Abstract:

To address the problem of robust parameter design for high-dimensional experimental data, the parallel partial Gaussian process (PPGP) model is adopted to construct the response surfaces between the test factors and the multivariate quality characteristics under the modeling framework of Gaussian process (GP). On this basis, the multivariate quality loss function is used as an optimization index to obtain the optimal parameter design values of the controllable factors. A classic simulation examples and two actual cases are used to verify the effectiveness and advantages of the proposed method. The research results show that compared with the independent modeling of the univariate GP model or the Kriging model, the proposed method can not only deal with the modeling and parameter optimization problems of high-dimensional experimental data effectively, but also obtain more robust optimization results and higher operational efficiency.

Key words: Gaussian process (GP), high-dimensional data, multivariate quality loss function, robust parameter design