Automatic feature selection and failure diagnosis for bearing faults

This study develops a novel dual-stage diagnosis scheme for accelerating bearing failure diagnosis. The schema integrates the intelligent methods, i.e., genetic algorithm, k-nearest neighbors, and neural network, in the featuring and modeling stages to automatically select the significant features from various feature candidates for modeling bearing failure modes. After applying the scheme to classify two cases of bearing faults, the mean training time for model diagnosis is reduced to 8.1% that of using a neural network model. In this work, case 1 indicates that training and testing accuracies of seven failure modes are 98.8% and 94.5%, respectively; in addition, case 2 shows that the training and testing accuracies are 96.2% and 91.8% while using the top seven features.