Use of data standardization to improve inverter - induction machine fault detection

Intensive research efforts have been focused on the signature analysis (SA) to detect electrical and mechanical fault condition of induction machines. Different signals can be used: voltage, current and flux. The characteristic frequency research by a current spectral analysis is a well-known method widely used. This method is valid when the motor is supplied by the three-phase main network. However nowadays, in industrials applications, the asynchronous motors are more and more supplied by converters, in particular for variable speed. The current spectral analysis is almost not exploitable because of appearance of multiple harmonics of the commutation frequency. This paper presents a diagnosis method applied to a set "converter-machine-load". This method is based on pattern recognition approach. The use of the data standardization makes it possible to free from the level of load and thus to represent an operating mode by only one class. This fact allows decreasing the number of initial data necessary to the training phase and improving the final diagnosis