Vibration Analysis of Gear Defects using Machine Learning Approach

Gear drives are considered as the most effective transmission method in automobiles as well as in various industries because of its high efficiency, reliability and high velocity ratio. As a result, from its trustful usage, failure in any part may lead to a large and unpredictable production loss along with massive service cost and safety concerns. Scheduled condition monitoring and Periodic maintenance are the only solution to avoid the above scenario. Vibration analysis is the most sounded term in fault detection due to its runtime condition monitoring and low cost. Nowadays, vibration analysis has been offset to the machine learning methods, which is a modern technique enabling us to automate such that the system can learn from the input data and make decisions with a nominal human interface whereas conventional methods are highly operator dependent. Here in this study, the effectiveness of a machine learning based gear fault diagnosis system is carried out. Gear defects like misalignment and broken teeth were considered. No combinations of these defects were considered. The experimental setup consists of a paired spur gear shafts, driven by a variable speed motor with a belt drive along with a vibrational data acquisition system. Vibration signals for three classes were collected including the signals from healthy gear too. TensorFlow is used to implement machine learning models. The proposed method is successful in detecting gear defects while the machine is in running condition. The new method is fast and can be automated. This reduces the human intervention to a minimum level