GB2244131A

GB2244131A - Spectra based regression analysis of reactants

Info

Publication number: GB2244131A
Application number: GB9110468A
Authority: GB
Inventors: Harold C Brashears; Junior Arthur Katzakian
Original assignee: Aerojet General Corp
Current assignee: Aerojet Rocketdyne Inc
Priority date: 1990-05-18
Filing date: 1991-05-15
Publication date: 1991-11-20
Anticipated expiration: 2011-05-15
Also published as: FR2662251A1; ITRM910339A0; GB2244131B; ITRM910339A1; JPH04231849A; IT1248326B; DE4116027A1; GB9110468D0

Abstract

Properties of a reaction product are predicted from the reactants from which it is formed, by performing spectral analyses on the reactants and performing a factor analysis of quantitative factors derived from spectra. The spectral analyses are performed on the reactants in the mixture in which they will be reacted, and the reaction is generally one which involves the chemical conversion of two or more species into a single, chemically distinct product, optionally in the presence of other, non-reacting components. The factor analysis is based on information derived from known mixtures which upon reaction produce products of known properties, and yet show sufficient variation among themselves to adequately predict the type and extent of variation to be detected in the test sample. Examples are given for predicting properties of rocket motor propellants.

Description

:2 -1 -9 1 3 J_ 1 PRODUCT QUALITY PREDICTION BY SPECTRA-BASED FACTOR

ANALYSIS OF REACTANTS This invention lies in the field of chemical processing, with special focus in the area of quality control by factor analysis.

BACKGROUND AND SUMMARY OF THE INVENTION

In many types of chemical processes, one can determine in advance whether a reaction product will meet specifications by monitoring the raw materials. As process engineers well know, this type of advance quality control can result in considerable savings in down time, energy usage, and raw material cost.

Reliable and meaningful-information are needed, however, regardless of whether it is the raw materials or the product being sampled. Contributing factors include how representative the sample is, whether or not it undergoes physical or chemical changes between sampling and analysis, and how often the samples are taken. The information must also be useful it must provide a fair indication of the property of interest and yet be obtainable from a small sample, and the analysis must be fast enough to permit remedial action before much of the reaction takes place. Predictions based on raw material analysis have the additional requirement that the property of interest must be one which can indeed be predicted without first forming the product.

It has now been discovered that an essentially unlimited variety of properties of a reaction product can be predicted by way of factor analysis on spectral data obtained on the raw material. Factor analysis of spectral information is a known technique, limited up until now to predicting performance characteristics of the material on which the spectra are performed rather than of a reaction 2 product of the material. The new and surprising discovery embodied in this invention is that one can successfully apply this technique to systems where there is a chemical reaction between the material whose properties are being predicted and the material being analyzed.

Literature of potential relevance to this invention includes Lowenhaupt, D.E., U.S. Patent No. 4,370,201, issued January 25, 1983, and Swinkels, D. A., et al., U.S. Patent No. 4,701,838, issued October 20, 1987, and the International Application corresponding to the latter, published under the Patent Cooperation Treaty, Publication No. WO 84/04594, November 22, 1984. Lowenhaupt discloses the use of factor analysis to monitor and control the blen4ing of coal. Swinkels, et al. disclose a control loop for the continuous blending of solid particles of diverse materials in general, utilizing factor analysis to govern the blend. Example 2 of Swinkels, et al. describes factor analyses on coals to predict micrbstrength indices of cokes made from the coals, but this is merely a change in crystal structure rather than a true chemical reaction. Additional literature relating to FTIR factor analysis may be found in the references listed on the cover page of Lowenhaupt and in the International Search Report associated with the Swinkels, et al. International Application.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a plot of predicted vs. measured values of modulus for samples of an Advanced Solid Rocket Propellant, based on FT-IR factor analyses of spectra taken on the propellant premixes in accordance with the invention.

FIG. 2 is a plot of predicted vs. measured values of strain for the same samples.

FIG. 3 is a plot of predicted vs. measured values of modulus for samples of a Peacekeeper Propellant, taken in a manner analogous to that used for FIG. 1.

DETAILED DESCRIPTION OF THE INVENTION

1 3 is AND PREFERRED EMBODIMENTS An analysis in accordance with the present invention is performed on a test mixture, which is either a small representative sample isolated from the reaction mixture or a part of the reaction mixture itself. The term "reaction mixture" is used herein to denote a mixture which includes the as yet unreacted species which upon reaction are converted to the product. The invention also contemplates application to systems which include other components in addition to the reactants, as well as those in which other components are either added subsequent to the analysis or removed (such as solvents, for example).

