25 / 2016-07-05 19:10:56

Internal Combustion Engine Fault Detection by Wavelet Packet and Softmax Regression

Wavelet Packet Transform,Softmax Regression,artificial neural network

An automated fault detection and classification technique is proposed using acoustic signal generated from IC (Internal Combustion) engines. This technique aims to remove unwanted features of acoustic data by pruning the wavelet packet transform (WPT) tree. This pruning is done by entropy based best basis algorithm. The FFT spectrum of the reconstructed signal from the pruned tree is used as the feature vector for the classifier. The FFT Spectrum of data removes the repetition of features and reduce the size of the data. This data is used by softmax regression classifier, which classifies unknown engines into healthy and faulty class. This technique is tested on IC engines, with overall classification performance of 142 correct classifications out of 143 test cases with five different fault classes with majority voting. The classification performance of softmax regression is also compared with widely used artificial neural network classifier and softmax regression is found to be superior in performance.