Process Analytical Chemistry - Analytical Chemistry (ACS Publications)


Process Analytical Chemistry - Analytical Chemistry (ACS Publications)pubs.acs.org/doi/abs/10.1021/ac00136a723by JB Call...

6 downloads 252 Views 13MB Size

Process Analytical James B. Callis Deborah L. Illman Bruce R. Kowalski Center for Process Analytical Chemistry Department of Chemistry, BG-10 University of Washington Seattle, Wash. 98195

T

he goal of process analytical chemistry is to supply quantitative and qualitative information about a chemical process. Such information can be used not only to monitor and control a process, but also to optimize its efficient use of energy, time, and raw materials. In addition, it is possible to simultaneously minimize plant effluent release and improve product quality and consistency. Although process analytical chemistry has long been an important endeavor in industry (i), it is now receiving increased attention because of the opportunities presented by technological and methodological advances, as well as changing needs within the chemical and allied products industries. As an example, consider the potential impact of process analytical chemistry on the traditional production of basic chemicals. The emphasis is shifting from adding capacity to making existing plants more efficient to meet the demands of increasing international competition. This competition involves not only pricing; in many cases the issues of product quality and consistency are equally important. Greater emphasis has been placed on lowering the "quality cost" of production—the loss suffered when a batch of product fails to meet specifications and has to be reprocessed, sold at reduced cost, or discarded. But more important are the possibilities posed by a decade of advances in materials science (2). With improved

theoretical understanding of the relationship of molecular and supramolecular structure to material properties and with new synthetic methodologies and analytical instrumentation, materials scientists are producing advanced engineering plastics, ceramics, and composites of widespread applicability. It is frequently stated that those countries capable of manufacturing the new materials in quantity and at a competitive price will prosper in the next decade. At the same time, we should not fail to mention two other recently emphasized areas of the chemical industry. The first is the use of living organisms to produce materials. Bioprocessing will continue to grow exponentially for at least the next two decades. The second is the production of specialty materials for the electronics industry. In some cases this might involve traditional chemicals that are of higher quality than those currently available (i.e., ultrapure solvents with low particulate counts). In other cases, new materials will be required for integrated optoelectronic semiconductor structures, ultramicrolithography, and ceramic substrates, to name a few examples. The key element underlying all of these endeavors is improved process monitoring and control. Advances in electronic hardware, especially microcomputers, and improved algorithms for feedback control have led to the rapid implementation of distributed,

O F F - L I N E intelligent control systems. Such networks already have vastly improved process optimization, materials accounting, resource utilization, and inplant management. Although process engineers will continue to be preoccupied by process control intelligence issues, they will readily admit that the current process controllers are literally starved for information, and that the continuing lack of sensors, especially chemical sensors, has become a major bottleneck. Indeed, it is now widely appreciated that the same tools that have revolutionized the electronics industry have the potential to further the scope of chemical sensing, and a flurry of activity in this

Table 1. Five eras of process analytical chemistry Era

Current example

Off-line

GC/MS

At-line On-line (intermittent) On-line (continuous)

Colorimeter Capillary GC Correlation IR

In-line

Conductivity sensor

Noninvasive

Diffuse reflectance near-IR

624 A · ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987

Future example Capillary electrophoresis with tandem MS detection Flow injection analyzer Microbore LC FT-IR using circle cell Micro UV-vis with fiber optic probe Multifrequency microwave radar 0003-2700/87/0359-624A/$01.50/0 © 1987 American Chemical Society

Chemistry Α Τ - L I Ν Ε area has begun (3, 4). However, one must realize that the practice of process analytical chemis­ try involves much more than improved chemical sensing (5). The issues of sampling (6), extraction of information from the sensor data, integration of the information into process control, and the sociological difficulties of gaining

REPORT process engineers' and plant operators' confidence in the new measurement tools must all be given equal consider­ ation. Clearly, process analytical chem­ istry should be considered a worthy subdiscipline of analytical chemistry, and it requires an interdisciplinary sys­ tems approach so that progress can be made. Our purpose is to acquaint the reader with the important technologi­ cal, methodological, and chemometric advances that are making possible a major revolution in this field. We par­ ticularly recommend that our academ­ ic colleagues make a closer study of this field, not only for the employment op­ portunities it creates for students, but also for the intellectual challenges it offers creative scientists. Five eras of process analytical chemistry As a conceptual framework, we find it useful to identify five eras of process analytical chemistry: off-line, at-line, on-line, in-line, and noninvasive (see Table I). Historically, there has been no systematic progression of the eras, and it is not uncommon to find exam­ ples from all stages simultaneously functioning in a single modern chemi­ cal processing plant. Off-line and at-line eras. The first two eras of process analytical chemis­ try are distinguished by the require­ ment of manual removal of the sample and transport to the measuring instru­ ment. In off-line process analysis, the sample is analyzed in a centralized fa­ cility with sophisticated, and perhaps even automated, instrumentation. The