The test mixture in accordance with the invention may be a single phase, such as a liquid mixture, solution.or emulsion or a mixture of solid particles of different species or composition. Also contemplated are test mixtures which include both liquid and solid phases, such as slurries, suspensions and pastes. The product may also consist of a.single or multiple phases.

The invention further applies to systems where t he test mixture and the product differ in their phase constitution, and is of particular interest with regard to systems where the test mixture contains a liquid phase and the product does not, particularly those where the reactants are liquid and the product is a solid. One such system is the binder mix used in the formation of solid propellants.

The reaction which converts the reactants in the test mixture to the product may be any of a wide variety of chemical reactions, involving twoor more reactants to produce a product distinct in chemical structure rather than one which differs from the reactants simply in terms of its crystalline phase. The reaction may be one which produces a product which is a single species or one which produces a mixture of products. Reactions producing a single- species product from two or more reactant species are of particular interest, with polymerizations being of further interest.

4 Polymerizations of particular interest are copolymerizations involving two or more chemically distinct monomers.

The analysis can be used to predict a wide variety of product qualities, properties, parameters, and characteristic attributes in general, limited only by the requirement that the values of these attributes relate back to the starting materials from which the product is made. The analysis is limited to systems where the chemical reaction itself can be performed in a reproducible, standardized manner which avoids introducing additional variables or factors which will affect the product attributes of interest. The reaction may thus be one in which the product is not affected by variations in the operating conditions, or one in which product variations arising in the reaction can be eliminated by standardizing the operating conditions.

The analysis is performed to detect defects or variations in the test mixture which will result in a product failing to meet specifications in one or more of these attributes. Such defects and variations may include deviations in the proportions of the reactants, as well as deviations in the quality or character of one or more of the.reactants themselves. For example, in cases where a reactant contains multiple functional groups and/or multiple types of functional groups, deviations and variations from one batch to the next which lead to variations in the product may be detected. In cases where a reactant is an oligomer or prepolymer with a molecular weight distribution rather than a single distinct species, deviations in the molecular weight distribution, including the average molecular weight as well as the spread, may be detected. High impurity levels may also be detected, as well as reactants having unusually low or high activity for one reason or another.

The standards used to develop the factors and factor loadings in the factor analysis are reaction mixtures which have first been subjected to the same spectral 1:

analyses used as the test mixture, then converted to the product by undergoing the chemical reaction in the reproducible manner referred to above, and the product then tested or measured to ascertain the values of the attributes in the specification. For best results, the standards will be mixtures of the same chemical species as those of the test mixtures, or at least the same as those believed to constitute, or considered acceptable for, the test mixtures.

The standards will be several such mixtures, varying among themselves in such features as the proportions or quantities of the components, functionality type and content, or any other characteristic which will produce a variation in the product attribute or attributes in question. The standards will bq sufficient in number and the variations over ranges of ' sufficient scope to permit one to assign factor loadings sufficiently reliable and precise to provide accurate predictions of the product characteristics. The greater the number of standards, the better the correlations. The optimum number however will vary with the complexity of the.system. As the number of components in a system increases., the number of standards required to establish a given accuracy in the correlation will also increase. In the most typical systems, the-number of standards used will range from 10 to 60, preferably from 20 to 40.

The attributes themselves may vary widely, and will depend on the product. Propellants and explosives are an example of one class of product. Attributes of interest for this class may include physical properties such as melting or freezing points, density, tensile strength, elongation, and modulus; thermal properties such as differential thermal analyses; performance characteristics such as burn rate, detonation velocity, and combustion temperature; safety characteristics such as impact sensitivity, friction sensitivity, electrostatic sensitivity, detonation sensitivity; and stability characteristics such as thermal stability and vacuum 6 stability. The analysis can be used to predict one such attribute, or two or more simultaneously.

