Thermal Image Based Fault Diagnosis of Gears using Support Vector Machines

Condition monitoring and fault diagnosis of working machines have gained significant attention due to their prospective benefits, such as enhanced productivity, decreased repair and maintenance costs and enhanced machine operation. In this paper, a thermal image based non-contact methodology has been proposed to diagnose the gear faults using support vector machines (SVM). The thermal images acquired from gearbox simulator were preprocessed using 2D-discrete wavelet transform to decompose the thermal images. The relevant features were extracted from converted thermal gray-scaled images followed by selecting the strongest feature using Mahalanobis distance criteria. Finally, the selected features were given to a SVM classifier for classifying the different gear faults. The experimental findings indicate that fault diagnosis using thermography for rotary machinery can be put into practice to industrial fields as a new smart fault diagnostic method with excellent prediction performance.