A Novel Soft Sensor Approach of Polymer Process for Resin Monitoring Based on Melt Pressure Spectrum

In the plastic extrusion process, most of the equipment is not production record and manufacturing information integrated, some production lines are not even installed sensors to monitor basic signals. This will cause the production line not being able to manufacture in optimal mode and cannot troubleshoot when exceptions occur. This research is mainly verified by the pressure instability of the extrusion system, and unmelted resin will lead to extrusion instability. The single-screw extruder is used as the experimental process machine (screw diameter is 25 mm, screw diameter ratio is 26 D, four temperature zones). Since there are many melt properties that cannot be measured by physical sensors, the melt pressure spectrum as input features, and used a regression model to design a soft sensor. To develop a monitoring system to monitor the melt state of the extrusion process, the pressure signal is captured on the adapter flow channel. The spectrum of the pressure signal is obtained using a fast Fourier transform (FFT) after analog signal processing. To apply the spectrum to the analysis extrusion instability, this research extracted the Top-10 frequency components in the spectrum. The machine-learning model of soft sensor predicts the perturbation frequency of the pressure and other related numerical results. The model dataset contains process parameters, variables, and spectral features. The models used are random forest classification and regression. According to the model results, the classification prediction accuracy is at least 0.91, and the
${R}^{{2}}$
score based on the proposed dataset can reach 0.97. The classification and regression results of the proposed model are satisfactory.