Chemical sensors and instrumentation intelligent

0 downloads 0 Views 693KB Size Report
formations, and analytical applications of Fourier transform in- frared spectroscopy. ... An intelligent analytical instrument is one that can. 'understand' its ... IEEEiGPIB (parallel) data transfer link. Facilities .... A free flowing junc- tion was used ...
trendsin analyticalchemistry,vol. 8, no. 2,1989

58

Exploratory Research under Cooperative Agreement CR-813621010) is gratefully acknowledged. Although the research described in this article has been funded wholly or in part by the U. S. Environmental Protection Agency, it has not been subjected to Agency review, and therefore does not necessarily reflect the views of the Agency, and no official endorsement should be inferred. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

10 M. Mossoba, U. S. Food and Drug Administration, Washington, DC, private communication, 1988; also presented at the 3rd Annual Cryolect User’s Meeting, Research Triangle Park, North Carolina, May 11, 1988.

11 J. C. Demirgian, Trends Anal. Chem., 6 (no. 3) (1987) 58. 12 D. F. Gurka, L. D. Betowski, T. A. Hinners, E. M. H&hmar, R. Titus and J. M. Henshaw, Anal. Chem., 60 (1988) 454A (and references cited therein). 13 S. L. Smith, Appl. Spectrosc., 40 (1986) 278 (and references cited therein). 14 B. M. Gordon, M. S. Uhrig, M. F. Borgerding, H. L. Chung, W. M. Coleman, III, J. F. Elder, Jr., J. A. Giles, D. S. Moore, C. E. Rix and E. L. White, .I. Chromatogr. Sci., 26 (1988) 174. 15 P. R. Griffiths, Mikrochim. Acta (Wien), 1988, in press.

References 1 C. J. Wurrey and D. F. Gurka, in J. R. Durig (Editor), Vibrational Spectra and Structure, Vol. 19, Elsevier, Amsterdam, 1989, in press (and references cited therein). 2 D. F. Gurka and S. M. Pyle, Environ. Sci. Technol., 22 (1988) 963. 3 T. T. Holloway, B. J. Fairless, C. E. Freidline,

4 5 6 7 8 9

H. E. Kimbal, R. D. Kleopfer, C. J. Wurrey, L.A. Jonooby and H. G. Palmer, Appl. Spectrosc., 42 (1988) 359. A. M. Haefner, K. L. Norton, P. R. Griffiths, S. Bourne and R. Curbelo, Anal. Chem., 60 (1988) 2441. R. S. Brown and C. L. Wilkins, Anal. Chem., 60 (1988) 1483. D. F. Gurka, M. Umana, E. D. Pellizzari, A. Moseley and J. A. de Haseth, Appl. Spectrosc., 39 (1985) 297. M. D. Erickson, Appl. Spectrosc. Rev., 15 (1979) 26. R. L. White, Appl. Spectrosc. Rev., 23 (1987) 165. J. Grainger, V. V. Reddy and D. G. Patterson, Jr., Appl. Spectrosc., 42 (1988) 800 (and references cited therein).

Charles J. Wurrey received his B.S. degree in chemistry and mathematics from Northern Michigan University in 1969, and his Ph.D. degree in physical chemistry from Massachusetts Institute of Technology in 1973, where he worked for Professor Richard C. Lord in the areas of infrared and Raman spectroscopies and molecular structure. He then conducted post-doctoral research in molecular spectroscopy with Professor James R. Durig at the University of South Carolina. In 1974, he accepted an academic appointment with the Department of Chemistry at the University of Missouri-Kansas City, and he is currently Professor of Chemistry there. At UMKC, his research interests have involved the use of vibrational spectroscopy in the determination of molecular conformations, and analytical applications of Fourier transform infrared spectroscopy. Professor Wurrey has consulted with the CT. S. Environmental Protection Agency since 1982, and he has been appointed an EPA Distinguished Visiting Scientist from 1986 to 1989. His address is: Department of Chemistry, University of Missouri, 5100 Rockhill Road, Kansas City, MO 64110, U.S.A.

