滚动轴承作为一种基础工业零部件，在国民经济的众多支柱产业中发挥着至关重要的作用。然而恶劣的工作环境使得滚动轴承故障率居高不下，因此，实现滚动轴承故障的精确识别，并从根源上降低故障发生的概率，对提高生产效率、保障财产和工人的生命安全具有非常显著的意义。本文围绕上述背景，从滚动轴承故障诊断和基于故障诊断结果的优化设计两个方面展开研究，主要的成果包括： 1. 针对轴承振动信号所具有的非线性、非平稳以及噪声强烈等特点，使得有效提取滚动轴承故障特征的难度较大的问题，提出了一种基于EEMD-PCA的故障特征提取与降维方法。首先使用集合经验模态分解(Ensemble Empirical Mode Decomposition，EEMD)对原始信号进行分解，进而基于相关系数法筛选出分解得到的有效分量，并在此基础上进行信号重构，然后基于重构信号提取出时频域特征，并基于有效分量的能量分布提取出能量分布特征；针对特征冗余会增加计算量却无法使故障识别准确度提高的问题，使用主成分分析(Principal Component Analysis，PCA)实现了初步提取的故障特征的降维。2. 针对多分类故障识别难度较大以及滚动体故障程度的识别在单层分类模型中存在的准确度较低的问题，建立和完善了基于极端梯度提升算法(Extreme Gradient Boosting，XGBoost)的双层故障识别模型，实现了滚动轴承故障部位及故障程度的识别。针对XGBoost算法参数选择困难的问题，通过网格搜索及交叉验证实现了XGBoost参数寻优。通过建立双层故障识别模型，显著提高了滚动体故障程度的识别准确度。利用实验数据进行验证，获得了98.5%的故障识别正确率。3. 针对现有的滚动轴承优化设计技术过于依赖理论公式、对轴承实际运行结果考虑较少的问题，提出了基于故障诊断结果的滚动轴承优化设计方法。基于Hertz弹性理论，推导了滚动轴承接触应力的计算；在此基础上，确立了联合接触应力的优化目标，结合粒子群算法建立了滚动轴承优化设计模型；通过6209型深沟球轴承的故障案例验证了该方法能够降低轴承内部接触应力，从而减小因疲劳剥落引起的轴承故障概率；利用AutoCAD及SolidWorks的二次开发实现了轴承二维图纸及三维模型的自动生成，提高了轴承设计的数字化水平。
As a basic industrial component, rolling bearing plays a vital role in many support industries of the economy. However, the harsh working environment makes the failure rate of rolling bearing remain high. Therefore, to realize the accurate identification of rolling bearing failure and reduce the probability of failure from the root is of great significance to improve the efficiency in manufacturing and guarantee the safety of property and workers. This paper studies rolling bearing fault diagnosis and optimal design based on fault diagnosis results. The main research results are shown as follows:1. To solve the problem that the bearing vibration signal has the characteristics of nonlinearity, nonstationary and strong noise, which makes it more difficult to effectively extract the fault characteristics of rolling bearing, a fault feature extraction and fault feature dimension reduction method based on EEMD-PCA is proposed. The original signal is decomposed by ensemble empirical mode decomposition, and then the decomposed effective components are selected through the correlation coefficient. On this foundation, the signal is reconstructed, and then the time-frequency domain features are extracted based on the reconstructed signal, and the energy distribution features are extracted from the energy distribution of the effective IMFs; In consideration of the problem that feature redundancy will increase the amount of calculation but can not improve the accuracy of fault recognition, principal component analysis is applied to realize the meaningful dimension reduction of the preliminarily extracted fault features.2. To solve the difficulty of multi classification fault identification and the low accuracy of rolling element fault degree identification in single-layer classification model, a double-layer fault identification model based on Extreme Gradient Boosting is established to realize the forecast of the position and degree of rolling bearing fault. To solve the difficulty of parameter selection of XGBoost algorithm, the optimization of XGBoost parameters is realized through grid search and cross validation. By establishing a double-layer fault identification model, the fault identification accuracy of rolling element fault degree is fully improved. The CWRU experimental data is selected to verify the model and the correct rate of fault recognition is 98.5%.3. Aiming at the situation that the current optimal design technology of rolling bearing relies too much on theoretical formula and considers less the actual operation results of bearing, a rolling bearing optimal design technology based on the results of fault diagnosis is proposed. Based on Hertz elastic theory, the calculation of contact stress of rolling bearing is derived; On this basis, the optimization objective of joint contact stress is established, and the optimal design model of rolling bearing is established combined with particle swarm optimization algorithm; Through the failure case of 6209 deep groove ball bearing, it is fully proved that this method can reduce the internal contact stress of the bearing, so as to reduce the bearing failure probability caused by fatigue spalling; Using the secondary development of AutoCAD and SolidWorks, the automatic generation of 2D drawing and 3D model of bearing is realized, and the digital level of bearing design is improved.
