry and processed using a chemometric or neural classifier. The results of the application of multivariate statistical techniques are given. The instrument, or electronic nose, is capable of discriminating between various commercial beers and, more significantly, between standard and artificially-tainted beers. An industrial version of this instrument is now undergoing trials in a brewery. Keywords: Odour detector; beer flavour sensor; conducting polymer sensor; sensor array; analytical instrum en tation Introduction Chemical Senses and Flavour The sensation of flavour is due to the simultaneous stimulation of all of the chemical senses together with an integration of the signals from the component senses by the higher brain centres.In humans there are three main chemoreceptor systems. These are gustation, or the sense of taste, olfaction, or the sense of smell, and the trigeminal sense. Taste is used mainly to detect non-volatile chemicals which enter the mouth while the sense of smell is used to detect volatile compounds. Receptors for the trigeminal sense are located in mucous membranes and in the skin, they also respond to many volatile chemicals and the trigeminal sense is thought to be especially important in the detection of irritants and chemically reactive species.In the perception of flavour all three chemoreceptor systems are involved but olfaction plays by far the grcatest role with the other two senses contributing much less to the overall perception. The sensation of smell arises from the stimulation of the olfactory neurones, the receptor cells located high up in the nose in the olfactory epithelium, by the odorant molecules. Odours can be simple or complex, a distinction which is based on the nature of the stimulus and not the quality of the sensation. A simple odour is one which consists of only one type of odorant molecule whereas a complex odour is a mixture of many, possibly many hundreds, of different types of odorant molecule.Simple odours, as defined here, are essentially man-made curiosities because virtually all naturally occurring odours are complex mixtures. Odorants are typi- cally small hydrophobic, organic molecules containing one or two functional groups and with a mass range from 34 to 300 Da. The relationships between the physico-chemical properties of the odorant molecules and the odours have been discussed by several workers172 and whilst it is clear that the * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18, 1992. t To whom correspondence should be addressed. $ Prcsent address: Department of Chemistry, University of Southampton, Highfield, Southampton, UK SO9 SNH. size, shape and polar properties of the molecule determine its odour properties the rules which govern this are poorly understood.As a result, classifications of odour type are empirical and the number of distinct odour descriptors required (the dimensionality of the problem), has not been established. Beer flavour is a complex problem because there are hundreds of compounds present. Some of these are at levels that exceed the sensory threshold ( i e . , down to parts per billion) but are below the detection limit of most gas chromatographs.3 Studies of beer flavour show that there are just over 100 separately identifiable flavour elements of which 39 or so are present in most beers with the others less common or flavour faults, i.e., off-flavours.4 Of these 39 key flavour notes in beer, 15 can be explained ( e .g . , alcoholic, estery and diacetyl), 20 partly explained (e.g., hoppy, malty and worty) and 10 cannot be explained at all (e.g., spicy, woody and grainy). The situation is further complicated by the fact that the beer flavour is unstable and its odour will change with time as the chemical composition of the beer changes. Beer is prepared commercially by batch processes and it is of concern to ensure consistency from batch to batch and overall product quality. Quality is currently assured in several ways including the use of analytical techniques such as gas chromatography (GC) or GC-mass spectrometry (GC-MS). However, the most important method remains the use of sensory panels of trained individuals who score the product on the basis of a number of flavour descriptors.All of these techniques are slow (i.e., it takes 2-3 d to obtain a result) and rather expensive. In this context the use of an electronic instrument, i.e., an electronic nose, which can assist in the monitoring of beer flavour, is highly attractive. The mammalian olfactory system makes use of a large number of non-specific receptors which show broad patterns of response. Typically, in the human olfactory epithelium there are about 50 million such receptors. These cells send their signals to secondary cells located in the olfactory bulb. There is a marked convergence at this stage with between 1000 and 20000 primary receptor cells connecting to each secon- dary cell. This suggests that the secondary cells are involved in processing and integrating the information from many input cells.This suggestion is consistent with the observation that while the primary cells are non-specific in their responses the372 ANALYST, APRIL 1993, VOL. 118 secondary cells respond to distinct categories of odours.5 The secondary cells, in turn, interact with each other and with higher cells. These interactions are reminiscent of those found in the vision system.6 In the electronic nose we have attempted to mimic some of the features of the mammalian system by combining an array of non-specific, chemical sensors with suitable data acquisition and processing software. The sensors in the electronic nose described here are conducting polymer chemoresistors fabricated by the electro- chemical deposition of this conducting polymer films across the gap between two thin gold electrodes.Conducting polymer chemoresistors, based on polymers such as poly- (pyrrole) (PPy), have been shown to respond to a wide varicty of gases including inorganic species such as ammonia and hydrogen sulfide7-9 as well as many organic vapours.l@-*2 A detailed understanding of the response mechanism is not available although it is generally believed that the adsorbed molecules affect both the inter-chain hopping electronic charge-transfer process and cause physical swelling of the polymer structure. Experimental Preparation of Odour Sensors The devices were fabricated by the physical evaporation of pure gold (99.99%) onto alumina tiles (12 x 12 mm), out of which an electrode pattern was etched by conventional ultraviolet (UV) lithography and then a final passivation layer was spin-coated and etched away to leave the areas where the polymer growth was desired. The active area of gold was approximately 1 mm2 per device, and had a 15 pm gap vertically along the centre of the gold pad.Each tile contained three independent devices and electrical contact was made to each of the devices by pads at the top of the tile. Fig. 1 shows a tile containing three electropolymerized areas seen as a black coating. All electrochemical procedures were carried out using a three-electrode system, controlled by a laboratory-construc- ted potentiostat. The reference electrode was a saturated calomel electrode (SCE), and all the potentials quoted are relative to this reference.The counter electrode was a large surface area platinum gauze, which was flamed prior to use. Tetraethylammonium tctrafluoroborate (TEATFB, Aldrich) was recrystallized from methanol, tetraethylammo- nium toluenesulfonate (TEATS, Aldrich) was recrystallized from acetone. All other background electrolytes were used as received: butanesulfonic acid (BSA, Aldrich), pentanesul- fonic acid (PSA, Aldrich) , hexanesulfonic acid (HxSA, Aldrich), heptanesulfonic acid (HpSA, HPLC grade BDH), octanesulfonic acid (OSA, Aldrich), decanesulfonic acid Fig. 1 Photograph of an alumina tile upon which three different conducting polymer chemoresistors have been selectively electropoly- merized (DSA, Aldrich), para-toluenesulfonic acid sodium salt [TSA(Na), Aldrich], para-toluenesulfonic acid monohydrate [TSA(m) Aldrich], and sodium hydrogensulfate monohydrate (NaHS04, Aldrich). Pyrrole (Py, Aldrich) was purified by passing it through an alumina filled Pasteur pipette, while aniline (AN, Aldrich) and 3-methylthiophene (3MT, Aldrich) were distilled at reduced pressure.All aqueous solutions were prepared using water from a Whatman RO 50 reverse osmosis de-ionizer, with a,. Whatman 'Still Plus' organic removal system. Aceto- nitrile (CH,CN, Aldrich HPLC grade) was distilled over calcium hydride, and propylene carbonate (PC, Aldrich) was percolated over molecular sieves, then distilled. To prepare the solution for the deposition of poly(ani1ine) (PAN), aniline was added to the background electrolyte and then the solution was acidified with concentrated sulfuric acid until the white precipitate dissolved.Prior to deposition, the devices were examined using a' low-powered microscope to check for any major mechanical defects, and the resistance of the devices was measured to ensure that there was no electrical shorting. The gold working electrodes on the devices were cleaned prior to polymer deposition by cycling in 2 mol dm-3 sulfuric acid followed by washing with water. The cleaned devices were kept under pure water until required to avoid recontamination of the surface. The 12 polymer systems used in this work along with their growth conditions are given in Table 1. The polymers were deposited by stepping the potential from 0 V to the required growth potential for a fixed time of 120 s.At the end of this time, the polymer-coated device was either stepped back to 0 V, and the current