Vibration condition monitoring of planetary gears based on decision level data fusion using Dempster-Shafer theory of evidence

In recent years, due to increasing requirement for reliability of industrial machines, fault diagnosis using data fusion methods has become widely applied. To recognize crucial faults of mechanical systems with high confidence, indubitably decision level fusion techniques are the foremost procedure among other data fusion methods. Therefore, in this paper in order to improve the fault diagnosis accuracy of planetary gearbox, we proposed a representative data fusion approach which exploits Support Vector Machine (SVM) and Artificial Neural Network (ANN) classifiers and Dempster-Shafer (D-S) evidence theory for classifier fusion. We assumed the SVM and ANN classifiers as fault diagnosis subsystems as well. Then output values of the subsystems were regarded as input values of decision fusion level module. First, vibration signals of a planetary gearbox were captured for four different conditions of gear. Obtained signals were transmitted from time domain to time-frequency domain using wavelet transform. In next step, some statistical features of time-frequency domain signals were extracted which were used as classifiers input. The gained results of every fault diagnosis subsystem were considered as basic probability assignment (BPA) of D-S evidence theory. Classification accuracy for the SVM and ANN subsystems was determined as 80.5 % and 74.6 % respectively. Then, by using the D-S theory rules for classifier fusion, ultimate fault diagnosis accuracy was gained as 94.8 %. Results show that proposed method for vibration condition monitoring of planetary gearbox based on D-S theory provided a much better accuracy. Furthermore, an increase of more than 14 % accuracy demonstrates the strength of D-S theory method in decision fusion level fault diagnosis.