A modified APLS for key performance indicator-related diagnosis in case of outliers

That dataset contains outliers is a popular problem in the multivariate statistical process monitoring. The standard PLS has a modified approach named as APLS to achieve fault diagnosis in case of outliers. It pays attention to the key performance indicators (KPI)-related fault diagnosis. However, the fault diagnosis logic of APLS places the whole residual space into the KPI unrelated part, the residual space still contains information related to KPI. This paper proposes a novel combination approach for KPI fault diagnosis in case of outliers, considering the impact of residual space on the KPI. It inherits the prediction scheme of APLS and modifies the KPI fault diagnosis logic. The effectiveness of the modified approach has been verified through the Tennessee Eastman (TE) benchmark process.