allowed to decay until it became stable, or the electrode switched to open circuit, thus leaving the polymer at the growth potential. This final step is important because it controls the oxidation state of the polymer and hence the resistance of the final device. After coating, the devices were removed from the growth solution, washed with solvent and allowed to dry. The base resistance of the dry devices was recorded. Full details of the fabrication and electrochemistry of the devices will be given elsewhere. Instrumentation Headspace sampling In the design of an electronic nose for monitoring beer flavour it is necessary to use a number of broadly tuned sensing elements combined with suitable multivariate analysis tech- niques.This principle of using sensor arrays for odour discrimination was originally demonstrated by Persaud and Dodd for a three-sensor system.13 Fig. 2 is a schematic diagram of the instrument developed to analyse the static headspace of beer samples. It consists of three separate elements. The chemical hardware consists of a glass vessel (2.0 dm3) to hold the analyte, immersed in a temperature-controlled water-bath set to 30 "C. It was found that this temperature was required to produce an odorous beer headspace from 100 cm3 of the analyte. A motorised fan was installed within the sample vessel to assist uniform mixing. This arrangement is shown in Fig. 3. The sensor head was designed to house four separate tiles, one on each side of the block, with each tile containing three separate polymer devices.This was fabricated using a brass block and poly- (tetrafluoroethylene) as relatively inert materials. Devices were wire bonded on one end and soldered onto metal posts in the block. A perspex disc was incorporated into the sensor head to seal the sensor vessel during testing. The following procedure was used to sample the beers: first, the sample vessel was lowered into the water-bath. Then 100 cm3 of beer were transferred into the sample vessel, the vessel was sealed and left for 20 min while the liquid and vapour phases of the analyte equilibrated. The lid was then removed from the sensor vessel and the sensor head loweredANALYST, APRJL 1993, VOL.118 373 ~~~~ ~ Table 1 Details of conducting polymers No. 1 2 3 4 5 6 7 8 9 10 11 12 Polymer system PPy-BSA PPy-PSA PPy-HxS A PPy-HpSA PPy-OS A PPy-DSA PPy-TSA(Na) PPy-TS A( m) PP y-TEATS PPy-TEATS PAN-NaHS04 P3MT-TE ATFB Monomer concentration/ mol dm-3 Py 0.1 PyO.1 PyO.l PyO.1 Py0.1 PyO.1 PyO.1 Py 0.1 Py 0.1 Py 0.1 AN 0.44 3MT 0.1 Electrolyte con- centration/mol dm-3 BSA 0.1 PSA 0.1 HxSA 0.1 HpSA 0.1 OSA 0.1 DSA 0.1 TSA(Na) 0.1 TSA(m) 0.1 TEATS 0.1 TEATS 0.1 NaHS04 0.5 TEATFB 0.1 Solvcnt Water Water Water Watcr Water Water Water EtOH Watcr PC Water CH3CN Growth potential/V 0.85 0.85 0.85 0.85 0.85 0.85 0.80 1.20 0.7.5 1.10 0.90 1.65 Final potential/\/ 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.90 1.65 Resistance/ !2 1650 193 27 16 35 37 19 70 34 37 44 13 User control Polymeric sensors File conversion I I Pre-processing I I Template matching University Response plots network Cali bration Fig.2 Basic arrangement of the electronic nose D Electrical connection to interface electronics I ’ ----- Perspex Polymeric sensors lid Sensor head Teflon/brass control led water - bath Beer sample (100 cm3) tional amplifier (U2A). This constant voltage offset ensures that the current through the precision scaling resistors (R3-R10) and hence through the sensor itself is only related to that set of precision resistors selected through the dual in-line scale switches (S3). At low concentrations, because the action of the conducting polymer sensors is virtually ohmic, we can assume that the voltage generated across the device is linear with conductance.The second stage of the circuit (U2D) is then simply to provide voltage offset nulling (via adjustment of CE2) and scaling of output voltage VA (via adjustment of CE1) during calibration. Hardware calibration takes place through the use of a shorted link in place of the sensor element (across RA and supply common) representing a nominal impedance to the circuit. The null offset potentiometer (CE2) is then adjusted to trim any zero error of output voltage VA. Standard precision resistors (accurate to 0.1% and possessing good long-term stability), suitable for the selected scale settings are then substituted as the input element, and the gain adjust potentiometer (CE1) is trimmed to scale the output voltage (VA) to be as close to the full dynamic range of the analogue to digital (A/D) input stage as possible without causing saturation. This ensures that the highest resolution of the A/D sub-system is exploited during data acquisition.Two printed-circuit boards were laid out using Orcad SDT IWRacal Redac Redboard, and constructed in such a way that each circuit could process the signals from six polymer sensors (module 2), see Fig. 5. These were assembled in a rack system along with another custom PCB (module 4) to synchronize the signals feeding the DT2811 A/D card in the PC. Fig. 3 Schematic diagram of the bcer headspace sampling system into the vessel. The resistances were monitored for 10 min after which time the sensor head was removed from the vessel. The vessel was cleaned with water and then blown with a clean air supply for about 2 min to remove any contaminants.The sensor head was subsequently replaced in the clean vessel. The sensars were left to recover for 30 min (maximum 1 h). Therefore, the total sampling time was typically about 40 min. Interface electronics The headspace sampling system was followed by the interface electronic circuitry that converts the polymer resistances into a 0-5 V analogue signal suitable for input to a DT2811 data acquisition card in a 286-based PC. The interface electronic circuitry for a single polymer is shown in Fig. 4. The conducting polymer sensor is connected across the input port RA and supply common. The operation of the circuit is based upon the principle that the first stage supplies a constant current to the sensor.This is achieved through the action of the precision voltage reference diode (D2) tied between the output (pin 1) and inverting input (pin 2) of the first opera- Data acquisition Data acquisition and processing software routines were written in TURBO PASCAL version 5.5 (8000 lines of code) using pull-down menus. The function of these software routines was to provide data collection and storage, softwar? calibration, response display, chemometric fingerprinting, pre-processing and communications with a local area network. Further facilities incorporated into the data acquisition software include file conversion utilities to allow acquired data to be stored and retrieved in a portable file format (Lotus 1-2-3). Data Processing It is convenient to consider the process of pattern recognition to have three stages.14 In the first stage the physical world can be represented as a continuum of parameters that are essentially infinite in dimensionality. The sensors describe a representation of that world in terms of R scalar variables.This then becomes the dimensionality of the pattern or sensor space. Secondly, the dimensionality of R is often high and so it is then convenient to reduce the dimensionality while still retaining the discriminatory power for classification purposes.374 ANALYST, APRIL 1993, VOL. 118 0.1 % Precision -- 8765 IT" I L SW DIP-4 I L Angle mount p+ vcc I I Sensor connection 10k c2 0.1 p: -vcc U2D UF 2 Null i* I offset R11 -vcc 100 - - Fig. 4 Intcrfacc circuit for polymer chemoresistors (for details see text) _ _ _ _ - _ _ _ _ _ Backplane Rack system I I r- I Module 4 I I Q+ - - I I System cards I DT 281 1 board A/D converter 1 L - - - - - - - - - - - - - - - - - Data acquisition PC Module 2 Module 2 - To sensor head Signal pre-processi ng Fig.5 Architecture of thc data acquisition system This then becomes the feature space of a lower dimensional- ity. Finally, the classification space is simply the decision space in which one of K classes has been selected. We can, therefore, consider the pattern recognition problem as a transformation (usually non-linear) from pattern spbce, through feature space, to classification space. By using this description, the response of our sensor array to an odour can be represented as a path followed in R dimensional pattern space, as shown in Fig.6 in 3D pattern space ( x l , x2, x3). The response parameter is usually defined as a function of the start and finish points, e:g., the change in sensor values. The starting point occurs just at the time when the rig is subjected to the odour and the end point is reached when all sensors have reached steady-state values. In this example the path followed and the speed at which it is travelled, is ignored. The broken line in Fig. 6 illustrates the use of a euclidean distance metric to map out pattern space, being linear this is the simplest metric. Fig. 6 sensor space (for details see text) Representation of an odour signal in R-dimensional pattern or In general a point in pattern space is a column vector of sensor responses x = (XI&, ... ) x,, ..., XR)T (1) In our instrument the response vector X corresponds to the time-dependent conductance of the polymeric sensor array. However, we pre-process the scalar terms in the response vector in order to reduce temperature effects. The pre- processed scalar x: is defined as the fractional change in response for each sensor1s maxos, s r,,b,> - mino,,,&,> mino<rst,,{x,> (2) x' = where t is time and t,, the time at which steady-state values are achieved. Many distance metrics have been used in the analysis of sensor data, including