EP1218165A1

EP1218165A1 - Method for control of extrusion process

Info

Publication number: EP1218165A1
Application number: EP00967004A
Authority: EP
Inventors: Michael Joseph Piovoso; Paul Thomas Shea; Chi-Kai Shih; Mark David Wetzel
Original assignee: EI Du Pont de Nemours and Co
Current assignee: EIDP Inc
Priority date: 1999-10-01
Filing date: 2000-09-28
Publication date: 2002-07-03
Also published as: JP2003511260A; WO2001024991A1

Abstract

An improved computer monitored continuous extrusion process for controlling the properties of an extruded material, wherein: 1) variable signals generated by an extruder drive and control system are transmitted to a data acquisition signal conditioning front end hardware, which isolates, amplifies and filters the variable signals and outputs variable analog signals; 2) the variable analog signals output by the data acquisition system signal conditioning front end hardware are converted to variable digital signals; and 3) the variable digital signals are transmitted to a control computer which calculates setpoints, which can be transmitted back to the extruder drive and control system or a feed control system used by the extrusion process, the improvement comprising selecting at least one variable signal having a coefficient of variability of greater than 0.05 and correlating said variable signal with variations in at least one property of the extruded material.

Description

TITLE METHOD FOR CONTROL OF EXTRUSION PROCESS FIELD OF THE INVENTION This invention relates generally to a method for controlling the extrusion of fluid materials. More particularly, the invention relates to control of the properties of the material being extruded, with or without the physical parameters of the extrusion process, using a twin screw extruder.

BACKGROUND OF THE INVENTION Ever since the advent of polymeric materials and the mixing or processing thereof using extrusion techniques, the art has tried to develop methods of controlling the properties of the composition or compound being extruded or the parameters of the extrusion process itself, such as temperature, throughput, etc. These methods ranged early on from simple analysis of the resulting extruded product, then attempting to adjust extrusion conditions to change the characteristics of the resulting product to very sophisticated mathematical methods such as those described by Costin et al. in Polymer Engineering and Science, mid-December, 1982, Vol. 22, No. 17, pp. 1095-1108. Costin et al. describes a method to control a plasticating extruder by running a "steady state, step, and time-series analyses" during extrusion, then derived a "transfer function" for purpose of process control. Specifically, Costin et al. describe a method to generate a transfer function between extruder pressure and the screw speed to regulate extruder pressure by using the screw speed as a manipulated variable. Costin et al. developed the control algorithm from the transfer function generated. Results were obtained by testing the effectiveness of the control in the face of a pressure surges and a disturbance in feed quality. Commonly assigned, copending patent application Provisional Serial No. 60/02758, filed October 2, 1998, now PCT US99/22579, filed on October 1, 1999, describes a process for dynamically controlling the properties of an extruded material by use of external perturbation of the extruded material. While such methods have merit, they do not address the need to dynamically control the extrusion process without the use of external perturbation of the system. An object of this invention is to provide a dynamic control method for the product properties of the extrusion process with externally perturbing the extrusion system. SUMMARY OF THE INVENTION

This invention provides an improvement in a computer-monitored process for controlling the properties of a material extruded in a continuous extrusion process. The improvement resides in the utilization of the product property information that can be present in variability of signals such as pressure, temperature, and amperage, and others normally acquired in an extrusion process.

More specifically the improvement can be described as follows. In a computer monitored continuous extrusion process for controlling the properties of an extruded material, wherein (1) variable signals generated by an extruder drive and control system are transmitted to a data acquisition signal conditioning front end hardware, which isolates, amplifies and filters the variable signals and outputs variable analog signals, (2) variable analog signals output by the data acquisition system signal conditioning front end hardware are converted to variable digital signals, and (3) the variable digital signals are transmitted to a control computer which calculates setpoints, which can be transmitted back to the extruder drive and control system or a feed control system used by the extrusion process, the improvement comprising selecting at least one variable signal having a coefficient of variability of greater than 0.05 and correlating said variable signal with variations in at least one property of the extruded material.

DESCRIPTION OF THE DRAWING Figure 1 is a schematic flow diagram of a means for practicing the invention.

DETAILED DESCRIPTION OF THE INVENTION During normal operation of an extruder, the fluctuations found in the process signals commonly acquired may be related to some physical properties. There are two steps involved in determining the relationship so that a control system can be used to maintain a certain property at a fixed level. The first step involves determining which signals that are currently collected are pertinent to the physical property that is being modeled. The second step is to generate the relationship.

