WO1996010009A1

WO1996010009A1 - A method for controlling polyol ester conversion using near or mid-infrared analysis

Info

Publication number: WO1996010009A1
Application number: PCT/US1995/013108
Authority: WO
Inventors: Larry Oliver Jones; Joe Randall Noles, Jr.; Robert John Louis Chimenti; Howard L. Fang
Original assignee: Exxon Chemical Patents Inc
Current assignee: ExxonMobil Chemical Patents Inc
Priority date: 1994-09-28
Filing date: 1995-09-28
Publication date: 1996-04-04
Anticipated expiration: 1997-03-28
Also published as: AU3831195A

Abstract

Monitoring an acid and polyol esterification reaction mixture by means of near infrared or mid-infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of the reaction mixture in the presence of excess carboxylic acid can be predicted, thereby allowing for real-time control of the overall esterification process.

Description

A METHOD FOR CONTROLLING POLYOL ESTER CONVERSION USING NEAR OR MID-INFRARED ANALYSIS

The present invention relates generally to the conversion of polyols in the presence of excess acids to their corresponding esters. A key step in this process is to measure the near or mid-infrared spectra of the reaction mixture and to estimate the degree of conversion from the spectra using a multivariate model. Knowledge of, for example, the initial acid concentration in the reaction mixture will also enable one to predict the ratio of reactants charged at the outset into the reaction vessel. Moreover, near or mid-infrared analysis may also be used in other parts of the polyol ester process, i.e., it may be used to predict other important process control variables such as water concentration, polyol concentration, acid concentration, and/or other compositional properties of the reaction, stripping and run-down steps.

BACKGROUND OF THE INVENTION The reaction conditions under which esterification is effected can be varied considerably. The reaction proceeds very slowly at room temperature. However, at approximately 200°C about 99% of the limiting reagent, such as polyols, is converted to an ester within a few hours. Limiting reagents are typically reagents which are not present in stoichio etric excess.

To facilitate the complete esterification of the reactants, it is desirable that the water which is formed during esterification be removed as rapidly as possible. It is known that water has a detrimental effect upon the rate of conversion. Conventionally, water has been removed by carrying out the reaction using an entrainer which forms an azeotrope having a boiling point that is lower than that of either component of the reaction. However, in many cases water can be removed without the need for an entrainer by, for example, co-distillation with one of the low boiling point reactants. If the resulting ester has a boiling point well above 100°C at atmospheric pressure, then the reaction temperature can be adjusted such that no liquid medium capable of forming an azeotrope is required. It is also critical that the concentration of water be known at all times in order to fine tune the esterification process.

For polyol esters, e.g., esters made from aliphatic acids and trimethylol propane (TMP), the commercially desirable conversions are at greater than 98%. In the case of polyol esters, the excess acid is generally removed by a combination of stripping or neutralizing and washing. The conversion level is determined by the product specification for the hydroxyl number, a measure of the number of residual hydroxyl groups in the ester. Typical product applications require conversions of about 98.5% of the original number of hydroxyl groups in the polyol.

In the course of the manufacture and sale of polyol esters, there are a number of costly (manpower, equipment, and cycle time debits) analyses which must be performed to insure product quality. Table 1 below sets forth the various analytical measurements which are required during polyol ester synthesis, the current analytical methods used to obtain those measurements, the frequency with which the measurements are needed, and the time required to perform such measurements. Table 1

(Analytical Requirements for Polyol Esters)

Measurement Current Method Freαuencv Time*

Reaction

Hydroxyl Conversion Gas Chromatography 10 per day 30-60 min.

Compositional/ Gas Chromatography 30-60 min. physical properties

H₂0 Concentration Karl Fischer Titration 10 min.

StrΪDDins

Acid Concentration GC/Titration 5 per day 30/10 min.

Run-down Tank

Flash Point ASTM-D92 2 per day 6 hours

Pour Point ASTM-D97 2 per day 6 hours

Viscosity ASTM-D445 2 per day 6 hours

OCS Stability FTM-5308.7 60 hours

SOD Pb Corrosion FTM-5321.2 6 hours

Hydroxyl Number OiT-Line Mid-IR 2 per day 30 min.

H₂0 Concentration ASTM-D1744 2 per day 10 min.

Concentration

TAN ASTM-D664 2 per day 10 min.

* Does not include sampling time.

It is well known that the amount of water in the esterification reactor at any time is a major factor in determining the overall rate of reaction. Currently, trained analytical personnel use a manual gas chromatography to measure the unconverted hydroxyl groups in order to detect the rate of conversion to the polyol ester. Due to difficulties in operating on-line manual gas chromatography devices, a sample is taken and manually analyzed in about 30 to 65 minutes. This manual gas chromatography procedure is labor intensive and requires the most rigorous of laboratory skills.

The measurement of low levels of water in organic solvents has previously been accomplished by gas chromatography, solid state capacitance sensors, photometers and Karl Fischer titration. The Karl Fischer titration is labor intensive and requires the most rigorous of laboratory skills. The photometer is the simplest of the technologies but lacks specificity and therefore frequently gives erroneous readings. The gas chromatography although more specific than the photometer is more maintenance intensive, requires sample lines, and has longer lag times.

