Application of Deep CNN-LSTM Network to Gear Fault Diagnostics
T. Haj Mohamad, Amirhassan Abbasi, Eun Hee Kim, C. Nataraj
Abstract
Condition-based maintenance (CBM) is an optimum predictive maintenance framework that proposes maintenance actions based on monitoring the state data of an asset. Diagnostics is a principle concept in this framework and deals with fault detection, identification and isolation. Improving performance of diagnostics methods is of importance since it can result in reducing downtimes, improving operation reliability, reducing operations and maintenance costs. On the other hand, development of computational resources and sensory facilities could contribute highly to data based diagnostics approaches. The current paper studies one of these approaches that is categorized under deep learning (DL) concepts for a fault classification problem. A convolutional neural network (CNN) is used along with a long short term memory (LSTM) network for fault classification of vibration data of a helicopter gearbox mockup system. In experimental tests, multiple gears at different conditions e.g. healthy gear and defective gears with root crack on one tooth, multiple cracks on five teeth and missing tooth, are taken into account. A deep learning model is built and its performance is evaluated using post processing techniques.