Effect of k-Nearest Neighbours (k-NN) in Classifying Planetary Gearbox Faults Using Statistical Features

The use of planetary gearboxes in heavy-duty industries is dominant due to their compact size, large transmission ratio and torque delivery capability with different configurations. Due to their harsh operating conditions, localised gear tooth faults such as cracking and chipping are more common in such gearboxes. Furthermore, localised gear tooth failure initiates distributed gear faults such as pitting and wear on the gear tooth. Therefore, it is necessary to monitor such localised gear faults continuously and detect them at an early stage to prevent sudden and catastrophic failure. In this study, gear tooth localised defects on various gear elements of the planetary gearbox are seeded using Electrical Discharge Machine (EDM). Then the vibration signals from the gearbox are captured. Afterwards, a decision tree algorithm selects the most prominent statistical features from many extracted features. Further, to automate the fault detection process, the k-nearest neighbours (k-NN) algorithm is selected to classify the faults. Finally, a study was carried out on the effect of the number of nearest neighbours to classify the faults accurately. The study outcome recommends using k-NN to predict the faults accurately.