To be useful, a signal must have sufficient variability and that variability can not be random white noise. The coefficient of variability is a statistical measure that can be used to define the degree of variation in the data. It is computed as the ratio of the standard deviation of a signal, x, to its mean.

CV = x

where σ_x is the variance and x is the mean. If the CV is greater than 0.05, the corresponding signal is a possible candidate for inclusion into a model for prediction of a product property.

Even if a signal has sufficient variability, it may be random fluctuations that are not related to anything. For the fluctuations to be meaningful, the autocorrelation of the signal should be significant. The autocorrelation is defined to be R_xx(m)=E[x_n* x_n+m]/σ_x ², where E represents the averaging or expectation operator. In the case of white noise, R_xx(m)=l for m=0, and 0 elsewhere. Of course, when computed from data, Rχχ(m) is indistinguishable from zero if be Rχχ(m) < l/(N-|m|) where N is the number of points used in doing the computation and |m| is the absolute value of the shift, m, used in the computation. Having identified potential signals that might be used to do the modeling, the next step is to determine if the variability present is consistent with the property that one wishes to predict. To establish if that is the case, a set of experiments must be run. If, for example, a particular property of interest has a value C and potential range d, experiments with known concentrations of C, C+d/2, C-d/2, C+d, and C-d are run and the corresponding signals of interest are acquired. A measure of variability is needed for each signal to correlate to the desired property. There are many such measures; a few of which are: - standard deviation or variance

- maximum signal value - minimum signal value

- mean absolute deviation from the signal median

- Energy within certain spectral bands of the signal.

Variance is defined by E[(x - E[x])²], and standard deviation is the square root of the variance. The mean absolute deviation from the mean is E[|x-median(x)|]. A reference for this method is Huber, Peter J., "Robust Statistics", John Wiley &

Sons, New York, 1981, ISBN 0-471-41805-6, at page 2.

To compute the energy within a specific band, Fast Fourier Transforms

(FFT) are used. Signals must first be preprocessed to remove trends and means since these components distort the precision of the computation. FFTs can be generated from a number of programs, one of which is MATLAB® Signal

Processing Toolbox. Once, the FFT is computed, the sum of the magnitude of the resulting spectra components is generated over specific bands of interest. These bands might be chosen based on some physical understanding of the system. For example, if the average residence time within an extruder is say 2 minutes, then, one might look at frequencies below 0.5 cycles per minute and above that.

Otherwise, one might try an arbitrary selection of frequency bands and iterating on that.

After a number of measures of variance are obtained, the correlation with the properties is established. Standard regression methods such as multi-linear regression, stepwise regression or subspace regression methods, such as Projection onto Latent Structures or Partial Least Squares (PLS), are possible. For the examples in this work, PLS was used. Subspace regression methods tend to give more robust models.

Referring to Figure 1, the extrusion process is running at its normal, steady-state operating condition. The feed system 2 delivers the steady state feeds, in normal steady state operation, i.e. an unperturbed, continuous operation of the extruder. The steady state feed system can consist of one or more metering devices to control ingredient addition to the extruder 3 in normal, steady-state operation. Additional feed systems 2 can be added along extruder 3, if desired. Extruder 3 is preferably a twin-screw (T/S) extruder, a mechanical device that melts, mixes and pumps polymeric or high-viscosity materials. A twin-screw, co-rotating extruder is a starve-fed device that consists of a barrel, die and screw. The process is the sequence of steps performed in the extruder from the feed addition through product forming in the die at the exit. The primary actions monitored by this invention are: feed transport, melting, mixing, pumping and chemical reactions. Working zones of the screw are defined as the melting zone 3a, the mixing zone 3b and the pumping zone 3c. The product 16 is the material exiting the extrusion process. Extruder motor drive and control system 4 generates motor amperage, voltage and screw speed dynamic signals. These signals are connected to the data acquisition system signal conditioning front end hardware 7. Process ingredient steady-state feed rates are regulated by a feeder control system 5. Feed rate dynamic signals are measured by this system. Control setpoints 15 are received by this system. Point measurements, such as the measurement of pressure, temperature and residence time distribution (RTD), using transducers 6, can be taken by inserting transducers 6 into the extruder to monitor the machine and process material. These signals are connected to the data acquisition system signal conditioning front end hardware 7. Pressure is a preferred variable to measure in this way in accordance with the preferred practice of this invention. The signal conditioning front end hardware 7 processes dynamic analog sensor signals and digital signals connected thereto. This hardware isolates, amplifies and filters source signals. It also provides excitation voltages and currents to active sensors. Human operators or certain equipment can initiate and control the analysis and control technique of this invention by requesting an update of the control adjustments. The start test signal 8a initiates the data acquisition test cycle. The analog signals from the signal conditioning front end hardware 7 are converted to digital sequences with analog to digital