difference16 (simply theANALYST, APRIL 1993, VOL. 118 375 difference in response due to stimulus), relative17 (the ratio of the resistance of the sensor in air to the resistance due to the odour stimulus) and fractional models defined above.18 By splitting this distance metric into its constituent scalar com- ponents across all channels we can define (3) This normalizing procedure helps reduce concentration errors in a triangular taste test.In terms of the vectorial representa- tion, the response vectors are mapped onto the R - 1 dimensional surface of an R-dimensional hypersphere of unit radius, centred at the origin. It is now possible to define a score 2, as being a standardized normal variate with mean = 0 and variance = 1, where with N,fe\t represents the normalized response components of our unknown odour under test. This assumes that the response scalars N,fe\f come from a normal distribution and are independent.The term p, is the average and or is the adjusted sample standard deviation of the rn samples carried out for each reference beer (as part of a class-conditional database of sensor responses) for sensor Y. The distribution of the scores is by definition the x’ distribution, i.e., We can use the computed value to rank the odour patterns to their proximity to known classes and assign a confidence level from standard statistical tables. In reality, the pattern space X ’ is both non-euclidean and non-normal. However, the analysis of two beers in close proximity in pattern space permits us to make a local linear approximation of pattern space and thus employ this simple method. The main disadvantage of the method is that it treats all sensor signals as equal, thus it may be necessary to weight sensors that are known to be more sensitive or reliable by inserting an additional factor in eqn.( 5 ) . This chemometric fingerprint, or template, method has bcen incorporated into the electronic nose for sensing beer flavours. Fig. 7 summarizes the two elements of this method, first the supervised template learning and class assignment process, and secondly the template matching or predictive classification of unknown test samples. The template learning process must be carried out for all data runs to be used as part of a class conditional database. First, an averaging filter (moving or block average) is applied to improve sensor response continuity. Next a statistical feature extractor then computes the component values of x’, or N , through the use of eqns.(2) and (3) for all the sensors. These reference values are then stored in a template file and assigned to one of the known classes. The classification database is then created by defining the classes of the template files, i.e., lager 1. In the work reported here crisp classification functions are used, although fuzzy organoleptic data could also be mapped onto feature space. Finally, an odour of unknown class is sampled and its class is assigned, using the template matching process, to a known class from the class-conditional database previously generated. The pattern vectors in R-dimensional space can also be analysed using a standard technique in cluster analysis (CA). Firstly, the distances between points are calculated for a euclidean metric and then their proximity is calculated and ranked hierarchically by a similarity index.Finally, the points are linked together by a simple rule applied to their similarity indices. Single linkage uses the nearest distance while com- plete linkage uses the furthest neighbour. These are the simplest linking methods although a variety of others are available. More details on the application of CA to odour discrimination are given in ref. 19. Results and Discussion Fig. 8 shows a plot of the typical percentage changes in conductance of three conducting polymers (polymers labelled 8, 9 and 11 in Table 1) in lager 1 (a standard strength lager), lager 2 (an extra strength lager), ale 1 (a low alcahol beer) and methanol. Batch variability of two of the polymers is shown as well as typical polymer stability via the response to air.Typical responses were of the order of a 1-10% change in the conductance of the beer headspace and a 5-20% change for methanol vapour. Most of the polymers listed in Table 1 gave reproducible stable responses except PPy-TSA-H20 which was the least responsive and P3MT-TEATFB which gave large responses but also showed considerable drift. After allowing the polymers to stabilize, the drift of the sensor baseline conductance in air ranged from 0.1 to 10% per month, but was typically 2%. However, the sensor response [i.e., relative change in conductance as defined by eqn. (2)] was much more stable (<1% per month), suggesting that the change in sensor conductance with beer is a function of the baseline conductance.This is an encouraging result as the stability of the sensor response space now exceeds that of the conductance space. Typical response times (tg0) of the Class known Class Template matching process unknown Match with class-conditional database t t Class-conditional database Fig. 7 Functional block diagrams of thc template learning and matching procedures376 $ 28 i 2 4 E 20 s 2 12 16 +I - 8 - ANALYST, APRIL 1993, VOL. 118 - - - - - polymeric sensors were about 250 s, with linear temperature coefficients of resistance of 1 X 10-5 K-1. A brand test was first carried out by sampling the three dissimilar beers, namely lager 1, lager 2 and ale 1, which should in principle be easy to discriminate. Fig. 9 shows the results of CA using a euclidean metric and a single-linkage procedure.Three distinct clusters (labelled A, B and C) can be seen with no incorrect class assignments. In fact two jack-knife classification planes would give a 100% success rate on this data set. Next a 12-element array was used to analyse five sample headspaces of two similar products, i.e. , lager 1 and lager 3 (lagers of similar alcoholic strength and taste). Fig. 10 shows the dendrogram of the same CA on the X' response vectors as before. Two clusters, labelled A and B, are observed with one sample in each class being wrongly identified. The template matching method confirmed this result by also giving a success rate of only 80%. This is very encouraging as the difference in flavour of the two products is slight (with alcoholic content the same) and the centroid of a set of, for example, five samples is easily discriminated.Moreover, the selectivity of this polymer array was found to exceed that based on commercial tin oxide sensors. A measurement was then made of a control lager, i.e., lager 1, and the same lager with a single taint (stale) artificially made. Ten samples of each were taken and there was a success rate of 90% using the template matching method. Fig. 11 shows the cluster graph which again shows the discrimination of the lagers from the normalized vectors, N' (the normalized data gave slightly tighter clusters, A and B). An examination of the points plotted in the figure shows that two control beers and one tainted beer were misclassified. A close examination of the sensor data for these three samples showed that sensor 11 was giving anomalous values and so upsetting the clustering result.This problem could be obviated by a simple sensor validation scheme in which the initial data are checked for outliers, or ameliorated by averaging the results from several samples. The difference between individual sensor responses to tainted and control lagers varied with the largest value of 15% (sensor 1) and an average value of about 4%. As the Fig. 8 Typical responses of conducting polymers to beer headspaces (the letters denote repeated sensor type) -1.2 -1.6 cv E -2.0 -2.4 f -2.8 8 -3.2 A1 I .I-aLlaL1 & -3.6 4- - 2 -4.0 -4.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 Cluster component 1 Fig. 9 Cluster graph of the response of a six element polymer array to samples from three lagers: ale 1 (A), lager 1 (B) and lager 2 (C).Euclidean metric, single linkage classification technique is biased towards the larger differ- ences, a value averaged over the five most sensitive sensors may be a more relevant measure and would be about 6%. Thus a calibration period of about 6 months could be expected on this test for a sensor response stability of typically 1% per month. Clearly the drift, and hence the precise calibration period, will be odour specific. However, some advantage may be found in the use of a neural predictive classifier that has been trained upon a data set which contains the effect of long-term systematic drift of the sensor responses. ' Finally, an experiment was carried out to ascertain whether intra-batch variation of lager 1 could be detected.Five samples from five batches of cans were analysed but no can batch could be identified at a significant level using the template matching procedure. Again this result is confirmed by examining the cluster graph, see Fig. 12, where all can batches are intermingled. No significant improvement was observed with either the use of non-euclidean metrics or other linkage methods in the CAs. This validates the use of a linear, G.8 I I 1 0.7 - 0.6 - 8 0.5 - 0.4 - 1 C 6 0.3 - I - 0.2 0.1 - - n Fig. 10 Dendrogram of the res onse of a 12 element polymer array to two similar lagers: lager 1 &luster B) and lager 2 (cluster A). Euclidean metric, complete linkage 0.4 I -0.31 I I I I I I I I I I -0.5 -0.3 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 -0.4 -0.2 Cluster component 1 Fig.11 Cluster graph of the res onse of a 12 element polymer array to 10 samples of tainted lager 1 (!luster A) and control lager 1 (class B). Euclidean metric, single linkage hl 32 36 1 4L 0 -2.25 Fig. 12 Cluster graph of five samples of five batches of cans ( ~ 1 x 5 ) of lager 1. Euclidean metric, single linkageANALYST, APRIL 1993, VOL. 118 377 multi-normal template matching routine as a simple predictive classifier of beer odours. In conclusion, an instrument based on polymeric chemo- resistors and associated pattern recognition techniques has been developed which is capable of discriminating the flavours of various commercial lagers, or identifying certain off- flavours in a standard lager.The principle application envisaged is the quality control of beers in breweries. The authors express their thanks to a number of colleagues in carrying out this work which was supported by a LINK programme: Dr. H. Shurmer and Dr. G. Dodd at Warwick University; Bass PIC (Dr. S. Molzahn, Dr. E. Hincliffe, R. Cope) and Neotronics Ltd. (Dr. J. Iredale, A. Sheard, D. Williams) and the Department of Trade and Industry and Ministry of Agriculture, Fisheries and Food for their financial support. This work is the subject of an International Patent applica- tion No. GB92/01401. References 1 2 Beets. M. G. J., Structure-Activity Relationships in Human Chemoreception, Applied Science Publishers, London, 1978. Methods in Olfactory Research, eds. Moulton, D. G., Turk, A., and Johnston, J. W.. Academic Press, London, 1987. 