摘要
针对滚动轴承故障智能诊断问题,提出一种基于时序模型和可拓学的滚动轴承故障诊断方法。利用时序模型中的AR(Autoregressive Model)模型对轴承振动信号进行特征提取,以AR模型的自回归参数和残差方差作为特征向量,再利用Fisher比对AR模型的特征向量进行选择,将最终所形成的特征向量作为可拓物元模型的特征参数,以此特征参数来建立轴承不同健康状态下物元模型的经典域和所有状态下物元模型的节域。将待测数据输入到已建立的滚动轴承不同健康状态对应的物元模型之中,通过关联函数来计算待测数据与滚动轴承不同健康状态的综合关联度,实现滚动轴承状态的可拓学定性与定量诊断。进行了滚动轴承包含不同故障类型和故障程度的十种不同健康状态识别实验,每次随机选取训练样本,100次测试的平均识别率达98.86%,较基于AR模型和BP神经网络的传统诊断方法效果要好。
An intelligent fault diagnosis paradigm is proposed for rolling bearings based on autoregressive models(AR)and extension theory.AR models were employed to extract feature vectors consisting of model parameters and variance of residuals.Fisher scores were then assigned to all features of which the ones with high scores constitute final feature vectors.These feature vectors severed as inputs in the context of extension theory to determine classic domains of different matter elements corresponding to different bearing health conditions as well as joint domain encompassing all those health statuses.Once the classical and joint domains are established,the feature vector of the signal to be tested is inputted to and then a dependence degree is yielded for each of the afore-constructed matter elements.Bearing health condition is deemed the one related to the matter element which engenders the maximum dependency degree.Experiment was conducted on a bearing test rig involving a total of 10 bearing health conditions.The results reveal an 98.86%averaged identification rate was achieved by the proposed approach,which outperformed the traditional method using AR and back-propagation neural networks.
作者
雷兵
张龙
吴荣真
易剑昱
LEI Bing;ZHANG Long;WU Rongzhen;YI Jianyu(Engineering Vocational College,Jiangxi Open University,Nanchang 3330025,China;School of Mechatronics&Vehicle Engineering,East China Jiaotong University,Nanchang 330013,China)
出处
《机械设计与研究》
CSCD
北大核心
2021年第5期88-93,105,共7页
Machine Design And Research
基金
国家自然科学基金资助项目(51665013,51865010)
江西省教育厅科学技术研究项目(GJJ200616)
江西省研究生创新基金项目(YC2018-S248,YC2019-S243,YC2020-S335)。
