Condition monitoring of induction motors : The detection of broken rotor bars in variable speed induction motor devices

The squirrel cage induction motor is the most common means of converting electrical energy to mechanical energy. As such they form a very important part of modem industrial plants. Adverse service conditions may cause faults to develop within a motor that eventually result in the motor failing. If warning of an impending failure can be obtained, the motor can be scheduled for repair or replacement before catastrophic failure occurs, thus avoiding costly excess downtime of plant. A fault which occurs in cage induction motors, is where a fracture occurs between the end ring and a rotor bar, or in an end-ring segment. These faults may be detected by examining the frequency spectrum of the stator current, while the motor is operating under loaded conditions, for the presence of characteristic frequency components. The basic theory is reasonably well known, however little work has been done on detecting faults when the motor is controlled by a variablefrequency drive, which causes extra frequency components to appear in the stator current spectrum. A variable speed drive controls the speed of an induction motor by changing the frequency of the supply voltage. Thus the problem of detecting faults becomes one of analysing a non-stationary signal. One approach to solve this problem is to synchronously sample the stator current waveform, such that the sampling process tracks any change in frequency, producing a useful frequency spectrum. A hardware system based on a phase locked loop circuit is developed in order to implement such a process. In order to determine which frequencies are produced by a pulse-width modulated (PWM) drive, a theoretical analysis of various PWM methods is carried out, with particular reference to fault frequency components. The change in frequency component amplitudes between mains operation and VSD operation and also with change in load is also examined experimentally. The uncertainty in amplitude due to the signal processing techniques employed, is also determined by experiment. The effects of changes in frequency component amplitudes on the detection of faults is discussed. A full-transient model for the induction motor is developed as an assembly of inductively coupled coils using a model that can represent the effects of individual rotor bars. The effect of a broken rotor bar on the frequencies that are introduced into the supply current can then be predicted. Finally, the possible application of time-frequency and continuous wavelet transform analysis to the problem of a non-stationary signal is examined. Various types of transform are compared to find the most suitable for tracking frequency components as they change