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Meilgaard, M. C., MBAA Tech. Q., 1991, 28, 132. Landas, E., Brauwelt International, 1991, 217. Holley, A., Duchamp, A., Revial, M. F., Juge, A., and MacLeod, P., Ann. N.Y. Acad. Sci., 1974, 237, 102. Lancet, D., Nature (London), 1991,353,799. Gustafsson, G., and Lundstrom, I., Synth. Met., 1987,21,203. Miasik, J. J., Hooper, A., and Tofield, B. C., J. Chem. SOC., Faraday Trans. I , 1986, 82, 1117. Hanawa, T., Kuwabata, S., and Yoneyama, H., J. Chem. SOC., Faraday Trans. I , 1988, 84, 1587. Slater, J. M., Webb, E. J., Freeman, N. J., May, I. P., and Weir, D. J., Analyst, 1992, 117, 1265. Bartlett, P. N., and Ling-Chung, S., Sens. Actuators, 1989, 19, 141. Topart, P., and Josowicz, M., J. Phys. Chem., 1992, 96, 7824. Persaud, K., and Dodd, G. H., Nature (London), 1982, 299, 352. Andrews, H. C., Introduction to Mathematical Techniques in Pattern Recognition, Wiley , New York, 1972. Gardner, J. W., Sens. Actuators B, 1991, 4, 109. Heiland, G., Sens. Actuators, 1982, 2. 343. Horner, G., and Hierhold, C., Sens. Actuators B , 1990,2, 173. Morrison, S., Sens. Actuators, 1982, 2,329. Gardner, J. W., and Bartlett, P. N., Sensors and Sensory Systems for an Electronic Nose, Kluwer Academic Publishers, 1991, Series E, vol. 212, ch. 11. Paper 21064.51 K Received December 3, 1992 Accepted January 18, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800371

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, APRIL 1993, VOL. 118 370 Multi-layer Conducting Polymer Gas Sensor Arrays for Olfactory Sensing* Jonathan M. Slater, J. Paynter and E. J. Watt Analytical Science Group, Birkbeck College, University of London, 20 Gordon Street, London, UK WCI H OAJ Conductivity sensors using the conducting polymer poly(pyrro1e) and its derivatives have been prepared with a novel electrode design which allows the probing of resistance changes within zones of a single sensor. It was found that the application of principal component analysis to the sensor responses allowed methanol, ethanol and propanol to be distinguished. The use of layered conducting polymers improved the discrimination of the sensor. A sensor consisting of four electrode pairs and two polymer layers is capable of separating the response of certain alcoholic beverages.Keywords: Array sensor; olfaction; pol y(p yrrole) gas sensor; pattern recognition; principal component analysis Array Sensing and Olfaction I t is difficult to realize an odour or gas sensor with a high sensitivity and selectivity. I t ic attractive, therefore, to make a sensor array and analyse thc output pattern to recognize the various types of gases. This has arisen partly because most solid-statc gas scnsors are non-specific and partly from the realization that an artificial analogue of the olfactory system would have many applications. Pattern recognition for odour sensing has more constraints than for gas sensing' but many of the principles are related and similar difficulties apply. Thc approach of using arrays of sensors with partially ovcrlapping sensitivities and associated pattern recognition (PARC) mcthods has been practiced with a range of sensor technologies including bulk wavc and surface acoustic wavc quartz crystal rcsonators,zJ metal oxide gas sensors,3 electro- chemical cells and, more recently, conducting polymer chemi- resistors.All of thcse approaches have specific problems: thcse include high power consumption for the heating elements of metal oxide sensors, poor stability and sensitivity of piezoelectric dcvices, and slow response times for clectro- chemical cells. A gencral difficulty is drift within the array sensor. The aim of this paper is to considcr methods of extracting information suitable for PARC methods from a single or composite sensor layer in order to minimize drift problems and maximize information from a small device.The human olfactory system employs about SO million olfactory receptors in a massively parallel neural architecture and it appears that four principal structural elements are deduced that define an odour or gas: the size and shape of the molecule, and thc nature and position of its functional groups. A successful artificial nose will nccd to be capable o f acquiring this information. Although many sensors are capable of giving information about functional groups the stereochemical parameters are more difficult to clucidate and are often more important . Conducting Polymer Sensors Conducting polymers such as poly(pyrrole) possess many of the characteristics required for an array sensor. They display rapid, reversible changes in conductivity at room temperature combined with a broad overlapping specificity that is readily modified by chemical tailoring.5--7 Polymers can be prcpared from modified monomer units and with different countcr ions to provide materials with different, broadband responses to * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Mceting, Dublin, Ireland, September 16-18, 1992.volatile compounds. Importantly, their response kinetics are highly reproducible and the limit of detection is comparable to the concentrations of thc volatiles we wish to measure (ppm).* Recently, combined conductivity and quartz crystal micro- balance studies have shown that poly(pyrro1e) appears to have a mixed response mechanism to gascs.6 Early reports in the literature related the observed changes in conductivity caused by electroactive gases to the p-type semiconducting nature of poly(pyrrole)." Exposure t o electrophilic gases, such as NO,, tends to attract clcctrons out of the polymer matrix, causing an increase in conductivity, whereas nuclcophilic gases, such as NH3, will have the opposite effect and increase the rcsistance of the material.In addition to this mcchanism, dual sensor measurements have shown that certain vapours have a solvent-type action on the polymer, causing it to swell, and these changes in dimension are accompanied by a conductivity change.8 This effect was related to polymcr thickness and it appears that thickcr layers of poly(pyrro1e) do not swell uniformly, i.e., the penetration of thc vapour is limited.This paper describes a sensor based on a polymer-coated array of conduction bands, which allows differential conduc- tivity measurements to probe thc relative resistances of zones of the polymer. The aim is to obtain broadband, overlapping conductivity responses that corrcspond to different zones of the same polymer film to allow the response of a partially swollen surface zone to be compared with that of an unswollen 7onc. Experimental Reagents and Materials All reagents were o f analytical-reagent grade. All aqueous solutions were prepared using de-ionized water obtained from a Whatman rcvcrse osmosis deionizer water purification system. Pyrrole-type monomers were obtaincd from Aldrich and were freshly distilled before use.Electrolyte salts were obtaincd from Fisons. Test alcohols wcre purchased from Aldrich, except for gin (Gordons), brandy (Remy Martin VSOP, Procupak-Belgrade) and Southern Comfort (Southern Comfort Co). Vapours were generated by a bubbler system described previously.