Application of wavelets and neural networks for the detection and classification of transients from machines

This study applies wavelet based multiresolution analysis (MRA) methods to the detection of transients produced in the electric and magnetic fields when machines are turned on or off. The use of neural networks for the automatic detection and classification of the machines producing the transients was also examined.The results from the wavelet based MRA showed the effective use of this method for the detection of transients. This method allowed the detection of the transients even when they were buried in ambient noise. The results of training the neural networks with the wavelet coefficients of the MRA showed the possibility of classifying machine transients based on the characteristic signatures of the machines.The electrical and magnetic fields were measured from six different machines as they were turned on and off. The MRA was performed on the measured signals to decompose them into separate frequency subbands. The wavelet coefficients from the first 8 low frequency subbands were than used as input to the neural networks.