(A/D) converters at the data acquisition hardware 9. The control setpoints 14 are converted to Analog signals with digital to analog (D/A) converters. Digital inputs 8 are converted to binary numbers. The analysis and control computer 10 controls performs signal processing 12 to analyze the extruder inputs from the data acquisition hardware 9, calculates a control decision 13 and outputs the control setpoints 14 to the data acquisition hardware 9 or the extrusion process operator 18. The analysis and control computer 10 may, optionally, input the product's 16 physical properties 17. The analysis 11 consists of all input signals measured the signal conditioning front end hardware 7 and converted to digital sequences in time by the data acquisition hardware 9. The analysis 11 contains the steady-state data 1 la. The steady state data 1 la represents the values ofthe measured inputs (4, 5, 6) under normal operating conditions. Signal processing 12 analyzes steady state response 11 to obtain physical property estimates 12a or set of values used in the control computation 13. A model relating the response 11 to physical properties 17 may be used here. The results ofthe signal processing 12 produces physical property estimates 12a as an input to the control model/decision 13. The control algorithm calculates new control setpoint(s) 14 for the process. That is, based on the analysis 11 , any error in properties requires a control adjustment. The control setpoints 14 are a sequence of numbers generated for the next time interval that is used to manipulate the process. The outputs can be sent to the extruder motor drive and control system 4 and the feeder control system 5 through a D/A converter or the data acquisition hardware 9 and the signal conditioning front end hardware 7. The setpoint signals 15 are the digital or analog signals to the extruder motor drive and control system 4 and the feeder control system 5 to manipulate the process. A human or machine operator may also manipulate the system directly based on setpoint information presented at operator interface 18. Generally, the so controlled extrusion process produces the product 16.

The setpoint signals 15 and the human operator manipulate the process to maintain product quality. The extruder motor drive and control system 4 and the feeder control system 5 change extrusion conditions, the extruder 3 responds, and the product 16 is altered. The product 16 has a set of physical properties 17 that can be measured on or off-line. The goal ofthe extrusion control process of this invention is to maintain these properties within specification limits. The physical properties 17 are correlated to analysis 11 by the signal processing 12 and the control model/decision 13. Physical properties 17 are then estimated by the control model/decision 13 for manipulating the control setpoint(s) 14. The physical properties 17 can be measured during closed loop control to improve the control setpoint(s) 14. As stated above, a human or machine operator can initiate the analysis and control test cycle through a request to the control computer 10, and obtain control setpoint(s) 14. In a typical industrial extrusion process, the analysis and control process of this invention is used to generate direct estimates of product physical properties on a frequent, periodic basis. That is, every time a control setpoint update is needed, the steady-state data is acquired. The responses are analyzed by the signal processing algorithm that estimates the physical properties to be controlled. The control model then calculates new setpoints or setpoint changes to be implemented by the operator or process control systems. The analysis and control is performed at regular time intervals so that the control setpoints can be adjusted as often as is required to meet quality specifications. In contrast, in conventional extrusion control processes, the goal is to reject disturbances detected in measured signals. The analysis and control process of this invention observes the steady state behavior ofthe process and produces an estimate when a product property measurement is required. The control setpoints computed from the analysis model adjust for changes in ingredient and the extrusion process over time scales determined by the analysis cycle.

