Condition monitoring of electric-cam mechanisms through motor current signature analysis

Predictive maintenance can be of use to maximise the availability of the system and minimise the maintenance costs. While diagnostics of induction motor systems have been traditionally handled with vibration-based techniques, multiple applications have successfully implemented Motor Current Signature Analysis to diagnose internal faults of the motor. In this application, we apply MCSA to detect faults of the electric driven mechanism, which can also be subject to faults. Condition Monitoring is implemented through Data-Driven methods such as system identification, exploiting the execution of electric cams on PLC’s to monitor the conditions of an electric motor driven mechanism. Auto-Regressive models of the system in different health conditions are estimated through the Least Squares algorithm, using the measured currents as the available data. Once we have computed the Power Spectral Density of each estimated AR model, we compute the spectral distances (Log Spectral Distance, CosH Distance, Itakura-Saito Distance) between the estimated models in faulty and healthy conditions, and the reference (healthy) model. In order to improve the results obtained from the original current signals, we have repeated this procedure on pre-processing currents. In the first pre-processing method, we undersampled the signals to reduce unwanted noise. The second method involved using the wavelet approximation of the signal as a baseline to filter the currents values, exploiting the properties of the Discrete Wavelet Transform. This process was repeated on two different databases, with different constant speed of the virtual master; in both cases we were able to obtain indicators that could allow us to detect and identify the considered faults from future, unseen data.