On-Line NIR Sensing of [CO.sub.2] Concentration for Polymer Extrusion Foaming Processes.
An on-line sensor using near infrared (NIR) spectroscopy is developed for monitoring [CO.sub.2] concentration in polymeric extrusion foaming processes. NIR absorption spectra are acquired by a probe installed at the foaming extruder die. The calibration curve relating the absorbance spectrum at 2019 rim to the dissolved gas concentration is derived so as to infer dissolved [CO.sub.2] gas concentration on-line from measured NIR spectra. Experimental results show the developed on-line NIR sensor can successfully estimate dissolved [CO.sub.2] concentration in the molten polymer and illustrate that the developed NIR sensing technique is among the more promising methods for quality control of polymeric extrusion foaming processes.
The use of polymeric foams is rapidly expanding because of their excellent properties such as light weight, high strength/weight ratio, and superior insulating abilities. The production scheme of foamed plastic products can be roughly divided into two categories: extrusion foaming and batch foaming. In extrusion foaming, foams are created in polymer products by dissolving gas into molten polymer at the middle of the extruder as a foaming agent, and then releasing pressure at the exit of the extruder die, where nucleation and subsequent bubble growth occur.
The properties of the foamed products totally depend upon their cellular structure such as bubble sizes and bubble density. One of the key factors determining the cellular structure is the gas concentration dissolved into the molten polymer inside the extruder. However, the current sensing technology does not allow us to measure the dissolved gas concentration on-line at the middle of the extrusion foaming process. The lack of a suitable on-line sensor has kept the practical method of monitoring the dissolved gas concentration in being empirical. That is, in practice, measurement of the concentration is available only for the steady state condition. The feed flow rate of gas to the extruder as well as extrusion rate of polymer are measured at the steady state condition, then, the gas concentration is estimated by the ratio of two measurements. Therefore, the estimates could not be used to control the dissolved gas concentration once the process becomes unstable or changes over time, which often occurs in pra ctical processes.
In the past decades, Freon gases, such as trichlorodifluoroethane (HCFC), have been used as the foaming agents. (Freon is a DuPont product.) However, most of them were banned because of environmental consideration. As a substitute for Freon gases, organic chemicals, such as butane, were once used in industry, but because of their explosiveness, they are not preferred in terms of plant safety. Therefore, recently, the foaming industries are interested in using super-critical carbon dioxide, which is non-explosive and soluble in the polymer, as a blowing agent [1, 2].
The aim of this study is to develop an on-line sensor that can measure [CO.sub.2] concentration in molten polypropylene for [CO.sub.2] extrusion foaming processes.
Near infrared (NIR) spectroscopy analysis has greatly increased sensor abilities with the help of recent advances in chemometrics. NIR spectroscopy has a number of advantages, such as remote data collection ability and ease of sample handling. It has been used for chemical process monitoring. In polymer production plants especially, it is used for on-line measurements of polymer composition , polymer viscosity , and concentration of 1,1,1 -trichlorodifluoroethane (HCFC) in polystyrene .
The following will demonstrate the feasibility of the NIR technique coupled with chemometrics methods to perform on-line measurements of [CO.sub.2] concentration in the molten polypropylene. In order to remove the baseline of NIR spectrum, the wavelet transform is employed. Then, two different calibration curves, which describe the relationship between NIR absorbance spectra and [CO.sub.2] concentration, are developed by simple regression and PLS methods. They are used in real-time estimation of [CO.sub.2] concentration during extrusion foaming.
Figure 1 shows the schematic diagram of the experimental setup of on-line Fourier transform NIR unit (Yokogawa Electric Co., FIR1000L). The extruder used in this study was the one with a twin screw. The material, polypropylene, was provided in pellet form from the hopper. At the middle of the extruder, liquid [CO.sub.2] was introduced to the molten polypropylene. The feed rate of [CO.sub.2] was measured by the change in weight of the [CO.sub.2] supply cylinder. At the exit of the extruder, polymer flow was split into the main and the sideflows by a throttle valve. The flow cell for NIR measuring was installed on the side-flow. Downstream of the flow cell, a gear-pump was equipped to realize a steady flow rate.