As shown in Figure 1 , the fundamental extrusion apparatus is comprised of a feed system 2, i.e., a means of metering feed ingredients into one or more port locations ofthe extruder. In the invention, it is preferred to use a starve-fed extrusion process, where the feed rates are set and controlled independently from the extruder screw speed (flood-fed single screw extrusion process rate is determined by the extruder screw speed). In the T/S extrusion process, ingredient feed rates are regulated with the feeder control system 5. The invention provides for the adjustment ofthe ingredient feeder controller setpoints. As far as the type of extruder to use, the invention is not limited to the use of a twin screw extruder, but preferred examples use modular, co-rotating, intermeshing T/S extruders. Modular twin screw extruders are typically comprised of:

• segmented barrels with 2 bores that enclose the screw, including feed ports, closed barrels and vent ports,

• 8-0 transition piece and a die to form product strands, film or other shapes,

• two intermeshing screws located inside each barrel bore separated by a center-to-center distance, as described in detail in the book entitled "POLYMER EXTRUSION" by Chris Rauwendaal, Hanser Publishers, 1990. Chapter 10, "Twin Screw Extruders", that are comprised of • screw shaft connected to the drive transmission,

• modular screw elements including a selection of one or more conveying bushings, reverse pumping bushings, kneading blocks, spacers, gear mixers, tip pieces and other geometry, • drive motor, transmission and gearbox to turn the screw,

• barrel heaters and cooling system,

• control system to regulate barrel temperature and screw speed, and monitor amperage (torque) or power and maintain safe machine operating conditions.

Auxiliary equipment including vacuum handling system, strand quenching and pelletizing hardware can be used to solidify and cut the product into its final form.

Examples of conventional screw design used to melt, mix and pump a polymeric material system are shown in many publications, for example, H. Cartier and G.-H. Hu, Polymer Engineering and Science, June 1999, Vol. 39, No. 6, p. 998.

Machine material of construction is generally from steel, surface treated steel and other metal alloys for wear resistance. Polymeric materials (low and high viscosity), monomers, plasticizers, fillers and other ingredients constitute the feed materials. Many feeds are melted or fluidized in the extruder to enable mixing, forming and chemical reactions.

In T/S extruders, machine size is an important factor in determining production throughput. In general, different processes utilize different machine size and barrel configurations. Most important, different combinations of screw elements mounted on the screw shafts are used make up the screw design. The screw design in the T/S machine usually has groups of working elements (kneading blocks and gear mixers), or working zones, separated by conveying elements. So, a conventional screw design may have a melting zone to transform solid pellet or powder feeds into molten liquid, mixing zones to mix and react ingredients and a pumping zone at the end ofthe screw to drive the liquid product through the die. In an extrusion process, product physical properties 17 are measured to maintain quality and consistency. Most properties, such as Durometer Hardness, ASTM test, D2240, are measured in a laboratory off-line. In some applications, on-line measurements are possible and are useful in closed- loop control ofthe process. Without a direct, real-time, on-line measurement, a alternative method of inferential measurement is required in order to perform closed-loop control ofthe extrusion process to maintain the physical property of interest.

The prior art extrusion control processes typically correlate a process parameter such as pressure to physical properties and use extruder sensor signals as part of a disturbance rejection control scheme. In industrial practice, control based on pressure in an extrusion process is not considered a practical method. Other variables in the process cause the measured pressure and temperature to change, making the controller vulnerable to error.

Extruder operating conditions depend on machine size. Machine size is determined by the barrel diameter. In the invention, machine sizes range from 30 mm and 40 mm (pilot scale) to 92 mm and 120 mm (plant scale). Typical throughputs range from 10 to 35 lb/hr (4.54 to 15.58 kilograms/hr)on the pilot scale, and 1000 to 6000 lb/hr (454 to 2724 kilograrns/hr) on the plant scale. Barrel temperature settings depend on ingredient and product thermal properties. They may be set from 100°C to 350°C for typical applications. Screw speed (RPM) typically varies from 100 to 500 RPM. Specific experimental conditions can be found in the Examples ofthe invention herein below.

EXAMPLES The following representative examples illustrate the process and catalyst compositions of this invention. All parts, proportions, and percentages are by weight, unless otherwise indicated. In each example, the following procedure was used unless otherwise noted.

In the specific examples which follow, the rubber is an SB S (Styrene- Butadiene-Styrene) or SIS (Styrene-isoprene-Styrene) resin, manufactured by Shell Chemical Corporation (Houston, Texas) under the trade name Kraton®. The monomer is hexamethylene diacrylate orhexamethylene dimethacrylate. The plasticizer is a polybutadiene oil, Nisso Oil manufactured by Nippon Soda Co, Ltd. Tokyo, Japan. The product property utilized in the examples which follow was hardness, as measured by a commercially available durometer using ASTM test, D2240. The durometer used was a Zwick 3123 & 3124 hardness tester available from Zwick of American located at 18 Thompson Rd., East Windsor, CT, telephone (203)-623-9475.

