Process monitoring for chemical process based on semi-supervised principal component analysis

The performances of methods based on principal component analysis (PCA) for process monitoring can degrade quickly when the abnormal samples included in samples for modeling. However, the labels of the process samples are difficult to obtain. Usually, we have many unlabelled samples and small labeled samples. In this paper, the semi-supervised PCA (SSPCA) is proposed by combining both labeled and unlabelled samples. The fault detection based on SSPCA is researched in this paper. The methodology presented was assessed on the Tennessee Eastman Process (TEP) benchmark. These results demonstrate the validity and superiority of this method and the promising potential for the diagnosis of industrial applications.