Feature Selection by Genetic Programming , And Artificial Neural Network-based Machine Condition Monitoring

177 Abstract—This Paper presents the performance of bearing fault diagnosis using Genetic Programming and artificial neural networks (ANNs). The experimental data is collected for four bearings conditions namely: Healthy, defective Outer race, defective Inner race and defective ball fault condition. Artificial neural network have been widely used for health diagnosis of rotating machinery using features extracted from vibration emission signals. One of the most important considerations in applying neural networks to condition monitoring of electrical machine is the proper selection of training features. Irrelevant or noisy features unnecessarily increase the complexity of the problem and can degrade modeling performance. A Genetic programming for feature selection is developed, based on the concept of dominance. GP is used for two purpose Feature extractor and feature selector but in this paper GP is used only for best feature selection from a large features data set. The algorithm is used to effectively select a smaller subset of features that together form a genetically fit family for fault identification and classification tasks.