基于灰狼算法优化极限学习机的中介轴承故障诊断方法

Method on inter-shaft bearing fault diagnosis based on extreme learning machine optimized by gray wolf optimizer

关键词:  中介轴承  模态分解  极限学习机  灰狼算法  故障诊断

Key words:  Inter-shaft bearing  Mode decomposition  Extreme learning machine（ELM）  Grey wolf optimizer（GWO）  Fault diagnosis

摘要:

针对中介轴承故障振动信号具有传递路径复杂、强背景噪声干扰等特点，其故障特征不易提取的问题，提出基于自适应噪声完全经验模态分解（CEEMDAN）与灰狼算法（GWO）优化的极限学习机（ELM）相结合的中介轴承故障诊断方法。利用CEEMDAN和相关系数-能量比-峭度准则（CEKC）对振动信号进行分解、筛选、重构；再提取重构信号的时域和频域特征构成特征矩阵；然后以平均错误率作为GWO的适应度值，对ELM的输入层与隐含层的权值和隐含层阈值进行优化后重新构建ELM；最后将特征矩阵输入ELM得到故障诊断结果。应用于中介轴承故障诊断中，ELM在GWO优化后故障诊断正确率有明显提升，其中45°方向传感器数据正确率由93.33%提升到99.17%。结果表明：该方法能够有效诊断中介轴承故障类型，表现出了较强的泛化能力。

Abstract:

For the problem that the vibration signal of intermediate bearing faults are disturbed by complex transmission path and strong background noise and the fault features are not easily extracted， an inter-shaft bearing fault diagnosis method based on a combination of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise （CEEMDAN） and Extreme Learning Machine （ELM） optimized by Gray Wolf Optimizer （GWO） was proposed. The vibration signal was decomposed， screened and reconstructed by CEEMDAN and the Correlation coefficient-Energy ratio-Kurtosis Criterion （CEKC）； then extract the time and frequency domain features of the reconstructed signal to form the feature matrix； then reconstruct the ELM after optimizing the weights of the input and implied layers and the threshold of the implied layer using the average error rate as the fitness value of GWO； finally the feature matrix was input to ELM to obtain fault diagnosis results. Applied to the inter-shaft bearing fault diagnosis， the ELM has significantly improved the correct rate of fault diagnosis after GWO optimization， in which the correct rate of sensor data at 45° direction increased from 93.33% to 99.17%. The results show that the method can effectively diagnose the inter-shaft bearing fault type and shows a strong generalization ability.