复杂装备轴承多重故障的线性判别分析与反向传播神经网络协作诊断方法

发布时间：2018-06-11 14:29

本文选题：机械学 + 轴承多重故障诊断　；参考：《兵工学报》2017年08期

【摘要】：由于复杂装备运行工作环境恶劣,导致其轴承多重故障诊断的准确率不高,为此提出一种基于线性判别分析(LDA)与反向传播(BP)神经网络协作下复杂装备轴承数据驱动的多重故障诊断方法。将无量纲指标作为轴承多重故障数据的反映指标,利用LDA对轴承多重故障的无量纲指标数据进行线性映射降维处理;通过拉格朗日极值法获得最佳投影向量,沿着该方向将轴承多重故障数据投影到类别最易区分的方向;将经投影处理后的样本作为BP神经网络的输入样本,通过训练测试网络,实现轴承多重故障的预测分类。对某型装备大型旋转机械机组进行仿真实验,验证了所提方法能够有效对轴承多重故障进行降维映射,并且能较好地实现多重故障分类诊断,具有良好的有效性和实用性。
[Abstract]:Because of the bad working environment of complex equipment, the accuracy of bearing multi-fault diagnosis is not high. In this paper, a multiple fault diagnosis method based on LDA and BP neural network for complex equipment bearing data drive is proposed. The dimensionless index is used as the reflection index of bearing multi-fault data, the dimensionless index data of bearing multi-fault is reduced by using LDA, and the optimal projection vector is obtained by Lagrange extreme value method. Along this direction, the bearing multi-fault data are projected to the most easily distinguished direction, and the samples processed by projection are used as input samples of BP neural network, and the prediction and classification of bearing multiple faults are realized by training and testing network. The simulation experiments on a large rotating machine unit of a certain type of equipment show that the proposed method can effectively reduce the dimension of multiple faults of bearings, and can better realize the classification and diagnosis of multiple faults, which has good effectiveness and practicability.
【作者单位】：重庆交通大学信息科学与工程学院;广东石油化工学院广东省石化装备故障诊断重点实验室;
【基金】：国家自然科学基金项目(61663008、61573076、61473094、61304104、61004118) 教育部留学归国人员科研启动基金项目(2015-49) 重庆市高等学校优秀人才支持计划项目(2014-18) 广东省石化装备故障诊断重点实验室开放式基金项目(GDUPKLAB201501、GDUPKLAB201604) 重庆市研究生教育教学改革研究重点项目(yjg152011) 重庆市高等教育学会2015—2016高等教育科学研究课题项目(CQGJ15010C)
【分类号】：TH133.3

【相似文献】