CA2174641A1 - Method for monitoring and controlling a chemical process - Google Patents

Abstract

A computer (14) implemented on-line method of monitoring (12, 26) and controlling (22) a chemical process is used for maximizing the yield of ethanol, sugars such as glucose and xylose and by-products such as silica or silicates from a biomass. The concentration of process reactants and products is sampled (10) and measured using spectrometric technology. The spectral data is stored and analyzed in the computer (14) using a modified chi-squared processing method to determine the unknown concentration of reactants and products in test samples (10) withdrawn from the process reactor (11) product stream. The need for the spectral data to conform to Beer's Law is avoided;
the best spectral range is determined automatically; and the physical parameters (12) of the process are monitored (12, 26) and altered (20, 22, 24) based on this determination as required to optimize biochemical actions in the process reactor (11).

Description

2 1 7464 1 ~- WO95/11485 PCT~S93/10170 Nh~O~ FOR MONITORING AND CONTROLLING A CHEMICAL PROCESS

RET~TED PATENT APPLICATIONS

This application is a continuation-in-part patent application of U. S. Serial No. 07/628,321, filed February 2, 1993, and entitled "Method For Monitoring And Controlling A Chemical Process," which is now U. S. Patent No. , and which is incorporated herein by reference and made a part of this patent Aprl;cAtion.

BACKGROUND OF THE INVENTION

Field of the Invention:
This invention relates to a method for monitoring and controlling a chemical process essentially instantaneously based on real time measurements of the concentration of reactants or products produced by the process, or both.

Wos5/ll~5 pcT~ss3llol7o 2~ 7464 1 Background Discussion:

There are many chemical processes which require physical parameters to be altered based on the concentration of reactants in the process or products produced by the process. Frequently, temperatures, pressures, flow rates, pH and other physical parameters of the process must be changed to optimize the operation of the process. (As used herein chemical process(es) shall mean any process wherein the concentration of reactants or products produced by the processes changes. It includes in vitro and in vivo biochemical, nuclear, metallurgical, petrochemical, etc, processes.) Conventional monitoring and control techniques are unable to measure the concentration of reactants, the products produced by the process, or both, with sufficient speed to then regulate process parameters based on the measured concentrations. One technique used to measure the concentration of reactants is electromagnetic radiation absorption technology. Over the range of the electromagnetic spectrum, all chemicals absorb or reflect "light" in a unique way that is characteristic of the structure of chemical being examined and its concentration in a mixture of different chemicals. For our purposes, light includes all regions of the electromagnetic spectrum from x-rays, W, visible, infrared to mic-o..aves. For example, one may determine the co~ce~Lration of sucrose in water by spectroscopy, a technology wherein, for example, an aqueous sucrose solution is exposed to infrared light (IR) at different, discrete wavelengths.
The absorption spectra (the level of light absorption over a range of different, discrete wavelengths of light) is characteristic of the aqueous sucrose solution. The spectra is usually described in terms of the wavelength of electromagnetic radiation, for example, from l to l00 micrometers (ym). In spectroscopy, one frequently finds it useful to use a slightly different measure for the spectral position known as the wavenumber. The wavenumber, in cm-l, is related to the wavelength in ym by l/l0,000, that is l ym is l0,000 cm-l and l0 ym is l,000 cm-l. Conventional Fourier Transform-IR technology is routinely capable of scanning from 1.2 ~lm to 100 ym in as little as 1/8 second with a resolution of 2.5x10-5 )lm (O.25 cm-l). For most purposes, including the purposes of the present invention (on-line chemical control and monitoring), a resolution of 2.0-4.0 cm and scan speed of 1-10 seconds is sufficient. This allows use of less expensive equipment since one pays a premium for speed and resolution.
The governing principle behind current quantitative analytical methods of transmission or absorption measurement instruments, in which realm IR analysis falls, relies on a relationship known as the Bouguer-Beer-Lambert Law. Many sources simply call this Beer's Law. In the simplest form it is written:
A = abc ~1]
where A is the absorbance, a is the molar absorptivity, b is the pathlength, and c is the concentration. Since the amount of energy absorbed is related to the number of molecules, the concentrations involved are molar quantities such as moles per liter or mole fraction. A mole of a material is a fixed number of molecules, e.g., 6.023 x 1023 if the weight (called the molecular weight in this case) is given in grams. The molar absorptivity is the absorbance expected when 1 mole of a particular compound is present at the part;r~ r wavelength that the measurement is made.
Equation [1] is normally assumed to hold for every discrete wavelength for which the instrument can distinguish adjacent wavelength intervals. For example, with a 2.0 cm-1 resolution, it is possible to distinguish reproducible differences as close as 1.0 cm-1. Thus, a spectra from 450 cm-1 to 4400 cm-1 would have

3,951 points and equation [1] would be assumed to apply to each of these points.
It is frequently useful to combine terms in equation 11] into the form:
A = kc [2]
where k now represents a "constant" which combines the molar absorptivity and the pathlength. By measuring the absorbance, A, of samples with differing amounts of material at various concentrations, c, one can calculate k. When k does not vary over wo gs~ll48s 2 1 7 4 6 4 1 pcT~ss3llol7o a range of concentrations, the samples and the material being measured are said to obey Beer's Law.
Once k is known, unknown samples of the material are determined simply by measuring the Absorbance, A, and dividing by k for each discrete wavelength for which the measurements are made and for which Beer's Law has been shown to apply.
One can measure transmission as well as absorbance. The two quantities are related by the expression:
A = ln (l/T) [3]
where T is transmission. The transmission is defined as the fractional reduction in intensity of a beam of electromagnetic radiation passing through the medium containing the absorbing material. Formally, it is:
T = I/Io [4]
where I is the measured intensity with the absorbing material in the beam and Io is the measured intensity without the absorbing material. Sometimes Beer's Law is written:
I = Io e-~C [5]
which is the algebraic combination of equations [l], [3] and [4].
It is much more convenient to analyze results using equations [l]
avoiding the use of exponentials.
When one has a mixture of materials, equations [l] or [2] is usually held to be applicable to each of the materials separately.
That is, for a mixture of 3 materials at each wavelength or ~._v~l.umber:
A = klcl + k2C2 + k3c3 [6]
where the subscripts refer to the three components. Since equation [6] holds at each wavelength (or wavenumber), there are as many equations as wavelengths so the values at the selected wavelength (m) can be written as Am = kmlcl + km2C2 + km3C3- This could be, for example, a mixture of three gases moving through a gas cell attached to an FT-IR instrument or two solutes (such as sucrose and ethanol) dissolved in a solvent (such as water). In current practice, again assuming that Beer's Law is valid, the k1 woulq have been determined from previous experiments and thus the cl can be calculated if three sets of measurements are taken at three or more wavelengths. In current practice, the measurements ~- woss/11485 PCT~S93/10170 over many wavelengths are used to find the "best fit" for c1 using least squares or partial least squares regression analysis. There is no way of determining which series or ranges of wavenumbers is the best to use. This is done exclusively by trial and error based on the analyst's experience.
For a discussion of the state-of-the-art, the book "Fourier Transform Infrared Spectrometry" by Peter R. Griffiths and James A. de Haseth, John Wiley & Sons, 1986 is recommended. Chapter 10 in this book discusses quantitative analysis. Of particular importance is section IV on multicomponent analysis beginning on page 355. On page 356, the authors note that Beer's Law is a requirement for the analysis techn;ques they present.
There is also a four volume series edited by John R. Ferraro and Louis J. Basile. The series is entitled "Fourier Transform Infrared Spectroscopy" and is published by Academic Press, Inc.
Volume 1 was published in 1978, Volume 2 in 1979, Volume 3 in 1982, and Volume 4 in 1985. The latest volume contains a contribution by P. C. Gillette, J. B. Lando and J. L. Koenig on "A
Survey of Infrared Spectral Processing Techniques." They again ~tate the requirements for Beer's Law as well as mention that least squares analysis is a preferred technique. Based on experimentation in connection with conducting chemical analysis which the present invention addresses successfully, the least squares and related techniques are highly overrated and are rarely the best techniques for looking at variable data or determining the "best fit" in analysis of spectral data. Further, Beer's Law rarely holds in practical systems, particularly in solvent systems or complex mixed gases or polymeric solids.

