Title: Online Bayesian optimization design-based closed-loop control with model parameter uncertainty and data quality
Authors: Linhan Ouyang - Nanjing University of Aeronautics and Astronautics (China) [presenting]
Abstract: Response surface-based design optimization has been commonly used to seek the optimal input settings in processes or products design problems. Focused on statistical modeling and numerical optimization strategies, most researchers typically assume that there is no model parameter uncertainty in modeling process or the data quality performs well in the online update process. However, if the estimated model parameter varies from the true one or the online observations contain substantial error in quality improvement, the resulting solution may be quite far from the optimal. A Bayesian modeling approach is proposed to closed-loop online optimization design that accounts for model parameter uncertainty and data quality in micro-manufacturing processes. The uniqueness of the proposed approach can provide the information of how and when to update the settings of the design variables based on the online observations. Therefore, it can avoid the danger of over-updating the process if online design approaches are used in quality improvement. The effectiveness of the approach is illustrated with simulation experiments and a micro-milling process. The comparison results demonstrate that the proposed approach with consideration of model parameter uncertainty and data quality can achieve better process performance than conventional design approaches, since it can make corrective adjustments by updating the model parameters or the target value of each run.