Mahalanobis Distance Based Dynamic Time Warping for Fault Detection

A data-driven fault detection framework based on MDDTW

We establish a novel data-driven fault detection framework for industrial processes, in which multivariate time series are used to represent the dynamic features of the measurement signals, and a multivariate dynamic time warping
method based on Mahalanobis distance is proposed. In order to obtain the Mahalanobis distance function, we propose a oneclass metric learning algorithm, which learns a distance metric where the normal samples have concentrated distribution while the faulty samples are far away from normal samples. The distinct boundary between normal and faulty signals helps to improve the fault detection performance. In the program, the TE process is used to verify the proposed data-driven fault detection method.