STUDY ON THE APPLICATION OF MULTI-SENSOR DATA FUSION IN GEARBOX FAULT DIAGNOSIS

摘要

The traditional method of gearbox fault diagnosis is based on the respective analysis of vibration signals, and does not fully employ the relativity of all vibration signals. However in the gearbox fault diagnosis system, usually more than one sensor is placed to acquire vibration signals. So the traditional method is difficult to judge the fault exactly. In the paper, multi-sensor data fusion is introduced to gearbox fault diagnosis. Different ways of data fusion are attempted to use in practical gearbox fault diagnosis. Based on the proper placement of sensors, several simple methods of data fusion in gearbox fault diagnosis are put forward firstly. In virtue of these methods, we can judge the source of fault, such as bearing, gear etc; noise disturbance can be reduced; gear local defects can also be judged, and rolling bearing fault can be distinguished from generic rotating machinery fault. At last, a model of multi-layer data fusion diagnosis is introduced in detail. The model consists of the pixel level fusion module, the feature and decision level fusion diagnosis module based on parallel BP neural network and Dempster-Shafer evidence theory. The first module is used to extract characters of typical gearbox faults. Then the second and third modules dispose the data in sequence. Finally diagnosis result can be gotten. Applications reveal that multi-sensor data fusion is successful and promising in gearbox fault diagnosis.