Modifying test conditions from the conventional titration or gas chromatography to achieve shorter testing times often leads to inaccurate data. Using near infrared spectral techniques combined with near infrared chemometric analysis as employed by the present inventors should eliminate the latter concern.

It would be highly desirable to avoid the additional personnel, time delay and inaccuracies which are inherent in the current titration and gas chromatography analyses. The delayed measurements cause difficulty in pinpointing the reaction endpoints. Also, these delays cause errors in real-time product quality prediction which results in extended batch times.

The present inventors have discovered that the use of a near or mid- infrared analyzer will allow the use of a single spectrum measurement to predict, on a real-time basis, chemical and physical properties of the esterification reaction mixture in the presence of large amounts of carboxylic acids (5 to 20 weight %). The compositional properties of the various product streams that exit the reaction mixture into the stripper and run-down tanks can also be accurately predicted using the analyzer of the present invention. The ability to predict chemical and physical properties of the reaction mixture and product streams allows for: (a) reaction cycle time reduction due to instantaneous conversion measurement; (b) reduced fixed costs such as manpower wages; and (c) consistent product quality through consistent conversion and an ability to adjust reaction mixtures (i.e., by measurement of compositional properties and through addition of fresh acids) during the course of the reaction to correct for compositional properties due to slight processing variations from batch to batch. Moreover, instantaneous knowledge of water concentration in the esterification mixture will allow real-time monitoring of water removal efficiency.

SUMMARY OF THE INVENTION A process for the esterification of a polyol with an acid which comprises the following steps: adding a polyol and at least one acid to a reaction vessel to form a reaction mixture; heating the reaction mixture and maintaining a pressure sufficient to convert the polyol and acid to a polyol ester and removing water as vapor; monitoring the reaction mixture by means of near infrared or mid-infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of the reaction mixture in the presence of excess carboxylic acid can be predicted; and adjusting the reaction conditions of the esterification reaction, the concentration of the acids, and/or the concentration of the polyol. The aforementioned are adjusted depending upon the degree of conversions, concentrations of water, acids, and/or polyol, and/or the other compositional properties in the reaction mixture as predicted by the near or mid-infrared analysis. There are several mathematical methods that can be used to model the relationship between near or mid-infrared spectra and the properties or chemical compositions to be predicted. Examples include multiple linear regression, principal component regression, partial least squares, and neural net.

The esterification process preferably includes the additional steps of removing excess acid from the polyol ester product by stripping (e.g., steam or nitrogen) and recycling the excess acid to the reaction vessel, monitoring the stripped ester product by means of near infrared or mid-infrared analysis and adjusting (1) the operating conditions of the stripper (e.g., temperature, pressure and steam rate), (2) the reaction conditions of the esterification reaction, and or (3) the concentrations of the acids and/or polyols. The adjustments are made depending upon the degree of the conversions, concentrations of water, acids, and/or polyol, and/or other compositional properties in the stripped ester product.

Additionally, the esterification process may include the steps of sending an enriched or concentrated polyol ester product to a run-down tank, monitoring the enriched polyol ester product by means of near infrared or mid- infrared analysis and adjusting (1) the operating conditions of the stripper (e.g., temperature, pressure and steam rate), (2) the reaction conditions of the esterification reaction, and/or (3) the concentrations of the acids and/or polyols. These adjustments depend upon degree of the conversions, concentrations of water, acids, and/or polyol, and/or other compositional properties in the enriched polyol ester product. The other compositional properties of the enriched polyol ester product are selected from the group consisting of: viscosity, viscosity index, flash point, cloud point, pour point and OCS stability.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a graph comparing the near infrared absoφtion spectra of trimethylol propane (TMP) and linear C₇ carboxylic acid mixtures;

Fig. 2 is a graph showing the dependence of absorbance on weight percent of TMP;

Fig. 3 is a graph showing the standard errors of estimate (SEE) for estimating the weight percent of TMP in the mixtures as a function of the wavelength;

Figs. 4 and 5 are graphs showing mid-IR absorption spec^*ra in the 400 to 4600 cm^'1 range;

Fig. 6 is a graph plotting the standard errors of estimate for the TMP and C7 acid reaction mixture using mid-IR;

Fig. 7 is a graph plotting the standard errors of estimate for the TMP and C₇ acid reaction mixture using near-IR;

Fig. 8 is a graph showing the spectra of the di-ester, tri-ester, and acid mixture of Cβ, C₇, C₈, and C10 linear acids;

Fig. 9 is a schematic representation of the esterification process according to the present invention; Fig. 10 is a flow chart of the near infrared analysis used to monitor and adjust the esterification process of the present invention;

Fig. 11 is a graph plotting predicted versus actual values for tetra-ester conversion;

Fig. 12 is a graph plotting predicted versus actual values for hydroxyl conversion;