SUM~L~RY OF THE INVENTION

It is the objective of this invention to measure in real time the concentration of reactants and products produced by a chemical process and essentially instantaneously control the physical parameters of the process based on the concentration measurements to opt;m;ze the process. The term instantaneous can mean from a few seconds to a few minutes, depending on the WO 95/11485 PCT/US93/10170 `--speed required for the system under control. For a biochemical reaction, which takes days to complete, a few minutes is sufficiently instantaneous. For a rapid gas phase reaction, a few seconds would be needed. Specifically, this patent application deals with the application of the modified Chi-Square technique disclosed in U. S. Serial No. 07/628,321 to a process for producing ethanol from a cellulosic material.
The method of this invention has several features, no single one of which is solely responsible for its desirable attributes.
Without limiting the scope of this invention as expressed by the claims which follow, its more prominent features will now be discussed briefly. After considering this discussion, and particularly after reading the section of this application entitled, "DE~ATTT~n DESCRIPTION OF THE ~K~nRED EMBODIMENT," one will understand how the features of this invention provide its advantages, which include:

1. The mixture of chemicals being analyzed do not need to obey Beer's Law.
2. The "best" wavelength range for determining concentration is automatically located.
3. Answers are obtained based on statistical criteria, i.e., stAn~rd deviations and Chi-square.

4. The process is fast. Typically, it takes a few seconds even with only a 12 Mhz 80826 microcomputer with math co~lG~essor and hard disk to perform the final analysis.
Each of the steps can also be automated and are also fast.

5. The method works for any spectrometer and with any combination of solvents and solutes to the precision set by the limits of the mathematical analysis employed.
That is, if a statistically significant answer is possible for a particular combination of instrument, cell, and mixture of chemicals to a desired level of precision, the method of this invention has the best mathematical chance of finding useful ranges of the spectra and solutions to the analysis problem.

~- wos5/1148s PCT~S93/10170 There are a number of different types of IR instruments available, including grating, prism, and Fourier transform (FT) instruments. Although grating or prism IR instruments may be used for the purposes of the present invention, the FT-IR instrument is preferred because it can scan a wide range of IR frequencies in a very short time. The methods to be described are also applicable to other portions of the electromagnetic spectrum besides IR, for example, W, visible and microwaves.

Statistical D~corlvolution In accordance with this invention, a modified Chi-Square fitting mathematical technique is employed in analyzing spectral data. For purposes of this spectral analysis, points are better weighted by the inverse of the stAn~Ard deviation at a particular wavelength rather then the minimum of the squares of the deviations as in the least squares analysis. The analytical technique used in this invention is related to the Chi-Square fitting mathematical technique and it provides a much more powerful method and is a major improvement in analysis of spectral data.
In the modified Chi-Square technique of this invention, obtAin;ng the spectrum is repeated many times, for example, 32 or 64 replicates. The average and standard deviation are then calculated at each wavenumber (or wavelength). The calculation to find the best values of the concentrations from an equation like equation [6] are fit by weighting the points in the spectrum with the lowest standard deviation more than points with the higher deviations. This results in improved accuracy and precision.
The modified Chi-Square technique employed in the present invention is new in the context of spectroscopy. The known Chi-Square technique has been particularly modified to fit into the method of this invention for analyzing spectral data using a conventional general purpose computer and FT-IR instrument. An excellent discussion of the Chi-Square technique, in general terms, appears in "Numerical Recipes, The Art of Scientific Computing," by William H. Press, Brian P. Flannery, Saul A.

wogs/ll~s 2 1 7 4 6 4 1 PCT~S93/10170 ~

Teukosky, and William T. Vetterling, Cambridge University Press, 1986. Chapter 14 in the book does a good job of explaining the short falls of the overused least squares techniques and provides a particularly robust (in the mathematical sense) technique for carrying out a Chi-Square fit known as Singular Value Decomposition (SVD). This technique assures meaningful answers even when the equations being solved are unstable.
One disadvantage of the Chi-Square technique is that the stAn~Ard deviations of the measured solution are needed at each wavenumber. This would normally mean taking many spectra of the same sample. Since one purpose of our process is for continuous and rapid monitoring, a means was developed to avoid making multiple determinations.
It was found that the errors in the spectra are mostly a function of the instrument, the cell in which the sample resides, and the particular nature of the chemical solution being analyzed.
That is, a sample of water, sucrose, ethanol, and glucose in a particular cell in a particular instrument will have similar errors over a wide range of concentrations. Thus, in determining unknown samples of sucrose, glucose, ethanol in water, one can use the standard deviations from previous measurements on known solutions made while calibrating the instrument to determine the k's of equation [6].
One can always improve on the algorithm, either the Chi-Square or the least squares, by selecting specific wavelengths atwhich the calculations are done. That is, by looking at the separate spectra of the materials to be determined in combination, one can find regions of the spectrum where there is just the right amount of overlap or interference to make the final absorbance be a "comfortable" sum of all the materials. To better understand this, consider a gas phase spectrum comprising a mixture of nitric oxide, carbon monoxide, and carbon dioxide. In the gas phase, it is possible to find regions of the spectrum where characteristic absorbance lines of only one of these items shows up. Since only one item shows up, one can use equation [l] separately for each component by only analyzing for that gas over that restricted wavelength region. It is this technique that made gas phase work ~- woss/1148s PCT~S93/10170 easier and this logic that made artisans in this field seek simpler analytical methods for liquid or solid phase problems.
The present invention provides such a simpler method using regions of the spectrum where the absorbance due to a complicated mixture is strongly influenced (overlapped) by many of all of the components of the mixture.
Surprisingly, in the case of accurate and precise multicomponent analysis using equation [6], or a similar equation for however many components one wishes, the best results appear in wavelength regions where the terms k1 and cl have approximately equal values in the expression (k1c1). One can determine the k1 separately for each component beforehand. The region of the spectra that has k values for each component multiplied by the-expecteq concentration, approximately equal to each other, will be the best range in which to work. For example, in a mixture of sucrose, glucose, and fructose in water, this region is 920 to 1250 cm-l. This type of conclusion, for each set of materials studied, is reached naturally by use of the methods of this invention, but cannot be predicted.
Even more surprising was the discovery that once one became used to exploring the values of k versus wavenumber (or wavelength) the files of k's could be kept as a function of concentration and the best k for the values nearest the concentration range being sought could be automatically used with a very simple computer algorithm. This means that adherence to Beer's Law is no longer a requirement for the analysis. One simply must know the value of the k's as a function of concentration for each material separately or as modified by interactions with each other as shall be explained below. These k values are stored in data files in the memory of the computer as ranges of interest and then equation [6] is solved using the k values that best meet the concentration range of the mixture.
After one calculation, when concentrations have been found, the k values closest to that concentration, or new k values found by non-l;ne~r cubic spline interpolation between k values above and below, are used to recalculate the concentrations. This procedure Woss/1148s 21 74~41 PCTtUS93tlO170 is repeated until the method converges to the desired degree of precision.