Chemical sensors and intelligent instrumentation Dermot Diamond Dublin, Ireland Over the next decade, instrumentation incorporating pattern recognition techniques pioneered for use in other fields will revolutionise the range and quality of information obtainable from chemical sensors. These ‘intelligent instruments’ will, however, require sensors with significantly better performance than those currently available.

Instrument

intelligence

An intelligent analytical instrument is one that can ‘understand’ its environment to some degree’. The basic components are illustrated in Fig. 1. An array of sensors enables the instrument to monitor its environment. Raw data from the sensors are converted into a suitable form by an interface before being passed to the microprocessor. This typically involves signal amplification, offsetting and digitisation. 0169936/891$03.00

The microprocessor interprets the incoming data using memory resident routines, displays it in a suitable form, and takes appropriate action if required. Getting signals into a computer is, in these days, relatively straightforward. Interface boards are now available at prices ranging from around US$ 500 to many US$ 1000’s depending on the specification. Most are designed for use with the IBM (or IBM compatible) range of microcomputers, either as internal boards slotted directly into one of the internal expansion board slots, or as external boards connected to the computer via an RS-232 (serial) or IEEEiGPIB (parallel) data transfer link. Facilities available on these boards can include multi-channel data acquisition, on-board memory, variable data sampling rate, programmable gain and offset, variable-voltage output, and low-level signal conditioning. To complement these impressive hardware specifications, powerful software data interpreta0

Elsevier

Science Publishers

B.V.

59

trends in analytical chemistry, vol. 8, no. 2,1989

LO

b 7

I

-i-----l

/I

Fig. 1. Schematic layout of an intelligent instrument: (1) Sensors, (2) actuators, (3) inputloutput interface, (4) microprocessor, (5) memory, (6) expert system shell, (7) knowledge base, (8) external expert, (9) periferals (VDU, printer, mouse etc.), (IO) microcomputer.

tion/manipulation packages have been developed which can enhance signal purity, carry out complex mathematical operations, and display the data in a multitude of different forms. Pattern recognition routines are available which can, in some applications, decipher the raw experimental data, usually by best fit using carefully defined algorithms or by comparison to a ‘look-up’ table or data library. Facilities may also be available to enable the instrument to expand a stored knowledge base through interaction with an experienced expert in the field. When an unrecognised data pattern is confronted, a question and answer routine is initiated with the expert who may then add an interpretation of the situation to the instrument knowledge base. In this way, the instrument gradually gains experience as its knowledge expands. The software required to perform this task is known as an ‘expert system’2>3. Besides passively monitoring its environment, the instrument may take an active role in controlling various conditions through actuators (valves, pumps, motors, heaters). Expert systems and chemical sensors Compared to instrumentation aimed at physical or electronic measurements, expert systems have made relatively little impact on chemical instrumentation, a situation which is set to change markedly over the next decade. The reason for this lack of progress lies with the performance of chemical sensors relative to

physical sensors (for heat, light, pressure, movement etc.). In order to allow correlation between stored information and experimental data, the sensors used to build an expert system must be extremely stable during extended use, and respond in a highly reproducible manner to changes in their environment. In addition, they should be of small dimensions, robust and have a fast response. ISFETS (ionselective field effect transistors) are an example of a chemical sensor which might be suited to this kind of work as they are small, multi-channel devices based on integrated circuit technology4. However, there are major problems involved in using ISFETS for the development of an intelligent instrument. Take, for example, an ISFET sensitive to the physiologically important ions H+ , K+ , Na+ , and Ca2+. This may be fabricated by coating a quadruple-gate FET with a polo (vinyl chloride) (PVC) layer responsive to each ion , and then encapsulating the device so that only the ion-sensitive areas are exposed (Fig. 2). Unfortunately, these devices tend to have a short effective life-time and suffer from drift. This is due, in part, to gradual penetration of the sensor by the sample (particularly at the PVC/encapsulate seams) and a gradual leaching of the active ingredient (typically a low-molecular weight ionophore dissolved in a PVC plasticiser) into the sample. Design modifications have extended the lifetime and stability of ISFETS, for example the suspended mesh gate approach initiated by Blackburn and Janata (Fig. 3)) and the ionophore doping technique recently reported by Bezagh et aL7. In the latter case, a dual-gate FET was completely coated with PVC and the gate regions then doped with ionophore solutions sensitive to Na+ and Cl- (Fig. 4). Cross-contamination of the different gate regions through lateral

Fig. 2. Cross-section of a typical ISFET: (I) p-type silicon substrate, (2) n-type silicon, (3) metal contact, (4) ion-selective PVC membrane, (5) insulator, (6) epoxy insulating encapsulate, (7) sample solution, (8) external reference electrode, (S) source, (0) drain.

