Key-performance-indicator-related Fault Diagnosis In Case Of Outliers Based On Partial Least Square

In the industry process,fault diagnosis technique is brought to the huge attention of the production personnel.People usually need to focus on the key performance indicators(KPI)changes of which the study is called the KPI-related fault diagnosis.The partial least square(PLS)method in the multivariate statistical fault diagnosis has shown remarkable results in the KPI-related fault diagnosis.The KPI-related fault diagnosis methods based on PLS has been widely studied in the recent years.However,standard PLS cannot be directly used to diagnose the KPI-related faults,and there are some problems in the current modified PLS methods.Based on the previous study,this paper proposes a new modified KPI-related fault diagnosis algorithm based on the standard PLS.At the same time,the proposed algorithm is improved in case of outliers in the process measurements.The main contents are as follows:This paper introduces the classification of current fault diagnosis technologies and their respective advantages and disadvantages.This paper sums up the current modified fault diagnosis algorithms based on PLS,which can be divided into preprocessing PLS and post-processing PLS.The research progress of PLS with outliers is introduced particularly.On the basis of T-PLS and IPLS,this paper explains the measurement space partition of the standard PLS.Then it introduces the respective relationship of different partition between the measurement variables and quality variables.In terms of the KPI-related fault diagnosis,both T-PLS and IPLS have problems in the processing of the residual space of PLS.To guarantee the simplicity of fault diagnosis logic,this paper proposes a fault diagnosis method of PLS,namely KPI-PLS,which has a more reasonable space division.Comparing with IPLS in a numerical example and TE process,the correctness and effectiveness of KPI-PLS have been verified.This paper combines the modified partial robust M-regression(MPRM)with KPI-PLS algorithm to come up with a modified PLS algorithm in case of outliers,which is named as KPI-APLS.The correctness of KPI-APLS has be verified through the simulation on the TE process.