EXAMPLE I

The modified Chi-Square technique of this invention is best explained with reference to a concrete example such as analysis of a mixture of sucrose, glucose, and ethanol in water. The method steps are:
l. Make up a series (5) of individual calibration solutions or samples of each of the components at a known concentration in water, spAnn;ng the concentration range of interest, for example, l-lO weight percent. Take the spectra of these calibration samples over a selected range of interest, for example 600 cm-l to 4400 cm-l~ and determine k's for each sample at each concentration and at each wavenumber from:

A = kwatercwater + kgamplec [7]
and Cwater + c = l.0 [8]

Equation t8] is true because the concentrations are in mole fractions which must add up to lØ The kWater is simply the absorbance of the background water. This is obtained by taking a spectra of pure water alone in the same cell. The range for the spectra can be in any portion of the spectra in which one expects to find significant information. In this case, based on general knowledge of the spectra of the components, the range of 600 cm-l to 4400 cm-l was chosen. This means that one has a file of k's for each of the compounds, sucrose, glucose, and ethanol for each wavenumber. This process is repeated 5-20 times, optimally lO, such that one can obtain statistically significant values of the k's at each concentration.

` 21 74641 --WO gS/11485 PCT/US93/10170 2. Use the same spectra from 1 to obtain a standard deviation file. For example, the 10 spectra of sucrose (without modification) are averaged and the standard deviation is obtained at each wavenumber. The standard deviation of the set of three (glucose, sucrose, ethanol) are then averaged (or summed) to obtain the standard deviation file to be used for the Singular Value Decomposition (SVD) fit of the data in step 3 below. Generally the average is preferred. However, the sum would give the most conservative estimate of the goodness of fit, which is also available from the covariance matrix in the SVD technique.

3. Make up one mixture of known composition and use this as a "mock unknown" or calibration sample mixture. Measure its spectrum and then analyze it according to [6] and the SVD technique over a range of wavenumbers covering all significant features of the absorbance spectra of the previous separate knowns used in 1 and 2. For example, the range 800 cm-l to 1500 cm-1 could be a good wide starting choice. The SVD technique provides a statistical measure of the goodness of fit in two ways.
First, it provides the Chi-Square statistic and second, it provides a standard deviation for each determined quantity. In this case, we are determining the mole fractions of water, sucrose, glucose, and ethanol, so each of these would have a stAn~Ard deviation associated with it which are compared to the actual values. The range of wavenumbers used is reduced and/or moved in the spectrum to find the range of wavenumbers that provide the lowest Chi-Square statistic and lowest standard deviations. In the example, this could be the range 994 cm-1 to 1100 cm-1. The exact range will differ from instrument to instrument and cell to cell, but the technique will find the best range for the particular combination one has available. The computer can be Wo95/l1485 21 7~64 1 rcT~s93llol7o '~

programmed to move the range randomly or systematically.
The method will always find the best range of those examined based on the Chi-square value and standard deviations. The range can be varied until a desired degree of precision is achieved.

4. one can now measure true unknowns. The measurement is made normally and the first analysis is made using the k's determined form knowns from step 1 that are close to the anticipated concentrations. The stAn~rd deviations are also used from the same files that provide the k's.
When the answers are obtained from the SVD technique, they are examined to see if there is a set of k's determined from any of the non-solvent unknowns (sucrose, glucose, ethanol) at a concentration closer to the answer just found than were those used in determining the answer. If a set of k's from a closer concentration is available, it is used and the concentrations recalculated. This procedure is repeated until the closest k's are used. It is preferred to refine this procedure by interpolating between k's such that the final answer is as close to the concentration corresponding to the k's as one would like it to be, e.g., within 1% or so. For mixtures of gases, equations [7] and [8] are replaced by the simpler expression of equation [2].

Use of the Modified Chi-Square Technique The modified Chi-Square technique of this invention is ideal for on-line monitoring and control of a chemical process. Any suitable computer using, for example, an 80826 microprocessor and equipped with analog to digital (A/D) and digital to analog (D/A) interfaces, may be use to implement this modified Chi-Square technique. The computer is attached to both the D/A and A/D
boards via a parallel port (D/A) and a serial port (A/D). A
second serial port is att~che~ to the FT-IR instrument.

--WO95/11485 ~- PCI`/US93/10170 A chemical stream from the process of interest is constantly flowed through the cell in the FT-IR instrument. For example, this could be the exhaust of a combustion device, the broth in a fermenter, water being discharged from a factory, a chemical process, etc. The same process is also monitored with various probes to measure temperature, pressure, flow, pH, dissolved oAyyen~ humidity, density, weight, etc.
The probes monitoring the physical parameters of the process are connected to the A/D board, which has an on board microprocessor to act as storage and shipping center of the information to the host 80826 microprocessor. The host computer need not use an 80826 microprocessor and, in fact, all of the hardware and software is designed to be portable to any computer environment. The A/D board collects the probe information until it is polled by the host computer. The FT-IR instrument is also commanded by the host computer. In this manner, the host computer can regulate the flow of both chemical and physical input information and create files and displays of all of these parameters as a function of time.
The host computer also contains outputs through the D/A board connected to the parallel port that allows the chemical and physical information to be used to control switches and valves of the control instrumentation for the process. The process of this invention performs the integrated tasks of monitoring and control based on measurements of chemical concentrations as well as physical criteria. Until now, processes that use chemical composition criteria have been limited to those which use electrodes (which are relatively inaccurate and reguire frequent replacement) or to those which use mass spectrometry or combination gas chromatography and mass spectrometry. These later cases are slower in response and more costly and frequently require either removal or pretreatment of the sample. Using FT-IR
instruments, the sample essentially never leaves the system being monitored, the sample requires no pretreatment, and the answers are as rapid as the scans. Including analysis processing, the answer is available in seconds compared to minutes or hours for the other instrument techniques and much more accurately and for a wo g5nl48s 2 1 7 4 6 4 1 rcT/usg3/l0l70 ~

much wider range of chemicals when compared to the very limited electrode technique.