advantages of this approach include the economy and efficiency of time­ sharing use of expensive instruments and the availability of an expert staff for consultation, methods develop­ ment, and maintenance. The disadvan­ tages, which include the delay between submission of sample and reporting of results, the additional administrative costs, and the competition among users for the resources, have led to a second era: at-line process analysis. In this type of analysis a dedicated instrument is installed in close proximity to the process line. The advantages include faster sample processing, closer control of the analysis by the process person­ nel, and employment of a simpler in­ strument with less cost and mainte­ nance and greater ease of use. The above eras of process analytical chem­ istry are so close to the usual practice of analysis that we will not discuss them further. On-line era. In the on-line era of process analytical chemistry the subdiscipline becomes distinct from its parent. An automated sampling system is used to extract the sample, condition it, and present it to an analytical in­ strument for measurement (6). Yet an­ other difference between process ana­ lyzers and their laboratory counter­ parts is the end point of the analysis. For the laboratory counterparts, one is generally trying to elucidate the chemi­ cal composition of a mixture (less fre­ quently, to understand molecular structure and interactions). For the process analyzers, the end point might be much closer to quality parameters, such as octane number or elongation index. We will have more to say on this topic later. It is possible to subdivide the on-line era into two categories: intermittent methods that require injection of a por­ tion of the sample stream into the in­ strument and continuous methods that permit the sample to flow continuously through the instrument. The classic example of an intermit­ tent on-line process analyzer is the gas chromatograph. Until recently, packed columns held at a fixed temperature

O N - L I N E

I N - L I N E

NONINVASIVE

ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987 · 625 A

generally were employed. This placed the burden of selectivity on the liquid phase and often led to lengthy develop­ ment times and adaptation of propri­ etary stationary phases. Alternatively, various types of column switching or backflushing arrangements were em­ ployed. The recently introduced wide-bore fused quartz columns seem nearly ideal for process chromatography. First, they yield more theoretical plates per unit of time. Second, they have such a light thermal mass that temperature programming is quite practical. Third, the columns are inert and yield almost ideal peak shapes. These advances translate into a requirement for fewer types of phases and less reliance on col­ umn-switching techniques, with conse­ quent decreases in method develop­ ment time. Two recent developments promise to advance process gas chromatography (GC) even further. The first is the ob­ servation by Phillips that the inlet to capillary columns can easily be ther­ mally modulated; this might, in favor­ able cases, permit the injection port to be eliminated and multiplex chroma­ tography to be implemented (7). The second is the introduction of very hightemperature capillary columns (8), which will extend the use of the tech­ nique to ever more polar, high molecu­ lar weight species and decrease the need for derivatization. Liquid chromatography (LC) has been used less on-line, but its advan­ tages are becoming apparent as more complex, water-soluble materials are synthesized. The advent of smaller di­ ameter packing materials has led to the use of very short columns with excel­ lent resolution and sensitivity. For bioprocessing applications, capillary zone electrophoresis (9) seems appealing, but new types of detectors and sample introduction technologies must be de­ veloped. Recently, an offshoot of this technique using micelles or tetraalkylammonium ions as a "pseudo reversed phase" has led to greatly increased ver­ satility for the technique and should provide impetus for accelerated techni­ cal development. The potential for eliminating expensive, unreliable highpressure pumps and replacing them with compact, solid-state, high-voltage power supplies is an important reason to explore capillary electrophoresis as an on-line technique. The recent introduction of a process supercritical fluid (SCF) chromatograph (10) will provide a new dimen­ sion of versatility to on-line chroma­ tography. This technique yields excel­ lent resolution at high speed on very nonvolatile polar, heat-labile mole­ cules. The SCF equivalent to GC tem­ perature programming is pressure pro­ gramming, which provides a much fast-

A T - L Ι Ν Ε er recycle time. Compared to an LC, the SCF system does not need expen­ sive solvents and is much faster, espe­ cially from the standpoint of recycling, when a solvent programmed run is em­ ployed. The final intermittent technique to consider is flow injection analysis (FIA), a technique that may be thought of as an on-line alternative to manual wet chemical analysis (11). The major advantages of FIA are high through­ put, reliability, and precision. Howev­ er, an even greater range for FIA will be forthcoming as new technologies are added. First is the capability to per­ form on-stream liquid-liquid extrac­ tions. Second is the incorporation of a chromatographic step. Third is the in­ clusion of sophisticated multichannel, rapid-scan spectroscopic detectors to provide a more robust single-compo­ nent analysis or simultaneous multicomponent analysis. As these capabili­ ties are added to FIA, it will become more apparent that this technology is an appealing paradigm for developing an automated universal microlaboratory that integrates the functions of sample acquisition, clean-up, concen­ tration, separation, sensing, and iden­ tification. Process engineers will find the approach particularly attractive because it so closely resembles the flow-stream concept used in automat­ ed chemical-processing lines. Continuous on-line process analysis is the first era to offer true real-time capability. The instrumentation is largely spectroscopic in nature. Until recently, spectroscopic devices were limited by the technological con­ straints imposed by extreme reliability requirements, the hostile nature of the plant environment, and the necessity to remain accurate through long peri­ ods of inattention. In the past, such constraints led to the development and use of dedicated process analyzers that achieved ruggedness and reliability at the expense of analytical power. As an example, consider infrared (IR) spec­