A fiber-optic probe was used to transmit NIR light through the polymer running in the flow cell. The optical path length between the probe windows was fixed at 8 mm. The probes were connected via single-fiber optic cables to an Analect Diamond-20 Fourier transform near-infrared spectrometer (FT-NIR). This spectrometer possesses an interferometer that works on the moving-wedge principle. For measurements in the NIR region of the electromagnetic spectrum, the wedges are made up of calcium fluoride ([CaF.sub.2]). The spectrometer uses a quartz halogen lamp as the light source and an indium arsenide (InAs) detector with [CaF.sub.2] windows.
In this study, on-line NIR measurements were performed using a commercial polypropylene (Sumitomo Chemical Y101) with highly purified [CO.sub.2]. The polypropylene was linear homo-polypropylene whose MFR was about 8 g/10 min. The relationship between NIR spectrum and [CO.sub.2] concentration was analyzed from experimental data obtained under nine different conditions as listed in Table 1. The conditions were realized by changing either the feed rate of [CO.sub.2] or flow rate of side-stream. The [CO.sub.2] concentration listed in Table 1 was calculated by
[CO.sub.2] concentration [wt%] =
flow rate of [CO.sub.2] in a steady state [g/min]/total extrusion rate in a steady state [g/min] X 100 (1)
The total extrusion rate is the sum of the flow rate of the main stream and that of the side-stream.
NIR transmission spectra in the wavelength range of 1000 to 2500 nm were obtained by averaging 128 scans. Each scan takes 0.4 sec. The resolution in wavenumber employed in the experiments was 8 [cm.sup.-1].
RESULTS AND DISCUSSION
Relationship Between [CO.sub.2] Concentration and Absorption Spectra
Figure 2 shows the overlaid nine NIR absorption spectra. The enlargement of these nine absorption spectra around wavelength 2000 nm is illustrated in Fig. 3. As illustrated in Fig. 3, the NIR absorbance changes drastically at 2019 nm as the [CO.sub.2] concentration changes. This absorbance comes from the combination of symmetric stretching vibrational mode, asymmetric stretching vibrational mode, and first overtone bending vibrational mode of [CO.sub.2] (6).
These raw spectra data could not be directly used for quantitative analysis because change in the flow rate shifts the baseline of spectra. The absorption spectra of runs No. 1 to 3, where [CO.sub.2] was not introduced but the extrusion rate was changed, are illustrated in Fig. 4. It shows that the baseline was shifted as the polymer flow rate was changed.
Development of Calibration Curve Using Chemometrics
Baseline Treatment by Wavelet Transform
First, the wavelet transform was performed to remove the baseline as well as noise from the original spectra. The basic concept of wavelet transform is to represent the given signal by the linear combination of known functions (basis functions), which is equivalent to the concept of Fourier transform. Wavelets, which are a new family of basis functions, are unique due to their local character in both time and frequency domain. The character enables us to analyze the signals at different levels of resolution in a two-dimensional, time-frequency domain (7). The discrete wavelet transform (DWT) is often used to analyze a regularly sampled data sequence in time. In this study, it was applied to the MR spectrum, which was a regularly sampled data sequence in wavelength. The resulting transformation is thus a wavelength and frequency domain. For example, a discrete signal composed of [2.sup.J] measurements can be decomposed into [2.sup.J-1] coefficients in the space spanned by the scale functions and [2.sup.J-1] coefficients in the complimentary space spanned by the wavelet functions. The coefficients in the domain spanned by the scale functions are called the approximation coefficients and these in the complimentary space are called detail coefficients. By applying the transformation to the approximation coefficients [J.sup.*] times, the [J.sup.*] sets of detail coefficients (i.e., [J.sup.*] different levels of resolution components) and one set of approximation coefficients are obtained as shown in Fig. 5.
The discrete wavelet transform was applied to nine spectra measured under different conditions listed in Table 1. All calculations were performed by MATLAB (8). In this study, the Symlet 4-wavelet function was used because it resembled to the shape of peaks found in NIR spectra.