The analyses are spectral analyses, including the various forms of spectroscopy, spectrometry and spectrophotometry. Notable examples are atomic absorption spectroscopy, flame emission spectroscopy, mass spectrometry, infrared spectroscopy, laser Raman spectroscopy, ultraviolet spectroscopy, visible spectroscopy, nuclear magnetic resonance spectroscopy, and electron spin resonance spectroscopy. The appropriate choice will depend on the reaction system, i.e., the chemical and physical nature of the substances and the type of reaction which they will undergo, and the properties being predicted. Preferred spectral methods are infrared spectroscopy, laser Raman spectroscopy, ultraviolet spectroscopy, and nuclear magnetic resonance spectroscopy., with infrared spectroscopy particularly preferred. This includes the various types of infrared spectroscopies, the appropriate choice for any particular system depending on the characteristics of the system itself. Examples are midrange infrared, near-range infrared and diffuse reflectance infrared.

The simultaneous use of an entire range of frequencies is preferred for the advantages which it provides in terms of sensitivity and speed. Thus, techniques involving Fourier transform treatment are preferred, notably Fourier transform infrared (FTIR). Conventional instrumentation for obtaining such spectra may be used.

The factor analysis may be performed according to known techniques, disclosed in the literature and used for the prediction of performance characteristics of such materials as coal, peat moss, mineral ores and preformed propellant mixtures based on analyses of the materials themselves. One of numerous such disclosures in the literature is found in Swinkels, D.A., et al., U.S. Patent No. 4,701,838, referenced above, and its corresponding

7 International Application published under the Patent Cooperation Treaty, Publication No. WO 84/04594, both of which are incorporated herein by reference.

Further in accordance with known techniques, the general procedure will involve deriving the factor loadings for a test mixture from selected features of the spectral analysis of the test mixture, followed by a regression analysis to calculate predicted values for the product C haracteristics. The regression analysis is preferably a linear regression analysis, most preferably a multiple linear regression analysis. The analysis is readily performed by computer, and any of the wide variety of computer systems commercially available with a sufficient capacity for the analysis may be used. Minicomputers are pirticularly well adapted to the analysis. Examples are HP 1000 series (Hewl ett Packard) and VAX systems (Digital Equipment Corporation).

The following examples are presented by way of illustration, and are intended neither to define nor limit the.invention in any manner.

-EXAMPLE 1

A solid rocket motor propellant based on aluminum powder, ammonium perchlorate (AP) and Fe 2 0 3 with a polybutadiene/acrylonitrile copolymer binder was prepared as a calibration set consisting of eighteen small batches of premixes, in which the amounts of AP, Fe 2 0 3 and binder prepolymer were varied from one batch to the next to mimic the typical range of manufacturing errors and variabilities.

FT-IR spectra were taken on each premix batch, and each was then cured to its final form as a solid propellant, for which mechanical and physical properties were then determined. In addition, a validation set was prepared, consisting of six batches of identical composition. For this set as well, spectra were taken on the premixes and 8 properties of the product after cure were determined for each batch.

The FT-IR spectra of all samples were transmitted to a VAX computer, where the spectra corresponding to the calibration set were used to derive factor loadings for a number of factors equal to the number of samples in the calibration set, and the spectra corresponding to the validation set were subjected to factor analysis based on these factor loadings. The result was a series of predicted values of the composition of the premix, as well as the physical and chemical properties of the cured propellant. These predicted values were averaged and compared with the averaged actual values, and are listed in the table below, which shows that the predicted and measured values correlate very closely. The entries in the-table which'.are properties of the post- cure propellant are the density, Shore "All hardness, tensile strength, elongation and modulus.

PROPERTY MEASURED PREDICTED LSBR (in/sec) 0.4115.0.4005 Density (g/mL) 1.767 1.763 Shore "All (Hardness Units) 49 48 Tensile, max. (PSI) 126 123 Elongation, max. (%) 31.4 32.3 Modulus, initial (PSI) 591 575 Prepolymer (weight %) 12.04 12.04 curing agent (weight 1.96 1.95 Fe2 0 3 (weight 0.060 0.063 Al (weight %) 16.00 15.97 AP, 200A (weight 50.44 49.18 AP, 20A (weight 19.50 20.53 AP, total (weight 69.94 69.71 liquid strand burning rate 1 9 EXAMPLE 2 using a procedure similar to that of Example 1, a series of samples of Advanced Solid Rocket Motor Propellant were prepared. Eighteen of these samples varying in their proportions of binder, AP and Fe203 and were thus used as a calibration set. Twenty additional samples, also varying in the proportions of binder, AP and Fe203 were used as test samples to compare predicted with measured values of the modulus and strain of the- product resulting after cure, the prediction based on FT-IR spectra taken prior to cure.