626 A · ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987

troscopy. Process IR analyzers previ­ ously were based on nondispersive, dual-cell correlation devices with mi­ crophone detectors. The state of the art in laboratory analysis is, of course, the Fourier transform infrared spectrome­ ter (FT-IR), an instrument that in the past would hardly have been thought suitable for use on the factory floor. Thanks to innovative design, the reli­ ability of laboratory instruments has so greatly increased that the FT-IR is be­ ing used on-line. At least two commer­ cial manufacturers have begun to offer instruments that were developed spe­ cifically for on-line process analysis. Another trend that has enhanced the versatility of FT-IR use on-line is the development of improved methods for continuous sampling. For liquid streams, attenuated total reflectance (12) offers a way to eliminate ultra-nar­ row path-length cells and their atten­ dant difficulties with blockage from particle matter in the sample and inter­ ference fringes. In addition, this ap­ proach is well suited to analysis of mul­ tiphase processes such as emulsions. The cylindrical geometry cell provides a simple method for implementing at­ tenuated total reflectance technology in a form well suited to high-pressure, high-temperature, mixed-phase proc­ ess applications (13). In the ultraviolet-visible (UV-vis) arena, technological developments have made scanning versions of these instruments far more rugged and reli­ able. Three key concepts are the adap­ tation of single-beam techniques that use stored baselines rather than me­ chanically sampled reference and sam­ ple compartments; the use of concave holographic gratings that yield accept­ able stray light rejection in a single monochromator and eliminate the need for collimating mirrors; and the introduction of photodiode arrays that eliminate mechanical scanning of the grating (14). The current generation of low-end UV-vis spectrometers based on these principles is almost cost-com­ petitive with filter wheel photometers and is certainly more cost-effective and just as reliable. In-line era. The major disadvantage of on-line analysis is the need to con­ struct a separate analytical line that properly samples the main stream and presents it to the instrument at a suit­ able temperature and pressure. This has led to the fourth era: in-line process analytical chemistry. Here, chemical analysis is done in situ, directly inside the process line, using a probe that is chemically sensitive. In its ideal form, the device might resemble a typical in­ dustrial temperature probe. Such an implementation obviously would be at­ tractive to process engineers. It is widely believed that the key to scaling down chemical sensors is to fab-

ricate them with tools that originate in the microelectronics revolution: microlithography and micromachining (3). Virtually all of the microsensors fabricated to date use a two-stage detection scheme. In the first stage, chemical selectivity is obtained by a chemical transformation or by physisorption or chemisorption to a chemically selective surface. Table II lists various schemes that have been used to obtain chemical selectivity for microsensing. In the second stage, some physical consequence of the chemically selective stage, such as release of heat or change of optical absorption, is converted to an electrical signal by a suitable microtransducer. Table III lists the various physical properties that can be sensed and the types of transducers that have been employed. Together Tables II and III form a matrix of possibilities for chemical sensing. Virtually all entries in the tables have been explored. The most noted of the microsensors are the CHEMFET and the ISFET (15). The ISFET, which was first reported by Bergveld some 15 years ago, arose from a desire to scale down common pH or ion-specific electrode technology. It is therefore important to note that such devices share all of the advantages and disadvantages of potentiometric electrochemical detectors. It is now widely appreciated that the development of long-lived, reliable

Table II. Chemical/physical transformations employed in microsensors to obtain chemical selectivity Chemical transformations • Electrochemical • Enzyme catalyzed Physisorption to a surface • From gas phase • From liquid phase Binding to an immobilized receptor • Immunoadsorption • Ion exchange Membrane transport • Electrolyte separation

Table III. Techniques for detection and transduction employed by chemical microsensors Physical property

Transducer(s)

Chemical potential Optical Heat Mass Mass Conduction Surface resistance Current flow

Field effect transistor Photodiode Thermistor, pyroelectric Piezoelectric balance Surface acoustic wave Dielectrometer Chemiresistor, Taguchi Ampometer

