Data-Driven Integration of Scheduling and Control With Closed-Loop Prediction for Industrial Flexible Production

Industrial flexible production, such as flexible assembly incorporating multiple processing machines, aims to operate in a high-quality and low-energy-consumption fashion. This article proposes a data-driven integration approach for scheduling and control to enhance product quality and energy efficiency. Closed-loop dynamic prediction is crucial for the integration of scheduling and control. Its accurate acquisition remains challenging in unknown nonlinear multimode production processes. First, the dynamic linearization (DL) method is used to transform the unknown nonlinear process into an equivalent linear form suitable for different time scales. Then, a closed-loop dynamic real-time optimization (CL-DRTO) problem is formulated at the scheduling layer to select the appropriate processing modes and assign cost-optimized production objectives to them. The control action for the closed-loop dynamic prediction is generated by a model-free adaptive predictive control. It is designed as a control subproblem based on DL and embedded into the CL-DRTO problem. Finally, the equivalent equality and inequality constraints for the control subproblem are established through the KKT conditions, simplifying the complexity of solving the optimization problem. The feasibility of the data-driven CL-DRTO algorithm is demonstrated through a thread-tapping process. The DL methods achieve a mean absolute error of less than 3%. The diversity of DL structures enables flexible controller design. Compared to open loop DRTO, the feedback adjustment mechanism of CL-DRTO enables the system to achieve the desired value more rapidly and accurately. On average, the CL-DRTO algorithm under single-mode operation ensures production quality with a 6% energy saving. The multimode operation further enhances energy efficiency by flexibly switching between modes.