Simulation-assisted Training of Neural Networks for Condition Monitoring of Electrical Drives: Approach and Proof of Concept

One crucial aspect of data based modeling is the availability of a sufficient amount of proper data. In the context of AI systems used for condition monitoring of electrical drives, to predict certain faulty conditions, also the corresponding faulty real world data has to be provided to teach an AI based condition monitoring system. But this is most likely linked to an enormous effort. In this paper an approach is presented, how such a condition monitoring system can be created by mainly using simulation data and mapping the simulation domain to the real world domain using fault-free measurements, which are usually easily accessible. After presenting the concept of simulation assisted training, prediction of a commutation angle error of a block-commutated 280W motor will serve to prove the concept.