O N - L I N E versions of these devices will require considerable progress in encapsulation and packaging. In the CHEMFET, the ion-selective membrane is replaced by a chemically selective layer. For example, a palladium-gate FET has been developed for detection of hydrogen at the parts-permillion level. Janata pioneered the use of this approach for biochemical detection by developing enzyme-based (ENFET) and antibody-based (IMMUNOFET) sensors. One of the major issues in the design of these microsensors is the optimization of their responses. (See the paper by Janata and co-workers that shows the multitude of considerations involved [16].) A second class of microsensors takes advantage of the electrooptical revolution: fiber optic sensors (4). With these microsensors, probe light is generated remotely and conveyed to the sensor end by a light guide, where it interacts with a chemical probe by absorption, scattering, fluorescence, or Raman emission. The light, which has been encoded with chemical information, is then returned by the same or a second light guide to an electrooptical transducer that creates the desired electrical signal. Such an optical technique is very versatile. One has the entire range of chemistries developed over the years as spot tests to provide inspiration. Many of these can be immobilized onto glass or polymer beads and encapsulated into microcuvettes to form optrodes of all sorts. Another use of fiber optics may prove to be of even greater value as a means of elevating optical spectroscopy to the fourth era. This is accomplished by developing various types of fiber optic probes that substitute for the conventional sampling streams. Still another advantage of this approach, as pointed out by Hirschfeld (17), is the ability to employ fiber optic multiplexing schemes so that a single expensive instrument can be placed in a safe, convenient location to sense

628 A · ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987

multiple points in a process line. At present, fiber optic sensing is limited to the UV-vis-near-IR, but new types of IR transparent glasses are rapidly being developed. Thus far, we have confined our thinking about chemical microsensors to the one-sensor/one-analyte concept. These microsensors, however, offer many other possible applications, and we find it useful to think in terms of a hierarchical complexity of devices that the microelectronics paradigm provides. Simple single sensors are at the first level. The second level in the hierarchy is the sensor array. Here, the ability to use microlithography to fabricate multiple sensors on one chip is an obvious way to avoid having multiple probes for multiple analytes. The straightforward approach to implement a multifunction sensor is to devote one sensor to each analyte of interest. This method, however, does not address the wellknown problem of interferences in the chemical selectivity step. An alternative strategy is to use a set of relatively unselective sensors, each of which has a different response profile to all the analytes of interest. In this case, the array generates a response pattern that may be analyzed by various powerful methods of multivariate statistics and pattern recognition. The combination of sensor array and data processing constitutes a robust analytical system that can yield simultaneous multicomponent analyses with the capability of recognizing and correcting for interferences and drift (18,19). The same multivariate methods also may be used to optimize specific multicomponent analyses problems, selecting the minimum number of sensors or sensing channels (e.g., wavelengths) to yield optimal performance (20). The third member of the microsensor hierarchy is the ultraminiature optical spectrometer. One possible approach employs the highly successful optical imaging device, which is a microfabricated array of photosensors. Previously these devices were used as electronic photographic plates in conjunction with conventional spectrographs. Alternatively, extremely small solid-state spectrometers have been constructed using tiny diffraction gratings and graded index of refraction lenses as collimators (21). Unfortunately, diffraction gratings do not scale down very small, because fewer and fewer grating lines are sampled and lower resolution results. A different approach is multibeam interferometry, which is widely known as the basis of interference filters. Here, wavelength analysis takes place over the dimensions of microns rather than centimeters. One suitable implementation of this concept is the linear interference

wedge filter in which the wavelength of maximum transmission varies continuously along one axis. A prototype compact planar spectrometer using a wedge filter and linear array detector gives acceptable results (22). The remaining member of the hierarchy is the chemical microlaboratory. The goal is to integrate all functions of a chemical analysis laboratory into a unit no larger than a credit card. The key to this development is the integration of microconduits, microvalves, and microsensors by various types of hybrid circuit fabrication techniques (23). Although this device is really a third era concept, it provides a pathway to the implementation of chemical microsensors. These devices should last much longer when incorporated into a microlaboratory because of extensive sample conditioning and intermittent use. Noninvasive era. The final era of process analytical chemistry, the noninvasive era, represents the ultimate in desirability. Because the probe does not physically contact the sample, the sampling problem is greatly alleviated. This era obviously has a great deal in common with remote sensing and nondestructive evaluation. Near-IR spectroscopy from 7001100 nm has much to offer (24) in this regard. First, the extinction coeffi-

I N - L I N E cients in this region are very small and allow path lengths of 0.5-20 cm for clear solutions. Generally, absorptions in this region arise from the second and third overtones of CH, OH, and NH stretches together with combination bands from other vibrations. Because these are highly forbidden transitions, most materials are very transparent (exceptions are metals and graphite containing composites). At these optical distances, a thin layer of adsorbed material in the windows does not fatally degrade the results. Also, quantitative measurements of highly scattering

Laser Light Scattering Systems for particle size analysis, molecular weight measurement and other applications The Malvern SYSTEM 4700c is an advanced modular system designed for research and development applications. It features the 8-bit Malvern 7032 Multi-8 correlator with Variable Time Expansion (VTE) allowing up to eight sampling times per measurement Fixed and expandable configurations ranging from 72 up to 264 data channels are available. A 16-bit IBM compatible processor plus full range of software for photon correlation spectroscopy and total intensity light scattering ensure optimum utilisation of the top quality manual or computer controlled goniometer.