FIG, 1 is a graphical representation of the predicted vs. measured values of the modulus (in PSI), and FIG. 2 is a graphical representation of the predicted vs. measured values of the strain (in PSI.). A line has been drawn in each figure to represent:c-n approximation of the correlation.

EXAMPLE 3..

Using a procedure similar to tat of Example 2, a series of Peacekeeper Propellant ANB-.3600 samples were prepared. Forty-nine unknown samples,of the premix for this propellant were subjected to the analysis, and the comparison of the predicted vs. measu red values for the modulus (in PSI) of the cured propellant is presented graphically in FIG. 3, where a line has been drawn to approximate the correlation. Each point on this graph represents an average of duplicate samples for each premix.

The foregoing is offered primarily for purposes of illustration. It will be readily apparent to those skilled in the art that additional variations, modifications and substitutions may be made in any of the various aspects of the invention without departing from its spirit and scope.

i

Claims

CIAIMS

1. A method of predicting values of selected characteristics of the product of a preselected reaction of a plurality of reactants contained in a test mixture, said method comprising:

(a) performing a spectral analysis on said test mixture to ascertain the values of each of a series of spectral parameters which vary with those of said selected cha.racteristics; (b) deriving from the values so ascertained a corresponding series of quantitative factors in a manner such that said quantitative factors so derived are representative of said values; and (c) performing a regression analysis on said quantitative factors with the use of regression coefficients to derive predicted values for said selected characteristics of said product; wherein said regression coefficients are representative of a correlation between (i) predetermined values of said selected characteristics of products resulting from sai ' d preselected reaction in a series of standard mixtures and (ii) quantitative factors derived in the manner of step (b) from spectral analyses performed on each of said standard mixtures.

3

2. A method in accordance with claim I in which said spectral analysis of step (a) is a member selected from the group consisting of infraredspectroscopy, laser Raman spectroscopy, ultraviolet spectroscopy, and nuclear magnetic resonance spectroscopy.

3. A method in accordance with claim 1 or claim 2 in which said spectral analysis of step (a) includes the simultaneous use of a plurality of frequencies of a spectrum and the determination of an inverse Fourier transform of emissions resulting therefrom.

1 11

4. A method in accordance with any one of claims 1 - 3 in which said regression analysis of step (c) is a multiple linear regression analysis.

5. A method in accordance with any one of claim 1 - 4, further rising:

(d) comparing said predicted values derived in step (c) with preselected values as a means of detecting deviations in said test mixtures from a predetermined standard.

6. A method in accordance with any one of claim 1 - 5 in which said reactants differ in p fran said products.

7.

A method in accordance with claim 6 in which said reactants are liquids, and'said product is a solid.

8. A method in accordance with any one of claims 1 - 7 in which said test mixture is a slurry and said product is a solid.

9 - A method in accordance with any one of claims 1 - 8 in which said preselected reaction is a pol_ymerisation reaction.

10. A method of predicting values of selected characteristics of a solid substance which is the product of a preselected reaction of a plurality of reactants contained in a test mixture which includes a liquid phase, said method comprising:

(a) performing a Fourier transform infrared analysis on said test mixture to ascertain the values of each of a series of spectral parameters which vary with those of said selected characteristics; (b) deriving from the values so ascertained corresponding series of quantitative factors in a a 12 manner such that said quantitative factors so derived are representative of said values; (c) performing a multiple linear regression analysis on said quantitative factors with the use of regression coefficients to derive predicted values for said selected characteristics of said product; wherein said regression coefficients are representative of a correlation between (i) predetermined values of said selected characteristics of products resulting from said preselected reaction in a series of known mixtures and (ii) quantitative factors derived in the manner of step (b) from spectral analyses performed on each of said known mixtures; and (d) comparing said predicted values derived in step (c) with preselected values as a means of detecting deviations in said test mixtures from a predetermined standard.

11. A method substantially as hereinbefore described, with refe.rence to the drawings.

Published 199 1 at The Patent Office. Concept House. Cardff Road. Newport. Gwent NP9 IRH. Further copies may be obtained from Sales Branch. Unit 6. Nine Mile Point, Cwmfelinfach. Cross Keys, Newport. NP I 7HZ. Printed by Multiplex techniques ltd. St Mary Cray, Kent.