EXAMPLE II

The modified Chi-Square technique of this invention has also been used in connection with a process for producing ethanol from a cellulosic material. Intermediate products are glucose and xylose, and valuable byproducts are silica or silicate compounds.
There are several stages in this process where the modified Chi-Square technique of this invention is utilized, and each will now be discussed briefly.

Acid Hydrolysis In the acid hydrolysis portion of the process, concentrated sulfuric acid is mixed with a biomass comprising predominantly cellulosic and hemicellulosic material to produce a liquid contAin;ng approximately 30% sulfuric acid and 15% sugars and a solid cake which contains predominantly lignin and inorganic materials with small quantities of unreacted cellulose and hemicellulose. The sugars are a mixture of glucose,-xylose, other C5 and C6 sugars, and to a much lesser degree cellobiose and other short glucose polymers. If the hydrolysis is carried on too long, the glucose and xylose are degraded. The objective of the process is to produce as much glucose and xylose as possible without degradation. It has been empirically found that the best way to do this is to perform the hydrolysis twice, stopping the first time before the sugars begin to be degraded, pressing the cake to remove the sugar and acid solution and then performing the entire process a second time. It is also important for commercial viability that the sugars be at least 15% concentration by weight.
In accordance with this invention, the process is monitored continuously to determine when to stop the reaction at the peak of sugar production in either the first or second hydrolysis. A small sample stream is mixed with sodium hydroxide to make sugars and sodium sulfate rather than having sugar and concentrated sulfuric 21 7~641 --WOgS/11485 PcT~ss3llol7o acid. This is necessary if a stainless steel spectrophotometer cell is used with an ATR crystal made of ZnSe because long term contact with sulfuric acid will attack the metal and crystal. If the cell is made out of Alloy 20, Carpenter 20, Hastalloy or tantalum steel and a silicon crystal the neutralization with sodium hydroxide is not needed. In either case the modified Chi-Square technique of this invention is used to measure sulfate, glucose and xylose simultaneously. Other sugars such as arabinose and xylitol can also be measured when present. The degradation products such as glucaric acid and xylaric acid can also be measured when present.
The number and kind of sugars and acids are identified for a given raw material mixture the first time it is hydrolyzed using either gas chromatography or liquid chromatography. Once this identification is made, the FT-IR instrument is calibrated according to the modified Chi-Square technique of this invention and then can be used to quickly and accurately measure the composition of the hydrolysis liquid. The results are used to modify the conditions of reaction and the further processing of the sugars.

Sugar Separation After the sugar and acid liquid is pressed from the cake, the acid and sugar are separated in a strong acid polystyrene resin bed. In order to make sure that the streams are separated properly and that the concentration of sugar is maintA;neA at least at 15%, it is important that the output at various places along the resin bed be measured. The acid, glucose, xylose and other C5 and C6 sugars are measured using the modified Chi-Square technique of this invention.

Silica Gel and Silicate Production -The cake left after hydrolysis may contain silica when the agricultural raw materials contain silica. High levels of silica make the cake unsuitable for use as a combustion fuel because WO95/11485 PCT~S93/10170 silica causes both operations and pollution problems. Since silica has a considerable value when converted to sodium silicate, potassium silicate or silica gel, the process removes the silica from the cake and purifies it. To achieve this, the cake is treated with a solution of 5-l0% NaOH and then heated for various times (e.g. 1-2 hours). The NaOH solution dissolves the silica as sodium silicate with the general formula (Na20)x(SiO2)y. It is necessary to know how much of this material is present and the ratio of the Na20 to SiO2 portions of the material. This allows the operator to know if the process is complete by comparing the total amount of SiO2 to the amount in the initial agricultural material corrected for other changes up to this point in the the e~s .
The modified Chi-Square technique of this invention can be adapted to determine the amount of (Na20)x(SiO2)y present and the ratio of the Na20 to SiO2 portions of this material by calibrating with solutions that contain various ratios of Na20 and SiO2 derived by using Na2SiO3 as a starting material for the calibration. It is surprising that despite the fact that the real solutions are inky black and the pure silicates are clear, the IR
region of the spectrum enables one, using the modified Chi-Square technique of this invention, to distinguish between the various silicate species. The material from the reacting solution can be placed directly in the cell without modification and the process stopped when all the silica is essentially in solution.
The next stage in the process is to purify the silica solution by adding a small amount of acid to reduce the pH of the NaOH solution to about pH l0 (from above 12). This precipitates virtually all of the SiO2 as silicic acid (H2SiO3) and allows it to be washed and treated to remove colored materials. By measuring the filtrate using the modified Chi-Square technique of this invention, we can judge the completeness of the purification method.
The silica itself can either be dried for sale as silica gel or it can be reacted with either NaOH (to form various commercial mixtures of sodium silicate) or ROH (to form various commercial mixtures of potassium silicate). In all cases the method of the 21 7464l - WO95111485 ~ - PCT~ss3/10170 invention can be used to assess product quality and to control the reactions to produce the desired products.

Fermentation one last area of the process is the fermentation to produce ethanol. The particular problem here is that the sugar solutions contain a mixture of sugars that are not normally fermented together to produce ethanol. For example, the solution can contain glucose and xylose. While glucose ferments easily, xylose ferments only with difficulty and only after the glucose has fermented. It becomes necessary to be able to monitor these sugars, along with the production of ethanol, during the course of the reaction to know both how well it is proceeding (and whether corrective steps need to be taken) and when it is completed.
The modified Chi-Square technique of this invention is used, with the addition of ethanol as the product to be monitored, to virtually continuously (at a desired number of measurements per unit time) measure the concentration of the sugars, ethanol and glycerol (a by-product of ethanol fermentation).

