Sensor fault identification and isolation for multivariate non-Gaussian processes

This paper addresses fault identification and isolation of multivariate processes for which the recorded variables follow non-Gaussian distributions. Recent work has demonstrated the effectiveness of independent component analysis to extract non-Gaussian source signal and support vector data description to determine control limits for associated monitoring statistics. This article extends this work by developing a fault reconstruction technique and introduces a fault identification index to diagnose abnormal process conditions. The utility of this work is demonstrated using a simulation example and the application to the Tennessee Eastman benchmark simulator.