基于压缩感知和盲源分离的风电齿轮传动系统早期故障特征提取与诊断方法研究

The gear transmission system is the important part of wind turbine generators. To identify and diagnose the incipient fault accurately and timely, which initiates and evolves during the operation of wind generator gear transmission system, would be of great significance for the safe operation of wind turbine and the avoidance of the heavy accidents. The project takes the wind generator gear transmission system as the research object and conducts basic scientific research on compression and reconstruction of fault signal, feature extraction and diagnosis of incipient fault. The specific contents are as follows: The overcomplete dictionary that matches the fault feature of the wind generator gear transmission system will be built. The sparse representation method of fault signal will be studied, and then the fault feature sparse representation of the wind generator gear transmission system will be realized. Then the compressed sensing reconstruction algorithm based on multiscale manifold learning is proposed. In this method, the compression and reconstruction for fault signal will be realized. Combining compressed sensing theory and blind source separation method, the blind source separation method based on compressed sensing will be studied and the fault source signal will be separated effectively. Then the information entropy will be introduced into time-frequency analysis. The method for weak fault feature extraction and diagnosis based on improved ensemble local mean decomposition and adaptive multiscale time-frequency entropy will be studied. The incipient fault feature extraction and diagnosis will be realized. Finally, the effectiveness of the theories will be tested through fault simulation experiments. To summarize, this project aims to establish a new incipient fault feature extraction and diagnosis method of wind generator gear transmission system. It will lay the theoretical foundation for fault diagnosis technology of wind generator gear transmission system.

齿轮传动系统是风电机组的重要组成部件，准确及时识别诊断风电齿轮传动系统运行过程中萌生和演变的早期故障，对风电机组安全运行，避免重大事故的发生具有重要意义。本项目以风电齿轮传动系统为研究对象，对故障信号压缩重构及早期故障特征提取与诊断方法进行基础科学研究。具体内容包括：构建风电齿轮传动系统故障特征过完备字典，研究故障信号稀疏表示方法，实现故障特征的稀疏表示，进而提出基于多尺度流形学习的压缩感知重构算法，实现风电齿轮传动系统故障信号压缩与重构；将压缩感知理论与盲源分离方法相结合，研究基于压缩感知的盲源分离方法，实现故障源信号的有效分离；将信息熵引入时频分析，研究基于改进总体局部均值分解和自适应多尺度时频熵的早期微弱故障特征提取与诊断方法；最后在实验台进行故障模拟实验，验证理论方法的有效性。本项目旨在研究一种新的风电齿轮传动系统早期故障特征提取与诊断方法，为风电齿轮传动系统故障诊断奠定理论基础。

随着风电产业的持续快速发展和风电机组规模的不断扩大，风电机组故障不断涌现。齿轮传动系统作为风电机组的重要组成部件，其工作状况影响整个风电机组的性能，准确及时识别诊断齿轮传动系统运行过程中萌生和演变的故障，对风电机组安全运行，避免重大事故的发生具有重要意义。本项目面向风电机组运行状态监测与故障诊断重大工程需求，以风电机组齿轮传动系统为研究对象，研究了风电齿轮传动系统故障特征提取与诊断方法。本项目按照研究计划完成了研究工作，并取得了一定的研究成果，完成的主要研究工作如下：.(1)建立了风电机组齿轮传动系统动力学模型，分析了不同类型故障下齿轮传动系统振动响应特征，建立了不同故障状态与齿轮传动系统振动响应之间的映射关系，揭示了风电机组齿轮传动系统振动响应随故障发生、发展的变化规律。.(2)通过挖掘故障振动信号中包含的故障特征信息，建立了稀疏信息与故障特征之间的关系模型，研究了信号稀疏表示中包含的特征信息与信号原始特征空间中故障特征的映射机制，基于自适应分块的故障特征稀疏表示方法，根据信号特征调整分块长度，达到均衡信号块稀疏的目的，实现了故障特征稀疏表示。.(3)研究了基于优化测量矩阵模型的故障特征增强方法，将测量矩阵稀疏表示为基测量矩阵和稀疏基测量矩阵形式，采用分段重构方法重构故障信号，通过对测量矩阵和故障敏感特征参量进行分析，利用测量矩阵以较低的平均采样率采集信号，直接对压缩采集信息进行敏感特征提取和故障识别诊断。.(4)研究了基于变分模态分解的欠定盲源分离方法，通过对观测信号进行变分模态分解，将分解得到的分量和观测信号组成新的观测信号，在压缩感知框架下，研究了基于最大类间方差优化密度峰值聚类的欠定盲分离方法，实现了故障信号欠定盲分离。.(5)研究了基于改进局部均值分解和瞬时能量分布的齿轮故障特征提取与诊断方法，利用局部时频熵表示故障信号在不同时频段的能量分布，研究了基于多尺度时频熵的故障特征提取方法，实现了故障特征提取与识别诊断。