* Preparation of Sensors The sensors were a mu1 ti-microband electrode array prepared by standard photolithographic techniques for another pur- pose. The electrode material was platinum; a chromium adhesion layer was used and the electrode substrate was Corning glass. Thc design and spacing of thc electrode structure is shown in Fig. 1. Each sensor consists of eight380 ANALYST, APRIL 1993, VOL. 118 microbands of 5 pm width, which were used as working electrodes for the electropolymerization of pyrrole, and two 100 pm wide electrodes, which were used as auxiliary electrodes. Connections were made to the electrodes with multi-filar wire bonded with silver-loaded epoxy (RS Com- ponents).The device was then encapsulated in epoxy resin leaving only the electrode surfaces exposed. The electropoly- merization was carried out using an EG & G Priilceton Applied Research 362 potentio/galvanostat under conditions described previously.6.8 A summary of the preparation condi- tions is given in Table 1. After electropolymerization of the conducting polymer sensor layer the polymerization solution was exchanged for a monomer-free solution and the electro- chemical cycling regime continued until no further change was observed in the shape of the voltammogram.The sensor connections were reconfigured to enable each individual electrode to be contacted. Apparatus The inputs from the sensor were multiplexed so that it was possible to measure the conductivity between any pair of electrodes. The first five microbands were used as conductiv- ity electrodes and measurements made between bands 1 and 2 (response coded El), bands 1 and 3 (coded E2), bands 1 and 4 (coded E3) and bands 1 and 5 (coded E4). The conductivity (resistance) measurements were made using a laboratory-built computer controlled data acquisition system based on a Keithley 617 electrometer and 500A series / / Auxiliary electrodes Micro bonds \Bond pads /' Fig. 1 Plan view of platinum electrodes on glass substrate.The microbands are 5 pm wide and separated by 5 pm; the over-all size of the device is 10 X 28 mm data acquisition equipment. The 617 electrometer allowed resistance measurements to be made in both constant current and constant voltage (V/i) mode. The autoranging constant current mode was used for all the measurements reported in this paper. An IBM PC was used to collect data and control the gas blender system. The normal measurement regime was to expose the sensors to alternate pulses of clean, scrubbed air and sample vapour. The duration of air of vapour pulses was software controlled and varied according to experimental requirements. All data were stored in a Microsoft Excel spreadsheet for subsequent processing. Computational Chemistry and Statistical Analysis of Data The molecular structures of strands of potential conducting polymers were constructed using a Tectronics CAChe mol- ecular modelling system, The system was used to execute an AM1 semi-emperical molecular orbital calculation using MOPAClO for each structure.Analysis of the data sets for vapour exposure experiments was carried out using a variety of statistical methods including the programs ARTHUR11 and Statistica Mac12 running on an Apple Macintosh Fx computer. Conductivity data were recorded as a matrix with time-resolved conductivity measure- ments (1 s-') for each pair of electrodes (four pairs) recorded as variables and each sample as a case. Each case, therefore, resulted in four raw data values for each second of experimen- tal run time. Results and Discussion Selection of Sensor Coatings A large number of conducting polymers based on pyrrole and substituted pyrrole monomers have been reported in the literature.13 Some of these have been utilized for gas sensing applications in several modes including mass balance sensors, chemiresistors and work function sensors.An attractive approach to selecting the most appropriate polymers for chemiresistor array sensors is to develop a model of how the sensing material responds and then to exploit the most important criteria. Unfortunately, the information available for materials such as poly(pyrro1e) is incomplete and suggests that the conduction response is fairly complex. Our studies suggest that the observed gas induced conductivity changes may result from different processes in different polymers and with different test gases.8.14 Important possible mechanisms include: (i) the direct generation or removal of charge carriers from the polymer chain; (ii) modification of the charge carrier mobility; (iii) interactions with the counter ions and associated solvent changing the polymer state; (iv) association with solvent in the polymer modifying interchain electron hopping; (v) changes in the rate of interfacial charge transfer between the polymer film and the metal contact; and (vi) physically changing the state of the polymer by processes such as Table 1 Preparation of array sensor coatings Coating Monomer* Solvent Electrolyte* Polymerization regime 1 Pyrrole Aqueous LiC104 8 cycles, sweeping between 0.0 and 0.9 V, 2 N-Methylpyrrole Aqueous LiC104 12 cycles, sweeping between 0.0 and 0.9 V, 3 2-Methylpyrrole Acetonitrile Et,NBF, 14 cycles, sweeping between 0.0 and 1.1 V, 4 2,3-Dimethylpyrrole Acetonitrile Et4NBF4 16 cycles, sweeping between 0.0 and 1.1 V, 5 2-Methylpyrrole Acetonitrile Et4NBF4 14 cycles, sweeping between 0.0 and 1.1 V, scan rate = 20 mV s-1 scan rate = 20 mV s-l scan rate = 10 mV s-1 scan rate = 10 mV s- scan rate = 10 mV s-1 scan rate = 10 mV s-1 and 2,3-dimethylpyrrole Acetonitrile Et4NBF4 4 cycles, sweeping between 0.0 and 1.1 V, * Solutions contain electrolyte (0.1 mol dm-3) and monomer (0.05 mol dm-3).ANALYST, APRIL 1993, VOL.118 381 swelling. As it is possible to synthesize a large range of substituted pyrrole monomers, a potentially large family of polymers and co-polymers exist.In some instances the addition of pendant groups, possibly with specific functionali- ties, will limit the access of potential analyte vapours to the polymer backbone and it is reasonable to expect that even simple steric effects will impart some degree of selectivity to the chemiresistor gas response. In other instances bulky substituents on the pyrrole ring will prevent polymerization of the monomer. A simple screening of a range of potential substituted poly(pyrro1e)s was carried out by constructing molecular models of short strands of the polymer chain. The structures were refined using semi-empirical and quantum mechanical methods to optimize the geometry. Although the predictive powers of the energy minimization models used were limited, the results indicated the general shape of potential polymers (Fig.2) and also showed which monomers would not polymerize ( e . g . , 3-methyl-N-methylpyrrole will Fig. 2 CAChe generated surface and cross-sectional electron density maps of ( a ) poly(pyrrole), (b) poly(N-methylpyrrole), (c) poly(N- ethylpyrrole) and (d) poly(3-methylpyrrole)382 ANALYST, APRIL 1993. VOI,. 118 not readily polymerize and this is confirmed by the energy minimization routines). These software-generated represen- tations are not real surfaces; nevertheless, they allow the user to visualize the different surfaces offered by different poly- mers and consider possible interactions with target gases. For instance, the progression from poly(pyrro1e) + poly(N- methylpyrrole) + poly(N-ethylpyrrole) shows an increasingly twisted polymer backbone where the substituents may steri- cally hinder vapour interactions with the polymer backbone.On the basis of these studies, polymers were prepared from monomers including pyrrole, N-methyl-, N-ethyl-, 3-methyl- and 2,3-dimethylpyrrole with a range of counter ions. The characteristics of the most promising polymers are recorded in Table 1. Conductivity Measurements With the Multi-electrode Sensor The aim of using a sensor with multiple electrodes (Fig. 3) was to allow conductivity measurements to be made on different zones of the polymer. The principle is based on the guard ring technique, 15 which is used to distinguish between surface and bulk conductivity. In this instance by making measurements between electrodes with different spacings, e .g . , band 1 and band 2 or band 1 and band 4, it is possible to relate to the conduction of different zones of the polymer. Hence the conduction path between band 1 and band 2 involves bulk polymer close to the electrode substrate while measurements between band 1 and band 6 reflect a conduction zone that includes more surface layers. An exact interpretation of the conduction zones is complicated because (a) the current path is distorted by intermediate electrodes (which act as short circuits by virtue of their high conductivity) and ( b ) the polymer structure is probably different over the gap and electrode regions. The initial resistances of the devices varied with preparation conditions (Table 1) but were usually consistent (&5%) for a batch of nominally identical devices (coating 2, rz = 5 ) .The appearance and thickness of coatings also varied; an examination of a sample of coating 1 by scanning electron microscopy indicated a film thickness of 85 pm. We have previously reported that poly(pyrro1e) swells when exposed to certain vapours such as methanol and, impor- tantly, that the swelling of thicker polymer layers does not appeaf to be complete, i.e., the vapour does not fully penetrate the polymer.6 For these reasons different relative responses would be expected from dii-ferent electrode pairs. In addition, as the diffusion process occurs on a relatively slow timescale (several tens of seconds), the relative time-resolved response should yield additional information.The conductiv- ity response of four pairs of electrodes, El-E4 (where El is the closest electrode pair and E4 is the most distant electrode pair), for the first sensor (coating layer 1, Table 1) to 1% methanol is shown in Fig. 4. In this experiment the sensor is exposed to alternating pulses of air and methanol for periods of 120 s each. Data are displayed as a time averaged signal (3 s averaging time) for the first 63 s of exposure, 21 data points. Although the data have not been processed, the difference in response between electrode pairs is clear. Conducting polymer layer / ‘1 ‘2 ‘3 Electrode bonds Glass/’ substrate Fig. 3 Cross-section of polymer-coated electrode array Signal Pre-processing A large number of pre-processing techniques such as autoscal- ing and feature weighting can be used to transform data with the aim of promoting and displaying underlying patterns within data sets.Often it is also important to ameliorate thc concentration dependence of the response signal if the aim is to classify a vapour. Several of the more popular feature extraction methods reported in the literature were considered in conjunction with normalization procedures before applying pattern recognition techniques. However, for linear pattern recognition techniques such as principal component analysis (PCA), some methods behave the same. The feature extrac- tion methods evaluated included: (1) difference mode1~,~~.17 R,,, - R,,,; (2) relative models,Ix R,,,/R,,,, GgJG:,,r; and (3) percentage change,“’ (RgCl, - f?