Machine learning assisted optimization of blending process of polyphenylene sulfide with elastomer using high speed twin screw extruder.

Random forest regression was applied to optimize the melt-blending process of polyphenylene sulfide (PPS) with poly(ethylene-glycidyl methacrylate-methyl acrylate) (E-GMA-MA) elastomer to improve the Charpy impact strength. A training dataset was constructed using four elastomers with different GMA and MA contents by varying the elastomer content up to 20 wt% and the screw rotation speed of the extruder up to 5000 rpm at a fixed barrel temperature of 300 °C. Besides the controlled parameters, the following measured parameters were incorporated into the descriptors for the regression: motor torque, polymer pressure, and polymer temperatures monitored by infrared-ray thermometers installed at four positions (T1 to T4) as well as the melt viscosity and elastomer particle diameter of the product. The regression without prior knowledge revealed that the polymer temperature T1 just after the first kneading block is an important parameter next to the elastomer content. High impact strength required high elastomer content and T1 below 320 °C. The polymer temperature T1 was much higher than the barrel temperature and increased with the screw speed due to the heat of shear. The overheating caused thermal degradation, leading to a decrease in the melt viscosity and an increase in the particle diameter at high screw speed. We thus reduced the barrel temperature to keep T1 around 310 °C. This increased the impact strength from 58.6 kJ m as the maximum in the training dataset to 65.3 and 69.0 kJ m at elastomer contents of 20 and 30 wt%, respectively.