张量数据驱动的高端机械装备故障诊断与预测方法研究

In the age of big data and artificial intelligence (AI), a tensor-data-driven approach for fault diagnosis and prognosis of large and complex mechanical equipment has been developed in this project based on the theories of tensor algebra, AI and mechanical dynamics. Novel multi-dimensional tensor data model and its order will be researched in the condition monitoring of mechanical systems. Low-rank tensor decomposition is applied to extract fault signatures in the multichannel signal processing. Evolution of coupling failures will be in depth interpreted with tensor decomposition technique combined with mechanism of faults. Subsequently, based on the distance of tensor data, supervised and unsupervised intelligent fault diagnosis approaches have been developed based on support tensor machine and density peak search clustering algorithm, respectively. Dynamic tensor completion technique is proposed to solve the issue of missing data. Degradation trend of mechanical equipment can be predicted based on dynamic tensor completion technique and the selected key performance indicator. It will be demonstrated that those developed techniques can well improve the accuracy of intelligent fault diagnosis and prognosis, because of the mined multi-mode information under the framework of tensor data. Those proposed techniques are under the guidance of theory combined with practice, and their effectiveness will be further verified using practical data collected in the urban rail transit vehicle bogies. The results of the project can be also used to the diagnosis, prognostics and health management of high-end mechanical equipment. Thus, this research has important meanings in scientific theories and practical applications.

大数据与智能时代，本项目以张量代数、人工智能和机械动力学理论为源泉，研究张量数据驱动的高端机械装备故障诊断与预测方法。专注高端装备动态监测数据的多维张量模型建立与参数估计方法，研究基于低秩张量分解的多路机械信号处理与多域特征提取，采用数据与耦合故障机理联动研究阐释张量分解物理意义及故障演化规律。在此基础上，构造张量数据间距离度量方式，提出了有监督学习的支持张量机与无监督学习的密度峰值快速搜索两类机械故障智能识别方法；寻求关键张量数据退化指标，研究面向不完备动态张量数据流的填充预测方法，旨在张量框架下充分挖掘系统各类信息的多模态特性，提高高端装备的智能诊断与状态退化预示精度。本项目采用理论与实验研究相结合的研究方法，并在城市轨道交通车辆装备走行部运维中进一步验证工程应用的有效性。项目研究成果将为高端机械装备诊断、预测与健康管理提供理论与技术支持，具有显著的科学意义和重要应用价值。

大数据与智能时代，本项目以张量代数、人工智能和机械动力学理论为源泉，研究张量数.据驱动的高端机械装备多种信号处理与故障诊断方法。专注高端装备动态监测数据的多维张量模型建立与参数估计方法：研究提出基于张量分解的机械多维监测信号自适应降噪；基于张量分解的滚动轴承复合故障多通道信号降噪；基于同步挤压提取变换（SST）进行变转速工况滚动轴承故障诊断；基于两步自适应调频模式分解(two-step ACMD)的变转速滚动轴承故障诊断方法；基于自适应调频模式分解和脊检测的滚动轴承故障诊断方法。在此基础上，开展有监督学习的支持张量机与无监督学习的密度峰值快速搜索两类机械故障智能识别：研究提出基于核支持张量机和多向主成分分析的旋转机械故障诊断；基于张量对齐的不变子空间学习和二维卷积神经网络(TAISL-2DCNN)的智能故障诊断新方法；基于张量表示的半监督部分标签学习(SSPLLTR)的跨域轴承智能故障诊断方法；基于迁移成分分析(TCA)的域自适应轴承智能故障诊断技术；基于自适应迁移密度峰值搜索聚类算法的机械智能故障诊断等方法。本项目采用理论与实验研究相结合的研究方法，并在城市轨道交通车辆装备走行部运维中进一步验证工程应用的有效性。项目研究成果将为高端机械装备诊断与健康管理提供理论与技术支持，具有显著的科学意义和重要应用价值。