60

trendsin analyticalchemistry,vol. 8, no. 2, 1989

6 3 s

5 D

I

I

Fig. 4. Ionophore doped ISFET:

1-6, S and D as for Fig. 2; (7)

blank plasticised PVC membrane. The entire device is coated with Fig. 3. Suspended mesh ISFET: 1-6, S and D as for Fig. 2; (7) suspended polymer mesh. The suspended mesh acts as an anchor for the PVC membrane, securing it more firmly to the gate region of the FET.

diffusion was prevented by using a plasticiser with an extremely low diffusion coefficient. Although a useful life-time of about two months was claimed, the sensitivity of each sensor decreased by 4 mV (Na+) and 5 mV (Cl-) for a ten-fold concentration change over a period of three weeks. Drift was reported at 0.05 mV/h in each case. While this performance compares favourably with other chemical sensors currently available, significant improvements will be needed if they are to be used along with an expert system (for monovalent ions, a 0.1 mV error in voltage measurement results in a 0.4% error in the estimated ion activity). Performance priority in future sensors An intelligent instrument requires information from many sources in order to gain as complete a picture of its environment as possible. This is a completely different situation from the normal single-analyte measurements carried out using, for example, an ion-selective electrode (BE) and high-impedance voltmeter. The main priority with these traditional measurements is to be sure that the electrode signal is overwhelmingly dominated by the primary ion, so that signal variations can be related with confidence to changes in analyte activity. Hence the search for highly selective ionophores has dominated ISE research over the past twenty years. However, no electrode is completely specific, and the range and extent of interference from other ions largely determines the practical usefulness of an ISE. In contrast, the expert system/multi-sensors approach requires only that the response pattern of the sensor array be unambiguously assigned to individual ions. This ‘fingerprinting’ of ions makes deduction of a sample composition possible and also removes the burden of producing individually highly selective

a blank, plasticised layer of PVC. The region above the gate is subsequently doped with the ion-selective compound dissolved in a PVCplasticiser. The PVClepoxy seam, which is the main source of sample penetration of ISFETs, does not come in contact with the external world.

electrodes. Indeed, some response to a range of ions could in fact be a bonus, as long as each ion produces its own characteristic response pattern in the electrode array. The feasibility of using moderately selective chemical sensors to quantify more than one analyte simultaneously has been demonstrated in work published recently by Kowalski et ~1.~. An array of electrodes was used to measure Na+ and K+ activities at blood serum levels through interpretation of the array response pattern with a simplextype algorithm. A similar approach was adopted by Haugen and Hieftje’ to analyse a hydrocarbon gas mixture using an array of moderately selective piezo-electric sensors. Commercial interest in this approach is demonstrated by the recent launch of an “analogue interface expert system” for clinical monitoring applications called ANNIE by the U.K. company Intelligent Applications Ltd. A concise review of the types of chemical sensors available and their use in environmental monitoring has been published recently”. To summarise, more emphasis will have to be placed on improving the reproducibility, lifetime and stability of sensors rather than the selectivity. Improving data quality Reproducibility Reproducible behaviour among batches of sensors can only be expected if the sensors are manufactured in an identical manner. All too often, research groups are forced, through lack of resources and funding, to adopt somewhat crude methods to construct their devices. There is an urgent need for much closer industrial and academic links to be formed. One possibility is the formation of flexible pilot sensor manufacturing plants which could un-

trends in&analytical chemistry,

vol. 8, no. 2,1989

dertake precommercial small-scale manufacturing of new sensors. Careful attention must also be paid to the purity of materials used in sensor production, as the presence of even trace impurities can often have a marked effect on sensor behaviour . Lifetime