Hardness was controlled by changing the percent monomer and/or rubber in the feed. The twin screw extruder utilized was a ZSK 30 mm manufactured by Werner-Pfleiderer. Rubber or a pre-compounded binder (rubber, plasticizer and wax) was added at barrel 1, a monomer solution (of monomer, initiator, inhibitor, and dye) at barrel 3 and the remaining plastisizer at barrel 5 in a nine barrel extruder. The screw configuration was a combination of pumping and kneading elements that allows melting ofthe rubber binder and mixing ofthe monomer solution and plastisizer into the binder matrix. Throughput was 20 pounds per hour (9.1 kilograms per hour) at 300 RPM.

The extrusion procedure per se which was utilized in the Examples below is described in U.S. Patent 5,135,837. Model Development and Usage A model to estimate the chemical composition is needed to do on-line control. This model was generated using a standard chemometric method, partial least squares, (PLS). A description of this method is found in the book by K. R. Beebe, R. J. Pell, and M. B. Seasholtz, Chemometrics - A Practical Guide. John Wiley & Sons, 1998, ISBN 0-471-123451-6, pp. 278-339.

Hardness was measured offline for the sake of model development. A model is needed to give on-line predictions ofthe composition and from that a hardness value that can then be used to control this product property. To generate the data model, a series of experiments were run on the above mentioned extruder. The feed rate was held constant at a constant 20 pounds per hour (9.1 kilograms per hour), but the ratio or rubber to monomer was varied as 100/0 - 90/10 - 80/20 - 70/30 - 60/40 - 50/50 - 40/60 - 30/70 -20/80 -10/90 -0/100. At each ofthe states, a pulse test was run with the exit pressure and barrel temperature measured by transducers 6 in Firgure 1, motor amps and screw revolutions per minute signals being sampled at 20 samples per second for a total of four minutes. The data were processed and a chemometric model was generated to predict the composition ratio based on only the amperage response to the pulse. The signals, other than motor amperage, did not markedly improve the performance in this case, but can be utilized, if desired, in other applications.

This model is used in controlling the extruder. For example, suppose that the ratio of rubber to monomer is unknown. A sequence of six pulse experiments is run and the motor amp signal is applied to the model. The average ofthe predicted composition for each ofthe six runs is used as the product composition. That number is compared to the desired composition ratio that is needed to produce the required hardness. Based on the difference between the desired composition and the estimated one, a correction is made to the flow rates of both the rubber and monomer so that the total flow remains constant but the composition ratio changes to the desired one. Signal Processing and Model Development

In the Examples below, the following steps were used in the model development. All software is in MATLAB, Version 5.2, sold by The Mathworks, Inc. 21 Elliot Street, South Natick, MA. 01769.

1) Data consisting ofthe signals (motor amps, exit pressure, melt temperature, motor revolutions per second (RPM)) were sampled at 20 samples per second for a total of 10 minutes by the data collection computer (data acquisition hardware 9 in Figure 1) The model was generated in the following way. The data were collected for different levels of oil in the product. Each file consisted often minutes worth of data for oil levels that spanned the nominal ±10%. There were a total of 5 levels: nominal, nominal +5%, nominal -5%, nominal +10%, nominal -10%. The data were processed by a program that a) use a median filter to eliminate outliers and some noise, b) decimate the signal for a sampling frequency of 20 Hertz to 2 Hertz, c) eliminate the mean value ofthe signal, d) use wavelet techniques to denoise or filter the signal, e) eliminate any trend that is the data, f) compute the spectral characteristics using FFT, and g) generate the energy in two bands: low, medium and determine the maximum ofthe spectral characteristics and the corresponding period. The Matlab code below is used to do these functions.

The Matlab code for preprocessing these data is as follows:

The "%" sign at the beginning ofthe line indicates a comment which explains the code, rather than a programming step. % function that aligns the pulses so that the beginning ofthe pulse location matches

% closely as possible.

