The Influence of Handling Imbalance Classes on the Classification of Mechanical Faults Using Neural Networks

This work proposes a simpler automatic fault classifier that uses multi-layer perceptron (MLP) to identify faults in rotating machines. For classification, only statistical features and rotation frequency was taken in to consideration. Synthetic Minority Over-sampling Technique (SMOTE) was incorporated to handle imbalance classes in the machinery fault database. The proposed approach was evaluated on Machinery Fault Database (MAFAULDA) database. The proposed system has achieved an accuracy of 96.2%, which is comparable with the reported results in the literature. Results indicate that handling imbalance classes greatly increases the generalizability power of the MLP classifier on the machinery fault database.