DATA-DRIVEN MODELING AND CONTROL FOR TIME-VARYING MULTISTAGE MANUFACTURING PROCESSES

Abstract

This thesis aims to develop a unified data-driven process modeling and control framework for quality improvement of nonlinear and time-varying Multistage Manufacturing Processes (MMPs). We first investigate the impact of modeling accuracy on the residual controls which are acknowledged as the main techniques for process monitoring of MMPs. The results confirm the importance of process modeling in the quality assurance activities. Gaussian Process Regression (GPR) based online modeling methods are further proposed to address the nonlinearity and time-variant properties of a manufacturing process. The applicability of the online GPR model in the model-based active control is also explored. Lastly, we extend the online GPR models for the dynamic and complex MMPs modeling. A hybrid control strategy that integrates both statistical process control and online GPR model based active control is further developed. The case studies confirm the online GPR method based control scheme as an effective error compensation tool for nonlinear and time-varying MMPs.