Application of data mining and feature extraction on intelligent fault diagnosis by Artificial Neural Network and k-nearest neighbor

In this paper the frequency domain vibration signals of the gearbox of MF285 tractor is used for fault classification in three class: Healthy gear, Worn tooth face and broken gear. The effect of applying statistical parameters to signals on accuracy is studied. In addition, Influence of feature selection using Improved Distance Evaluation on classification performance and training speed is another target of present research. Two classification methods are used; Artificial Neural Network with variable neuron count for hidden layer in 2 layer network and k-nearest neighbor with variable k number. Using variable settings for classifier is due to make effect of statistical parameters and IDE independent from classifier settings. Results show that, accuracy improved from 86.6% to 100% by applying statistical parameters and 100% and 95.5% performance gained by applying IDE on ANN and kNN simultaneously but influence of IDE on kNN was higher than ANN.