基于非平稳信号组合分析的故障诊断方法

Non-stationary signal combined analysis based fault diagnosis method

鉴于深度学习、频谱、时频分析方法间的优势互补,设计了由卷积网络、傅里叶变换和小波包分解组合的多流分析处理框架,对非平稳信号进行组合分析。提出了一种基于非平稳信号组合分析的故障诊断方法,提取信号的多属性特征并加权融合。应用于故障诊断的实验结果表明,所提出的信号组合分析方法能够更加稳定、准确地刻画故障类型,在不显著增加计算复杂度的前提下有效提高了故障诊断的分类准确率。

Considering the complementarity between the deep learning,spectrum and time frequency analysis methods,a multi-stream framework was designed by combining the convolutional network,Fourier transform and wavelet package decomposition methods,with the aim to analyze the non-stationary signal.Accordingly,a none-stationary signal combined analysis based fault diagnosis method was proposed to extract features in difference aspects.The fault diagnosis experiments demonstrate that the combined analysis method can efficiently and stably depict the fault and significantly improve the performance of fault diagnosis.