Data-driven approaches for impending fault detection of industrial systems: a review

Industrial systems operating under harsh and stochastic conditions are vulnerable to anomalies that degrade its performance and subsequently lead to unexpected breakdown. With the advent of Internet of Things (IoT), intelligent sensors have enabled maintenance managers to collect system data and analyze its behavior accurately in real-time. Based on this, many data-driven early fault detection approaches have been developed that try to detect anomalies associated with impending faults. Despite its advantages, only limited and scattered applications of anomaly detection approaches can be seen in system health monitoring of mechanical systems. One of the possible reasons could be scarcity of a comprehensive literature review that presents evolution of the field, highlighting key challenges and open questions to be addressed for future developments. This study narrows this gap by presenting a state-of-art review of data-driven approaches employed to early fault detection of various industrial systems. After critical analysis, challenges found in the previous studies and open questions for future research are also discussed. This study can be a reference point for researchers interested in addressing the critical challenges faced by maintenance practitioners in the industry.