Anomaly detection without a pre-existing formal model: Application to an industrial manufacturing system

Some faults in manufacturing systems that are evident in event-based data cannot be detected by existing solutions. This paper summarizes a method for identifying anomalies in event-based data using model generation. The method is based on knowledge of events and resources of the system and generates a set of Petri Net models to detect the anomalies. The method is applied to an industrial machining cell that has been experiencing a gantry waiting problem. The anomaly detection solution is able to accurately identify the gantry waiting anomaly and another anomaly that occurred right before the gantry waiting issue, indicating a possible cause.