混沌萤火虫参数优化去噪自动编码器在轴承故障诊断中应用

Application of chaotic firefly parameter optimization denoising autoencoder in bearing fault diagnosis

为提高滚动轴承的故障诊断效率,设计了一种混沌萤火虫参数优化去噪自动编码器(DAE),标记为CFADAE。在DAE基础上建立得到深度网络,通过DAE从中提取出首层特征,再利用自动编码器(AE)堆叠的方式构建深度网络,提取得到更深层特征参数。研究结果表明:相比较标准萤火虫算法(FA)数据,混沌萤火虫算法(CFA)方法的准确率更高,标准差更小。混沌改进萤火虫算法能够达到更准确诊断的效果,同时降低了波动程度。通过DAE网络进行诊断所得的平均准确率可以达到95.84%,推断结合去噪AE和L2惩罚项使模型获得更强的鲁棒性和泛化能力,提升诊断性能。本研究实现准确诊断轴承故障信号,可以拓展到其他的机械传动领域。

In order to improve the efficiency of rolling bearing fault diagnosis,a chaotic firefly parameter optimization denoising auto-encoder(DAE)was designed.The deep network is established on the basis of denoising autoencoder,and the features of the first layer are extracted from the denoising autoencoder.The deep network is constructed by stacking autoencoders,and deeper feature parameters are extracted.The results show that compared with standard FA data,CFA method has higher accuracy and smaller standard deviation.Chaos improved firefly algorithm can achieve more accurate diagnosis and reduce the degree of fluctuation.The proportion of diagnosis obtained through DAE network can reach an average accuracy of 95.84%.Inferring to denoising autoencoder and L2 penalty terms contributes to stronger robustness and generalization ability of the model,thus improving the diagnostic performance.This research can realize accurate diagnosis of bearing fault signals,which can be extended to other mechanical transmission fields.