Partial discharge detection techniques in inverter-fed induction motors

Low-voltage induction motors are known to experience premature breakdowns when supplied by power converters (PC) [1, 2]. The impulsive waveforms generated by PC, in fact, may give rise to enhanced electric stresses with respect to sinusoidal supply. In particular we can observe: i) large overvoltages and ringings at each voltage commutation due to impedance mismatch between motor and cable/inverter; ii) uneven potential distribution along the winding turns having the highest voltage drop located in the first turns. This may incept partial discharge (PD) activity, especially in the phase-to-phase insulation which is subjected to the highest overvoltages [3, 4]. Indeed, PD are claimed to be the main cause of accelerated degradation of winding insulation, particularly if manufactured with fully-organic materials [5]. The measurement and analysis of PD activity under pulsed voltages is, thus, an important task for preventing motor insulation failures. With this purpose, off-line and on-line techniques were proposed recently by literature and IEC standards [3, 4, 6]. Off-line techniques provide phase-resolved PD (PRPD) patterns from which defect location could be performed. However, the motor must be removed from service and a HV square generator is needed to feed the machine. On-line detection is clearly cheaper, but PD identification and location become a hard task [4]. A PRPD pattern similar to that obtainable under sinusoidal condition, in fact, cannot be provided easily under high-repetitive variable-width impulses. Maps showing impulse rise time vs. PD amplitude [7] were proposed in the past to collect PD data providing information only about the existence/ absence of PD activity. Recently an effective technique using the Park transform of the supply waveform was introduced [8] to obtain maps with the aim of identifying PD activity under impulse voltages. The interpretation of the maps, however, is not as easy as it could be for a traditional PRPD pattern. The purpose of this paper is to improve result analysis obtained from on-line PD measurements on low voltage inverter-fed motors. In order to strengthen the diagnostic inference and identify PD source location, a method for building and interpreting a PRPD pattern under repetitive impulse voltages is proposed. PD measurements on induction motors were carried out and compared with results obtained on twisted pairs aiming at finding pattern features useful for PD identification. Finally, the bi-dimensional PD map obtained by means of Park transform is discussed to integrate the information coming from PD patterns to achieve more effective diagnosis.