In this paper, a new method is proposed for incremental identification of Programmable Logic Controller (PLC) controlled tool changing process using available binary event logs obtained from the PLC. The identified discrete event model identified takes the form of a modified Timed Petri Net (TPN). A real time anomaly detection system is then constructed by synchronizing the identified TPN model with the actual tool changing process through the event sequence. Any discrepancies between the model and actual system are recognized as anomalies. The test results show that the diagnostic system automatically constructed using the newly proposed procedure is able to detect anomalies, such as incorrect timing and illegal event sequence. The same procedure has been successfully applied to monitor other PLC controlled automation processes.
Multi-Scale One-Class Recurrent Neural Networks for Discrete Event Sequence Anomaly Detection
