面向动轴齿轮传动故障诊断的振动模型驱动稀疏表征理论研究

Since the gear damage fault mechanism of moving-axis gear transmission systems is still not clear, the dynamic response signal is complex and its fault feature is difficult to extract, which bring the great difficulties to the fault diagnosis of moving-axis gear transmission systems, it is urgent and necessary to research the fault diagnosis theory and method for moving-axis gear transmission systems, in order to make the mechanical equipment work safely and reliably. Firstly, the fault mechanisms of NGW planetary gear transmission systems, planetary gear transmission systems with small tooth number difference and cycloidal pin gear epicyclic transmission systems will be revealed respectively, and the fault vibration signal model will be set up. After that, the overcomplete flexible wavelet transform with a customized time-frequency sampling which leads to high denoising performance will be explored, and its wavelet basis construction method will be developed. Moreover, by using vibration signal model, the construction method of the wavelet basis with high approximation accuracy will be proposed. Then the influence law of the construction parameters on the wavelet basis properties will be illustrated. With the overcomplete flexible wavelet basis and the vibration model-driven wavelet basis, the corresponding signal sparse representation theory will be proposed. This model-driven sparse representation theory will be used to extract fault feature, and then the feature fusion method based on the sparse manifold clustering and embedding algorithm and the fault classification method with least square support vector machine will be researched, so as to realize accurate and intelligent fault identification. The achievements obtained by this project will hopefully establish the key scientific foundation for the fault diagnosis of moving-axis gear transmission systems.

由于动轴轮系行星齿轮传动系统轮齿损伤的故障机理尚不明确，动态响应信号复杂且故障特征难以提取，给其故障诊断带来了很大的困难，因此急需开展关于动轴轮系行星齿轮传动系统故障诊断理论与方法的研究，以保证机械装备安全可靠运行。在系统揭示NGW行星齿轮传动系统、少齿差行星齿轮传动系统和摆线针轮行星传动系统的故障机理以及建立其故障振动信号模型的基础上，提出可自定义时频采样率并具有很高降噪性能的过完备柔性小波变换与其小波基构造方法，以及具有很高逼近精度的振动信号模型驱动的小波基构造方法，阐明构造参数对小波基特性的影响规律，构建基于过完备柔性小波基和振动模型驱动的小波基的信号稀疏表征理论体系；然后将该模型驱动稀疏表征理论用于提取故障特征，并通过研究监督式稀疏流形聚类嵌入特征融合算法和最小二乘支持向量机故障分类方法，实现高精度的故障智能识别。本项目取得的成果有望为动轴轮系行星传动系统故障诊断奠定关键科学基础。

动轴轮系行星传动系统在直升机、风力发电机、重型车辆、坦克等重要装备中得到了广泛应用。这些传动系统中一旦出现故障，就可能引发连锁反应，导致整个装备停机，从而造成重大事故和巨大的经济损失。本项目针对动轴轮系行星齿轮传动系统故障诊断中故障机理不明、信号异常复杂、故障特征提取困难等关键难题，开展了故障机理、特征提取、智能诊断等研究，取得了以下代表性成果：1）建立了动轴轮系行星齿轮传动系统的故障动力学模型，分析了齿轮磨损、裂纹、点蚀、剥落、断齿等典型故障下的动态响应，为故障诊断提供了指导；2）提出了M子带柔性小波变换，设计了频域小波滤波器组，形成了基于快速傅里叶变换的快速实现算法，提高了对行星齿轮传动系统振动信号的降噪性能；3）构造了基于振动模型的冲击小波，建立了冲击分量提取的稀疏优化模型，提出了基于变量分裂和可分离代替函数的迭代阈值收缩算法，提高了强谐波和噪声干扰下微弱冲击故障特征提取的精度；4）提出了自适应初始字典K-SVD算法，攻克了从强干扰中快速自适应地提取故障冲击分量的难题，为一维信号快速K-SVD处理奠定了理论基础；5）创新性地构建了Isigmoid与ReLTanh两个梯度稳定的函数，建立了相应的深度学习诊断模型，以及提出了多尺度迁移投票机制，提高了复杂工况下故障诊断的精度和鲁棒性。相关成果在IEEE TIE/TII/TIM、MSSP、JSV、MMT等国内外著名期刊和会议上发表SCI收录论文25篇，EI收录论文1篇，ISTP收录会议论文2篇；获授权发明专利2件，申请发明专利5件；获重庆市科技进步一等奖1项、教育部科技进步一等奖1项和重庆产学研创新成果奖一等奖1项。通过本项目培养了博士研究生3名与硕士研究生6名。项目成果在直升机主减行星传动系统和风电机组行星齿轮箱故障诊断中得到了应用，为保证重要装备安全可靠运行提供了支撑。