Fault Diagnosis in Industrial Process Based on Locality Preserving Projections

Abstract

A new fault diagnosis based on Locality Preserving Projections (LPP) is proposed in this paper. The recently developed LPP is a linear dimensionality reduction technique for preserving the neighborhood structure of the data set. It is characterized by capturing the intrinsic structure of the observed data and finding more meaningful low-dimensional information hidden in the high-dimensional observations compared with Principal Component Analysis (PCA). In this study, LPP is used to extract the intrinsic geometrical structure of the process data. The Squared Prediction Error (SPE or Q) and Hotelling's T <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> statistics charts for monitoring are used to detect the diagnosis. The reasons that arouse the faults can be found out by the SPE contribution plot of the process variables. The effectiveness and advantages of the LPP monitoring approach are tested with the data based on a Tennessee Eastman (TE) process.