Fig. 13 is a graph plotting predicted versus actual values for acid mole

Fig. 14 is a graph plotting predicted versus actual values for acid #; and

Fig. 15 is a graph plotting predicted versus actual values for water ppm.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The conversion of polyols and acids to polyol esters typically comprises: (a) the esterification of the starting acid with a polyol, and optionally, a catalyst, at a temperature and pressure which permits boiling of the mixture in a reactor; (b) monitoring the reaction mixture by near or mid- infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of the reaction mixture in the presence of excess carboxylic acid can be predicted; and (c) adjusting the reaction conditions and/or the concentrations of the acids and/or polyol in the response to the predictions obtained in step (b) above. The esterification process may also include one or more of the following steps: removal of excess acid by stripping (e.g., nitrogen or steam); addition of absorbents such as alumina, silica gel, activated carbon, clay and/or filter aid to the reaction mixture following esterification before further treatment, but in certain cases absorbent treatment may occur later in the process following steam stripping and in still other cases the absorbent step may be eliminated from the process altogether; addition of water to hydrolyze the catalyst (if present); filtration of solids from the ester mixture containing the bulk of the excess acid used in the esterification reaction; removal of excess acid by steam or nitrogen stripping under vacuum and recycling of the acid to the reaction vessel; and removing solids from the stripped ester in a final filtration and transporting to a run-down tank for storage.

As such near infrared analysis can be used to predict the polyol conversion to 0.02%, and di-ester and tri-ester formation to 0.04% and 0.5%, respectively. The mid-infrared spectra can be used to predict the polyol and water concentrations during the esterification reaction.

The preferred esterification catalysts, acids and polyols are set forth in U.S. Patent No. 5,324,853 (Jones et al.) which issued on June 28, 1994, and which is incoφorated herein by reference.

The following acids are preferred during the formation of polyol esters: neopentanoic acid, neoheptanoic acid, neo-octanoic acid, neononanoic acid, neodecanoic acid, 2-ethyl hexanoic acid, oxo-heptanoic acid (i.e., a mix of isomers derived from oxonation/oxidation of hexenes), oxo-decanoic acid (i.e., a mix of isomers derived from oxonation/oxidation of mixed nonenes), oxo- octanoic acid (i.e., a mix of isomers derived from oxonation/oxidation of mixed heptenes), 3,5,5-trimethylhexanoic acid, linear C₅-Cι₈ alkanoic acids, and blends thereof.

Polyols (i.e., polyhydroxy compounds) are represented by the general formula:

R(OH)_n

wherein R is an alkyl, alkenyl or aralkyl hydrocarbyl group and n is at least 2, and can be used in place of the mono alcohols when polyol esters are desired. The hydrocarbyl group may contain from about 2 to about 20 or more carbon atoms, and the hydrocarbyl group may also contain substituents such as chlorine, nitrogen and/or oxygen atoms. The polydroxyl compounds generally will contain from about 2 to 10 hydroxy groups and more preferably from about 2 to 6 hydroxy groups. The polyhydroxy compound may contain one or more preferably from about 2 to 6 hydroxy groups. The polyhydroxy compound may contain one or more oxyalkylene groups and, thus, the polyhydroxy compounds include compounds such as polyetheφolyols. The number of carbon atoms and number of hydroxy groups contained in the polyhydroxy compound used to form the carboxylic esters may vary over a wide range.

The following alcohols are particularly useful as polyols: neopentyl glycol, 2,2-dimethylol butane, trimethylol ethane, trimethylol propane, trimethylol butane, mono-pentaerythritol, technical grade pentaerythritol, dipentaerythritol, ethylene glycol, propylene glycol and polyalkylene glycols (e.g., polyethylene glycols, polypropylene glycols, polybutylene glycols, etc., and blends thereof such as a polymerized mixture of ethylene glycol and propylene glycol). This method, with or without the titanium, zirconium, or tin based catalysts, is useful in forming polyol esters, such as, neopolyol esters, from polyols and excess fatty acids. The polyol or polyol mixture is preferably technical grade pentaerythritol (PE), trimethylol propane (TMP), and neopentylglycol each which can be admixed with monopentaerythritol and/or trimethylol propane or other neopolyols. The preferred acid component is typically a mixture of straight chain acids having five to ten carbon atoms, or a branched chain acid having from five to eighteen carbon atoms preferably five to nine carbon atoms, namely 2-methylhexanoic, 2-ethylpentanoic, 3,5,- trimethylhexanoic acids or mixtures thereof. Generally, the acids are monocarboxylic acids. Suitable straight chain acids include, but are not limited to, valeric acid (C₅), oenanthic acid (C₇), caprylic acid CC₈), pelargonic acid (C9), and capric acid (C10).

In the reaction used to form esters, the acid mixture is pres-..; in an excess of about 10 to 50 mole percent more than the stoichiometric equivalent of the hydroxyl group on the polyol used. The excess acid is used to force the reaction to completion. The composition of the feed acid is adjusted so as to provide the desired composition of product ester. After the reaction is complete, the excess acid is removed by stripping and additional finishing.

This process utilizes an optical cell or any other suitable fiber optical probe to illuminate the esterification stream with optical radiation, determining the differences in the optical absoφtivity of the stre: ^* at predetermined wavelengths, and for determining the conversions ar^ or concentrations of water, acids, and or polyol, and/or other compositional properties of the esterification reaction mixture, polyol ester product stream and enriched polyol ester product stream from the difference in the optical absoφtivity. We define absoφtivity as the negative of the logarithm of the ratio of the transmitted light intensity to the incident light intensity, the logarithm being divided by the path- length through the absorbing material by the light. The absorbance is the absoφtivity multiplied by the path-length.