‘i,r)/R~l,r x 100% (relative difference), where Rill,, Rga,, G,,, and Gg,,, are the sensor resistances and conductances in air and gas, respectively. It was found that in pattern recognition studies using data input from four electrode pairs the best separations were generally obtained using the relative difference model for each individual electrode pair.The application of this model to the data set shown in Fig. 4 results in Fig. 5. I t has the clear effect of accentuating underlying features in the data. 600 4- .- c 3 2 500 F 5 400 2 300 lu - .- > .- 4- s 200 (3 100 -0 0 E Fig. 4 Conductivity responsc in arbitrary units offour electrode pairs (EI-E4) to 1% methanol vapour over 21 time periods. Each column represents a 3 s mcasuring period Fig. 5 Fractional rcsponse of electrodes El-E4, re-codcd as VS-VX, to 1% methanol vapour over 21 consecutive timc pcriods of 3 s durationANALYST, APRIL 1993, VOL. 118 -1.5 r*1 383 Ah A A A AA In odour sensing, a small signal caused by a dilute or weakly interacting component may be as important as a large signal from a major component.Scaling is a pre-processing tech- nique that gives each variable equal weighting by adjusting the original data to remove inadvertent weighting on extremely small or large values.20 In our data set this will be necessary to remove the systematic weighting given to the data sets by different electrode pairs having increased electrode spacing and hence raw resistance. The data were autoscaled so that each variable is mean centred with a standard deviation of unity.2’ Principal Component Analysis of Alcoholic Vapours Principal component analysis, sometimes referred to as vectoral decomposition, is a powerful data reduction tech- nique that reduces the high dimensionality of a multivariate problem in which the variables are partly correlated, and allows the information to be displayed in two or so dimen- sions.It is clear that the output from our array is highly correlated so there should be some advantage in applying PCA to the data. A typical sensor with a type 2 coating (Table 1) was tested for its discrimination towards samples of alcohols. The test set comprised 15 exposures to samples of three alcohols (methanol, ethanol and propanol). Fig. 6 shows the results of PCA on a data set comprising the response obtained 30 s after exposure to the alcohol vapour for each of the four electrode pairs.The model used was the fractional difference model but the data were not autoscaled. The first principal component successfully separates methanol from the other alcohols but neither component resolves ethanol and propanol. This result is not surprising as no averaging of data has been performed (so the noise level is high) and the amount of information for each sample is limited. The data set can be substantially expanded by including time-resolved responses and the effect of noise reduced by averaging. The time response profiles (Fig. 5 ) indicate that resistance between different electrode pairs at different time intervals may yield additional informa- tion. In order to test this observation, PCA was performed on a data set comprising the resistance measurement of each electrode pair (four pairs) averaged for 2 s and input every 5 s for a period of 20 s [Fig.7(a)]. Two components were retained and the eigenvectors were varimax rotated. The data sampling method had generated 20 data points for each vapour sample and it is clear that the resulting two principal components now separate the data into three distinct groups that correspond to each of the alcohols. This is a particularly encouraging result 3.0 2.5 2.0 1.5 1 .o L 0.5 CU + g o LL -0.5 -1.0 - 1.5 -2.0 -2.5 4 A 2.0-1.5-1.0-0.5 0 0.5 1.0 1.5 2.0 Factor 1 Fig. 6 PCA of data for the response of four electrode pairs to 0, methanol; +. ethanol; and A, propanol at a response time of 30 s.Data points for methanol form the group on the left for data generated from a single sensor layer. The test set are chemically similar compounds and essentially differ only in size. It is known that the polymer swells when exposed to methanol and that this appears to be a solvent-type effect with the swelling process taking place over a period of several seconds;* in this instance it is possible to envisage that similar molecules of slightly different sizes (a homologous series of alcohols) cause the polymer to swell at different rates and may also penetrate the polymer to different extents. If either of these effects were to occur, one would expect to observe a difference in the relative responses of different pairs of electrodes and at different periods of time during the exposure cycle.It is likely that such effects can be maximized by varying the composition of the sensing polymer layer, so as to promote or restrict the permeability of the polymer layer. Each of the polymer formulations described in Table 1 was investigated with the three vapour test set as single sensor layers with the aim of maximizing the discrimination between the samples. The best separation [Fig. 7(6)] was obtained with a base layer of poly(2-methylpyrrole) coated with a thin layer of poly(2,3- dimethylpyrrole), i.e., coating 5. It was found to be necessary to control the thickness (number of cycles) and oxidation state (and hence the conductivity) of the polymer fairly carefully to ensure that pairs of electrodes gave usefully different responses.It is not yet possible to prepare sensor arrays with sufficiently closely matched characteristics to negate indivi- dual application of PCA ‘training’ for an array. However, once characterized, an array of devices (n = 3, coating 5 ) is able to classify alcohol vapours for a period of 6 d before there is some intersection of the original vapour feature space. The procedure used for these trials was to expose the sensors to 2 min pulses of a random test vapour every 30 min. 2.0 1.5 1 .o 0.5 0 -0.5 4- m L L 2.0 1.5 1 .o 0.5 0 -0.5 -1.0 -1.5 -2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 Factor 1 Fig. 7 Results of analysis of time-rcsolvcd data for thrce alcohol vapours: ( a ) for a single sensor layer and (b) for a two-layered sensor. 0, Methanol; +, ethanol; and A, propanol384 ANALYST, APRIL 1993, VOL.118 2.0 1.5 1 .o 0.5 cv L , o o (D L -0.5 -1.0 -1.5 -2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 Factor 1 Results of PCA on a data set of four alcoholic beverages: a, Fig. 8 brandy 1; D, brandy 2; A, gin; and +, Southern Comfort Application to Alcoholic Beverages Principal component analysis was also applied to the response of the sensor array to two brandies, a gin and a spirit. In each instance the sensor was exposed to seven samples of beverage vapour with a 2 min expose/purge cycle. The fractional difference model was used in conjunction with the normaliza- tion procedure to process the data prior to PCA. The best separated response was obtained when the signal for ethanol vapour (prepared by a bubbler to match the concentration in the beverage) was used as the background conductivity value in the fractional response model.In this instance the data are separated into four classes (Fig. 8). Conclusions It is possible to use a sandwich of conducting polymers as part of a single sensor to separate and classify various vapour types. In order to maximize the discrimination of the sensor it is necessary to use a time-resolved response. The problems associated with drift of individual array components appear to be minimized with this type of device as measurements are essentially always based on the difference in behaviour of a single polymer or polymer sandwich. By refining the parti- cular properties of the polymer layers it should be possible to develop sensors with enhanced discrimination of particular target analytes.The authors thank Dr. D. Totterdell and E. M. Wittam, AEA Technology, Hanvell, for their help in producing a sensor base and Dr. N. Freeman, GEC-Marconi, for many useful discus- sions. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 References Persaud, K., and Dodd, G. H., Nature (London), 1982, 299, 352. Zaromb, S . , and Stetter, J. R., Sens. Actuators, 1984, 6 , 225. Gardner, J. W., Sens. Actuators B., 1991, 4, 109. Shurmer, H. V., Gardner, J. W., and Chan, H. T., Sens. Actuators, 1989, 18, 361. Pelosi, P., and Persaud, K., in Sensors and Sensory Systems for Advanced Robots, ed. Dario, P., NATO ASI Series, Springer, Berlin, 1988, vol. F42, pp. 361-381. Slater, J. M., and Watt, E. J., Analyst, 1991, 116, 1125. Slater, J. M., and Watt, E. J., Anal. Proc., 1992, 29, 53. Slater, J. M., Watt, E. J., Freeman, N. J., May, I. P., and Weir, D. J., Analyst, 1992, 117, 1265. Nylander, C., Armegreth, M., and Lundstrom, I., in Proceed- ings of the International Meeting on Chemical Sensors, Fukuoka, 1983, eds. Seiyama, T., Fukei, K., Shiowkawa, J., and Suzuki, S., Elsevier, Amsterdam, 1983, pp. 203-207. MOPAC, Version 5 .O, Quantum Chemical Exchange Program, University of Indiana, Bloomington, IN. Carey, W. P., Beebe, K. R., Kowalski, B. R., Illman, D. L., and Hirshfeld, T., Anal. Chern., 1986, 58, 149. Statistica Mac, Version 2.1, Statsoft, Tulsa, OK, 1991. Feast, W. J., in Chemical Sensors, ed. Edmonds, T. E., Blackie, Glasgow, 1988, pp. 295-317. Topart, P., and Josowicz, M., J. Phys. Chem., 1992, 96, 7824. Reucroft, P. J., Rudyj, 0. N., and Labes, M. M., J. Am. Chem. SOC., 1963,85, 2059. Heiland, G., Sens. Actuators, 1982, 2, 343. Mokwa, W., Kohl, D., and Heiland, G., Sens. Actuators, 1985, 8, 101. Horner, G., and Hicrold, C., Sens. Actuators B , 1990, 2, 173. Cranny, A. W., and Atkinson, J. K., Sens. Actuators B, 1991,4, 169. Dobson, A. J., An Introduction to Statistical Modelling, Chapman and Hall, London, 1983. Gardner, J. W., and Bartlett, P. N., in Techniques and Mechanisms in Gas Sensing, eds. Mosely, P. T., Norris, J., and Williams, D. E., Adam Hilger, Bristol, 1991, p. 355. Paper 2105840E Received November 2, 1992 Accepted December 11, 1992

