On the learning of timing behavior for anomaly detection in cyber-physical production systems

Model-based anomaly detection approaches by now have established themselves in the field of engineering sciences. Algorithms from the field of artificial intelligence and machine learning are used to identify a model automatically based on observations. Many algorithms have been developed to manage different tasks such as monitoring and diagnosis. However, the usage of the factor of time in modeling formalisms has not yet been duly investigated, though many systems are dependent on time. In this paper, we evaluate the requirements of the factor of time on the modeling formalisms and the suitability for automatic identification. Based on these features, which classify the timing modeling formalisms, we classify the formalisms concerning their suitability for automatic identification and the use of the identified models for the diagnosis in Cyber-Physical Production Systems (CPPS). We argue the reasons for choosing timed automata for this task and propose a new timing learning method, which differs from existing approaches and we proof the enhanced calculation runtime. The presentation of a use case in a real plant set up completes this paper