The near and mid-infrared analyzers comprise hardware and software to determine the absoφtivity of the process stream over a range of wavelengths in the near or mid-infrared, respectively, followed by analysis of the spectrum by means of a multivariate statistical calibration model, the output of which provides predictions of the physical and/or chemical properties and composition of the process stream. The analyzer is sufficiently accurate, precise, reliable, and rapid to be suitable for providing process control information. The use of fiber optics make remote application of near and mid- infrared analyzers possible in varying process conditions and in hostile environments.

Multivariate analysis of near infrared spectra is usually necessary to extract chemical or physical information since there is often significant overlap between the spectra of different compounds in the near and mid-infrared range. In order to differentiate between hydroxyl groups of a polyol versus water, it is necessary to pass spectral data through a mathematical model which has been generated from calibration samples which represent known variability of the parameters to be predicted and the process conditions to be encountered. Thus, when properly constructed, the model provides statistically accurate predictions even under varying conditions, such as temperature.

There are several mathematical methods that can be used to model the relationship between near or mid-infrared spectra and the properties or chemical compositions to be predicted. Examples include multiple linear regression, principal component regression, partial _^st squares, and neural net.

One such mathematical model is based on a data analysis scheme which employs partial least squares (PLS) as implemented in the Unscrambler™ software (sold by Guided Wave, Inc.). The partial least squares scheme is a multivariate solution for distinguishing broad and overlapping absorbance bands in a near infrared spectroscopy. This software generates a statistical model that can relate some measurable quantity (i.e., spectrum, chromatogram, sensory data) to an unknown quantity (i.e., factors).

Additional sample and alternative modeling techniques may improve prediction capability of the near or mid-infrared analyzer beyond that demonstrated in the examples to follow.

A typical esterification synthesis syste is set forth in fig. 9, wherein a polyol limiting reagent is mixed with at least one acid within reaction vessel 1. Optionally, the esterification reaction within vessel 1 occurs in the presence of a catalyst. Acid and water vapors are taken overhead via conduit 3 and acid is returned to vessel 1 after condensation and separation from the water.

Optionally, the reaction conversion of polyol to the polyol ester, the polyol concentration and the water concentration of vessel 1 contents can be measured by infrared analyzer 56.

A polyol ester product solution is taken as bottoms via conduit 5 and delivered to reaction vessel 7 wherein the polyol ester product solution is treated either simultaneously or sequentially as follows:

(a) water is added to vessel 7 vi; ..onduit 13 in order to hydrolyze the catalyst; (b) an absorbent is added to vessel 7 via conduit 15, the absorbent being selected from the group consisting of: alumina, silica gel activated carbon, clay and/or a filter aid;

(c) partial vacuum and heat are applied to remove the water used in the hydrolysis step, wherein most of the water is removed and some of the acid is removed via conduit 9 and returned to vessel 1 ; and

(d) after water removal, the bottoms are passed via conduit 11 through filter unit 17 to remove solids (i.e., hydrolyzed catalyst, absorbent, filter aids, carbon, etc.) and thereafter passed on to stripping tower or reactor 21.

Optionally, and prior to step (d) above, reaction vessel 7 can be used as a stripping vessel. Steam or nitrogen is introduced via conduit 10 and the remaining acid and water is removed with the application of heat and vacuum. When vessel 7 is also used as a stripping vessel, the water concentration and acid concentration can be monitored by near or mid-infrared analysis. Following step (e), the bottoms are filtered through filter 17 to remove solids.

Any acid taken overhead in conduit 9 is recycled to succeeding reactor batches in vessel 1.

Optionally, during steps (a) through (c), the properties of the contents in vessel 7, the concentration of the polyol ester product, the acid concentration, and the water content can be monitored by means of near or mid-infrared analyzer 56. The measurement of the water content is particularly important in determining the ease of filtration in filter 17. The water concentration in conduit 11 should be less than 1000 ppm and preferably below 500 ppm. The solids are taken as bottoms via conduit 19 and the filtered polyol ester product is sent to stripping tower 21 through heater 30 via conduit 23. In stripping tower 21, excess acid is removed from the polyol ester product by steam or nitrogen stripping under partial vacuum and recycled to successive batches in reaction vessel 1 via conduit 20. Both the excess acid and the residual water concentration are reduced in this step to meet product specifications. The polyol ester enriched-product is taken as bottoms from tower 21 via conduit 25 and delivered to a second filter 27 wherein solids are taken out as bottoms via conduit 29 and further enriched polyol ester product is sent via conduit 26 to run-down tank 32.

Optionally, the residual water content, the residual polyol content, and the residual acid content may be measured in the final stripped polyol ester product using near infrared analyzer 56 on-line in conduit 25.

Additional equipment such as vacuum jets, pumps, separators, holding drums, and condensers are preferably added to the process for the practical application of the various steps.

In accordance with the present invention an in-line near or mid-infrared analyzer 56 is in fluid contact with reaction vessel 1 via conduit 54. Analyzer 56 comprises a means for generating a near or mid-infrared spectra of the reaction mixture and a means for applying a mathematical model and the spectrum data, that model embodying the relationship between the near or mid- infrared spectra and the properties or chemical compositions to e predicted.