Since each spectrum was composed of 1555 absorbance measurements covering the wavelength range from 1000 to 2500 nm, it was decomposed by successive application of DWT into 7 different scales of detail and one approximation coefficients. Then, the spectrum data was reconstructed by taking convolution of the detail coefficients from scale 1 to [J.sup.*] as illustrated in Fig. 5.
The correlation coefficients between [CO.sub.2] concentrations and the reconstructed absorption spectra at 2019 nm were calculated by changing [J.sup.*] value as illustrated in Fig. 6. As can be seen in Fig. 6, the correlation coefficients between [CO.sub.2] concentration and the reconstructed absorption spectra at 2019 nm take the highest value when [J.sup.*] is set at 4. Figure 7 and Fig. 8 illustrate the approximation coefficients on the 4th scale and the spectra reconstructed by taking convolution of the detail coefficients from scale 1 to 4, respectively. As illustrated in Fig. 8, the nine spectra, which were measured under rime different conditions (three different [CO.sub.2] concentrations X three different flow rates), converged on the three spectra. Namely, by leaving out the approximation and reconstructing the spectrum, the effect of flow rate could be removed. Furthermore, the reconstructed spectra clearly shows the change in absorbance caused by [CO.sub.2] not only at 2019 nm but also 1972 nm. The absorbance at 1972 nm comes from the combination of the first overtone symmetric stretching vibrational mode and asymmetric stretching vibrational mode of [CO.sub.2] .
The relationship between [CO.sub.2] concentration and the reconstructed spectra of 2019 nm is illustrated in Fig. 9. The calibration curve in the form of a simple regression model was given by
y = 67.9 [x.sub.1] - 0.0855 (2)
where y denotes [CO.sub.2] concentration, and [x.sub.1] is the absorbance of the reconstructed spectrum at 2019 nm. The correlation coefficient between y and x was 0.997.
The calibration curve derived from the reconstructed spectra at 1972 nm was also calculated. The resulted simple regression model was given by
Y = l65 [x.sub.2] - 1.40 (3)
where [x.sub.2] is the intensity of the reconstructed spectrum at 1972 nm. The correlation coefficient was 0.996.
Development of a PLS Model
In the NIR spectroscopic analysis, multivariate techniques such as partial least squares (PLS) and principal component regression (PCR) are often employed (3-5). In this study, the PLS method was applied to improve the prediction accuracy of [CO.sub.2] concentration from the spectra.
The reconstructed spectra of 130 measurements over the wavelength region from 1900 nm to 2100 nm were used for developing a PLS model. The cumulative PRESS values were calculated using the leave-one-out cross-validation for each number of PLS factors (9) and the results were illustrated in Fig. 10. Evaluating the PRESS values, the optimal number of PLS factors was decided to be two. Figure 11 illustrates the correlation between the [CO.sub.2] values predicted by the PLS model and actual steady-state measurement of [CO.sub.2] concentration. The correlation coefficient was 0.9998.
On-line Monitoring of [CO.sub.2] Concentration
Using the obtained simple regression model Eq 2 as well as the PLS model, on-line monitoring of [CO.sub.2] concentration was performed at the extruder. Continuously feeding both polypropylene and [CO.sub.2] to the extruder, NIR were measured on line every 50 seconds. The baseline was removed by DWT as described in the previous section.
The [CO.sub.2] concentration was estimated using two models during the startup operation, where the [CO.sub.2] concentration was changed from 0 to 5 wt%. The change in the reconstructed spectra during the operation is illustrated in Fig. 12. The estimate of [CO.sub.2] concentration for the operation is illustrated in Fig. 13. In Fig. 13, the solid line represents the estimate given by the simple regression model (Eq 2) and the broken line represents the estimate by the PLS model. As can be seen in Fig. 13, both estimates were progressively changed from 0 to 5 wt% like a step response of a first order lag system. The estimates of both models could reach the actual steady-state value of [CO.sub.2] concentration. The differences between two models were not so trivial. In other words, the results show that the simple regression model could provide us enough accuracy in monitoring the [CO.sub.2] on-line.