SYSTEM 4700c

Optical units are also available for use with SYSTEM 4700c for fixed angle PCS and electrophoretic zeta potential measurements.

Malvern Instruments Inc 1 200 Turnpike Road. Southborough, MA 01772, USA Tel. (617)480 0200

MALVERN

CIRCLE 94 ON READER SERVICE CARD

630 A · ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987

materials can be made because scattering coefficients are much greater than absorption coefficients. Consequently, diffuse log inverse reflectance and diffuse absorbance measurements show good linear correlation with composition variation. Diffuse reflectance is particularly useful, because only a single window into the process need be established. In addition, the high degree of light scattering results in the sampling of a large volume. The hardware is inexpensive and employs widely available fiber optic components, conventional monochromators, tungsten lamps, and silicon detectors. Finally, the spectra are characterized by an extremely high signal-to-noise ratio. Thus subtle shifts in the spectra, which might go undetected by the unaided eye, can become the basis for highly successful analytical procedures developed by computer learning methods. A second remote analysis technology is IR emission spectroscopy. This approach starts from the observation that all bodies at a temperature higher than their surroundings will emit radiation. For simple black bodies the spectral distribution of radiation is very smooth and is given by Planck's law. For real substances, however, discrete lines are observed on top of the Planck emission. These lines correspond well with those observed in the normal IR absorption spectrum of the emitting substance. Thus it is certainly possible to obtain qualitative information. But obtaining quantitative information is more difficult because most of the emission originates at the surface, severe reabsorption effects are evident, and the lines are present on a large background. Nevertheless, the potential advantages of a truly remote in situ technique justify further effort to understand the quantitative aspects. At lower energies, in the microwave and radiofrequency bands, objects become rather more transparent, and therefore information can be obtained from deep within. Whereas the microwave absorption spectroscopy of gasphase molecules is characterized by high-resolution unique spectra, condensed phases exhibit a much broader "dielectric" behavior. Nevertheless, different materials do exhibit different enough spectral characteristics to obtain compositional analysis, and even to form the basis for imaging (24). Other advantages of microwave technology for condensed phases include the ability to perform spectroscopy directly in the frequency domain with high-speed GaAs integrated circuitry and the ability to sample large, well-defined volumes with near-field resonators. Yet another noninvasive technology involves X-rays. Most analytical chemists will immediately think of X-ray fluorescence, which has been adapted

to both on-line and in situ use to yield high-speed elemental analysis. At the same time, it has been shown that differential X-ray absorption can be used to give compositional selectivity. Formerly, tunable X-ray sources were extremely rare, but major advances in Xray optical coatings, materials, and sources will remedy this difficulty (25). The final, fifth era of technology to consider is ultrasound. The major use of ultrasound has been for out-line imaging of flaws, employing reflections from index of refraction boundaries. However, a recent paper makes a strong case for further research on the mechanisms for ultrasound absorption, which is based at least in part on chemical composition (26). In reviewing the remote era, we can see how much commonality the field has with nondestructive evaluation and noninvasive medical diagnosis. As an example of the use of the process analytical chemistry paradigm, consider the current practice of clinical chemistry. For the most part, its status is similar to the first and second eras of process analytical chemistry. Blood is drawn from the patient and is either sent to a remote central laboratory (off-line) or analyzed by a simpler doctor's office instrument (at-line). At the University of Washington, we are developing the fifth era of clinical analysis (real-time, noninvasive) by exploiting new concepts in spectroscopy and data reduction. A new role is sure to emerge for the clinical chemist who becomes a full partner in supplying critical data to the doctor and, with multivariate statistics, interprets it in terms of the patient's current physiological status and prognosis. Role of chemometrics We have completed an outline of some of the possibilities for chemical sensing in process analysis. How does one analyze the data from the array of instruments and sensors that have been placed on-line? At this point, a major cultural gap becomes evident between the analytical chemist and the process engineer. Most analytical chemists regard their function as supplying chemical composition information, which is, after all, what they have been trained to do. Unfortunately, this does not always meet the needs of the process engineer, whose job is to provide a product that meets quality specifications. In some cases, especially for chemicals production, composition data suffices. In many situations, however, and especially for consumer products, the specifications are not obviously chemical in nature. An example of this is gasoline testing. Here, the most important consumer quality parameter is the antiknock property as expressed by the octane number. Presently, this parame-

