Research on Multivariable System Identification and Optimal Control Mathematical Model for Manufacturing Process Control

Abstract

Aiming at the complexity of dynamic characteristics, control target conflict and environmental disturbance in manufacturing process control, a hybrid modeling method combining subspace identification and sparse learning is proposed, and a multi-objective robust optimal control algorithm is designed. The hybrid modeling method effectively reduces the risk of over-fitting of high-dimensional data, and the simulation results show that the model order can be reduced by 40% while maintaining the fitting accuracy of more than 95%. The multi-objective robust optimization control algorithm transforms the nonconvex problem into semi-definite programming by introducing relaxation variables, which shortens the molding cycle and reduces the rejection rate in injection molding cases. The experimental results show that the application of this method in semiconductor etching equipment and injection molding machine effectively improves the control performance and shows good real-time and robustness.