Simultaneous Fault Diagnosis Method Based on Improved Sparse Bayesian Extreme Learning Machine

In order to identify simultaneous faults of the main reducer, an adaptive threshold denoising was adopted for intrinsic mode functions (IMFs) with high frequency and an interval threshold denoising was adopted for IMFs with low frequency, which are obtained from ensemble empirical mode decomposition (EEMD) of vibration signal, to eliminate noises. Then, a paired multi-label classification (PMLC) strategy was established, and probability classifiers based on PMLC and sparse Bayesian extreme learning machine (SBELM) were constructed with single fault samples; an optimal decision threshold was generated by using the grid searching method to convert the probability output obtained from classifiers into final simultaneous fault modes. On this basis, a simultaneous fault diagnosis (SFD) method based on adaptive threshold de-noising and SBELM was proposed. Its performance was verified using real samples of the main reducer. The experiment results show that the diagnostic accuracy of the method is 96.1%, which is 5% higher than that of methods based on probabilistic neural network (PNN) and support vector machine (SVM); its training time and execution time are 131.4 and 61.3ms, respectively, approximately 70% shorter than those of the method based on SVM.