Fault detection and classification with artificial intelligence techniques on worm gearboxes

Worm gearboxes (WG) are frequently used in many areas of the industry. WG is different from other gearbox types and due to their working principles, they are under high risk of wear and fault. Therefore, detection of faults that may occur in WG and taking measures accordingly especially are important for systems and facilities that require uninterrupted operation. For this purpose, in this study an experimental setup which simulates different working conditions has been developed for condition monitoring studies of WG. Fault detection and classification were performed based on vibration, sound and thermal images data features which were acquired and processed from the healthy and faulty WG in the test rig. Apart from classical studies, time and frequency domain features vibration and sound signals and thermal images features were extracted and evaluated singularly, dual or triple forms. Commonly effective ANN (Artificial Neural Network), SVM (Support Vector Machines), k-NN (k-Nearest Neighbor), ANFIS (Adaptive Neuro Fuzzy Inference System) and deep learning classifiers were selected to detect fault and to classify types of faults. It has been determined that the fault detection and classification performances are low with the use of single feature sources. The highest classification performances for fault detection were observed when the features of all three sources used. The experimental results indicated that the selection of features is an important step to maximize the performances of classifiers.