ISSN:0003-2654    

DOI:10.1039/AN9931800379

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST. APRIL 1993. VOL. 118 385 Fibre Optic Oxygen Sensor Based on Fluorescence Quenching of Evanescent-wave Excited Ruthenium Complexes in Sol-Gel Derived Porous Coatings* Brian D. MacCraith, Colette M. McDonagh and Gerard O'Keeffe School of Physical Sciences, Dublin City University, Glasnevin, Dublin 9, Ireland Emmetine T. Keyes and Johannes G. Vos School of Chemical Sciences, Dublin City University, Glasnevin, Dublin 9, Ireland Brendan O'Kelly and John F. McGilp Department of Pure and Applied Physics, Dublin University, Trinity College, Dublin 2, Ireland A simple, low-cost technique for the fabrication of optical sensors for oxygen is described and preliminary results obtained using these sensors are reported. The technique is based on coating a declad portion of an optical fibre with a microporous glass film prepared by the sol-gel process. A ruthenium complex [Rut'-tris- (2,2'-bipyridine) or Ru~~-tris(4,7-diphenyl-l,1O-phenanthroline)] is trapped in the nanometre-scale cage-like structure of the porous film.In this sensor configuration the complex is excited by the evanescent field of the 488 nm radiation guided by the optical fibre. The luminescence from such complexes is known t o be quenched by oxygen and the sensors exhibit repeatable quenching behaviour when exposed t o various concentrations of oxygen. The ratio R = lo/lloo where lo and llo0 represent the detected signals from a sensor exposed t o 100% nitrogen and 100% oxygen, respectively, is used as a measure of the sensitivity of the sensor. Sensors based on the diphenylphenanthroline complex exhibit greater sensitivity than those based on the bipyridine complex, in accordance with theoretical predictions.More importantly, however, the design potential of the sol-gel process for sensor fabrication is demonstrated by the achievement of a substantial increase in R when the process parameters are adjusted t o increase the pore volume. Keywords: Optical fibre sensor; oxygen; ruthenium complex; sol-gel; evanescent wave The determination of oxygen concentration is important in many situations in industry, medicine and the environment. Optical sensors for oxygen offer advantages over conventional polarographic and titration methods in that they are fast, do not consume oxygen and are not easily poisoned. The most common method adopted in optical sensing is based on the quenching of luminescence from a range of chemical species.In homogeneous media and when the quenching process is purely dynamic (collisional) the variation in luminescence intensity, I , with oxygen partial pressure, p 0 2 , is described by the Stern-Volmer equations M I = 1 + Ksvp02 (1) K,, = kT0 (2) where 1" and zo are, respectively, the luminescence intensity and excited state lifetime in the absence of oxygen, K,, is the Stern-Volmer quenching constant and k is the bimolecular quenching constant. Most of the luminescent indicators used for oxygen sensing suffer from the disadvantage of having both shortwave excitation and small Stokes shift, thus adding complexity to the required measurement instrumentation. I Luminescent transition metal complexes, especially ruthenium poly- (pyridyl) compounds, are particularly attractive as oxygen sensing species.2 These complexes exhibit strong absorption in the blue-green region of the spectrum, for which a range of suitable laser and light emitting diode (LED) sources is available, and have relatively large Stokes shifts.Further- more, these species have long unquenched excited-state lifetimes (microseconds) resulting in high sensitivity as indi- cated by eqns. (1) and (2) above. The relatively large values of q) also facilitate the design of relatively inexpensive sensing systems based on decay-time measurement.3 In the work * Presented at the Sensors and Signals Symposium at The Royal Society of Chemistry Autumn Meeting, Dublin, Ireland, September 16-18, 1992.reported here two different ruthenium complexes, Ku"- tris(2,2'-bipyridine) = Ru( bpy)32+ and Ru"-tris(4,7- diphenyl-1 ,10-phenan throline) = Ru(Ph2phen)32+, were evaluated as oxygen sensing materials in a novel sensor design. Reagent-mediated optical chemical sensors (optodes) generally require immobilization of the appropriate reagent on or near an optical fibre or waveguide in one of a range of possible configurations. Although many methods of immobil- ization have been reported for such applications these are often complex and, in some cases, lack reproducibility.4 A novel, generic technique for fabricating low cost, reliable optical waveguide chemical sensors has recently been deve- loped by this group.536 The technique is based on coating an optical fibre or waveguide with a microporous glass film prepared by the sol-gel process.Generally the porous coating contains an analyte-sensitive dye which is trapped in the nanometre-scale cage-like structure. This approach offers a number of advantages over other methods of sensor fabrica- tion: (i) sensor preparation is simple and avoids the complex chemistry often associated with other immobilization methods; (ii) the coating produced is tough, inert, intrinsically bound to the waveguide substrate, and considerably more resistant than, for example, polymer films in aggressive environments. Although distal-tip and longitudinal coating are both feasible with this process, the latter, which relies on interaction between the analyte-sensitive dye and the evanes- cent wave of the waveguided radiation, is more desirable.The evanescent wave, which decays exponentially with distance from the waveguide-coating interface, defines a short-range (approximately 0.5 h) sensing zone within which the guided radiation may interact with absorbing species. The longitudi- nal coating approach provides the sensor designer with much greater flexibility by enabling adjustment of sensitivity-deter- mining parameters such as the length, thickness, porosity and refractive index of the coating. Furthermore, the problem of photobleaching is considerably reduced and comparable signal levels are achieved with relatively short (centimetre range) coating lengths. The evanescent wave scheme is also386 ANALYST, APRIL 1993, VOL.118 Fig. 1 coating and opaque epoxy on distal tip Typical sensor configuration showing declad region, sol-gel I n~ B Fig. 2 Gas sensor characterization system: A, Air-cooled argon-ion laser (488 nm); B , lock-in amplifier and microcomputer; C, photo- multiplier tube; D, monochromator or glass filter combination; E. holographic edge filter; F, gas cylinders, mass-flow controllers and gas mixer; G, gas cell with sensing fibre and pressure transducer; and H, optical chopper compatible with planar waveguide devices and enables (quasi-) distributed sensing when substantial lengths of optical fibre are coated at discrete locations or continuously along the length of the fibre. Although the bchaviour of organic dyes trapped in sol-gel- derived porous glass7 and the potential for sensor applica- tions,s had been discussed previously, an intrinsic optical fibre sensor using the sol-gel approach was reported for the first time only recently.' This consisted of a pH sensor based on a declad optical fibre coated with a thin film containing fluorescein. However, the technique is particularly suited to gas sensing because the high specific area of the continuous porous structure confers enhanced sensitivity. In this work preliminary results obtained from oxygen sensors based on fluorescence quenching of ruthenium complexes trapped in sol-gel derived porous coatings on optical fibres are reported.Experimental Chemicals The RuC13,3H20 was obtained from Johnson Matthey. Tetraethylorthosilicate (TEOS) and the ligands (bpy) and (Ph2phen) were purchased from Aldrich.These chemicals were used directly with no further purification. [ R ~ ( b p y ) ~ ] C l ~ and [ Ru(Ph2phen)3]C12 were synthesized and purified as described in the literature.9 Sol-Gel Process The sol-gel process is a method of material preparation by room temperature reaction of organic precursors and has been applied most often to the production of glasses and ceramics.1') Typically, the process involves a metal alkoxide, water and a solvent which are mixed thoroughly to achieve homogeneity on a molecular scale. To make silica glass by this method a silicon alkoxide in solution (the sol) undcrgoes hydrolysis and condensation polymerization reactions at room temperature to produce a loose network (the gel). Drying and heat treatment will then progressively densify the gel by elimina- tion of solvents and water.By adding a sensor dye, for example, to the sol and controlling the densification process by appropriate choice of starting pH and temperature programme, dye molecules can be trapped in the nanometre- scale cages formed by the cross-linking silicon and oxygen units. This ensures that the dye molecules cannot be leached out but smaller analyte molecules can permeate the intercon- nected cages. Declad optical fibres or other appropriate optical waveguiding substrates may be