The measured near or mid-infrared spectral data is processed by a computer 58 (CPU) which contains a calibration or multivariate model (e.g., multiple linear regression, principal components regression, partial least squares, or neural net); whereby an accurate prediction of either the conversions and/or 16

concentrations of water, acids, and/or polyol, and/or other compositional properties in the reaction mixture is obtained. Typical on-line near or mid- infrared analyzer 56 includes a pump 52, heat exchanger 50, and thermometer 51. The resultant predictions are sent to CPU 58 where they are processed and if the predicted property/composition value meets a predetermined target value, then the results are recorded and the process stream is analyzed again. If, however, the predicted value is less than or greater than the target value, then a signal is sent to the process control system instructing it to adjust the reaction conditions and/or concentrations of acids and/or polyols depending upon the degree of conversions and/or concentrations predicted by analyzer 56.

In the case of reaction vessel 7, or other configurations where solids are present, a small filter will be placed in conduit 54 just after pump 52 and before heat exchanger 50 to remove solids from the stream which is fed to infrared analyzer 56.

Near or mid-infrared analyzers may also be used to detect concentration and conversion properties in other parts of the esterification process, e.g., finishing vessel 7 (e.g., a flashing vessel), stripper reactor 21, and run-down tank 32. The near infrared analyzer associated with stripper reactor 21 is preferably disposed about conduit 25 which is downstream of the bottoms discharged from stripper reactor 21.

Near or mid-infrared measurement of the solution spectra through the placement of an in-line or on-line analyzer in or about the esterification reaction, stripper reactor and run-down tank allows the esterification process to be followed in real-time for better control of product quality and more efficient utilization of the respective reactors. The esterification process may be configured in several ways to meet individual processor requirements. However, the near infrared analyzer 56 is adapted by easily engineered sampling techniques to measure reaction conversion and the concentrations of acids and water in any of several typical process configurations. One example offlexibility was demonstrated above with the optional step (e). In addition, the infrared analysis technique can be readily adapted to a completely continuous process for all steps or for a process where all the steps take place consecutively in a single reactor. Th *s, the infrared analysis technique is particularly well suited for monitoring any process ,J in real-time for more efficient utilization of the equipment tin the particular esterification process. In batch processes, the acid is recycled to the next batch; whereas during a continuous process the acid is recycled to the esterification reaction vessel.

The prediction of the conversion and/or concentration values of the constituents within the esterification process is demonstrated in Fig. 10, attached hereto. The esterification, stripping and run-down processes 70 are continuously monitored by means of near or mid-infrared analyzer(s) 72 to determine the absoφtivity of the process stream over a range of wavelengths in the near or mid-infrared spectrum. The spectral data from analyzer 72 is thereafter sent to a microprocessor where the data is pre-processed 74, and analyzed by a multivariate statistical calibration model 76. Pre-processing may include determining the quality of the spectrum, correcting the spectrum for instrumental distortions, differentiating the spectrum one or more times, and taking the Forier transform of the spectrum. Model 76 may include any known mathematical modeling technique which is capable of providing accurate i iictions 78 of the physical or chemical properties and composition of the process stream, e.g., multiple linear regression models, principal component regression models, partial least squares models, or neural net models. Thereafter, the predicted property/composition of the process stream is compared against predetermined property/composition values 80. If the predicted values meet the predetermined values the results are recorded 82 and a signal is sent to repeat the infrared analysis 72. If the predicted values to not meet the predetermined values, then process control system 84 either increases or decreases the concentration of the respective reactants in process 70.

The present inventors have discovered that near infrared pH stretching spectra can readily distinguish water hydroxyl groups from polyol hydroxyl group transitions. Near infrared analysis allows one to evaluate water concentration in the presence of carboxylic acid and polyols.

Example 1

This example demonstrates the use of optical spectra to predict the conversion and composition of reaction mixtures of trimethylol propane (TMP) and linear carboxylic acids in the C₇ range.

Absorbance measurements were made on two sets of samples. The first set was comprised of five samples: pure heptanoic acid (C₇ acid) and four mixtures of TMP, C acid, and water. The component concentration of these five samples, as prepared gravimetrically are shown below in Table 2.

Table 2

Sample No. TMP (wt.%) C₇ Acid (wt.%) H₂O (ppm)

~ϊ 0^~55 99.45 669.4

2 1.10 98.91 953.9

3 2.20 97.80 1197.8

4 5.50 94.54 2141.9

5 0 100.02 607.6

The measurement objective of this experiment was to determine whether the hydroxyl groups of the acid, polyol and water in mixtures can be distinguished by their near infrared spectra, and also how well the TMP content could be estimated on the basis of these spectra.

The second sample set comprised four reaction mixtures containing fully-converted tri-esters (TE), partially-converted di-esters (DE), excess carboxylic acids (Cβ, C , C₈, and Cio), and water. These compositions were determined using gas chromatography and Karl Fischer titration (for the water). The component concentrations and TMP conversion are given below in Table 3. The objective of these measurements was to estimate the conversion of the TMP to the tri-ester.