DESCRIPTION OF THE DRAWING

The preferred embodiment of this invention, illustrating all its features, will now be discussed in detail. This embodiment depicts the novel and non-obvious method of this invention depicted in the accompanying drawing, which is for illustrative purposes only. This drawing includes the following figures (Fig.), with like numerals indicating like parts:
Fig. l is a schematic diagram of a chemical process employing the monitoring and control method of this invention.
Figs. 2A through 2E is the absorption spectra for a 0.25 weight percent aqueous fructose solution taken over a range of ~ wavenumbers beL~/~en 800 and 2800.
Figs. 3A through 3G is the absorption spectra for a 0.50 weight percent aqueous glucose solution taken over a range of wavenumbers between 800 and 4400.

wog~/ll485 2 1 7 4 6 4 1 PCT~Ss3/10170 Figs. 4A through 4H is the absorption spectra for a 3.0 weight percent aqueous sucrose solution taken over a range of wavenumbers between 800 and 4400.
Fig. 5 is the absorption spectra of an aqueous solution containing known concentrations of sucrose (2.5%), glucose (0.59%), and fructose (0.30%), a calibration mixture sample, taken over a range of wavenumbers between 900 and 1150 using k values from data files 3% sucrose/ 0.5% glucose/ 0.25% fructose over the scan range 1114-1032. The solid line shows actual data points, the dotted line shows calculated data points.
Fig. 6 is the absorption spectra of an aqueous solution containing known concentrations of sucrose (2.5%), glucose (0.59%), and fructose (0.30%), a calibration mixture sample, taken over a range of wavenumbers between 900 and 1150 using k values from data files 3% sucrose/ 0.5% glucose/ 0.25% fructose over the scan range 1114-902. The solid line shows actual absorbances points, the dotted line shows calculated absorbance points.
Fig. 7 is the absorption spectra of an aqueous solution containing unknown concentrations of sucrose, glucose, and fructose taken over a range of wavenumbers between 1000 and 1160.
According to the method of this invention, the concentrations were determined to be sucrose 1.99%, glucose 0.48%, and fructose 0.36%.
The solid line shows actual absorbance points from the unknown measured spectrum, the dotted line shows calculated absorbance points using the values of concentration determined by the method of the invention.
Fig. 8 is a schematic diagram of an acid hydrolysis process employing the monitoring and control method of this invention depicted in Fig. 1.
Fig. 9 is a schematic diagram of a chromatographic separation process to separate sugar and acid, employing the monitoring and control method of this invention depicted in Fig. 1.
Fig. 10 is a schematic diagram of a fermentation process employing the monitoring and control method of this invention depicted in Fig. 1.

~ WO95/11~5 2 1 7 4 6 4 l pcT~ss3llol7o Fig. 11 is a schematic diagram of a process for making silica gel employing the monitoring and control method of this invention depicted in Fig. 1.
Fig. 12 is the absorption spectra for an aqueous solution of 2 weight percent glucose, 1 weight percent xylose, and 3 weight percent sodium sulfate taken over a range of wavenumbers between 900 and 1500.
Fig. 13 is the absorption spectra for an aqueous solution of 2 weight percent glucose taken over a range of ~avenumbers between 900 and 1500.
Fig. 14 is the absorption spectra for an aqueous solution of 1 weight percent xylose taken over a range of wavenumbers between 900 and 1500.
Fig. 15 is the absorption spectra for an aqueous solution of 3 weight percent sodium silicate taken over a range of wavenumbers between 900 and 1500.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

GENER~L

Fig. 1 illustrates a typical biochemical process wherein sucrose is added periodically to sustain the growth of living plant cells in a reactor 11. The sucrose inverts to glucose and fructose and some is consumed by the cells which are undergoing transformation in the process. To maximize the yield of the desired product of the process, in this case cell mass and metabolic byproducts, the temperature, pressure and pH of the reaction must be carefully regulated based on the concentration of the sucrose, glucose, and fructose.
A conventional FT-IR instrument 10 (Perkin Elmer Model 1640) is used to obtain absorption spectra on a sample stream from the ~ process. Samples are passed through the transmission cell (not shown) of the FT-IR instrument continuously, with readings taken ~ every 5 to 10 minutes over a period of 72 hours. More frequent readings may be made as required in order to optimize the physical wogs/ll485 2 1 7 4 6 4 1 PCT~Ss3/10170 parameters of the process. Conventional monitoring probes 12 are used to monitor the physical parameters of the process and a conventional general purpose computer 14 stores data and provides control functions as required based on the concentration of the 5 reaction ingredients. An output 16 from the FT-IR instrument is connected through an RS-232 port 18 to memory of the computer 14 and the data collected by the FT-IR instrument lO is stored in the computer's me ry. According to the method of this invention, data files are created which enable the computer 14, based on measurements being made as the process proceeds, to analyze the data and determine the concentration of reactants accurately and rapidly. Based on this determination, the process parameters are adjusted to optimize the process. As discussed above, a digital to analog converter (D/A) 20 is coupled between 15 the computer and a controller 22 which opens and closes a shut off valve 24 for the process. If the sucrose, fructose, and glucose drop below predetermined ranges, more sucrose is added to the system. If they remain high, either pH or temperature is adjusted to increase the rate of cell growth. If unwanted new chemicals 20 are observed as measured by a sudden increase in Chi-square value, the operator is warned. If either the concentrations or physical parameters fall outside dangerous limits for cell viability, the system can be automatic~lly shut down. Temperature, pressure, pH, flow rate are monitored by the conventional sensors or probes 12 25 and the electronic output measurements are feed to an analog to a digital A/D converter 26 connected between the computer l4 and these probes as discussed above. The output of the A/D converter 26 is connected through an RS-232 port 29. Thus, both the concentration of reactants, and if desired the concentration of 30 the reaction products, are monitored in real time along with the physical parameters of the process to control the process to optimize it. Hitherto this has never been achieved.
Consequently, conditions may now be controlled precisely to optimize the process. This has been achieved using this invention 35 for a wide variety of plant and yeast cells.

CREATION OF DATA FILES

In accordance with this invention, calibration samples are prepared before starting the process and data files are created and stored in the memory of the computer 14. The first step of the method of this invention is to prepare a number of cAl;hration samples spAnn;ng the concentration range of interest. Ten samples at each concentration were thus prepared. Calibration samples of aqueous solutions of 0.5, 1.0, 3, 5, and 7.0 weight percent sucrose, 0.5, 1, 2, and 4 weight percent glucose, and 0.25, 1, 2, and 4 weight percent fructose were initially prepared and spectra obtA; n~, The electromagnetic absorption of these calibration samples at a selected number of wavelengths (wavenumbers) over a predetermined range were measured to obtain a spectra for each sample, k values and stAn~Ard deviations thereof were calculated at each ~dvellumber, and data files consisting of these values were created and stored in the memory of the computer.

Using one of the following equations:

k = A [9]
c (for gases) or k = A - k olvent (1.0 - C) [10]
c (for solvent system -liquids or solids) -A is the absorbance measurement of each individual calibration sample, and WO 95/11485 PCT/US93/10170 ~

c is the concentration in molar units of the ingredient in the calibration sample, k~olvent is the absorbance value for pure solvent, for example, water, an average k value for each calibration sample is calculated at each of the selected number of different wavelengths over the predetermined range of wavelengths selected. The standard deviation value S of k values were determined according to the following formula.

