Domain asymmetry factor weighted rolling bearing fault depth local migration diagnosis method

Abstract

The invention discloses a domain asymmetric factor weighted rolling bearing fault depth local migration diagnosis method. The method comprises the following steps: firstly, extracting depth migrationfault features in source rolling bearing and target rolling bearing monitoring data by utilizing a depth residual error network; secondly, training a domain confusion network by using deep migration fault features, and calculating domain asymmetry factors; calculating the multi-core maximum mean value difference of the fault features of the deep residual network adaptation layer again, and suppressing the contribution degree of invalid fault features in the source rolling bearing by using domain asymmetry factor weighting; and finally, constructing a target function by using the weighted multi-kernel maximum mean value difference, and training a deep residual network. The trained local migration diagnosis model composed of the domain confusion network and the deep residual network can effectively overcome the adverse effect of domain asymmetry factors on migration diagnosis. The unbalanced health state of the target rolling bearing is identified by using the local diagnosis knowledge of the source rolling bearing, and the diagnosis precision of the migration diagnosis model is significantly improved.

Description

technical field [0001] The invention belongs to the technical field of rolling bearing fault diagnosis, and in particular relates to a rolling bearing fault depth local migration diagnosis method weighted by a domain asymmetry factor. Background technique [0002] Rolling bearings are one of the core components of large-scale rotating machinery. Once a failure occurs, it will cause huge economic losses, and even endanger the lives of personnel. Therefore, its healthy service is very important. Fault intelligent diagnosis uses advanced machine learning technology to construct the mapping relationship between bearing monitoring data and health status, which greatly reduces the excessive reliance on expert prior knowledge in the diagnosis process, especially with the vigorous development of deep learning technology in recent years, The intelligent level and diagnostic accuracy of fault intelligent diagnosis have been significantly improved, and it has become an important means ...

AI Technical Summary

Problems solved by technology

[0003] The existing rolling bearing migration diagnosis technology has significant limitations: the diagnostic knowledge domains of the source bearing and the target bearing are symmetrical, that is, it is required that ① the data of the target bearing must be evenly distributed in each health state; The size is equal to the tag space size of the target bearing data

However, in engineering practice, these two points are often untenable: the target bearing is in a normal state for a long time during service, and the frequency of fault states is significantly less than that of the normal state. In addition, the fault state generated by the source bearing may not occur on the target bearing. , therefore, the data distribution of the target bearing is severely unbalanced (contains a large amount of normal information and a small amount of fault information), and the label space of the source rolling bearing data often covers the label space of the target bearing, finally forming an asymmetric source bearing and target bearing diagnosis knowledge area

[0004] Affected by the asymmetric factors of the diagnostic knowledge domain, it is difficult for existing migration diagnostic techniques to effectively use the diagnostic knowledge of the source bearing to identify the unbalanced health status of the target bearing

Images