NONINVASIVE ter is measured by a knock engine that is extremely expensive, requires a subjective judgment (listening for pinging) to make the measurement, and must be exactingly maintained (1). Even worse, the measurement takes 20 min to perform and consumes a full pint of gasoline. Finally, the poor precision and large interinstrument variability cost the refiner a lot of money, because the quality specifications must be set high enough to allow for octane measurement uncertainty, and they cause a lot of irritation, because the regulatory agency's octane determinations may not agree with the company's. At the Center for Process Analytical Chemistry, we have been exploring a new approach to on-line quality evaluation. We begin with the observation that vibrational absorption spectroscopy can yield information about functional groups, molecular structure, and conformation, as well as intermolecular interactions. In principle, one should be able to obtain chemical and physical material properties from the spectrum. Unfortunately, such information often is present in subtle variations in the spectrum and might not be recognized by the naked eye, or it might be dependent upon an unknown relationship among several spectral features. Accordingly, to obtain the s t r u c t u r e property relationship, one must either develop a first principles theory, which is generally impossible, or use an empirical method such as multivariate statistical calibration (27). In the latter approach, the computer is provided with a training set consisting of the spectrum of each of the samples and the property, measured by an independent reference method. It is important that the training set consist of representative samples with variations spanning that of the set of all samples that might reasonably be encountered by the method. Once the data set is acquired, the computer searches for correlations between the spectra and the sought-for property. Eventually a

632 A · ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987

transformation vector is produced that, when multiplied by the spectral vector, yields an estimate for the property. Once the derived transformation vector has been validated with a double-blind study of a second set of samples for which spectra and property have been measured, the method is ready for use. In the case of gasoline, a study of 43 samples of unleaded gasoline of known octane number was made using nearIR spectroscopy over fiber optics. By simple step-forward multilinear regression, a linear equation involving only the absorption intensities at three wavelengths was found that correlated to the reference octane number with a standard error of 0.3 octane, the known precision of the octane engine (28). Further studies involving more samples clearly will be required to verify the method and to assure its robustness. But the effort is justified because the spectroscopic test takes only 20 s to perform, and the instrumentation costs far less and requires considerably less maintenance than that used in other methods. Another major advantage of on-line spectroscopy is that a number of chemical and physical properties can be measured simultaneously from one spectrum. For gasoline, we have been able to correlate the spectrum to both research and motor octane, API density, Reid vapor pressure, distillation points, and total aromatics, olefins and aliphatics—eight tests in 20 seconds! One of the most intriguing aspects of multivariate calibration is the capability to generate the correlation spectrum that is associated with a paticular property (29). This may prove to be an extremely valuable tool in materials research. For example, the correlation spectrum of tensile strength might reveal the presence of specific supramolecular structures or unique intermolecular bonds that determine the specific property. Such knowledge could lead to new materials with significantly improved performance. Thus far, no algorithms have been formulated that use multiwavelength spectroscopy in real-time process control. A study by Wise (30), however, shows how multiple physical measurements (temperature at various points, applied power, substrate resistance) together with multivariate statistics can be employed to prevent runaway foaming during the operation of a liquid-fed ceramic melter. Another potential use for multivariate calibration is to qualify incoming raw materials. Many types of processes are bedeviled by batch-to-batch variations in the raw materials. A recent study (31) showed that multiblock partial least-squares regression could be used to form a model with which one

could use the quality control data—in this case, IR spectra and sonic spectra—to predict the outcome of the process in terms of the performance of the final product. In the future it should be possible to combine information about the properties of incoming raw materials with a process model to adjust conditions and accommodate variations in raw materials, thereby producing a product of more consistent quality. Implementing process analytical chemistry In industry, process engineers and analytical chemists all too often are separate groups, each with its own goals and distinct culture. Thus it is not uncommon for the analytical group to convince managers to install a sophisticated process analyzer but to avoid the issues of maintenance and calibration. Frequently the shiny new instrument becomes rusty while the two groups argue over responsibility. We have observed that companies that are most successful in implementing process analytical chemistry have formed interdisciplinary working groups of engineers and scientists devoted solely to that task. Another difficulty in implementation arises from barriers created by horizontal integration in a company. Seldom does the bench chemist who develops a new synthetic method participate actively in scale-up operations. Even pilot plant personnel seldom migrate to the full-scale operation. This communication gap should be eliminated by encouraging vertical movement within a company. The benefits of this strategy are well known in a few instrument companies, where a research engineer or scientist might actually move through the design process to developmental engineering, to manufacturing, to marketing, and, finally, to participating in selling the product. The vertical concept can be readily applied to process analytical chemistry. The process analytical chemist should become involved with a new synthetic project at its earliest stages of development. The on-line tools supplied to the bench chemist will greatly aid the process of converging to a useful synthetic strategy. When the new process is worked out, ideally the synthetic chemist and process analytical chemist alike will move vertically to the pilot plant level. The bridge between these levels will be the analytical instruments, which should greatly decrease the pilot plant time. In the final step, the on-line instrumentation will migrate onto the plant floor. The above scenario has embedded within it a means whereby academic analytical chemists may also become involved in on-line analysis. The time is right for analytical chemists to ap-

