An Executive Summary Monitoring Emulsion Polymerization by Raman Spectroscopy Why process analytical matters to process development R&D. Serena Stephenson, PhD Senior R&D Analytical Manager Kishori Deshpande, PhD Research Scientist Ravi Dixit, PhD Retired Fellow Introduction Analytical scientists work closely with research scientists to qualify and support process development. This manuscript demonstrates the importance of coordinating process development R&D with analytical method development from batch through large-scale manufacturing. The initial goal was to demonstrate the capability of Raman spectroscopy for tracking unreacted monomers and comonomers during an olefin-based polymerization reaction in a batch reactor. Raman spectroscopy (785-nm Raman system from Kaiser Optical Systems Inc.) was selected from among several possible analytical tools because of its capabilities with emulsions. The group s analytical method development process, with initial failures and subsequent successes, demonstrates why analytical development must synchronize with process development to create value for manufacturers. Polyolefins: Structure and Why We Care Polyolefins are extremely versatile thermoplastics present in items ranging from adhesives to shoes to cars. These macromolecules are created from the polymerization of olefin monomer units. Two of the most common polyolefins are polyethylene and polypropylene. Like most polymeric materials, there is a strong structure-function link in polyolefins. The chemical structure of a polyolefin is optimized to impart the desired mechanical and functional material properties. For example, a basic linear polyethylene structure is composed of a long chain of carbon and hydrogen atoms. Meanwhile, branched polyethylene with mixed comonomers offer new and interesting possibilities. For instance, combining ethylene, octene, butene, or other complex olefins in various ways creates different chain lengths and branching configurations, thereby yielding specific polymer properties (see Figure 1). Knowing the chemical structure of the polymer is necessary to ensure that the desired material properties are met. From a process control perspective, it is important to have detailed information about a polyolefin s molecular architecture (such as the weight percentage of components) to have a reliable process and a consistent product. SPONSORED BY
Polymerization Process and Monitoring A process development group from The Dow Chemical Company designed a project to first demonstrate the feasibility of producing polyolefins with higher molecular weights and densities than allowed by currently world-scale processes. The group also wanted to show that its new process could generate these higher molecular weight polyolefins more reliably and with less energy consumption than the current method. The hope was that this process analytical work would one day be used as the foundation of a continuous process in a world-scale plant. Pilot-Scale Process Trials with Raman The group s first step was to investigate a method for continuous analysis in a batch reactor, with the goal of determining the weight percentage of comonomer incorporated into a polymer. Semi-batch process overview. The semi-batch process included a batch reactor and impeller for mixing. In this study, comonomer was added along with solvent and 2 g of ethylene to maintain the correct pressure during the reaction. Agitation began and the mixture was heated to create the optimal conditions for catalysis. Hydrogen was then added and once the set point was reached, the catalyst was injected and the equipment was run for 1 minutes. After agitation stopped and the temperature and pressure decreased, polymer was collected and analyzed (see Figure 2). In the current single-phase solution polymerization process, solvent was selected to solubilize the polymer. By contrast, the goal of the present work was to make the same polymer in a two-phase system using a different solvent that would not solubilize the polymer. Rather, an emulsion would form that would make it very easy to separate the polymer phase from the solvent phase during dump and rinse process. Figure 1: Varieties of plastics. Figure 2: Semi-batch process overview. To monitor the process in situ using an immersion probe to avoid extracting sample from the batch reactor and/or looping it back in; To get insight into multiple components and monitor beyond simply one comonomer incorporation to include monomer and other comonomers as well; To be robust enough for 24/7 operation; and To provide information on both the polymer and solvent phases of the two-phase system. The first option that was considered for achieving these goals was Fourier-transform infrared (FT-IR) attenuated total reflectance (ATR) spectrometry. The system was found to be susceptible to window fouling, which made it impractical for continuous monitoring. Another option was to use an FT-NIR immersion probe, which is most suitable for single-phase systems. However, it gure 2: Semi-Batch Process Overview Instrument selection. The group also had several preferences for the analytical method:
is difficult to account for light loss by scattering or absorption and is not preferable for emulsions. Next, the group considered Raman spectroscopy, which has a history of successful use for aqueous and latex emulsion systems. The use of Raman spectroscopy is not without some challenges, however. With Raman, analysts must account for laser power drift, it is historically more expensive to operate than FTIR or FT-NIR instruments, and it requires more maintenance to upkeep than other instruments. But given its capability with emulsions, the group chose Raman spectroscopy for its work. A Kaiser Optical System Inc. Raman 785-nm system with a computer-controlled laser shutter was used for the analytical semi-batch feasibility trials. The shutter provided an important safety feature, allowing the group to avoid stray laser radiation in a facility where scientists and technicians may be present. In addition, the Kaiser Raman spectrometer software was used to apply partial least squares (PLS) models for quantitation which allowed for flexibility during the experiments. Semi-batch reactor set up. The semi-batch reactor was located inside a walk-in hood enclosure with tubing and controls for raw materials, pressure and temperature monitoring, and safety-release valves. The safety features included a system to vent to the roof in the event of over-pressurization so that ethylene and other flammables would not vent directly into the hood. T h e R a m a n p robe wa s inserted through the top of the reactor in good contact with the reaction mixture (~.25 inches above the top impeller blade and ~1 inch from the wall). The placement of the probe was limited to fixed distances from the side of the reactor due to the availability of ports in the batch reactor vessel. In addition, the impeller speed was found to be fast enough for good mixing, but not so fast that it created a vortex that could eliminate the probe s contact with the mixture. 