The Effect of Temperature on NIR-[CO.sub.2] Relationship
In order to examine the effect of temperature as well as the effect of polymer grades on the NIR-[CO.sub.2] relationship, extra experiments, where a different grades of PP was processed by changing the flow cell temperature, were carried out. The polypropylene grade used in this additional experiment was EPR-block polypropylene whose MFR is about 1.8 g/10 min and weight fraction of EPR was about 9.9%. In these experiments, a single screw extruder was used. The NIR flow cell was equipped at the exit of extruder. By controlling the temperature at the flow cell, total extrudate and [CO.sub.2] flow rate, the NIR spectrum was measured under the conditions listed in Table 2. The wavelet transform used to remove the baseline and to reconstruct the spectrum. The obtained NIR-[CO.sub.2] relationship is illustrated in Fig. 14. As can be seen in Fig. 14. the effect of temperature is removed by the wavelet reconstruction.
In this study, a new methodology of on-line monitoring of [CO.sub.2] concentration is developed using NIR spectroscopy. The experiments showed that the NIR spectrum was strongly correlated with the [CO.sub.2] concentration. By performing the wavelet transformation and removing baseline from raw NIR spectrum, the effects of temperature and flow rate were erased. Either the simple regression model or PLS calibration model could estimate [CO.sub.2] concentration dissolved in the molten polymer from the NIR absorption spectrum data. However, from the viewpoint of ease of use, the simple regression model is recommended.
In this paper, the effect of dispersed material such as talc was not stated precisely. When dispersed material is increased, the baseline of spectrum is shifted as we observed at the flow rate effect. Especially, when the pressure is low and bubble nucleation occurs at the point where NIR is measured, the baseline is shifted prominently. Even for those cases, the baseline correction by wavelet's can elucidate the linear relationship between NIR absorbance and [CO.sub.2] concentration. However, there are limitations for using the sensing scheme in practice. If the dispersed material is too much, the incident light is scattered out. Then, the absorbance might become too weak to be analyzed precisely. Furthermore, under the current scheme, the calibration curve should be developed when the polymer and/or the processing unit are changed. However, these limitations do not disqualify the NIR sensing technique from holding great promise as a sensor of [CO.sub.2] concentration for use in controlling extrusion foamin g processes.
(*.) To whom correspondence should be addressed.
(1.) C. B. Park, M. Pan, and A. K. Yeung, Proc. of Cellular Polymers 4th International Conference, 13-1 (1997).
(2.) A. N. Paquet and C. V. Vo, Proc. of Foams 99 Conference. 5 (1999).
(3.) A. Khettry and M. G. Hansen, Polym. Eng. Sci., 36, 1232 (1996).
(4.) M. G. Hansen and S. Vedula, J. Appl. Polym. Sci., 68, 859 (1998).
(5.) Y. Thomas, K. C. Cole, and L. E. Daigneault, J. Cell. Plast., 33, 516 (1997).
(6.) W. J. Moore Physical Chemistry, Fourth Edition, Prentice-Hall, Inc., Englewood Cliffs, New Jersey (1972).
(7.) C. K. Chui, Introduction to Wavelets, Academic Press, Boston (1991).
(8.) D. Doncho, X. Huo, and T. Yu, http://www-stat.stanford.edu/[sim]wavelab/(1999).
(9.) M. A. Sharaf, D. L. Than, and B. R. Kowalski, Chemometrics, John Wiley & Sons, Inc., New York (1986].
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||near infrared|
|Author:||NAGATA, TAKEFUMI; TANIGAKI, MASATAKA; OHSHIMA, MASAHIRO|
|Publication:||Polymer Engineering and Science|
|Date:||Aug 1, 2000|
|Previous Article:||Theoretical Analysis of the Devolatilizing Performance of an Intermeshing Co-Rotating Twin-Screw Extruder With a Type of Square-Channel Flighted...|
|Next Article:||Effect of Supercritical Carbon Dioxide on Morphology Development During Polymer Blending.|