proach their colleagues involved in synthesis concerning a collaborative effort with a goal of optimizing new synthetic approaches, using analytical instruments to give real-time information about how a given reaction is proceeding. A recent contribution from academia describes a simple mass spectrometer interface to reaction mixtures that uses semipermeable capillary tubing and illustrates its use in reaction monitoring (32). The future Process analytical chemistry is a discipline poised to have a major impact both on industrial processing and on the parent discipline. The savings to be obtained from installation of the present generation of process analyzers has been quantified (33), and it can be enormous. The new generation of online spectroscopic instruments and the accompanying multivariate analysis promise even more capabilities, with the possibility of developing new processing methods that previously could not be considered because of the lack of adequate control. Perhaps the most exciting promise of process analytical chemistry is to make process lines so efficient and so well controlled that effluent release is decreased or even eliminated altogether. Imagine the good will such efficient plants might generate in the community. The changing practice of process analytical chemistry is sure to influence the discipline as a whole. It is a leading force behind the migration of analytical instruments from the protected laboratory into the real world. The use of the micro- and optoelectronics paradigms is leading to a new generation of personal, portable instruments and sensors that can be used as tools even by those who are inexperienced. The parallel with the impact of personal computers should be obvious and should be welcomed by analytical managers who see the trend as a way to free their personnel from the drudgery of routine analysis. References (1) Clevett, K. J. Process Analyzer Technology; John Wiley: New York, 1986. (2) Set Am. 1986,255(4), 1-212. (3) Wohltjen, H. Anal. Chem. 1984, 56, 87 A-103 A. (4) Seitz, R. Anal. Chem. 1984, 56, 16 A34 A. (5) Hirschfeld, T.; Callis, J. B.; Kowalski, B. R. Science 1984, 226, 312-18. (6) Cornish, D. C ; Jepson, G.; Smurthwaite, M. J. Sampling Systems for Process Analyzers; Butterworths: London, 1981. (7) Phillips, J. B.; Luu, D.; Pawliszyn, J. B.; Carle, G. C. Anal. Chem. 1985, 57, 27792787. (8) Lipsky, S. R.; Duffy, M. L. LC-GC1986, 4(9), 898-907. (9) Jorgenson, J. W. Anal. Chem. 1986, 58, 743 A-760 A. (10) Levy, G. B. Am. Lab. 1986, 28(12), 6271.

HPLC Systems For superior HPLC analyses, ESA provides a complete line of HPLC components and systems-pumps, injectors, detectors, and automatic sampling equipment. Use these modular components to design an HPLC system to meet your exact applications and budget needs. Or add highperformance components to your existing HPLC for analysis results you can trust. For example, our systems-



Advanced Modular HPH System includes random a c c e s s variable volume injection autosampler with most sensitive dual wavelength UV/Vis detector available and completely automatic dualchannel data acquisition and control system.



The World Standard Coulochem' Electrochemical Detector and Solvent Delivery System makes electrochemical detection at the picogram level as easy as UV detection at the microgram level.

Send today for our complete HPLC products catalog.

esa ESA, Inc. 45 Wiggins Avenue Bedford, Massachusetts 01730 (617)275-0100 CIRCLE 41 ON READER SERVICE CARD

ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987 · 635 A

(11) Ruzicka, J.; Hansen, E.H.H. Flow In­ jection Analysis; John Wiley: New York, 1981. (12) Harrick, N . J. Internal Reflection Spectroscopy; Harrick Scientific Corp.: Ossining, N.Y., 1979. (13) Rein, A. J.; Wilks, P. Am. Lab. 1982, 24(10), 152-55. (14) Miller, J. C ; George, S. Α.; Willis, B. G. Science 1982,57, 1917-20. (15) Solid State Chemical Sensors; Janata, J.; Huber, R. J., Eds.; Academic: New York, 1985. (16) Caras, S. D.; Janata, J.; Saupe, D.; Schmidt, K. Anal. Chem. 1985, 57,191720. (17) Hirschfeld, T. Adv. Instrum. 1985, 40, 305-18. (18) Kalivas, J. H.; Kowalski, B. R. Anal. Chem. 1981, 53, 2207. (19) Kalivas, J. J.; Kowalski, B. R. Anal. Chem. 1982, 54, 560. (20) Carey, W. P.; Beebe, K. R.; Kowalski, B. R.; Illman, D. L.; Hirschfeld, T. Anal. Chem. 1986,58,149-53. (21) Fuh, M. S.; Burgess, L. Anal. Chem., in press. (22) Pfeffer, J. C ; Skoropinski, D. B.; Cal­ lis, J. B. Anal. Chem. 1984,56, 2973-74. (23) Ruzicka, J. Anal. Chem. 1983, 55, 1040 A-1053 A. (24) Rao, P. S.; Santosh, K.; Gregg, E. C. Radiology 1980,135, 769-70. (25) Low-Energy X-Ray Diagnostics; Attwood, D. T.; Henke, B. L., Eds.; Amer­

