基于声学信号与视觉转换器的滚动轴承故障诊断方法研究

Study on Rolling Bearing Fault Diagnosis Method Based on Acoustical Signal and Vision Transformer

航空机载设备的可靠性对航空运输安全至关重要。针对航空机载设备上的滚动轴承故障,本文提出一种声学信号与视觉转换器(ViT)相结合的滚动轴承故障诊断方法。首先,将采样获得的滚动轴承声信号通过短时傅里叶变换转换为时频图。其次,将时频图按时序分割,作为ViT的输入。ViT通过多注意力机制提取图像块中的信息并输出数据。最后,输出数据通过多层感知机实现对不同类别的滚动轴承故障识别。试验表明,相较于传统的基于卷积神经网络和长短时记忆网络的滚动轴承故障诊断方法,本文所提方法的滚动轴承故障诊断准确率更高,为航空机载设备的轴承故障诊断提供了一类新方法。

The reliability of airborne equipments has a significant impact on aviation safety.Aiming at the characteristics of rolling bearings that are prone to faults and have unstable fault signals,a rolling bearing fault diagnosis method combining short time Fourier transform and Vision Transformer(ViT)is proposed.Firstly,the sampled rolling bearing sound signal is converted into a time-frequency map containing the time of frequency occurrence through the short time Fourier transform.Secondly,the time-frequency map is segmented chronologically as input to the ViT.The ViT extracts the information in the image block through the multi-attention mechanism and outputs the output data.The output data is matched by multiple perceptrons to realize the recognition of different types of rolling bearing faults.The experiment shows that the proposed method has higher accuracy in rolling bearing fault diagnosis compared to CNN and CNN+LSTM rolling bearing fault diagnosis methods.A new approach for the diagnosis of bearing faults in airborne equipments is proposed in this work.