旋转机械耦合故障微弱信号特征提取与诊断方法研究

As the power source and transmission mechanism for the fundamental and critical facilities, such as turbines, generators and etc., rotating machinery is widely applied in petrochemical, power generation, transportation and other industries. Real-time online condition monitoring and fault diagnosis are considered critical to the safety of rotating machinery. However, weak feature extraction and simultaneous fault diagnosis under the context of strong noise restrict the application of online health monitoring technology to complex rotating machinery. In this project, multi-stage gearbox is selected as the subject of this research with the focus placed on simultaneous diagnosis and identification method for rotary machinery based on a combination of experiment, finite element analysis and dual extreme learning machines. Firstly, the numerical simulation method that combines finite element model with experimental correction is studied to derive fault samples with low costs to conduct experiment. Subsequently, the method developed based on synchrosqueezed wavelet transforms and singular value decomposition is applied to achieve simultaneous fault feature extraction under the context of strong noise. Finally, a research into the method of fault diagnosis modeling based on dual extreme learning machines is carried out, and a novel method to enable weak feature extraction and simultaneous fault diagnosis of rotating machinery based on a combination of experiment, finite element analysis and dual extreme learning machine is proposed. The experiments are conducted to validate the accuracy of the proposed method. The research results are of a massive scientific significance and value for engineering application to address the problems with online monitoring and fault diagnosis for rotating machinery.

旋转机械作为汽轮机、发电机等重大基础关键设备的动力源和传动机构，广泛应用于石油化工、动力发电、航空航天等领域，实时在线状态监测和故障诊断是保障其安全运行的关键技术。强噪声下耦合故障微弱信号特征提取与诊断是制约旋转机械在线健康监测技术工程应用的难题。本项目拟以多级齿轮箱为研究对象，开展实验/有限元/双极限学习机协同的耦合故障微弱信号特征提取与诊断方法研究。首先，研究有限元建模与实验校正相结合的故障数值模拟方法，实现故障样本低实验代价获取；然后，研究基于同步压缩小波变换和奇异值分解的故障特征提取方法，实现强噪声下耦合故障微弱信号特征提取；最后，研究基于双极限学习机的故障诊断模型构建方法，形成实验/有限元/双极限学习机协同的耦合故障微弱信号特征提取与诊断新方法，并开展试验验证。研究成果对解决旋转机械在线监测、故障诊断问题具有重要的科学意义和工程应用价值。

旋转机械作为汽轮机、发电机、电动机等重大基础关键设备的动力源和传动机构，广泛应用于石油化工、轨道交通、航空航天等领域，实时在线状态监测和故障诊断是保障其安全运行的关键技术。强噪声下耦合故障微弱信号特征提取与诊断是制约旋转机械在线健康监测技术工程应用的难题。本项目以多级齿轮箱为研究对象，开展实验/有限元/双极限学习机协同的耦合故障微弱信号特征提取与诊断方法研究。首先，研究了有限元建模与实验校正相结合的故障数值模拟方法，实现了故障样本低实验代价获取。其次，研究了基于最大平均脉冲能量比反褶积的故障信息增强方法、基于同步压缩小波变换和奇异值分解的故障特征提取方法，实现了强噪声下耦合故障微弱信号特征增加及提取；在此基础上，进一步开展了基于自适应变分模态分解、自适应加权和重构的辛几何模态分解和增强梅尔线性频率倒谱系数的故障特征提取方法研究。再次，研究了基于自适应加权粒子群优化极限学习机、蚁群算法优化极限学习机、天牛须搜索算法优化核极限学习机及双极限学习机的旋转机械故障诊断模型构建方法，形成实验/有限元/双极限学习机协同的耦合故障微弱信号特征提取与诊断新方法；在此基础上，研究了基于模态归一化及通道注意力优化卷积神经网络的故障诊断方法，实现了数据不平衡和工况变化下轴承故障准确诊断。最后，以多级齿轮箱轴承为实验对象，设计并搭建了单一/耦合故障诊断实验系统，开展了单一/耦合故障诊断实验，验证了强噪声、多故障耦合条件下实验/有限元/双极限学习机协同的旋转机械故障微弱信号特征提取与诊断方法的准确性与实时性。研究成果在临沂中联水泥有限公司水泥生产线上斗开展了初步应用验证，对解决旋转机械在线监测、故障诊断问题具有重要的科学意义和工程应用价值。