基于残差网络和时频域特征融合的滚动轴承故障诊断方法

Rolling Bearing Fault Diagnosis Method Based on Residual Network and Time-Frequency Domain Feature Fusion

针对单一时频域方法对振动信号特征提取能力有限的问题,提出一种基于残差网络和时频域融合的滚动轴承故障诊断方法。通过采集轴承振动信号,计算滚动轴承振动信号量纲一和有量纲特征统计指标、能量算子指标共12个时域特征;将时域信号转化成频域信号,提取出4个频域特征指标;使用离散小波变换进一步提取信号特征,得到16个时频域特征。构建残差网络,将原始振动信号输入残差网络提取特征,在全连接层将提取的时域、频域、时频域特征连接得到32个特征,并与残差网络提取的时域特征融合。最后,将融合的特征输入分类网络得到故障诊断结果。根据在某故障诊断重点实验室数据集以及企业真实运行数据集上的实验验证,提出的方法相对于其他经典分类模型拥有更好的性能。

Aiming at the limited feature extraction ability of vibration signals using a single time-frequency domain method,a fault diagnosis method of rolling bearing based on the fusion of residual network and time-frequency domain fusion was proposed.Vibration signals from the bearing were collected,and a set of 12 time domain characteristics of rolling bearing vibration signals were calculated,including dimensionless,quantitative feature statistical indexes and energy operator indexes.The time-domain signal was transformed into frequency-domain signal,from which four frequency-domain feature indexes were extracted.Discrete wavelet transform was used to extract the signal features,and 16 time-frequency domain features were obtained.The residual network was constructed and utilized to extract features from the original vibration signal.The extracted features in the time domain,frequency domain,and time-frequency domain features were connected in the full connection layer to obtain 32 features,and then the fused features were combined with the time-domain features obtained from the residual network.Finally,the fused features were fed into a classification network to obtain fault diagnosis results.According to the experimental verification on the dataset of a key laboratory and the real operation dataset from enterprises,the proposed method has better performance than other classical classification models.