Scholarlink

Background <p>The operating status of bearings directly affects energy consumption and carbon emission levels, and plays a key role in the “dual carbon” strategy. Improving the accuracy of bearing life prediction is an important means to achieve this goal. However, existing methods generally ignore the importance of the prediction start time of remaining useful life (RUL). This work aims to enhance RUL prediction performance by proposing an effective method with adaptive dual-domain features and initial degradation point identification.</p> Methods <p>First, the original one-dimensional acoustic emission (AE) signal is transformed into a two-dimensional grayscale image and a wavelet transform spectrum image, respectively, and the two are mixed to extract complete bearing degradation characteristics. Then, the two-dimensional mixed feature image is imported into the two-dimensional grey-wavelet convolutional neural network (2D-GWCNN) model to extract deeper features, and the recognition of initial degradation (ID) points is achieved through the discrimination mechanism. Finally, the degradation features of acoustic bearing and a reliability factor are prepared for RUL prediction.</p> Results <p>The proposed method is validated using the acoustic bearing degradation datasets obtained from the accelerated bearing life tester-1A (ABLT-1A) machine. The results show that the proposed method has good accuracy in RUL prediction compared with other recent RUL prediction methods.</p> Conclusion <p>The adaptive dual-domain feature extraction and initial degradation point identification effectively address the bias in RUL prediction start time, thereby improving the accuracy and reliability of bearing remaining useful life prediction, which provides a valuable reference for predictive maintenance of rotating machinery under the dual carbon strategy.</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

An Effective Method for Predicting Acoustic Bearing Remaining Useful Life with Adaptive Dual-Domain Features and Initial Degradation Point Identification

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Yang Yu,
  • Yue Zhang,
  • Zheng Yang,
  • Guojun Bai

摘要

Background

The operating status of bearings directly affects energy consumption and carbon emission levels, and plays a key role in the “dual carbon” strategy. Improving the accuracy of bearing life prediction is an important means to achieve this goal. However, existing methods generally ignore the importance of the prediction start time of remaining useful life (RUL). This work aims to enhance RUL prediction performance by proposing an effective method with adaptive dual-domain features and initial degradation point identification.

Methods

First, the original one-dimensional acoustic emission (AE) signal is transformed into a two-dimensional grayscale image and a wavelet transform spectrum image, respectively, and the two are mixed to extract complete bearing degradation characteristics. Then, the two-dimensional mixed feature image is imported into the two-dimensional grey-wavelet convolutional neural network (2D-GWCNN) model to extract deeper features, and the recognition of initial degradation (ID) points is achieved through the discrimination mechanism. Finally, the degradation features of acoustic bearing and a reliability factor are prepared for RUL prediction.

Results

The proposed method is validated using the acoustic bearing degradation datasets obtained from the accelerated bearing life tester-1A (ABLT-1A) machine. The results show that the proposed method has good accuracy in RUL prediction compared with other recent RUL prediction methods.

Conclusion

The adaptive dual-domain feature extraction and initial degradation point identification effectively address the bias in RUL prediction start time, thereby improving the accuracy and reliability of bearing remaining useful life prediction, which provides a valuable reference for predictive maintenance of rotating machinery under the dual carbon strategy.