A data-driven multidimensional visualization technique for process fault detection and diagnosis

Abstract

A multidimensional visualization technique is described for fault detection and diagnosis of a multivariate process by principal component analysis (PCA) of historical data. The technique uses a parallel coordinate system to visualize data that allows for monitoring of abnormal process events that lead to process faults, enabling the visualization of multiple principal components effectively and facilitating the study of how each principal component varies with respect to time. The principal component and residual space control limits are established for fault detection and the Random Forests machine learning tool is adopted for fault diagnosis. The key features of the methodology are demonstrated through a study of the benchmark Tennessee Eastman process.