m S = 1 ~ (ki ~ k)2 1/2 m-l i=l where ki are the m individual values at each wavenumber, k is the average k at each wavenumber, and m is the number of replicates performed at each wavenumber (10 for example).
Figs. 3A-3G is the spectra of the 0.5 weight % glucose-water solution and Table I (Exhibit A) is illustrative of the date file for this solution. Figs. 2A-2E is the spectra of the 0.25 weight % frutose-water solution and Table II (Exhibit B) is illustrative of the date file for this solution. Figs. 4A-4H is the spectra of the 3.0% sucrose-water solution and Table III
(Exhibit C) is illustrative of the data file for this solution.
In Tables I, II and III, the column designated # presents the wavenumber; the columns designated A through J presents the k values for the ten replicated calibration samples at each wavenumber in the column designated #; the column designated AVG
presents the average of the ten k values at each wavenumber; and the column designated STD presents the standard deviation of the Wos~/11485 P~ s~/10170 average of the ten k values at each wavenumber. Because ten replicated calibration samples were taken for each calibration sample, statistically significant results are obtained. For greater accuracy a greater number of replicas are required. For most purposes twenty replicas are sufficient. The data in the Tables I, II and III is stored as data files in the memory of the computer 14.
After data files for individual reactants have been created, data files for the mixture of reactants is created:
First, a sample mixture of the reactants at known concentrations is prepared and then the electromagnetic absorption of the sample mixture is measured at each of a selected number of different wavelengths over a range of wavelengths of the electromagnetic spectrum anticipated to be best representative of the absorption characteristics of the sample mixture based on collected data from the individual components.
Second, it is determined which wavelengths within the range (scan range) of wavelengths of the electromagnetic spectrum shall provide a solution to the following equations to an acceptable level of precision. This is accomplished by solving the following equations to determine the respective concentrations of the ingredients in the calibration sample mixtures using (i) an arbitrarily selected number of wavelengths within the range of wavelengths, (ii) the lowest stAn~Ard deviation among the average k values as determined from the data files for the individual calibration samples as set forth in Tables I, II, and III, and (iii) the singular value decomposition mathematical technique to determine which of the arbitrarily selected number of wavelengths (scan range) provide the lowest chi-squared statistic between calculated and known values of the concentration of ingredients in the CAl; hration sample mixtures.

Al = kllcl + kl2c2 + kl3C3 + klnCn - A2 = k2lcl + k22c2 + k23c3.. ...k2ncn A3 = k3lcl + k32c2 + k33c3.. ...k3ncn An = klcl + k2c2 + k3C3..... kncn Am = kmlcl + km2c2 + km3C3.. ...kmncn woss/11485 2 1 7 4 6 4 1 PCT~S93/10170 where Al, A2, A3....An are the values of the absorbance measurements at said arbitrarily selected wavelengths, kml, km2~ km3....kmn are the average k values from the data files for the individual calibration samples which most closely correspond to the k values for the concentration of ingredients in the c~l;hration sample mixtures (the first subscript is the w_~enumber and the second matches the subscript for the unknown chemical species being determined), and cl, c2, c3....cn are the concentrations (either known or unknown) expressed in molar units, of the ingredients in the sample mixtures.
Figs. 5 and 6 are illustrative of these later steps of the method. Fig. 5 is the spectra over a wavenumber range between 900 and 1150 for a calibration sample mixture of 2.5% sucrose, 0.59%
glucose, and 0.30% fructose. Concentrations of ingredients in the calibration samples were determined using k values from the data files of 3% sucrose, 0.5% glucose, and 0.25% fructose aqueous solutions over a scan range of 1114-1032 wavenumber. Fig. 6 is similar to Fig. 5 except for the very important difference that the scan range was 1114-902 wavenumber rather than 1114-1032 wavenumber. Calculated results (dotted line) and actual reading (solid line) from the FT-IR instrument are displayed together. In Fig. 5 the calculated values match more closely to the actual reading than in Fig. 6. Therefore, the scan range of 1114-1032 wavenumber is used to determine unknown concentrations in a sample stream from the process.

5 wogs/ll48~ 2 1 7 4 6 4 1 PCT~ss3/10170 ON-LINE MONllO~ING OF CHEMICAL PROCESS

Using the data files created for the calibration samples and calibration mixture samples, the reactants and products (sucrose, glucose, and fructose) of chemical process shown in Fig. 1 are monitored by measuring the spectra of a sample stream from the process. This spectra is shown in Fig. 7. First, the chemical process is continually monitored to collect individual samples in which the concentration of ingredients is unknown and the electromagnetic absorption of each individual samples is measured over the scan range which provided the lowest chi-squared statistic beL~Iocn calculated and known values of the concentration of ingredients in the c~l;hration sample mixtures.
Second, the following equations are solved in accordance with singular value decomposition mathematical technique to determine the respective unknown concentrations of the ingredients in the test samples using the average k values at the wavenumbers determined above. Specifically, the following equations are solved at 1114-1032 wavenumbers.

Al = kllcl + kl2c2 + kl3c3....klncn A2 = k21cl + k22c2 + k23c3....k2ncn -A3 = k31c1 + k32c2 + k33c3....k3nCn .

Am = kmlcl + km2c2 + km3c3 + --- + kmnCn where A1, A2, A3....An are the values of the absorbance measurements of the test samples (the abscissa of Fig. 7), km1, km2, km3... ~kmn are the k values from Tables I, II, and III, and WO 95/1148!; PCI~/US93110170 c1, c2, c3....Cn are the concentrations expressed in molar units of the unknown ingredients in the test samples, in this case, four values representing sucrose, glucose, fructose, and water.

Forth, the second step is repeated using k values which corresponds most closely to the k value for the concentration of the unknown ingredient as determined in the second step.
Fifth, using the concentration of ingredients as determined in the forth step calculate the absorption of the test sample and compare the calculated absorption with the actual measured absorption. The forth and fifth steps are repeated until the values of k used in determining the unknown concentrations of ingredients in the test samples provide the statistically best results. Specifically, repetition is mandated so that the results obtained in repeated calculations of the unknown concentrations of ingredients in the test samples have a percentage deviation of less than about 1 percent. When this is achieved the concentration of unknown ingredients has been determined with the desired accuracy.
Table IV presents two sets of calculated values for an unknown mixture using the k values for two different-mixtures:

-wog5n1485 rcT~ss3llol7o TABLE IV
(Unknown Sucrose, Glucose, Fructose Solution) M1~1UK~ l M1X1UK~ 2 Using k file Using k file 3%Suc, 0.5%Glu,.25%Fru 1%Suc, 0.5%Glu,.25%Fru Mass % Mole % Mass % Mole %

Sucrose l.99 0.108 2.00 0.108 15Glucose 0.48 0.049 0.45 0.046 Fructose 0.36 0.037 0.35 0.036 Chi-Square Value 2.3 4.5 In TABLE IV, the I~ UK~ 1 comprises 3.0 weight % sucrose, 0.5 weight % glucose, and 0.25 eight % fructose, and MIXTURE 2 comprises 3.0 weight % sucrose, 0.5 weight % glucose, and 0.25%
fructose. The results are expressed as mole percent (as mole fractions are used in the calculations) as well as in the more usual engineering units of mass percent. The method of the invention obt~ine~ the results using the k files corresponding to mixture two with a Chi-square value of 4.5 and the results using the k files corresponding to mixture l with a Chi-square value of 2.5. These were the last two iterations zeroing in on the best values for the unknown concentrations at the desired level of precision (better than 0.1%). The results are taken corresponding to the answers given by the k files associated with mixture l because the Chi-square value is lowest.

