Application of neural networks to acoustic screening of small electric motors
Shawn Murphy, Samir I Sayegh · 2003
A three-layer backpropagation neural network trained to differentiate good versus bad electric motors based on aural cues is described. Training performance of 100% and test performance of greater than 95% is achieved. Motors are classified as 'good' (pass) and 'bad' (fail) by a human operator. Acoustic data constitute a continuous signal in the form of a sound pressure level processed into nine bands from 1 kHz through 10 kHz. The Galatea neural network simulator is used to model two common neural network paradigms (linear and backpropagation) for suitability in this problem. Preprocessing of data is necessary. Training takes place quickly with good results after the data are conditioned.>