Bayesian optimized adaptive control of injection molding machines for plastic recycling material

This study proposes a Bayesian-optimized adaptive model predictive control (MPC) strategy for the injection port drive motor of an all-electric injection molding machine using recycled plastic materials. To address the uncertainties introduced by variable properties of recycled plastics, a control framework integrating MPC with Bayesian optimization (BO) is developed. A Gaussian Process (GP) surrogate model, combined with the Upper Confidence Bound (UCB) acquisition strategy, dynamically tunes PID parameters to ensure robust performance under fluctuating material conditions. The proposed method adaptively regulates key process variables—including position, speed, pressure, and temperature—based on real-time feedback, supported by an MPC loss function and performance verification indices. Experimental evaluations demonstrate that the MPC-BO method achieves rapid convergence, high adaptability, and efficient computational cost compared to conventional adaptive controllers. In experiment results, the control accuracy of the proposed control system can be improved by 12% over RBF-PID and 25% over fuzzy PID, while maintaining computational efficiency equivalent to RBF-PID and reducing computational cost by 70% compared to DT-APID. In terms of convergence success rate, MPC-BO-PID achieves 90%, outperforming traditional MPC (72%) and approaching RL-MPC-PID (92%). These results confirm that the proposed MPC-BO framework provides a stable and reliable adaptive control scheme, balancing accuracy, adaptability, and efficiency, and offering a practical solution for sustainable all-electric molding systems using recycled materials.