Multilayer Gated Recurrent Unit for Spur Gear Fault Diagnosis

Abstract

As an important transmission component, the spur gearbox may cause great losses if it fails, so fault diagnosis is the key to ensure the normal operation of the equipment. In this paper, a multilayer gated recurrent unit (MGRU) method is proposed, which uses a three-layer gated recurrent unit (GRU) to deal with the fault diagnosis of the spur gear. Due to the complexity of the spur gearbox fault, vibration measurement is carried out separately at first. Then the vibration signals are extracted from the time domain and time-frequency domain. Finally, MGRU is used to learn representation and classification. By using this model, fault features can be deeply learned layer by layer, and feature types can be identified with higher accuracy. The proposed method was applied to two spur gears (number of teeth Gear1= 53, and Gear2= 80), which were installed on the input and the output shafts of the gearbox, and there are 10 state modes in total. To evaluate the method's performance, four methods were applied to compare, which are (GRU, multilayer long short-term memory (MLSTM), long short-term memory (LSTM) and support vector machine (SVM)) respectively. The classification result of MGRU model shows that it is effective for spur gear fault diagnosis.