WO95/11485 21 74641 pcT~ss3llol7o `~

ETHANOL PRODUCTION PROCESS

As depicted in Figs. 8 through 13, the monitoring method of this invention is used to maximize the yield of the desired products of ethanol and silica. As illustrated in Fig. 8, first a biomass comprising from 50-85 weight percent cellulose and hemi-cellulose is treated with concentrated sulfuric acid. The biomass is first treated with 70% acid to decrystallize the biomass, and then the mixture is reduced to 30% acid for the hydrolysis. The ratio of cellulose to hemi-cellulose ranges between 2:1 and 100:1.
The concentrated sulfuric acid is mixed with the biomass water added to reduce the concentration to 30% acid by weight and heated to a temperature ranging from 95 to 105 for 45 minutes to 1 hour to yield approximately 15 weight percent sugars comprising about the maximum of 20% glucose and 10% xylose and 10 % other combined sugars. A small sample stream of the reaction product is forwarded to a spectrometer to conduct the monitoring in accordance with this invention. A typical spectra for mixture of glucose a, xylose, and sodium sulfate is depicted in Fig 15.
As long as the percentage of glucose and xylose is increasing in the product stream, the reaction is maintained.-Once the yield of glucose and xylose begins to drop, the reaction is stopped. The biomass is washed and then again treated with a solution of acid to make the concentration 30% by weight sulfuric acid. Typical spectra for glucose and xylose are depicted, respectively, in Figures 13 & 14.
As depicted in Fig. 9, the products of the acid hydrolysis reaction are then separated from each other in a chromographic separation column, for example, a suitable ion exclusion resin provided by the DOw Chemical Company may be used to conduct this separation. Such a column results in a 98% separation of the sugars from the sulfuric acid. The acid runs through the column substantially more rapidly than the sugars. The column is equipped with a suitable detector that is located along the column to detect when the percentage of sugar in the stream from the column - WOgS/11485 PCT~S93/10170 exceeds about 2%. At that point a valve is manipulated so that the sugar stream is diverted.
The sugars, as illustrated in Figure 10, are then fed to a fermenter which uses yeast or bacteria in the presence of nutrients and buffering agents to convert the sugars to ethanol and glycerol primarily. The buffering agents and nutrients are added before the microorganisms to maintain the optimal environment for these microorganisms, For example, the desired pH
for yeast is about 4.5 and nutrients such as N, P, R, Mg and S Are added. Again the objective is to maximize the production of ethanol. Process conditions are monitored and controlled based upon this ethanol production. When the yield begins to decrease, conditions are adjusted (e.g.pH, dissolved oxygen) to maintain ethanol production at maximum.
As illustrated in Figure 11, the biomass cake remaining after the acid hydrolysis is then treated with sodium hydroxide to produce sodium silicate. The objective is to maximize the yield of silica or silicates. Typical spectra for sodium silicate is depicted in Fig 15.
SCOPE OF THE INVENTION

The above presents a description of the best mode contemplated of carrying out the present invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains to make and use this invention. This invention is, however, susceptible to modifications and alternate constructions from that discussed above which are fully equivalent. (For the purposes of this invention absorbance and transmission and wavelength and wavenumbers are equivalent terms, and light and electromagnetic radiation are equivalent terms.) Consequently, it is not the intention to limit this invention to the particular embodiment disclosed. On the contrary, the intention is to cover all modifications and alternate constructions coming within the spirit and scope of the invention as generally expressed by the following claims:

Claims (8)

CLAIMS:

1. using a general purpose computer having memory and an electromagnetic spectrum analysis instrument, a method of determining the concentrations of a plurality of different ingredients present over a concentration range in a chemical process, comprising (I) first creating data files in the computer memory by (a) preparing a number of calibration samples at different concentrations spanning said concentration range for each individual ingredient being monitored, (b) measuring the electromagnetic absorption of the calibration samples at a selected number of different wavelengths over a predetermined range of wavelengths of the electromagnetic spectrum and storing the measurements in a first data file in the memory of the computer, (c) repeating steps (a) and (b) a sufficient number of times to obtain statistically significant data composed of these absorbance measurements for the known concentrations of each of the ingredients and storing said data in a second data file in the memory of the computer, (d) using the computer, the data in said first and second data files, and the following equations (for gases) or (for solvent system -liquids or solids) where A is the absorbance measurement of each individual calibration sample, and c is the concentration in molar units of the ingredient in the calibration sample, ksolvent (for liquids and solids) is the absorbance value of the component designated as a solvent in which the other components are distributed measured in its pure form, calculating for each calibration sample an average k value at each of said selected number of different wavelengths over said predetermined range of wavelengths, and a standard deviation value thereof, and storing said calculated k values and standard deviation values thereof in a third data file in the memory of the computer, (e) preparing a plurality of calibration sample mixtures of the ingredients at known concentrations and measuring the electromagnetic absorption of the calibration sample mixtures at each wavelength within said range of wavelengths of the electromagnetic spectrum, (f) using the computer and the data in said third data file determining which wavelength within said range of wavelengths of the electromagnetic spectrum shall provide a solution to the following equations to an acceptable level of precision by solving said following equations to determine the respective concentrations of the ingredients in the calibration sample mixtures using (i) an arbitrarily selected number of wavelengths within said range of wavelengths, (ii) the lowest standard deviation among the average k values as determined in step (d), and (iii) the singular value decomposition mathematical technique to determine which of the arbitrarily selected number of wavelengths provide the lowest chi-squared statistic between calculated and known values of the concentration of ingredients in the calibration sample mixtures A1 = k11c1 + k12c2 + k13c3....k1ncn A2 = k21c1 + k22c2 + k23c3....k2ncn A3 = k31c1 + k32c2 + k33c3....k3ncn .
.
Am = km1c1 + km2c2 + km3c3 + .... + kmncn where A1, A2, A3....An are the values of the absorbance measurements at said arbitrarily selected wavelengths, km1, km2, km3....kn are the average k values from step (d) which most closely correspond to the k values for the concentration of ingredients in the calibration sample mixtures for each wavenumber or wavelength, and c1, c2, c3....cn are the concentrations(either known or unknown) expressed in molar units, of the ingredients in the sample mixtures, (II) second conducting on-line monitoring of the chemical process by (i) continually sampling the chemical process to collect individual samples in which the concentration of ingredients is unknown and measuring the electromagnetic absorption of said individual samples at the arbitrarily selected number of wavelengths which provide the lowest chi-squared statistic between calculated and known values of the concentration of ingredients in the calibration sample mixtures as determined in step (f), (j) using the computer solving the following equations using the singular value decomposition mathematical technique to determine the respective unknown concentrations of the ingredients in the samples taken in step (i) using the average k values at the wave lengths determined in step (f) A1 = k11c1 + k12c2 + k13c3....k1ncn A2 = k21c1 + k22c2 + k23c3....k2ncn A3 = k31c1 + k32c2 + k33c3....k3ncn .
.
Am = km1c1 + km2c2 + km3c3 + .... + kmncn where A1, A2, A3....An are the values of the absorbance measurements taken in step (i), km1, km2, km3..., kn are the k values from step (d) at each wavenumber or wavelength n, and c1, c2, c3....cn are the concentrations expressed in molar units of the unknown ingredients in the samples, (k) repeating step (j) using k values which corresponds most closely to the k value for the concentration of the unknown ingredient as determined in step (j), (l) using the concentration of ingredients as determined in step (k), calculating using the computer absorption of the unknown sample and comparing said calculated absorption with the actual measured absorption, and (m) repeating steps (k) and (l) until the statistically best values of k used in determining the concentrations of unknown ingredients so that the results obtained in repeated calculations of the unknown concentrations of ingredients in the samples have a percentage deviation of less than about 1 percent.

