Fault diagnosis method of rolling bearing under variable load based on unsupervised feature alignment

Abstract

A rolling bearing fault diagnosis method under variable load based on unsupervised feature alignment belongs to the field of rolling bearing fault diagnosis. Aiming at the lack of certain load data in the actual work of rolling bearings, the data in the source domain and the data in the target domain belong to different distributions and the samples in the target domain do not contain labels. Using variational mode decomposition combined with singular value decomposition to obtain the time-frequency characteristics of the vibration signal, and then combining the time-domain and frequency-domain characteristics of the vibration signal to construct a multi-domain feature set; introducing a subspace alignment algorithm that can achieve unsupervised domain adaptation in transfer learning and An improvement is made to combine the kernel mapping method with the SA algorithm. Map the training data and test data to the same high-dimensional space, perform feature alignment in the subspace of the high-dimensional space, and realize the alignment of source domain features to target domain features under different loads. In the case that there is no label in the target field, using the known load data of rolling bearings to identify the state corresponding to other load data has a high accuracy of fault diagnosis.

Description

technical field [0001] The invention relates to a rolling bearing fault diagnosis method under variable load, belonging to the field of rolling bearing fault diagnosis. Background technique [0002] Rolling bearings are key components of rotating machinery and are widely used in industrial production. Fault diagnosis of rolling bearings will effectively ensure the normal and smooth operation of equipment and prevent major accidents [1]. Rolling bearings often work under variable load conditions, resulting in the lack of or inability to obtain training data with the same distribution as the data to be tested in actual work [2]. Fault diagnosis of unknown tag vibration signals under other loads based on known tag vibration signals has important practical significance [3]. [0003] Machine intelligent fault diagnosis mainly includes feature extraction, fault diagnosis and prediction [4]. The time-frequency feature extraction methods of rolling bearing vibration signals have b...

AI Technical Summary

Problems solved by technology

[0007] In view of the lack of some load data in the actual work of rolling bearings, which makes the data in the source domain and the data in the target domain belong to different distributions, and the samples in the target domain do not contain labels, a rolling bearing fault with multi-domain feature construction and unsupervised feature alignment is proposed. diagnosis method

Images