Sensors with lifetimes extending to at least months are required. While progress with ISFETs has been slow, it is encouraging to see collaboration between major groups in an effort to overcome this problem7. With less emphasis on selectivity, a wider range of gate sensor materials may be considered for ISFETs, concentrating instead on compatibility with FET/encapsulant materials. For example, ion-selective glasses, although generally less selective, may graft more firmly than the PVC/ionophore membranes currently favoured. Three-dimensional fast ion conductors (e.g. NASICON) have also been shown to produce working ISFETsr ‘. Advances in materials science may produce entirely new substances with potential as solid state sensors (e.g. new superconductors or conducting polymers). A strategy which has already been tried, with some success, is to attach the ionophore in the PVC/liquid membrane sensors covalently to the polymer backbone, thus preventing a gradual loss of the active component into the sample solution12. Stability

Drift in ISEs and ISFETs arises from three main sourcesr3: (a) Leaching of membrane components from the sensor to the sample. This is particularly problematic

with sensors based on liquid-membranes. The best way to prevent loss of active components by gradual dissolution (which will occur with liquid membranes even if very hydrophobic membrane components are used) is to bind them covalently into a giant structure e.g. a glass or polymer. Hence a solid-state sensor would seem desirable. (b) Temperature related effects. Temperature related drift may be reduced by introducing a temperature sensor into the array, and compensating the individual sensor output for fluctuations in the measured temperature. Once again, this will work best if the sensor responds in a reproducible manner to temperature changes. (c) Variations in reference electrode junction potential. The production of a stable reference electrode junction potential is extremely important, as any fluctuations will affect the output of the sensor array. Recently progress has been reported with macro reference electrodes14. A free flowing junction was used, with extremely low flow-rates (OS-2

61

pllh). Measurements taken with the electrode were very reproducible in comparison to a commercial reference electrode utilising ceramic plug and ground glass junctions. The reference electrode was used successfully when applied for measurements of K+ and Na+ in undiluted blood serum over a period of several months, without any visible signs of clogging of the liquid junction bore. Miniaturisation of this type of electrode should be possible using modern microlithographic techniques. It is interesting to note that at a flow-rate of 1 pllh, a lo-ml reservoir would have the capacity to maintain an electrolyte bridge for over one year. Another way of improving stability is to improve the design of the measuring circuitry (see, for example, the operational transducer or optran developed by Sibbald”). Automatic identification of sensor malfunction Signals emanating from damaged or faulty sensors will obviously cause enormous problems for intelligent instruments, so a facility for automatic sensor malfunction will be necessary in critical applications. Methods of doing this include: (i) Activating a polling routine to compare the signal obtained from a number of identical sensors. A

sensor which consistently fails to agree with the others can be switched out of the array. For example, the sensors could have to produce a result to an arbitrary degree of precision. If the result fails to meet this requirement, the procedure could be repeated without accepting the result from the most dissenting sensor until the precision limits are met. These actions should be accompanied by messages relayed to the external expert warning of possible sensor malfunction, and indicating which sensor is at fault. (ii) Checking sensor resistance or leakage of current using integrated circuitry and software routines.

A very low resistance with PVC membrane ISEs indicates membrane rupture, whereas an extremely high resistance suggests loss of active components or the formation of an impermeable coating on the membrane16. Perhaps the simplest method of doing this would be to switch a constant voltage between a reference electrode and each ISE in turn and measure the resulting current. This can be related to the membrane resistance which is assumed to be the dominating factor in the overall circuit resistance. It should be possible to carry out these tests during continuous monitoring without having to replace the sample with standard solutions as any large variations in sample resistance would be superimposed on each sensor resistance. Problems may arise, however, due to electrolysis or redox reactions occurring at the electrode surfaces. Another approach would be

trendsin analyticalchemistry,vol. 8, no. 2,1989

62

to measure the voltage drop which occurs across a large value resistance as each sensor/reference is switched in series with it. Problems arising from electrode polarisation may be reduced by using an AC voltage source. (iii) Exposing the sensor(s) periodically to standard solutions and checking the response. In this case, the

sample is removed or flushed from the measurement cell and replaced by well defined standard solutions. Using two different standards has the advantage that the response slope of each sensor can be checked to ensure adequate sensitivity. Again, each sensor would have to meet arbitrarily set limits of acceptable performance.