% Code to do the noise filtering. clear lowfreq midfreq period maxfreq % Fixed parameters (Set in GUI and passed into function call): fft_size = 4096; low_freq_limit = 0.05; high_freq_limit = 0.35; wavelet_level_n = 5; median_filter_MWTN = 3 ; knt=0;

[kn,km]=size(name) ;

% is=[2 4 6 7];

% is =[2 3 4 5 6 7 9 10 11]; % texdat=[' AQS Con KB ';' AQS KBF KB ';' AQS Con TME';' AQS TME

TME';' AQS TME TME']; for kl=l :kn; eval(['load ', name(kl,:)]); x = Pulse_Data_Matrix; tit =Tag_Names';

% xname=strcat('X',name(kl,l :8));

% eval([* x =',xname,';']);

Fs=l/T; [kll kmm]=size(is); for mn=l:kmm

% knt=knt+l;

% figure(knt) xnn =x(:,is(mn)); ynn=medfiltl (x(:,is(mn)), median_filter_MWTN); y=decimate(ynn,round(Fs/2)) ;

% plot(y)

% title([name(kl,:) tit(mn,:)]) MWIN = median_filter_MWTN; mdβ>7 = medfiltl(y,median ilter_MWIN)*; len=length(mdf97);

% L = 4;

% QMF8 = MakeONFilter('Symmlet',8); m97 = mean(mdf97); dm97 = mdf97 - m97;

Ntot=length(dm97);

% dm97= mdf97;

% dmp97 = SymPadO(mdf97-mean(mdf97)); % scale97 = NormNoise(dmp97,QMF8);

% sfac97 = mean(max(dmp97)/max(scale97));

% [xscale97,wscale97] = WaveShrink(scale97,Nisu',L,QMF8);

% xdmp97 = sfac97*xscale97; % wen = wc;

% wcn(l:4)=zeros(l,4);

% en = xscale97(l :len)-mdf97;

% err=err-mean(err);

% dat=en(512:1023); % xn = iwt_po(wcn,L,QMF8);

% if(length(xn)>512),

% xnl=xn(l:1024)-mean(xn(l:1024));

% xn=xnl ;

% else, % xnl=xn(l:512)-mean(xn(l:512));

% xn=xnl;

% end lev = wavelet_level_n; xn = wden(dm97,'heursureVs',One',lev,'sym8'); tn=0:Ntot-l; tn=tn-mean(tn); xn=xn(l :Ntot); a = sum(xn.*tn)/(tn*tri); dat = xn-a*tn;

% plot(dat)

% title([name(kl,:) tit(is(mn),:)])

% pause w = fft(dat-mean(dat),fft_size); f = 0:2/fft_size:l-2/fft_size; wmags(mn, : )=abs(w( 1 : (fft_size/2)))/Ntot; wmag=wmags(mn, :); knt=knt+l; figure plot(f,wmag) title([name(kl,:) tit(is(mn),:)])

[maxfreq(kl,mn),loc] =max(wmag(find(f<=high_freq_limit))); maxfreq(kl,mn)=maxfreq(kl,mn); period(kl,mn)=l/f(loc); lowfreqind = find(f<=low_freq_limit); midfreqind = find(f >low_freq_limit & f< high_freq_limit); lowfreq(kl,mn)=sum(wmag(lowfreqind)); midfreq(kl,mn)=sum(wmag(midfreqind)) ; text(0.3, max(wmag)/3,[num2str(lowfreq(kl,mn)),'

',num2str(midfreq(kl,mn)),' ', num2str(maxfreq(kl,mn)),' ', num2str(period(kl,mn))])

% text(0.3, max(wmag)/2,texinfo(kl,:));

% plot(wcn(l :128)) % pause end clear Engg_Units Fs Ntot Pulse_Data_Matrix Tag_Names a dat clear dm97 f mdf97tn w wmag wmags x xn xnn yn ynn

% close all end

Data, typically motor amps and exit pressure, are collected off the extruder at 20 samples per second. Data are collected for a total of 10 minutes without interrupting the process in anyway. In the practice of this invention, the data are nedian filtered using function "medfiltl" used by software Matlab® by The Mathworks, Inc. The filetered data is decimated to 2 samples per second using the Matlab command "decimate". The decimated data are median filtered on additional time for added noise reduction. Next, the mean value ofthe signal is removed from the data. This is done using the Matlab function, "mean". The mean conected signal is then filtered using a function "wden" found in the Wavelet Toolbox ofthe Matlab software. This technique is refened to as "denoising" in the literature. Although there are many techniques for correcting the baseline, a straight line approach is used herein. Since we are working with steady-state data, a straight line is fitted to entire data, and the observed data are subtracted from the evaluation ofthe straight line at the same conesponding times to produce a baseline conected data. These data are then used in the Matlab "fft" function to generate the spectral characterization. The energy in different bands are evaluated for further processing. 2). The model was built using multi-linear regression on the low frequency energy. Evaluations with other characterizations ofthe variations did not conelate well with the properties. The low frequency energy did. These data, that consist ofthe low frequency energy in the amperage and pressure signals were normalized using the mean and standard deviations by subtracting the mean and dividing by the standard deviation. The multi-linear regression method provided coefficients that predicted the change in the oil from the target value.

