多源强噪下高铁齿轮箱强时变非平稳故障特征提取及定量诊断研究

Gearbox is a key part of high-speed train power transmission. Timely and quantitative diagnosis of gearbox operating status is one of the important guarantees for the safe operation. However, quantitative diagnosis is difficult due to the complicated structure and poor conditions. This project intends to study the dynamic evolution mechanism and weak fault feature quantitative extraction of the high-speed rail gearbox, reveal the relationship between fault evolution and feature representation, and solve the problems of quantitative diagnosis and degradation prediction. The main research includes: (1) studying the nonlinear rigid-flexible coupling dynamics model of fault gearbox, and exploring the quantitative relationship between fault degree and fault characteristics under complex conditions, (2) constructing the coupling sparse dictionaries and the compound dictionary, by the nonlinear weak signal enhancement method, (3) studying multi-information coupling model and information-fusion decision algorithm, establishing a hybrid intelligent forecasting model based on the weak fault feature extraction, information fusion and different intelligent algorithms to realize the quantitative diagnosis and degradation trend prediction of the gearbox.

高铁齿轮箱是高速列车的关键动力传动零部件，及时、定量诊断其运行状态是高速列车安全运行的重要保障之一，但由于其结构复杂、运行条件恶劣，其故障状态下动力学特性、定量诊断等关键问题尚需解决。课题拟以高铁齿轮箱为研究对象，研究其早期故障动态演化机理、微弱故障特征量化提取理论和方法，揭示其故障演化与特征表示的关联，以解决定量诊断与退化趋势预测的关键科学问题。主要研究工作有：1）研究故障高铁齿轮箱非线性刚柔耦合动力学模型，探索复杂工况下故障演化过程与故障特征间的对应量化关系；2）研究基于损伤程度的故障信息特征的耦合稀疏字典构建方法，建立与故障程度相对应的级联复合字典，探索基于时频系数结构之间相互匹配和协同作用的非线性微弱信号增强方法；3) 研究多信息耦合模型和信息融合决策算法，建立基于微弱故障特征提取、信息融合以及不同智能算法相融合的混合智能预测模型，实现齿轮箱运行状态的定量诊断与退化趋势预测。

本项目围绕多源强噪下高铁齿轮箱强时变非平稳故障特征提取及定量诊断开展研究，并以高铁齿轮箱为研究对象，研究了其早期故障动态演化机理、微弱故障特征量化提取理论和方法，揭示了其故障演化与特征表示的关联，解决了定量诊断与退化趋势预测的关键科学问题。针对强噪声背景和干扰源信号影响下齿轮箱故障诊断信号特征难以提取等问题，提出了一种基于变分模态分解和深度支持向量机的齿轮箱故障诊断方法，实现了强背景噪音下的信号特征准确提取；通过实验及仿真手段，建立了基于多参数融合的自适应动态趋势预测模型，提出了一种基于深层极限学习机的故障诊断方法，提高了故障诊断的准确率和稳定性；针对高铁齿轮箱信号噪音强、故障响应弱以及故障诊断准确率低等问题，提出了一种基于深层迁移结构和深度稀疏最小二乘支持向量机的故障诊断方法，建立了基于微弱故障特征提取、信息融合以及不同智能算法相融合的混合智能预测模型，实现了齿轮箱运行状态的定量诊断与退化趋势预测。