Investigations into the Use of Wavelet Transformations as Input into Neural Networks for Condition Monitoring

Abstract

Condition monitoring has become widely accepted in industry, with vibration providing the most useful condition parameter. Analysis of these vibration signals usually involves Fourier transforms and, occasionally, neural networks. However, a problem when using Fourier transforms as input into a neural network is the size of the input data set. Wavelet transforms provide an alternative which allows for a dramatic reduction in the size of the data set. This paper will explore the feasibility of using wavelet transforms as input into neural networks for condition monitoring.