基于多核超球支持向量机的滚动轴承状态定量评估方法研究

A multi-condition intelligent quantitative assessment method is studied for different fault types (inner raceway, outer raceway, rolling element) and different degrees of performance degradation of rolling bearing. The study result can show the performance degradation law of rolling bearing more accurately and has important scientific significance. The amplitude coefficient determined method is proposed for band-limited white noise that added to modified ensemble empirical mode decomposition (MEEMD) algorithm, the improved extracting sensitive component method for intrinsic mode function (IMF) obtained by MEEMD is proposed. The feature vectors obtained by several feature extraction methods are reduced by using manifold learning. Single kernel hypersphere support vector machine is optimized by combining with multiple kernel functions, and new spatial distance function of multi-kernel mapping is defined. On the basis of above results, the distance formula of hypersphere spherical centers is derived for multi-kernel mapping, and is regarded as separation index to determine the optimal kernel parameter selection range of multi-kernel classifier for further optimizing the classifier. Combining the minimum spatial distance rule of multi-kernel mapping with fuzzy membership function and spatial distance function of multi-kernel mapping, multi-condition assessment model is built for rolling bearing and is used to quantitatively assess the rolling bearing condition intelligently. By using theoretical study and experimental research to validate whether the proposed assessment method is an effective method, and it can lay the theoretical foundation of achieving rolling bearing and other equipments maintenance to "predict and prevent" methodology.

研究滚动轴承不同故障类型（内环、外环、滚动体）及不同性能退化程度的多状态智能定量评估方法，其结果可更准确地揭示滚动轴承性能退化的规律，具有重要科学意义。提出在改进集合经验模态分解（MEEMD）算法中加入带限白噪声的幅值系数确定方法、MEEMD所得的固有模态函数提存的改进算法，并利用流形学习对多种特征提取方法所得特征向量进行约简；采用组合多种核函数的方式优化单核超球支持向量机，建立新的多核映射的空间距离函数。在此基础上推导多核映射的超球球心间的距离公式，以此作为分离指数确定多核分类器核参数的最优选取范围，进一步优化分类器；提出多核映射的最小空间距离准则、模糊隶属度函数、多核映射的空间距离函数相结合的方法，建立滚动轴承多状态评估模型，智能定量评估滚动轴承的状态。通过理论和实验研究相结合的手段，验证该评估方法的有效性，为实现滚动轴承及其它设备的主动维护奠定理论基础。

为实现滚动轴承不同故障类型及不同性能退化程度的不同状态智能定量评估，更准确地揭示滚动轴承性能退化的规律，提出了一套滚动轴承不同状态定量评估方法。该套方法中提出了集合经验模态分解算法的改进方案，设定了基于能量标准差的加入白噪声准则，避免了人为确定分解参数，提高了其分解效率；提出了一种峭度值与归一化相关系数相结合的敏感固有模态函数（IMF）提取算法，可剔除虚假IMF分量和噪声干扰，凸显故障特征；提出了采用多维特征融合结合流形学习的适用于性能退化评估的特征提取与约简方法，与现有方法相比消除了特征之间的冗余性、凸显了振动信号的内在特征信息；提出了采用组合多种核函数的方式优化单核超球支持向量机（SVM），建立了新的多核映射的空间距离函数，推导了多核映射的超球球心间的距离公式，确定了核参数的最优选取范围，建立了多核超球SVM分类模型，解决了滚动轴承各状态数据分布不均匀、数据异构的问题，降低了模型训练时间，提高了分类准确率；提出了多核映射的最小空间距离准则、多核映射的空间距离函数和余弦定理相结合的评估指标，建立了滚动轴承不同状态评估模型，可定量评估滚动轴承故障位置及性能退化程度，与支持向量数据描述算法指标进行对比，验证了所提评估模型的有效性和稳定性。通过理论和实验研究相结合的手段，验证了所提该套评估方法中的振动信号特征提取、约简、智能分类方法和评估模型建立的有效性，为实现滚动轴承及其它设备的主动维护奠定了理论基础。