Transfer learning has been widely used in train bearing fault diagnosis. However, most existing methods are hindered under different working conditions in practical engineering applications. To fill this gap, this paper proposes an improved joint distribution adaptation algorithm (IJDA) based on Wasserstein distance. It first extracts multi-dimension features from time domain, frequency domain and entropy domain of raw signal to express the individual information of different fault types while reducing the number of input dimension. Meanwhile, by using Wasserstein distance as the metric of the K nearest neighbor algorithm, the distance between the source domain and target domain samples in the feature space is effectively pulled in, which improves the classification accuracy of JDA significantly. Experiments from two dataset with different probability distribution is designed, and results show that the effectiveness and robustness of the proposed method is superior to that of other state-of-art transfer learning methods.
Fault Diagnosis of Train Wheelset Bearings Based on Improved Joint Distribution Adaptation
Lect. Notes Electrical Eng.
International Conference on Electrical and Information Technologies for Rail Transportation ; 2021 October 21, 2021 - October 23, 2021
Proceedings of the 5th International Conference on Electrical Engineering and Information Technologies for Rail Transportation (EITRT) 2021 ; Chapter : 61 ; 542-549
2022-02-23
8 pages
Article/Chapter (Book)
Electronic Resource
English
Fault Diagnosis of Train Wheelset Bearings Based on Improved Joint Distribution Adaptation
British Library Conference Proceedings | 2022
|Wheelsets - Reconditioning wheelset bearings
Online Contents | 2004
European Patent Office | 2015
|