The model was implemented in the examples below in the following way. The amperage and pressure were collected and process with the Matlab script given above. The low frequency energy is normalized with the previously calculated mean and standard deviation. The normalized results were used in the multi-linear regression model to give a prediction ofthe deviation ofthe oil. That value then defined the change in oil needed to get back to a prescribed level.

EXAMPLE 1 Compounding of Rubber Binder and Acrylic Monomer An extrusion was performed by mixing two rubber binders with monomer and a polybutadiene oil at 120 pounds per hour (PPH) and 300 RPM. This was done on a ZSK 40 extruder, a commercially available twin-screw extruder manufactured by Krupp, Werner & Pfleiderer Corp., Ramsey, New Jersey. Binder #1 - Styrene-butadiene-styrene block copolymer Kraton® DX1000; Shell Chemical Co., Houston, TX

Binder #2 - Styrene-butadiene block copolymer

Kraton® 2105; Shell Chemical Co., Houston, TX Polybutadiene Oil - 1,4-Polybutadiene, MW3000 Monomer- Composed of by Wt. %

75.2% HMDA - hexamethylenediol diacrylate - 16.2% Initiator - 2-phenyl-2,2-dimethoxy acetophenone 5.8% Inhibitor - butyrated hydroxy toluene 2.7% HEMA - hydroxy ethyl methacrylate

0.1% Dye - Neozapon® Red dye; BASF Wyandotte Corp., Holland, MI Component Wt %

Binder #1 61.9

Binder #2 10.6 Polybutadiene Oil 18.8

Monomer 8.7

The banel temperatures (in Celcius) ofthe extruder were set at: BRL # 1 , #2 , #3 , #4 , #5 , # 6 , #7 , # 8 , # 9 , Die 180 , 180 , 160 , 150 , 140 , 120 , 100 , 100 , 100 , 100

The objective of this test was to control the polybutadiene oil content ofthe extruded polymer. The oil content determines the final polymer properties. The target oil content was 18.8%, and that conesponded to a targeted oil flow rate of 85.5 grams per minute at the first feed port, and 85.5 grams per minute at the second oil feed port. The results ofthe test are summarized below:

New definition of K:

K = target Oil flow rate - predicted Oil flow rate

1. Target Oil rate = 85.5 g/min .

Run Pred. Composition Obs. Composition K

1 82.5g/min +3.0g/min

2 81.25 +4/2

3 85.5 85.2

Final 85.2

Final Oil feed rates were 0.3 g/min below the targeted 85.5 g/min. The final product conesponds to a oil content] of 18.7%. Run 2. Target Oil rate = 85.5 g/min.

Run Pred. Composition Obs. Composition K

1 98.2g/min -12.8-5.1 g/min

2 83.5 +2.0+0.8

3 85.5 82.2 Final 82.2

Final Oil feed rates were 3.2 g/min below the targeted 85.5 g/min. The final product conesponds to a oil content of 18.1%.

Run 3. Target Oil rate = 85.5 g/min.

Run Pred. Composition Obs. Composition K

1 99.0 g/min -13.5 g/min

2 84.7 +0.75

3 Final 87.8

Final Oil feed rates were 2.3 g/min above the targeted 85.5 g/min. The final product conesponds to a oil content of 19.3%.

Claims

Claims:

1. In a computer monitored continuous extrusion process for controlling the properties of an extruded material, wherein (1) variable signals generated by an extruder drive and control system are transmitted to a data acquisition signal conditioning front end hardware, which isolates, amplifies and filters the variable signals and outputs variable analog signals, (2) the variable analog signals output by the data acquisition system signal conditioning front end hardware are converted to variable digital signals, and (3) the variable digital signals are transmitted to a control computer which calculates setpoints, which can be transmitted back to the extruder drive and control system or a feed control system used by the extrusion process, the improvement comprising selecting at least one variable signal having a coefficient of variability of greater than 0.05 and conelating said variable signal with variations in at least one property of the extruded material.