Wayside Acoustic Fault Diagnosis of Train Wheelset Bearing Based on Improved Frequency Sparsity Bayesian Learning

Important train component like wheelset bearing endure severe mechanical loads and alternating stresses for an extended period when the train is in operation. The cracks due to bearing fatigue are often difficult to detect in time, which is the biggest hidden problem for train safety. The wayside acoustic detection of train bearing faults plays an important role in maintaining the safe operation of trains. However, the acoustic signals collected by the wayside microphones are usually complex, mixed with other irrelevant signals with strong background noise when in the condition monitoring of wheelset bearings. Moreover, when the train passes through the detection area at a certain speed, the microphone picks up the acoustic signal for a short time, which makes it difficult to obtain accurate fault signal characteristics. Trackside acoustics often uses microphone arrays to overcome the problem of short sampling time. To diagnose wheelset bearing faults more effectively with fewer microphones, this paper proposes a wayside acoustic fault diagnosis method based on improved sparsity Bayesian learning. Advanced sparse representation (SR)-based approaches usually include two primary stages: recovery of fault pulses in the time domain and frequency transformation of the estimated signal envelope. However, any inaccurate signal recovery in the time domain can introduce the problem of error accumulation in the subsequent frequency transform, which suffers from the disadvantage of low resolution, especially for short time sampled acoustic signal. Taking the limitations into consideration, a new algorithm has been developed by combining the group sparsity and periodic structure of fault pulses with variational Bayesian inference (VBI) technology. This advanced method effectively extracts the fault pulse signal, even in the presence of high noise levels, which leads to a significant enhancement in the fault detection performance. With the aid of this method, in conjunction with the Doppler distortion preprocessing correction algorithms, the proposed approach has been successfully applied to the diagnosis of faults in train trackside acoustic wheelset bearings. The experimental results show that the proposed method can achieve good performance in fault diagnosis compared with conventional analysis method, making it more effective to monitor wheelset bearing faults.