Fault feature selection for the identification of compound gear-bearing faults using firefly algorithm

The occurrence of compound faults in real-time conditions leads to the early failure of components. However, identifying compound faults in a rotor system is more complex because extracting the fault information from the vibration signals is challenging. Hence, it is inevitable to devise a reliable strategy for predicting compound faults in a rotor system to ensure its life. This work proposes a novel feature selection method using the firefly algorithm (FA) to identify compound gear-bearing faults. The statistical features are extracted from the time domain vibration signals. The firefly algorithm is employed to select the features that have the most pertinent information about the faults. The classification potential of the selected features is tested with an optimized feed forward fully connected neural network (FFFCNN) architecture. Further, the performance of the proposed approach is compared with the genetic algorithm, relief-based feature selection techniques and without the feature selection approach. The FA-based feature selection combined with the FFFCNN achieves a prediction accuracy of 94.86% in identifying the compound gear-bearing faults.