| Abstract |
Fault diagnostics of induction motor drive system can be achieved online through the Motor Current Signature
Analysis Method. The advantages are obvious: the algorithm can be implemented in the control system scheme with
the existing DSP controller and current transducers. Therefore no extra cost is required. However, owning to the
fact that the monitored signal is rich in harmonics, with frequent dynamics, and the fundamental frequency of the
drive changes within a wide range, the traditional method based on FFT analysis, does not meet the requirement.
A novel online fault diagnostic algorithm for electrical faults of induction motors fed by variable speed drive is
studied. The innovative approach features wavelet analysis and artificial neural network method. A new set of
feature coefficients of the mechanical faults is extracted from the stator current by wavelet packet decomposition.
The features are represented with different frequency resolutions. And because of the wavelet function, such a
feature extraction method can be used for current signal with transients. It is also found that as long as the samples
of each cycle is kept constant, the node numbers of the feature coefficients for the rotor bar breakage will always be
around some of the certain nodes at certain Depths, despite the change of the fundamental frequency. These features
are advantageous for the fault detection for induction motor drive system where there are many transients, rich
harmonics distortion, and variable fundamental frequency. Multiple-layer perceptron network is employed as a tool
for the detection algorithm. The feature coefficients with multiple frequency resolutions and the slip speed are used
as the inputs of the artificial neural network. The proposed algorithm is evaluated on a 5 HP induction motor drive
system and is proved to be able to distinguish between healthy and faulty conditions with high accuracy. |
