A two-phase approach for model-based predictive control to improve process quality

The proposed model is developed to control production processes with highly varying input variables and stringent quality requirements for the process output. The input variables include, for example, changes in the quality of raw materials or wear-and-tear effects of the tools used. We present a two-phase modelling approach for dynamically adapting controllable machine parameters in production processes to improve the quality of the production output and reduce the variability of the process. In the first phase, a hybrid model combining linear regression and time-series analysis is developed to account for how feature-based influences and temporal dependencies–such as wear and tear or environmental fluctuations–affect the output. The second phase involves model-based predictive control to achieve target output values by optimising control parameters. The framework is validated using a simulated manufacturing scenario, demonstrating its ability to maintain dimensional tolerances while adapting to temporal variations. The results show improved forecasting accuracy and control precision, demonstrating that the hybrid model performs better than traditional regression methods. The achieved control precision has the same order of magnitude as the training error for the prediction model. This scalable solution ensures consistent product quality and reduces process variability, making it applicable to diverse production environments.