Feature selection for defect classification in machine condition monitoring

As the sensitivity of various parameters to a defect condition of a machine differs, it is imperative to devise a feature selection scheme that selects the best parameters to maximize the accuracy of the defect classification scheme. A feature selection scheme based on principal component analysis (PCA) is proposed in this paper. A methodology was developed for bearing defect classification using neural networks. The scheme has shown to provide more accurate defect classification with less parameter inputs than using all parameters initially considered relevant.