2. The method of Claim 1 where the chemical process produces sugars from a cellulosic material, with the sugars being fermented to ethanol.

3. Based on real time measurements, an on-line method of controlling a chemical process producing an effluent stream in which a plurality of ingredients are present at different concentrations over a concentration range depending on predetermined variable conditions occurring in the process, said method comprising monitoring the concentrations of the ingredients present in said effluent stream, and altering said process conditions as determined by absorption data of the effluent stream taken with an electromagnetic absorption instrument and using a general purpose computer having memory to adjust control devices that regulate said process conditions to produce a desired concentration of ingredients in the effluent stream, said absorption data reflecting essentially accurately the concentration of the ingredients in the effluent stream as determined by a method, comprising (I) first creating data files in the computer memory by (a) preparing a number of calibration samples at different concentrations spanning said concentration range for each individual ingredient being monitored, (b) measuring the electromagnetic absorption of the calibration samples at a selected number of different wavelengths over a predetermined range of wavelengths of the electromagnetic spectrum and storing the measurements in a first data file in the memory of the computer, (c) repeating steps (a) and (b) a sufficient number of times to obtain statistically significant data composed of these absorbance measurements for the known concentrations of each of the ingredients and storing said data in a second data file in the memory of the computer, (d) using the computer, the data in said first and second data files, and the following equations (for gases) or (for solvent system -liquids or solids) where A is the absorbance measurement of each individual calibration sample, and c is the concentration in molar units of the ingredient in the calibration sample, ksolvent (for liquids and solids) is the absorbance value of the component designated as a solvent in which the other components are distributed measured in its pure form, calculating for each calibration sample an average k value at each of said selected number of different wavelengths over said predetermined range of wavelengths, and a standard deviation value thereof, and storing said calculated k values and standard deviation values thereof in a third data file in the memory of the computer, (e) preparing a plurality of calibration sample mixtures of the ingredients at known concentrations and measuring the electromagnetic absorption of the calibration sample mixtures at each wavelength within said range of wavelengths of the electromagnetic spectrum, (f) using the computer and the data in said third data file determining which wavelength within said range of wavelengths of the electromagnetic spectrum shall provide a solution to the following equations to an acceptable level of precision by solving said following equations to determine the respective concentrations of the ingredients in the calibration sample mixtures using (i) an arbitrarily selected number of wavelengths within said range of wavelengths, (ii) the lowest standard deviation among the average k values as determined in step (d), and (iii) the singular value decomposition mathematical technique to determine which of the arbitrarily selected number of wavelengths provide the lowest chi-squared statistic between calculated and known values of the concentration of ingredients in the calibration sample mixtures A1 = k11c1 + k12c2 + k13c3....k1ncn A2 = k21c1 + k22c2 + k23c3....k2ncn A3 = k31c1 + k32c2 + k33c3....k3ncn Am = km1c1 + km2c2 + km3c3 + .... + kmncn where A1, A2, A3....An are the values of the absorbance measurements at said arbitrarily selected wavelengths, km1, km2, km3....kn are the average k values from step (d) which most closely correspond to the k values for the concentration of ingredients in the calibration sample mixtures for each wavenumber or wavelength, and c1, c2, c3....cn are the concentrations (either known or unknown) expressed in molar units, of the ingredients in the sample mixtures, (II) second conducting on-line monitoring of the chemical process by (i) continually sampling the chemical process to collect individual samples in which the concentration of ingredients is unknown and measuring the electromagnetic absorption of said individual samples at the arbitrarily selected number of wavelengths which provide the lowest chi-squared statistic between calculated and known values of the concentration of ingredients in the calibration sample mixtures as determined in step (f), (j) using the computer solving the following equations using the singular value decomposition mathematical technique to determine the respective unknown concentrations of the ingredients in the samples taken in step (i) using the average k values at the wave lengths determined in step (f) A1 = k11c1 + k12c2 + k13c3....k1ncn A2 = k21c1 + k22c2 + k23c3....k2ncn A3 = k31c1 + k32c2 + k33c3....k3ncn Am = km1c1 + km2c2 + km3c3 + .... + kmncn where A1, A2, A3....An are the values of the absorbance measurements taken in step (i), km1, km2, km3..., kn are the k values from step (d) at each wavenumber or wavelength n, and c1, c2, c3....Cn are the concentrations expressed in molar units of the unknown ingredients in the samples, (k) repeating step (j) using k values which corresponds most closely to the k value for the concentration of the unknown ingredient as determined in step (j), (1) using the concentration of ingredients as determined in step (k), calculating using the computer absorption of the unknown sample and comparing said calculated absorption with the actual measured absorption, and (m) repeating steps (k) and (1) until the statistically best values of k used in determining the concentrations of unknown ingredients so that the results obtained in repeated calculations of the unknown concentrations of ingredients in the samples have a percentage deviation of less than about 1 percent.

4. The method of Claim 3 where the chemical process produces an effluent stream including glucose and xylose.

5. An on-line method of monitoring and controlling a chemical process having physical parameters and where the concentration of process components is measured, comprising (a) measuring the concentration of process components in samples from the process using a spectrometric instrument to obtain spectral data characteristic of the process components, (b) analyzing the spectral data using a chi-squared mathematical technique to determine the unknown concentration of process components in said samples, and (c) monitoring the physical parameters of the process and altering said physical parameters based on the determination of concentration of process components in step (b) as required to optimize the process.

6. The method of Claim 5 where the process produces ethanol from sugars, and includes monitoring the physical parameters of the process and altering said physical parameters based on the determination of concentration of ethanol and sugars in step (b) as required to maximize the yield of ethanol.

7. The method of Claim 5 where the process produces silica or silicates, and includes monitoring the physical parameters of the process and altering said physical parameters based on the determination of concentration of silica or silicates in step (b) as required to maximize the yield of silica or silicates.

8. The method of Claim 5 where the process produces glucose and xylose, and includes monitoring the physical parameters of the process and altering said physical parameters based on the determination of concentration of glucose and xylose in step (b) as required to maximize the yield of glucose and xylose.