RESEARCH PAPER
A Multi-scale Attention Mechanism Diagnosis Method with Adaptive Online Updating Based on Deep Learning under Variable Working Conditions
More details
Hide details
1
College of Automation Engineering,, Nanjing University of Aeronautics and Astronautics, China
2
Institute of Intelligent Manufacturing, Nanjing Tech University, China
3
College of Electrical Engineering and Control Science, Nanjing Tech University, China
Submission date: 2024-06-14
Final revision date: 2024-07-31
Acceptance date: 2024-09-04
Online publication date: 2024-09-26
Publication date: 2024-09-26
Corresponding author
Ningyun Lu
College of Automation Engineering,, Nanjing University of Aeronautics and Astronautics, No. 169 Sheng Tai West Road, Jiang Ning District,, 211106, Nanjing, China
HIGHLIGHTS
- Multi-scale attention method with adaptive online updating for variable conditions.
- Flexibly updates diagnostic models based on online data status.
- Adaptive weight random undersampling balances inter-class data uniformly.
KEYWORDS
TOPICS
ABSTRACT
With the advance of industrial systems, the online equipment fault diagnosis has encountered many challenges such as data drift and data imbalance under varying operating conditions, thus making stable and accurate diagnosis increasingly critical. Considering the above issues, a multi-scale attention mechanism diagnosis method with adaptive model that can be updated based on deep learning has been proposed. The method is composed of four main steps: training the multi-scale offline diagnosis model, transferring the parameters of the offline model, assessing the degree of data drifting, and adaptively updating the diagnostic model. A data balance strategy with adaptive weight balances both inter-class and intra-class data. The method updates the diagnostic model flexibly according to online data status, to reduce the impact of data drifting. The method was verified on a bearing test rig, which can reproduce the common bearing faults under variable working conditions. The experimental results have shown that the proposed method can accurately and reliably identify the bearing faults.
ACKNOWLEDGEMENTS
This work was supported in part by the National Natural Science Foundation of China under Grant 62020106003, Grant 61873122, Grant 62203213, and Grant 62303217, in part by the Aero Engine Corporation of China Industry-university-research cooperation project under Grant HFZL2020CXY011, in part by the the Research Fund of State Key Laboratory of Mechanics and Control of Mechanical Structures under Grant MCMS-I-0121G03, in part by the National Key Research and Development Program of China under Grant 2021YFB3301300, in part by the Natural Science Foundation of Jiangsu Province under Grant BK20220332, in part by the China Scholarship Council under Grant 202206830124, and in part by the Natural Science Foundation of the Jiangsu Higher Education Institutions of China under Grant 23KJB510006.