16 14 12 1 8 6 4 2 3 Raman spectral feasibility and model building studies. The Raman spectral feasibility results using peak area models suggested the peak at 164 cm -1 was ideal for monitoring comonomer s incorporation into the polymer (see Figure 3). The comonomer peak was sharp and well separated from the complex polymer background. These encouraging results led to the next step of performing baseline studies and model building experiments in the batch reactor to optimize the method. The initial batch reactor modeling began with a non-reactive system. Temperature and pressure were set to mimic what would happen if a catalyst was added, without actually adding it. This allowed the group to generate truly steady state Figure 3: Raman spectral feasibility. 3: Raman Spectral Feasibility Figure 4: Comonomer/solvent solutions. Arb 8 7 6 5 4 3 2 1 2 16 1 14
values and develop phase diagrams to determine partitioning between headspace and solution. This information would be helpful in establishing the reference method for the model building. Some experiments in pure solvent were run, looking specifically at the effects of temperature, pressure, and purification beds to understand baseline fluctuation. Comonomer and solvent mixtures were studied at 125 C with 5 psi pressure (obtained by putting a nitrogen pad on the reactor). Figure 4 shows a Raman spectrum that clearly illustrates a decrease in comonomer concentration. Considering peak area by concentration provided nice linearity in the spectra, this system was used as a starting point for solvent concentration studies. Additional studies on an ethylene comonomer solvent mixture were more complicated, but the plot in Figure 5 clearly points to the effect of temperature and pressure on the spectra. Additional studies were run to determine the optimal stirring speed for contact with the tip of the probe. Analytical Model Development Information collected in initial models that tracked comonomer and ethylene concentration during a reaction were used to calculate how much comonomer was incorporated into the polymer by mass balance. Using the univariate peak area comonomer and ethylene models, the group averaged the last 1 points before adding catalyst and the last 1 points before the end of the run, and used that difference to calculate how much comonomer was consumed during the process (see Figure 6). Figure 7 shows one of many runs using this method, comparing the final polymer s weight percent incorporation numbers determined with an offline laboratory reference method against the weight percent comonomer consumed. The group found a poor correlation between the two. Upon further investigation, they determined that the presence of product polymer as well as the levels of ethylene, and Figure 5: Ethylene/comonomer/solvent. Arb 6 5 4 3 2 1 17 Figure 6: Calculation of monomer incorporation. poration even hydrogen, in the reaction vessel affected the results calculated by peak area models. Regrouping, the team moved away from the idea of determining weight percent comonomer consumed, and toward the creation of a model for predicting how much comonomer is present in the polymer and create correlation in that manner. Success with Raman and PLS Modeling The researchers switched to a PLS chemometric modeling approach using the GRAMS PLS-IQ program, and directly predicted the weight percent incorporation based on the Raman spectra. An off-line lab method was used as the reference method at the end of the batches. As shown in Figure 8 this approach simulates how monitoring would be performed in a world-scale continuous process, where it is wt% Eth 165 16 Wt% Comonomer 155 Temp (C) Pres (psi) 5 28.7 17 51 9.3 27.4 125 51 3.7 16 17 51 9.7 15.1 125 53 4.3 1.9 17 5 9.9 1.2 125 52 3 3.9 17 42 3.9 3.9 17 57 9 3.6 125 517
Figure 7: Example process reaction experiment. Figure 8: Tracking by PLS modeling. Calib. Type: PLS-1 Diagnostic: Cross Validation # Regions: 1 # Samples: 54 # Constituents: 1 # Points: 1177 Max. # Factors: 15 # Files Out: 2 File Ordering: Sequential Region 544-172 cm -1 Figure 9: PLS model for comonomer incorporation. SECV 8 1 6.8.6 4.4 2.2 1 2 3 4 5 6 7 8 9 1 11 12 13 14 15 # of Factors.4.38.36.34.32.3.28.26.24.22.2.18.16.14.12.1.8.6.4.2. 16 15 14 13 12 11 1 9 Summary of Fit R2 RSquare RSquare Adj Root Mean Square Error Mean of Response Observations (or Sum Wgts).892859.89798 1.98561 31.36511 54 critical that the operator know the amount of comonomer incorporated in the polymer at a particular point. The PLS model was tested with 54 samples; one region of the spectra contained 1,777 points. The entire region, from 544 cm -1 to 172 cm -1, was used and the results were cross validated. The PLS model generated from Raman spectra was the best approach for predicting the weight percent of comonomer in the polymer, with very clear polymer growth indicators. The PLS model results correlated with the validated method results (see Figure 9). Conclusions This research demonstrated feasibility of direct, in-line, Raman-based monitoring of an emulsion polymerization from laboratory to batch-scale. The PLS model using Raman spectra demonstrated sensitivity to both comonomer and ethylene concentrations and that the group could build a model for measuring comonomer incorporation in the polymer. The method requires further development before it would be suitable for a world-scale manufacturing operation, but the concept and feasibility were successfully demonstrated. This study also demonstrates why the coordination of method development with process R&D is so important. Analytical development must synchronize with process development to have a successful outcome and create value for commercial plants and manufacturing sites.