基于强制特征适配的卷积抗噪模型的滚动轴承故障诊断

Fault Diagnosis for Rolling Bearings Based on Convolution Anti-Noise Model with Forced Feature Adaptation

针对滚动轴承在噪声背景下的故障诊断研究不足的问题,提出一种基于卷积-注意力机制-强制特征适配(AMCNN-MFA)模型。卷积网络用于对振动信号进行高层特征提取,网络首尾加入通道注意力机制(CAM)用于动态分配特征通道的权重,过滤部分无效信息以减小干扰;强制特征适配(MFA)用于将原始样本和噪声样本中相同故障标签的特征进行领域重合,获取不变特征,实现了模型的适应噪声环境的能力。在不同轴承数据集中添加-10~10 dB噪声的测试集的试验结果表明:AMCNN-MFA模型的平均故障分类准确率高于96%且波动不超过0.4%,优于其他抗噪模型,具备较好的分类性能和鲁棒性,能够应对复杂噪声干扰场景下的滚动轴承故障诊断。

Aimed at insufficient research on fault diagnosis for rolling bearings under noise background, a model is proposed based on convolution-attention mechanism-forced feature adaptation(AMCNN-MFA). Convolution network is used for high-level feature extraction of vibration signals, channel attention mechanism(CAM) is added at head and tail of network to dynamically allocate the weight of feature channels and filter some invalid information to reduce interference. Mandatory feature adaptation(MFA) is used to overlap the features of same fault label in original sample and noise sample, to obtain the invariant features, and to realize the ability of model to adapt to noise environment. The test sets with SNR from-10 dB to 10 dB are added to different bearing data sets, the experimental results show that the average fault classification accuracy of AMCNN-MFA model is higher than 96% and the fluctuation is less than 0.4%, which is superior to other anti-noise models and characterized by good classification performance and robustness, and can deal with fault diagnosis for rolling bearings under complex noise interference scenarios.