Kantorovich Distance based Fault Detection Scheme: An Application to Wastewater Treatment Plant

A novel data-driven fault detection scheme based on Kantorovich Distance (KD) is proposed for monitoring sensor faults in wastewater treatment plant (WWTP). Since WWTP is highly dynamic in nature, the dynamic principal component analysis (DPCA) modeling framework is used to incorporate dynamics of the process. In this paper, the Kantorovich Distance metric is combined with dynamic principal component analysis modeling framework. The KD metric computes the difference between two data sets and uses the difference as a measure of fault. The KD metric is computed between the residuals of normally operating data and the abnormal data. The effectiveness of KD fault detection metric is compared with T², Q and generalized likelihood ratio(GLR) based fault indicators to detect bias, intermittent and drift faults in WWTP benchmark. The simulation results indicates the superiority of KD metric over T², Q and GLR based fault indicators.