Rolling Bearing Fault Diagnosis Method Based on Multilayer Noise Reduction Technology and Improved Convolutional Neural Network
Shaojiang Dong, Xuewu Pei, Wenliang Wu, Baoping Tang, Zhao Xingxin
Abstract
摘要: 针对滚动轴承微弱故障在强噪声下难以实现有效诊断的问题,提出了一种基于多层降噪技术及改进卷积神经网络(Improved convolution neural network,ICNN)的轴承故障诊断新方法。首先,对滚动轴承的一维振动信号进行预处理,得到标签化的数据样本,分为训练集和测试集;然后采用奇异值分解(Singular value decomposition,SVD)处理训练样本,通过二分之均值法选择有效奇异值个数,获得原始降噪信号和带噪信号;为了避免丢失微弱故障细节特征,将带噪信号经过SVD进一步去噪消除模态混叠并输入经验模态分解(Empirical mode decomposition,EMD)得到内禀模态函数,根据方差贡献率大小选出IMF分量并与原始降噪信号叠加得到最终信号;将处理后的训练集数据输入到引入注意力机制(Attention mechanism,AM)的ICNN中进行学习;最后将得到的诊断模型应用于测试集,输出故障类别诊断结果。通过滚动轴承故障诊断模拟试验,在强噪声环境下进行测试,结果表明所提方法能更准确的在强噪声环境中实现轴承的故障诊断。
Topics & Concepts
Convolutional neural networkBearing (navigation)Fault (geology)Reduction (mathematics)Noise reductionComputer scienceNoise (video)Artificial neural networkArtificial intelligencePattern recognition (psychology)MathematicsGeologySeismologyImage (mathematics)GeometryGear and Bearing Dynamics AnalysisAdvanced Algorithms and ApplicationsAdvanced Sensor and Control Systems