Adaptive modulation time-domain fault diagnosis of three-phase induction machines

This paper proposes a new time-domain method to detect all four major types of induction machine faults without added sensing or model information other than that needed for closed-loop regulatory control. The proposed method applies artificial modulating signals, which are self-adapting to contain the estimated fault frequencies based on the real-time machine speed and torque, to the stator current feedback. However, these signals are not part of the closed loop and therefore do not interfere with the stator current's harmonics. When a fault occurs, there is a very slow component, or ideally DC, in the product resultant of modulating and current feedback signals. The proposed method identifies this very low-frequency fault-indicative signal component by checking the time between the experimental or simulation runtime and the instance of the last zero-crossing point of the resulting signal. Substantial simulation tests have been performed to detect air-gap eccentricity and broken rotor bar faults under different speed, torque and noise conditions to check the successful detection rate of the proposed method. Moreover, the method has been verified with a hybrid experiment, in which virtual fault signals are injected in a real motor drive system using dSPACE DS1104 platform. The proposed method shows excellent fault sensitivity and is robust to noise, current harmonics and possible DC offset in current feedback.