Meta-Modeling and Robust Parameter Design of Multi-Response Gaussian Processes with Mixed Variables  
Author Huiting Xu


Co-Author(s) Jianjun Wang; Zebiao Feng; Cuihong Zhai


Abstract To address the problem of response surface modeling and robust parameter design for mixed variables, in other words, qualitative and quantitative process parameters data, this paper considers the meta-modeling with the unrestricted correlation structures based on multi-response Gaussian process. Specifically, we perform design of experiment according to the maximum projection criterion and then conduct experiments to collect the response values of interest at first. Then, we will use an alternative representation of the unrestricted correlation structures thereby incorporating the calculation of correlation between qualitative and quantitative variables into the framework of a multi-response Gaussian process model. Besides, we construct the quality loss function as the optimization index based on the prediction function. Finally, the genetic algorithm to solve the mixed-integer nonlinear programming problem is a good tool to find the target value that minimizes the quality loss function in the feasible region. The actual case study based on Jiangsu Province Engineering Research Center of Quality Improvement for High-end Equipment shows that compared with other Gaussian process models, the proposed method can not only effectively fit the response surface model between process parameters with mixed variables and multi-objective quality characteristics, but also obtain more robust optimization results.


Keywords Multi-response Gaussian Process, Mixed Variables, Unrestricted Correlation Structures, Indicator Variables, Robust Parameter Design
    Article #:  RQD27-12

Proceedings of 27th ISSAT International Conference on Reliability & Quality in Design
Virtual Event

August 4-6, 2022