基于自适应稀疏表征的声振多源信号分离方法研究

The intershaft bearing is a substantial component of advanced equipment, such as aero-engine and double-rotor centrifuge. Due to the transmission path of vibration signals is complex because of the special structure of the equipment, the effective fault signals of intershaft bearing are difficult to be measured. Therefore, the condition monitoring and fault diagnosis of main shaft bearing are difficult. This project aims at the research of feature extraction method and fault diagnosis method of intershaft bearing. In this project, sound signal and vibration signal are combined with verification experiments. This project will make a study on the separation method of vibro-acoustic multi-source signals based on adaptive sparse representation and blind source separation method. The distribution model of sound pressure will be established by acoustics numerical simulation method to probe how did the signals mixed, and optimize the position of sensors. Hybrid sparse dictionary and adaptive sparse optimization algorithm will be established, and combined with peak hold down-sampling method, in order to execute the signal feature-based sparse representation. The study on estimation algorithm of the matrix and signal sources will be carried out. The estimation accuracy of the mixing matrix will be improved. Sparse optimization algorithm will be combined with hybrid sparse dictionary and estimation algorithm to achieve effective separation of multi-source signals. The implementation of this project will enrich the signal blind source separation method and fault diagnosis methods, and provide the research foundation and technical support for the development of the monitoring and diagnosis methods of the aero-engine main shaft bearing.

航空发动机、双转子离心机等高端机械装备部分关键部件（如中介轴承），由于装备结构特殊，振动信号传递路径复杂、难以直接采集，造成关键部件的状态监测与故障诊断困难。为此，本课题拟以航空发动机中介轴承为研究对象，将声音信号与振动信号相结合，开展基于自适应稀疏表征的多源信号分离、故障诊断方法及相关实验研究。应用数值声学方法建立声音信号场强分布模型，探索信号混叠规律，实现故障特征信息感知优化；构建联合稀疏字典，建立自适应稀疏优化求解算法，结合峰值保持降采样算法，实现基于信号特征的自适应稀疏增强表示；研究面向稀疏增强的源信号数及混合矩阵估计方法，提高混合矩阵估计精度，结合稀疏优化求解算法，实现欠定多源信号的有效分离与故障诊断。项目的实施将丰富信号盲源分离与故障诊断方法，为拓展航空发动机中介轴承的监测与诊断方法提供研究基础和技术支撑。

中介轴承是双转子离心机、航空发动机等高端装备的关键部件，也是薄弱环节，因其故障导致的严重事故时有发生，其运行状态直接影响设备的性能和安全。本项目面向我国高端装备健康监测的重大战略需求，以中介轴承为对象，开展基于自适应稀疏表征的声振多源信号分离方法研究。重点研究了自适应稀疏字典构建方法，实现信号的理想稀疏表征；发展多源故障信号盲源分离模算法，完成强鲁棒性混合矩阵的准确估计；研究基于声音信号的故障特征提取及状态识别技术，提出多传感信息空间变换策略，建立多种高鲁棒智能诊断方法，实现故障特征信息感知与诊断。本课题的研究成果可为研发航空发动机中介轴承等高端轴承的健康监测系统提供技术支撑与应用基础。