Multivariable Model-Based Predictive Control for Injection Molding

Abstract

The rigorous quality criterion and intricate shapes of plastic injection molded parts require molders to improve process control systems in order to keep their competitive status in the market. In recent research, various advanced control algorithms are employed to develop in-line process controllers. In modem controllers design, in-mold process variables play a very important role in connecting machine variables and quality variables. Model-based predictive control (MPC) is used to investigate the controllability of cavity pressure and cavity temperature within a cycle or cycle-to-cycle. The objective of the present work is to demonstrate a procedure to develop MPC controllers based on simulation results. Moldflow® was used to simulate the injection molding process for a thin-wall cell phone cover. Cavity pressure profiles and part surface temperature profiles were extracted to develop the dynamic model for controller design. Thermal analysis for the cooling stage was investigated by ANSYS® FEM software. Mold surface temperature profiles were used for controller design. Dynamic matrix control, a type of MPC control, was developed by using Matlab® MPC Toolbox. A single-input/single-output MPC controller was developed to control cavity pressure in filling stage by manipulating injection flow rate. Simulation studies were then used to develop a MPC controller to implement a closed-loop control. The controller performed very well to control the pressure profile to trace the set-point, even with melt temperature or mold temperature change. Two MPC controllers were developed to control cavity surface cycle average temperature by manipulating coolant flow rate and coolant temperature. Both controllers show good controllability for cycle average temperature control. A two-input/two-output DMC controller was implemented to control cavity pressure and part surface temperature in the packing stage. Packing pressure and mold temperature were manipulated to trace the controlled profile set-points in each sampling time. Results shows that the controller was able to meet the set-point very well, for an unmeasured disturbance, based on a closed-loop test. All the controllers were developed based on simulation results, which will have some differences with real production data. Therefore, the model parameter and controller tuning parameter should be validated and modified if needed before real-time application.