Data-driven learning assumes that the collected data are independent and identically distributed.
However, data from industrial facilities cannot be trusted to follow this assumption.
The reason is that the equipment is operated under various operating conditions (irregular load patterns, different environmental conditions).
Anomaly detection in a variety of operating conditions can be a very challenging problem.
For example, when data from different conditions are mixed together, similar data points can cause outliers to be ignored or false alarms to occur.
The goal of our laboratory is to perform robust anomaly detection under various operating conditions, and we are conducting research related to this.