Classification of Machine Fault using Principle Component Analysis, General Regression Neural Network and Probabilistic Neural Network

Abstract

As a major industry prime mover, induction motor plays an important role in manufacturing. In fact, production can cease its operation if there is some error or fault in the induction motor. In the industry, bearing, stator and rotor fault are the highest among other faults. Thus, this paper is to compare the accuracy of bearing, stator and rotor fault classification between General Regression Neural Network (GRNN) and Probabilistic Neural Network (PNN) with the previous work using Principle Component Analysis (PCA). The accuracy of fault classification for each method is improved by the selection of features extraction and number of classification. The features extraction used are mean, root mean square, skewness, kurtosis and crest factor. The sample data has been taken from Machinery Fault Simulator using accelerometer sensor, logged to text file using Labview software and analysed by using Matlab software. The accuracy of fault classification using GRNN method is higher than PNN because the sample data is classified through the regression of data as long as the sample data is redundant and lies on the regression distribution.