复杂轮系微弱故障特征的群稀疏识别与感知方法研究

In a complex gear transmission system, the running condition is complicated and changeable, the multivariate vibrations are cross-coupled, and the signal is randomly missed during the sampling and transmitting process inevitably, which all make early fault signals difficult to be recognized, bringing hidden troubles to the system. This project studies the coupling mechanism of the multi-source signals of the complex gear system to obtain vibration symptoms of different types of faults by building dynamic models which could reflect the development mechanism of the faults. Based on the mechanism research, a sparse dictionary applicable to characterize fault signals under time-varying conditions is constructed, which could be used to sparsely decouple the vibration signals to obtain weak fault components without disturbance of normal gear vibrations. Compressive Sensing method is utilized to reconstruct the decoupled fault components in which some original information has lost, then the group relevant method and the group sparsity constraint method are applied to fuse and enhance the fault features of the weak fault signal after compression. A pre-estimation method of the compressive threshold value and signal sparsity is studied based on the methods above. Furthermore, a compressive sensing and early fault diagnosing method suitable for the complex gear transmission system is constructed. Experiments are implemented on wind turbine gearboxes to complete the experiment and the engineering verification of the proposed method. Finally, a new triune online monitoring method consisting of the decoupling of the complicated signal, the reconstruction of the deficient weak features and the warning of early faults is presented.

复杂齿轮传动系统运行工况复杂多变、多元振动交叉耦合，且在信号采集与传输过程中不可避免地存在数据随机缺失现象，使早期故障信号难于辨识，给轮系的安全运行带来隐患。本项目通过建立反映故障演变机理的动力学模型，研究复杂轮系多源信号耦合机理，得到不同类型故障的振动响应形式；基于机理研究建立适用于时变工况下稀疏表示故障分量的字典，对复杂轮系振动信号进行稀疏解耦得到去除正常轮系振动干扰的微弱故障分量；采用压缩感知理论对解耦后有缺失的故障分量进行感知重构，利用群相关和群稀疏约束对压缩降维后的微弱故障信号进行特征融合和增强；在此基础上，研究压缩阈值和信号稀疏度预估计方法，建立适用于复杂轮系振动信号压缩感知和早期故障诊断的方法。以风电齿轮箱为具体研究对象开展实验研究，完成理论方法的实验和工程验证。形成“复杂信号解耦—微弱特征缺失重构—早期故障预警”三位一体的复杂齿轮传动系统在线监测新理论。

针对复杂齿轮传动系统运行工况复杂多变、振动信号多元耦合难于进行故障辨识，以及轮系在进行长期健康监测时海量数据存储和远程传输压力大，数据随机缺失等问题，对复杂轮系多源故障信号的表征机理和不同类型故障的振动响应形式，以及复杂轮系多源耦合信号的稀疏解耦、特征提取和增强、数据压缩与重构等一系列问题开展理论和实验研究。.建立以柔性传递路径为前提的信号传递理论模型，提出了行星齿轮系统振动信号的数据模拟方法，分析了行星齿轮信号的幅值调制和频率调制产生机理，得到齿轮发生平稳型故障和冲击型故障时的不同振动调制边频带规律，为复杂轮系故障诊断奠定了基础。建立表征不同类型故障的稀疏字典，提出一种能对冲击型故障进行最稀疏表示的移不变学习字典，用于对不同工况和不同转速的局部故障信号进行稀疏表示，从而对复杂信号进行稀疏解耦。基于故障表征机理，建立了故障结构特征增强矩阵，对隐藏在噪声中的微弱故障冲击特征进行增强，提出基于冲击特征增强和非凸规则化的滚动轴承故障诊断方法。结合重叠群稀疏算法，提升了局部故障信号的稀疏性，提高了对多源故障进行特征提取的能力；基于移不变字典的循环结构特性推导了多种稀疏系数快速求解的方法，大幅提高长信号的稀疏系数求解效率。通过多数据压缩追踪方法对多通道压缩信号进行融合，在保证重构效率同时提高机械振动信号的压缩效率；通过快速迭代收缩阈值算法从压缩信号代理中提取出稀疏系数，进行信号不完全重构。得到适用于机械信号的鲁棒性观测矩阵优化框架，可以在大幅提升原始信号的压缩率的同时保证信号重构质量。归纳总结出一套齿轮传动系统冲击故障特征稀疏表示和压缩降维与重构的理论，为复杂装备的安全可靠运行和早期故障预警提供新的理论和方法支撑。.相关研究内容，共发表17篇学术论文（其中15篇SCI收录，2篇EI收录）；申请了6件发明专利（其中5件已授权，1件已受理）；相关研究共培养了1名博士和4名硕士研究生。