James B. Callis (right) is co-director of the Center for Process Analytical Chemistry (CPAC) and professor of chemistry and adjunct professor of bioengineering at the University of Washington. He received his B.S. in chemistry from the University of Cali­ fornia at Davis and his Ph.D. in physi­ cal chemistry from the University of Washington. Callis's major research interest is instrumentation for optical spectroscopy. He is developing ultraminiature spectrometers based on novel transform principles, laserbased chromatography detectors, and noninvasive diagnostic devices. Deborah L. Illman (center) is associate director of CPAC. She received her B.S. in chemistry (1976) from the Uni­ versity of Washington and her Ph.D. in chemistry (1982) from the State University of Campinas, Brazil. She then accepted a lectureship and post­ doctoral position at the University of

ican Institute of Physics: New York, 1981. (26) Slutsky, L. J. IEEE Trans. Ultrason. Eng. 1986, UFFC-33, 156-61. (27) Martens, H.; Naes, T. NATO Ado. Study Inst. Ser. C. 1984,138, 147-56. (28) Kelly, J. J.; Barlow, C. H.; Jinguji, T. M.; Callis, J. B., unpublished results. (29) Honigs, D. E.; Hieftje, G. M.; Hirsch­ feld, T. Appl. Spectrosc. 1984,38,317-22. (30) Wise, B. M.; McMakin, A. H., submit­ ted for publication in Glass Industry Journal. (31) Frank, I. E.; Feikema, J.; Constantine, N.; Kowalski, B. R. J. Chem. Inf. Comput. Sci. 1984,24, 20-24. (32) Brodbelt, J. S.; Cooks, R. G. Anal. Chem. 1985,57,1155-57. (33) Reeves, P. Proceedings of the ANA­ TECH '86 International Conference; Norwijkerhout, The Netherlands, April 1986.

Suggested reading Huskins, D. J. Quality Measuring Instru­ ments in On-Line Process Analysis; Halstead Press: New York, 1982. Mix, Paul E. The Design and Application of Process Analyzer Systems; Wiley Interscience: New York, 1984.

HPLC Autosomplers For superior HPLC analyses, ESA provides a complete line of HPLC components and systems-pumps, injectors, detectors, and automatic sampling equipment. Use intelligent components to design an HPLC system to meet your exact applications and budget needs. Or add these highperformance components to your existing HPLC system for analysis results you can trust. For example, our autosamplers-

This work was supported by the Center for Process Analytical Chemistry from grants provided by the National Science Foundation ( # ISI-8415075), the 30 sponsors, and the University of Washington.

Washington, where her research inter­ ests focus on chemometrics: multivari­ ate methods of pattern recognition and statistics applied to analyzing chemical data. Illman manages the operations, industrial relations, and policy development at CPAC and con­ tributes to its research program. Bruce R. Kowalski (left) is professor of chemistry and co-director of CPAC. He received his B.S. with majors in chemistry and mathematics (1965) from Millikin University and his Ph.D. in chemistry (1969) from the University of Washington, Seattle. His research interests include the ap­ plication of methods of multivariate analysis to the resolution and calibra­ tion steps in analytical instrumenta­ tion in order to develop new analytical methods and improve existing ones. Kowalski is editor-in-chief of the J o u r ­ nal of C h e m o m e t r i c s . In 1974 he cofounded the Chemometrics Society.

I Model 460-110 position random access, variable volume autosampler for maximum chromatographic versatility. Requires only 5 [XL more than injection volume to preserve valuable samples. Node! 660-60 position, fixed-loop autosampler for reliable, routine processing of samples at a reasonable cost.

Send today for our complete HPLC products catalog.

esa ESA, I n c . 45 Wiggins A v e n u e Bedford, Massachusetts 01730 (617)275-0100 CIRCLE 42 ON READER SERVICE CARD

ANALYTICAL CHEMISTRY, VOL. 59, NO. 9, MAY 1, 1987 · 637 A