Table 3

Sample Acid DE TE H₂O Conversion*

No. (wt.%) (wt.%) (wt.%) (ppm) (wt.%)

1 24 71 68.9 482 93.1

2 22 45 73.6 347 96.0

3 21 22 76.9 594 97.5

4 22 16 76.4 695 98.2

Conversion of TMP to the Tri-Ester Near infrared spectra were collected for both of the sample sets. In addition, mid-infrared spectra of the first sample set was also taken. Information on the measurements is provided below in Table 4.

Table 4

(Infrared Analysis Procedures] ,

Spectrometer: LTI-1200 (grating) Mattson (FT)

Spectral Range: Near-IR Mid-IR

800-1600 nm 5000-400 cm^'1

Resolution: 8 nm 2 cm^"1

Path length: 1 cm 0.005 cm

Reference: Empty cell + air Air

Temperature: 23°C (controlled) Ambient

No. of Scans: 256 100

Two separate measurements were made on each sample and the spectra were analyzed as if they were from different samples. The near infrared absorbance spectra of the TMP/C₇ acid mixtures comprising the first sample set are shown in Fig. 1. Two broad spectral features, consisting of several unresolved peaks, can be seen at 1200 nm and 1420 nm. The 1200 nm peak shows little visually discernible variation between the samples. This band is predominantly due to the second overtone of the unresolved vibrational stretching modes of the CH groups.

The 1420 nm band is dominated by the first overtone of the OH vibrational stretching modes of the alcohol (polyol), acid, and water. Combination bands of the first overtone of the CH stretch and fundamental CH bending modes also contribute in this wavelength region. Clearly, the spectral information relevant to the compositional variation of the samples is contained in the 1300-1600 nm wavelength region, consistent with the OH absoφtion bands.

Perhaps the simplest model relating the sample's absorbance and TMP content is one in which the absorbances at just two wavelengths are sued. The absorbance at one of the wavelengths may be chosen as a reference to compensate for instrumental drifts and other effects not related to the TMP, while the absorbance at the second wavelength provides information on the TMP content. It can be seen from fig. 1 that a suitable reference wavelength may be chosen at 1300 nm. The different spectra, A(x nm) minus A(1300 nm), where x takes the value of each wavelength between 1300 nm and 1600 nm, are shown in fig. 2. It can be seen that there is no obvious best choice of an analytical wavelength for the TMP content.

In order to determine the optimum analytical wavelength for the TMP content, least-square regressions were calculated in which the single independent variable for each regression was the value of different spectra at each wavelength between 1300 nm and 1600 nm. This resulted in 451 regression calculations.

Fig. 3 shows the standard errors of estimate (SEE) for estimating the wt.% of TMP in the mixtures as a function of the wavelength. The best SEE was 0.044 wt.% using the absorbance difference at 1437 nm and 1300 nm. However, good estimates of the TMP content could be obtained by using the absorbance at any wavelength between about 1420 nm and 1540 nm as an analytical wavelength. The absorbance at 1437 nm, referenced to the heptanoic acid, and the TMP content of the samples is shown in Table 5 below. A linear-dependence of absorbance with TMP may be seen, indicating Beer's Law behavior and the absence of discernible hydrogen-bonding at these concentration levels.

Table 5 Absorbance at 1437 nm TMP (wt.%)

0.000 0~00

0.020 0.50

0.033 1.25

0.034 1.25

0.065 2.00

0.068 2.00

0.155 5.50

0.157 5.50

Beyond the simple two-wavelength model, full-spectrum multivariate analysis using a Partial Least Squares (PLS) algorithm (sold by LT Industries), gave similar excellent results. The PLS analysis was applied to both the raw absorbance spectra and the second derivative of the spectra. Using second derivative spectra is one technique to remove offsets due to instrument effects. It is most useful for near infrared spectra, such as have been obtained here, which do not contain shaφ features and, therefore, do not suffer significant loss in resolution due to the numerical differentiation process. The results of the various chemometric analyses are summarized in Table 6 below for the TMP and water content of the samples using near infrared spectra.

Table 6

TMP (wt.%) H₂O (ppm)

Two- Wavelength Model A(1437 nm) - A(1300 nm) 0.044

PLS using Absorbance

3 -Factor Model 0.046 21

PLS using Second Derivative

3-Factor Model 0.031 21

Mid-infrared absoφtion spectra are shown in figs. 4 and 5. Major spectral differences can be seen in the 500-1200 cm^"1 and 2500-3700 cm^*1 regions. In the 1000-1200 cm^*1 region, all bands are primarily those involving the CO stretching modes of TMP, consistent with the observed increase in absorbance with increasing alcohol content. The band at 950 cm^'1 decreases with alcohol (polyol) concentration, reflecting a depletion of the OH deformation band of heptanoic acid. In the 2500-3700 cm^'1 region, the band at 3500 cm^"1 can be attributed to the OH stretching modes of TMP consistent with its observed increase with polyol content. The decrease of the 2730 cm^"1 band is related to the depletion of the acid.

Mid-IR absoφtion spectra also show good correlation to both the TMP and water contents of these mixtures. The standard errors of estimate for PLS models based on spectra in selected wavelength regions show better estimation of the TMP and water than obtained in the near infrared by factors of 10 and 3, respectively.

The present inventors have discovered that both near and mid-IR spectroscopy, coupled with multivariate modeling can be used to estimate the polyol and water content of mixtures in the presence of excess heptanoic acid. Using near infrared data, standard errors of 0.04 wt.% TMP and 21 ppm H₂O were obtained.

