Automatic real-time fault detection for industrial assets using metasensors

Large-scale industrial plants require physical sensors to continuously measure quantities such as temperatures or pressures. A large number of sensors is required to accurately describe the operating state of the plant, which unfortunately makes it very difficult for them to be effectively monitored by human operators. In this work we present a method to construct so-called metasensors, virtual sensors that compress the information from several sensors in an optimal manner. These metasensors are used as inputs to a novel anomaly detection system that automatically alerts operators to abnormal operation behaviour.