References 1 N. Ford, How Machines Think: A General Introduction to Artificial Intelligence, Wiley, New York, NY, 1987. 2 F. Hayes-Roth, A. Waterman and D. Lenat (Editors), Building Expert Systems, Addison-Wesley, Berks., 1983. 3 A. M. Harper and S.A. Liebman, J. Res. Natl. Bur. Stand. (U.S.), 90 (1985) 453. 4 G. F. Blackburn, in A. P. F. Turner, I. Karube and G. S. Wilson (Editors), Biosensors: Fundamentals and Applications, Oxford University Press, Oxford, 1987, Ch. 26, p. 481. 5 A. Sibbald, A. K. Covington and R. F. Carter, Med. Biol. Eng. Computing, 23 (1985) 329. 6 G. F. Blackburn and J. Janata, J. Electrochem. Sot., 129 (1982) 2580. 7 K. Bezagh, A. Bezagh, J. Janata, U. Oesch, A. Xu and W. Simon, Anal. Chem., 59 (1987) 329. 8 K. Beebe, D. Uerz, J. Sandifer and B. Kowalski, Anal.

Chem., 60 (1986) 66.

Summary The coming decade will undoubtedly see an enormous expansion in the area of intelligent chemical instrumentation. These instruments will have features such as automatic fault detection, calibration and temperature compensation as well as the capacity to interpret sample composition. Increasing user confidence in the results obtained from these instruments will lead to more applications which will in turn stimulate more progress. It would seem that some movement towards solid (or semi-solid) state sensors is required in order to obtain the required extended sensor lifetimes and long-term stability. Given the complex media often involved in chemical analysis, it is likely that the production of efficient chemical sensors will not be easy. However, the new flexibility in sensor specifications introduced by developments in computing and electronics may make the task a little easier.

9 G. Haugen and G. Hieftje, Anal. Chem., 60 (1988) 23A.

10 T. E. Edmonds, Trends Anal. Chem., 4 (1985) 220. 11 M. Kleitz, J. F. Millon-Brodaz and P. Fabry, Solid State Zonits, 22 (1987) 295.

12 L. Ebdon, A. T. Ellis and G. C. Corfield, Analyst (London), 104 (1979) 730. 13 B. J. Birch and T. E. Edmonds in T. E. Edmonds (Editor), Chemical Sensors, Blackie, London, 1988, Ch. 9, p. 221. 14 R. E. Dohner, D. Wegmann, W. E. Morf and W. Simon, Anal. Chem., 58 (1986) 2585.

15 A. Sibbald, Sensors and Actuators, 7 (1985) 23. 16 G. J. Moody and J. D. R. Thomas in T. E. Edmonds (Editor), Chemical Sensors, Blackie, London, 1988, Ch. 3, p. 111. Dermot Diamond obtained a BSc (1976) and later an MSc (1983) in Analytical Chemistry, from Queen’s University Belfast. He obtained his PhD from the same university in 1987 for research into the development of new liquid membrane ISEs under the supervision of Dr. Gyula Svehla. He is currently lecturing in Analytical Chemistry at the National Institute for Higher Education, Glasnevin, Dublin 9, Ireland. His research interests include the development of new chemical sensors, intelligent instrumentation, and computer applications in chemistry.

Contemporary wet-chemical flow analyzers for process control M. Gisin and C. Thommen Basle, Switzerland This article gives an outline of an alternative view on the underlying principles of FZA and indicates some directions for the future development of contemporary wet-chemical process analyzers.

Introduction Flow injection analysis (FIA) has emerged as a widely accepted analytical technique and an increasingly used laboratory tool in various branches of chemical analysis’22. Employment of the technique 0165-9936/891$03.00.

for on-line measurement in chemical process control is a growing trend334. The first few years of struggle for the acceptance of FIA were dominated by repeated attempts to differentiate it from the well established technique of continuous flow analysis (CFA)‘. Both techniques allow the automation of wet-chemical analysis in a system of flowing streams of carrier and reagent solution into which the sample solution is introduced, reacted and detected. In CFA, a sample is aspirated and its integrity conserved by segmenting the flow stream with air bubbles, whereas in FIA segmentafQElsevier

Science Publishers

B.V.