Scholarlink

Purpose <p>Aero-engine bearings often operate under varying working conditions, and obtaining large amounts of data for training purposes is challenging in practice. To address these challenges and enhance the generalization capability of intelligent fault diagnosis (FD) models, a reconstructed meta ensemble learning (RMEL)-based fault identification technique is proposed.</p> Methods <p>First, the original data are reconstructed using an improved denoising autoencoder (IDAE) to remove noise and ensure the quality of the original data. Then, the one-dimensional vibration signals are converted into time-frequency maps, which are fed into a network incorporating the squeeze-and-excitation network (SENet) for feature extraction. At this stage, a model-agnostic meta-learning (MAML) strategy is employed for training, and parameter optimization is continuously performed to enable the model to adapt quickly when facing tasks under unknown working conditions. Finally, the concept of ensemble learning is introduced to train multiple meta-learners, and a voting mechanism is used for FD.</p> Results and Conclusion <p>The proposed method is validated through two cases, achieving fault identification accuracies of 99.14<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\varvec{\%}\)</EquationSource> </InlineEquation> and 96.24<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\varvec{\%}\)</EquationSource> </InlineEquation> under limited samples and varying conditions, outperforming other methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Few-Shot Aero-engine Bearings Fault Diagnosis Based on Reconstructed Meta Ensemble Learning Under Varying Working Conditions

  • 期刊论文

  • 发表日期：

    2025年10月30日

查看全文
  • Jiajun Pan,
  • Ke Zhang,
  • Xue Li

摘要

Purpose

Aero-engine bearings often operate under varying working conditions, and obtaining large amounts of data for training purposes is challenging in practice. To address these challenges and enhance the generalization capability of intelligent fault diagnosis (FD) models, a reconstructed meta ensemble learning (RMEL)-based fault identification technique is proposed.

Methods

First, the original data are reconstructed using an improved denoising autoencoder (IDAE) to remove noise and ensure the quality of the original data. Then, the one-dimensional vibration signals are converted into time-frequency maps, which are fed into a network incorporating the squeeze-and-excitation network (SENet) for feature extraction. At this stage, a model-agnostic meta-learning (MAML) strategy is employed for training, and parameter optimization is continuously performed to enable the model to adapt quickly when facing tasks under unknown working conditions. Finally, the concept of ensemble learning is introduced to train multiple meta-learners, and a voting mechanism is used for FD.

Results and Conclusion

The proposed method is validated through two cases, achieving fault identification accuracies of 99.14 \(\varvec{\%}\) and 96.24 \(\varvec{\%}\) under limited samples and varying conditions, outperforming other methods.