dip-(or spin-) coated by the sol at a suitable stage in the densification process. Further curing produces a tough dye-doped microporous silica coating. In this work porous silica sol-gel glass doped with a Ru" complex was used to coat an optical fibre.The ruthenium complex is added to a precursor solution of water, TEOS and alcohol, and the pH is adjusted to have a value of 1 or 7 depending on the microstructure required, as explained below. The ruthenium complex concentration was typically 10 mmol 1-1 in the precursor solution. Sensor Fabrication Although a number of sensor configurations and substrates have been investigated by this group, the results presented here were obtained with polymer-clad silica (PCS 600) optical fibre with a 600 pm core diameter (TSL, Lceds, UK). The typical sensor structure used in this work is shown in Fig. 1 together with relevant dimensions. The fibre primary coating is removed by mechanical means followed by immersion in a proprietary etchant (Lumer, Ragnolet, France) to remove the cladding.The declad section of the fibre is then dip-coated by slow withdrawal from the coating solution. Withdrawal speeds of approximately 1 mm s-1 yield coating thicknesses in the region of 300 nm as determined by ellipsometry and scanning electron microscope measurements. Care is taken to prevent direct excitation of the fluorescent coating by either cleaning the fibre tip immediately after coating or by covering the tip with an opaque epoxy resin prior to coating as shown in Fig. 1. The ease of fabrication of these sensors is of particular importance. Furthermore, the flexibility of the sol-gel process facilitates adjustment of those coating properties which determine critical sensor parameters such as sensitivity and response time.For example, the coating thickness, d, can be controlled accurately by the adjustment of the withdrawal speed, v, as expressed by the relationshipl" d vn ( 3 ) where n lies between 0.5 and 0.7 depending on the solution viscosity. In addition, the coating porosity can be modified by ad.justing the starting pH and/or thermal treatment. For example, silica gels prepared at low pH (<3) are generally less porous than those prepared under more basic conditions (pH 5-7) although the addition of appropriate precursors is often necessary to reduce optical scattering in the latter example. Instrumentation The experimental system used to characterize the multimode optical fibre oxygen sensors in terms of fluorescence intensity variation is shown in Fig.2. Light at a wavelength of 488 nm from an air-cooled argon-ion laser (Cathodeon, Oxford, UK) is reflected from a holographic edge filter (Physical Optics Corp., Torrance, CA, USA) and coupled via a microscope objective to the coated PCS 600 fibre which is mounted in a gas cell. A combination of cylinders of oxygen, nitrogen and air, mass-flow controllers and a mixing unit enables precise gas mixtures to be passed at selected speeds through the gas cell. The evanescent field of the guided radiation in the fibre excites the entrapped ruthenium complex and a fraction of the resultant fluorescence is captured by the fibre. Some of this guided fluorescence passes back through the holographic edge filter, which rejects with high efficiency scattered or reflected laser radiation in a narrow band around 488 nm, to the spectral filtering unit (monochromator or glass filter combination).The fluorescence is detected by a photomultiplier, the signalANALYST, APRIL 1993. VOL. 118 N 2 N 2 Air t 0 2 I [ 0 2 - I I I I I I I I 387 v) +- ' E 1200, Fig. 3 Response ot sensor coated with sol-gel film containing Ru(bpy),*+ complex showing detected fluorescence signal for various oxygen concentrations 0 100 200 300 400 500 600 700 Time/s Fig. 4 Ru(Ph2phcn)32+ complex (precursor solution pH = 1 ) Response of sensor coated with sol-gel film containing from which is digitized by an analogue-to-digital converter and stored in a microcomputer. Results and Discussion All measurements were made at a pressure of 101.3 kPa in the gas cell.Data obtained from sensors with three different coatings are shown in Figs. 3-5. These plots illustrate a number of features which are important in the optimization of the sensor design. The ratio R = lo/lloo, where IlO0 represents the detected fluorescence signal in 100% oxygen, can be used as an approximate measure of the sensitivity of the sensor. In Fig. 3 the performance of a sensor coated with a film containing the Ru(bpy)3*+ complex is shown. Although the sensor response varies in a repeatable manner with oxygen concentration, the sensitivity is relatively low with an R value of approximately 2.0. The R ~ ( P h ~ p h e n ) ~ Z + complex is known to have a longer excited-state lifetime than the bipyridine complex and would therefore be expected to show higher sensitivity." This is confirmed by the results shown in Fig.4; where the response of a sensor coated with a film containing the diphenylphenanthroline complex is shown. In this case R has increased to about 2.6. Such R values are still considerably less than those reported in other ruthenium-based sensor configurations where values of R greater than 20 have been reported.12 These low values may reflect the fact that only a proportion of the ruthenium complex molecules is contained within the pores of the coating and therefore accessible to oxygen, the remaining molecules being enclosed in the densified regions of the microporous structure which are inaccessible to oxygen. With this in mind the precursor solution pH was increased from 1 to 7 in order to increase the 4000 ' E 3600 2 3200 2 .+i 2800 e .- 2400 2000 .p 1600 v) 4- 3 A - I c a, a, 2 1200 2 $ 800 2 400 - u.0 t I I I I I I I 100 200 300 400 500 600 700 Time/s Fig. 5 phen)32+- complex (precursor solution pH = 7) Response of sensor coated with film containing Ru(Ph2- 4.0 I 1 3.5 3.0 - 5 2.5 2.0 1.5 1 .o 0 20 40 60 80 100 Concentration of oxygen (%) Fig. 6 Typical Stern-Volmcr plot obtained with sol-gel oxygen sensors over-all pore volume. Using the (Ph2phen)32+ complex again, this approach resulted in a significant increase in sensitivity as shown by the data in Fig. 5 where R is approximately 3.7. All three plots show high values of signal-to-noise ratio with little evidence of photobleaching. The plots may not be used to deduce the sensor response time as the transition regions are mostly indicative of the time taken to achieve manually selected stable gas concentrations.Non-optimized measure- ments indicate response times for step changes in oxygen concentrations of at most 5 s. A typical Stern-Volmer plot obtained with these sensors is shown in Fig. 6. The plot shows the non-linearity previously seen with quenched dyes pre- pared in solid films.11 Such non-linear behaviour may reflect the combined response of those ruthenium complex molecules accessible to oxygen and those, in the densified region of the support structure, which are not. Fig. 6 also indicates that the highest sensitivity is obtained in the range 0-20% oxygen. Although further characterization and development work remains to be carried out, this work has established the great potential of the sol-gel approach for the fabrication of oxygen sensors.The approach enables sensor fabrication in a range of configurations, from disposable compact sensors to distri- buted sensing fibres, which can be applied to the monitoring of both gaseous and, possibly, dissolved oxygen. Potential problems include the usually hydrophilic nature of the porous support material, which may limit the sensor to gas phase measurements, and its high specific area leading potentially to a susceptibility to interferents. The design flexibility accorded by the sol-gel process, however, may allow these problems to be minimized. The facility to adjust the sensitivity of the sensor, as illustrated by comparison of Figs. 4 and 5, is an important feature of the sol-gel approach and will be used in388 the optimization of sensor design.Further work is underway to elucidate the impact of changing particular process parameters on film porosity as deduced from ellipsometric and other measurements. This work was supported by EOLAS, the Irish Science and Technology Agency, under their 1991 Strategic Research Programme. References Sharma, A., and Wolfbeis, 0. S . , Appl. Spectrosc., 1988, 42, 1009. Demas, J . N., and DeGraff, B. A., Anal. Chem., 1991, 63, 829A. Lippitsch, M. E., Pusterhofer, J . , Leiner, M. J . P., and Wolfbeis, 0. S . , Anal. Chim. Acta, 1988, 205, 1. Wolfbeis, 0. S . , in Molecular Luminescence Spectroscopy. Part I I , ed. Schulman, S. J . , Wiley, New York, 1988, pp. 129-281. 5 6 7 8 9 10 11 12 ANALYST, APRIL 1993, VOL. 118 MacCraith, B. D., Ruddy, V., Potter, C., O’Kelly, B . , and McGilp, J. F., Electron. Lett., 1991, 27, 1247. McGilp, J. F., MacCraith, B. D., O’Kelly, B., and Ruddy. V., PCT Pat. Appl., 1992, PCT/GB92/00428. Avnir. D., Kaufman, V. R., and Reisfeld, R., J. Non-Cryst. Solids, 1985, 74, 394. Badini, G. E., Grattan, K. T. V., Palmer, A. W., and Tseung, A. C. C., in Optical Fiber Sensors 1989, eds. Arditty, H. J . , Dakin, J . P., and Kersten, R. T., Proceedings in Physics Series, vol. 44, Springer-Verlag, Paris, 1989, pp. 43-40. Watts, R. J . , and Crosby, G. A., J. Am. Chem. Soc., 1971,93, 3184. Brinker, C. J., and Scherer. G. W., Sol-gel Science, Academic Press, New York, 1990. Demas, J. N., and Bacon, J . R., Anal. Chem.. 1987,59, 2780. Carraway, E. R., Demas, J. N., and DeGraff, B. A., Langmuir, 1991, 7, 2991. Paper 2/06485 E Received December 7, 1992 Accepted January 28, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800385

出版商:RSC    

年代:1993

数据来源: RSC