For the second sample set, comprised of TMP and acid reaction mixture samples, near-IR spectra were used to estimate the weight percent acid, weight percent di- and tri-esters, ppm H O, and the percent conversion of TMP to tri- ester. The absorbance differences, A(x nm) minus A(1300 nm), were again used as regression variables to estimate the composition and conversion. The standard errors of estimate for each of the composition and conversion quantities are shown in figs. 6 and 7 (on an expanded scale). It can be seen that for any wavelength between 1420 nm and 1600 nm, all of the quantities can be estimated to well below 1%. The conversion, for example, can be estimated to below 0.05 wt.% using the simple two-wavelength difference model.

To provide some quantitative measure of prediction, Principal component (PC) analysis followed by a "leave-one-out" cross-validation procedure were used to determine an average prediction error. The PC transformation of the absorbance spectra was carried out first to obtain a few independent regression variables that represent the deterministic part of the variance in the spectra. The relationships between the spectral and compositional properties or variances of the samples are the models that we seek to assess. A reduction in possible variables from 451 (the number of

SUBSTTTUTE SHEET (RULE 26) wavelengths between 1300 nm and 1600 nm in each absorbance spectrum) to about 4, at most, is obtained by the PC transformation.

The "leave-one-out" cross-validation procedure involves leaving the spectrum and properties of one of the samples out of the data set, using the principal component analysis to obtain the best model to estimate the properties of the remaining samples, and then to use this model to predict the properties of the excluded sample.

This procedure is repeated until each sample is excluded once. The standard deviation of the residuals between the predicted and actual values, called the PRESS (for Predicted Residual Sums of Squares), was calculated. This quantity is used as a measure of how well the composition and conversion can be predicted, and which of the PC variables should be included in the model. The final step is to use the "best" predictive model, as determined by the PRESS statistic, to estimate the properties of the entire data set in order to obtain the conventional standard error of estimate.

Models utilizing the fewest principal component variables and having comparable PRESS prediction and SEE estimation statistics are generally the most robust and reliable. Table 7 below summarizes the best results of the analysis.

Table 7

Acid DE TE H₂O Conversion*

Analysis (wt.%) (wt.%) (wt.%) (ppm) (wt.%)

PC var. 2 3 2 3 3

PRESS 0.31 0.04 0.5 24 0.02

SEE 0.27 0.03 0.42 19 0.013

* Conversion of TMP to the Tri -Ester

It can be seen that for the composition and conversion data, models involving 2 or 3 principal component variables have average prediction errors that do not differ substantially from the corresponding standard errors of estimate. Tables 8-12 compare the weight percent acid, weight percent di- ester, weight percent tri-ester, percent H₂O, and percent OH converted as estimated by both near infrared analysis according to the present invention and the conventional gas chromatography method.

Table 8

Wt.% Acid Wt.% Acid

Sample (Near Infrared) (Gas Chromatography)

1 24.1 24.0 2 22.1 22.0 3 21.4 21.0 4 22.0 22.0

SUBSTTTUTE SHEET (RULE 26) 27

Table 9

Wt.% Di-ester Wt.% Di-ester

Sample (Near Infrared) (Gas Chromatography)

1 7.1 7.1

2 4.5 4.5 3 2.2 2.2 4 0.6 0.6

Table 10

Wt.% Tri-ester Wt.% Tri-ester

Sample (Near Infrared) (Gas Chromatography)

1 69.2 68.9 2 73.8 73.8 3 77.1 76.9 4 76.8 76.2

Table 11

% H₂O % H₂O

Sample (Near Infrared) (Karl Fischer Titr.)

1 0.049 0.049

2 0.036 0.035 3 0.061 0.059 4 0.070 0.069

Table 12

% OH Converted % OH Converted

Sample (Near Infrared) (Gas Chromatography)

1 93.1 93.1 2 96.0 96.0 3 97.5 97.5 4 98.2 98.2

Fig. 8 shows the spectra of the di- and tri-esters and the acid. The acid in these samples is a mixture

of CQ , Cη , CQ , and C^Q linear acids. The spectrum of heptanoic acid is shown for comparison and overlays the acid spectrum. In computing these spectra the water component was not included. Thus, it may be possible to follow the molecular composition of the reaction, on-line.

EXAMPLE 3

This example compares actual compositional variables versus the use of in-line/on-line near infrared to predict compositional variables in the process of esterification of pentaerythritol (PE) in the presence of excess carboxylic acids.

This example successfully demonstrated that near infrared analysis can be effectively used to predict: (1) the distribution of molar fractions of all mono-, di, tri-, and tetra-esters of pentaerythritol; (2) the degree of conversion in tetra-ester formation and carboxylic acid depletion; and (3) the ether by¬ products and their corresponding tri-, tetra-, penta- and hexa-esters. The total acid number as well as the remaining water content can also be predicted within reasonable accuracy.

Both absorbance and second derivative of absorbance (A2d) spectra were taken and analyzed in this example. Figs. 11-15 show the results (predicted versus actual values) for some of the reaction parameters. Table 13 shows the SEE (Standard Error of Estimate) for some of the ingredients. A "leave-one- out" validation algorithm was also applied to find the SEP (Standard Error of Prediction) as listed in Table 14. For most ingredients, only three or four factors were used indicating a good correlation between the spectra and the corresponding chemical or physical properties.

