Abstract:
The final quality of product is mainly decided by those critical variables in production process,so the quality prediction ability is closely dependent on the selected process variables.This paper proposes a critical-variable-based OPLS prediction method,CV-OPLS model,for the quality prediction of industrial processes with multi output variables.First,according to the selection criteria of critical variables,we choose critical process variables for each quality variable in modeling.In order to reduce the number of final models,the data matrix composed of quality variable and its critical variables is recombined,in which disturbing variation irrelevant with quality variable will be removed by means of OSC method.And then,PLS models are formed on the corrected data matrix,and the regression coefficients are computed such that the final quality prediction results are obtained.Compared with the traditional PLS and OPLS,the proposed method can effectively simplify model structure and attain superior prediction performance.Finally,the feasibility and effectiveness of the CV-OPLS method are further verified through experiments in Tennessee Eastman (TE) process.