基于Teager能量算子和深度置信网络的滚动轴承故障诊断

Fault damage degrees diagnosis for rolling bearing based on Teager energy operator and deep belief network

针对传统的分类器对滚动轴承早期微弱故障进行诊断时泛化能力不强的问题,提出基于Teager能量算子(TEO)和深度置信网络(DBN)的滚动轴承故障诊断方法。先用TEO提取滚动轴承振动信号中的瞬时能量,构造相应的特征向量;采用层次优化算法调整DBN结构参数,生成合适的分类器。应用美国西储大学轴承实验振动信号,对不同类型、不同损伤程度的滚动轴承进行故障诊断,对比分析DBN、支持向量机(SVM)和邻近算法(KNN)的分类准确性。研究结果表明:DBN能更准确、稳定地识别滚动轴承各种故障,具有较强的泛化能力。

Considering that the traditional classifiers＇ generalization ability is not strong in the early fault diagnosis of rolling bearings, the fault diagnosis method based on Teager energy operator （TEO） and deep belief network （DBN） were put forward. Firstly, the instantaneous amplitudes of the vibration signal were calculated by TEO, and the instantaneous energies of the signal were extracted. Then the characteristic vectors were constituted with the instantaneous energies. DBN classifiers were used to identify the faults of rolling bearing. For different types of fault diagnosis, DBN structure parameters were adjusted according to the classification error rate of training sets. Using the bearing fault experiments＇ data of American West Storage University, the classification accuracy of SVM and KNN was compared. The results show that the suggested methods are more effective and stable for the identification of rolling bearing fault diagnosis in various situations.