Table 13

(Standard Deviation of Fitted Ingredients )

Model Acid (mole%) Acid # Water (ppm) Ester % OH%

Pentaerythritol Absor 0.0047 1.323 32.19 0.0072 0.0042

Pentaenihπtol A2d 00048 2.066 30.70 00035 0.0029

Table 14

(Validation Result of Pentaerythritol)

Parameter Factor # SEP

Acid (mole% ) 4 0 . 0037

Acid # 2 0 . 2440

Water 4 22 . 7

Ester % 3 0 . 0100

OH % 2 0 . 0100

In summary, the models, based upon near-IR and mid-IR absorbance spectra, can be used to predict the hydroxyl conversion to about 0.02%, and the di-ester and tri-ester and water contents to 0.04, 0.5, and 0.002 wt.%, respectively.

Claims

WHAT IS CLAIMED IS:

1. A process for the esterification of a polyol with at least one acid which comprises the following steps: a. adding a polyol and at least one acid to a reaction vessel to form a reaction mixture; b. heating said reaction mixture and maintaining a pressure sufficient to convert said polyol and acid to a polyol ester and removing water as vapor; c. monitoring said reaction mixture by means of near infrared or mid-infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of the reaction mixture in the presence of excess carboxylic acid can be predicted; and d. adjusting the reaction conditions of step (b) and/or the concentration of acids and/or polyol depending upon the degree of conversions and/or concentrations of water, acids and/or polyol, and/or the other compositional properties in the reaction mixture as predicted by the near or mid-infrared analysis of step (c).

2. The process according to claim 1 further comprising the addition of a catalyst to said reaction vessel such that said polyol and acid are catalytically converted to said polyol ester.

3. The process according to claim 1 wherein said n-- r or mid-infrared analysis comprises a means for gtiarating spectral data and a means for modeling the relationship between said spectral data and said conversions and/or concentrations to be predicted.

4. The process according to claim 3 wherein said modeling is at least one model selected from the group consisting of: principal component regression, multiple linear regression, partial least squares, and neural net.

5. The process according to claim 1 wherein said other compositional properties are selected from the group consisting of: viscosity, viscosity index, flash point, cloud point, pour point and OCS stability.

6. A process for the esterification of a polyol with an acid which comprises the following steps: a. adding a polyol and at least one acid to a reaction vessel to form a reaction mixture; b. heating said reaction mixture and maintaining a pressure sufficient to convert said polyol and acid to a polyol ester product and removing water as vapor; c. removing excess acid from said polyol ester product by stripping; d. monitoring said stripped ester product from step (c) by means of near infrared or mid-infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of said stripped ester product can be predicted; and e. adjusting the reaction conditions of step (b), the reaction conditions of step (c), and/or the concentration of acids and/or polyol depending upon the degree of conversions and/or concentrations of water, acids and/or polyol, and/or the other compositional properties of said stripped ester product.

7. The process according to claim 6 further comprising the addition of a catalyst to said reaction vessel such that said polyol and acid is catalytically converted to said polyol ester product.

8. The process according to claim 6 wherein said near or mid-infrared analysis comprises a means for generating spectral data and a means for modeling the relationship between said spectral data and said conversions and/or concentrations to be predicted.

9. The process according to claim 8 wherein said modeling is at least one model selected from the group consisting of: principal component regression, multiple linear regression, partial least squares, a* 1 neural net.

10. The process according to claim 6 wherein said other compositional properties are selected from the group consisting of: viscosity, viscosity index, flash point, cloud point, pour point and OCS stability.

11. A process for the esterification of a polyol with an acid which comprises the following steps: a. adding a polyol and at least one acid to a reaction vessel to form a reaction mixture; b. heating said reaction mixture and maintaining a pressure sufficient to convert said polyol and acid to a polyol ester product and removing water as vapor; c. removing excess acid from said polyol ester product by stripping to form an enriched polyol ester product; d. passing said enriched polyol ester product to a run-down tank; e. monitoring said enriched polyol ester product of step (d) by means of near infrared or mid- infrared analysis such that the conversions and/or concentrations of water, acids, and/or polyol, and/or other compositional properties of said enriched polyol ester product can be predicted; and f. adjusting the reaction conditions of step (b), the reaction conditions of step (c), and/or the concentration of acids and/or polyol depending upon the degree of conversions and/or concentrations of water, acids and/or polyol, and/or the other compositional properties of said enriched polyol ester product.

12. The process according to claim 11 further comprising the addition of a catalyst to said reaction vessel such that said polyol and acid is catalytically converted to said polyol ester product.

13. The process according to claim 11 wherein said near or mid-infrared analysis comprises a means for generating spectral data and a means for modeling the relationship between said spectral data and said conversions and/or concentrations to be predicted.

14. The process according to claim 13 wherein said modeling is at least one model selected from the group consisting of: principal component regression, multiple linear regression, partial least squares, and neural net.

15. The process according to claim 11 wherein said other compositional properties are selected from the group consisting of: viscosity, viscosity index, flash point, cloud point, pour point and OCS stability.