A new Approach for Electrical Machine Winding Insulation Monitoring by Means of High Frequency Parametric modelling

This paper investigates a new approach for stator winding insulation diagnosis. Unlike conventional methods such as insulation resistance tests or partial discharge monitoring, this approach is based on the effective use of parametric modelling and identification. The guiding principle consists in modelling the high-frequency input-output behavior of the winding ground-wall insulation system. The originality and the difficulty of our work are to find the best tradeoff between simplicity for system identifiability and complexity for diagnosis abilities. More specifically, we try to distinguish between winding pollution (moisture, coal or metallic dust) and insulation material ageing because the maintenance procedures and their induced costs are very different. Experiments have been carried out in order to appreciate the feasibility of this approach. A high voltage MOSFET pulse generator with ultra fast rise time has been specially designed. It allows to apply an input excitation covering the wide frequency range required for the ground-wall insulation system identification. In order to isolate numerous phenomena which could be mixed together in the whole machine, the experiments first deal with one bar and one coil taken from a 5 kV induction machine. Diagnosis model parameters are identified by means of advanced identification algorithms with different moisture contents. The first results are promising