Scholarlink

<p>In modern industry, the efficient operation of equipment is of critical importance. Unexpected failures can result in losses and risks. Therefore, Remaining Useful Life (RUL) prediction is extremely vital for equipment maintenance and cost control. Existing RUL prediction techniques encounter efficiency bottlenecks when processing long-span data and exhibit limited feature extraction capabilities within a single domain, which restricts the model’s ability to learn from long-time-series data. To address this issue, this paper designs the GDRP-Net to improve the Informer model, significantly enhancing its capacity to learn long-time-series features. The study extracts a total of 23 features from the time domain, frequency domain, and time-frequency domain, forming a comprehensive original feature set. These features are then filtered based on criteria of monotonicity, correlation, and robustness. Subsequently, Principal Component Analysis (PCA) is employed for feature fusion and dimensionality reduction to obtain key features and generate life span labels. Finally, the design-optimized PE-Informer model, which is tailored for attention mechanisms and long-time-series dependency processing, is utilized to model and predict the processed health indicators. Experiments on the XJTU-SY dataset demonstrate that the proposed method outperforms the Convlstm, Transformer models and SGACNN, with reductions in Root Mean Square Error (RMSE) by 45.54%,32.2% and 59.27%, respectively, and Mean Absolute Error (MAE) by 52.22%,29.03% and 65.78% respectively. This provides an efficient solution for predictive maintenance of industrial equipment.</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

A rolling bearing remaining useful life prediction method based on the PE-informer model

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Liyang Bai,
  • Yuheng Liu,
  • Hailong Duan

摘要

In modern industry, the efficient operation of equipment is of critical importance. Unexpected failures can result in losses and risks. Therefore, Remaining Useful Life (RUL) prediction is extremely vital for equipment maintenance and cost control. Existing RUL prediction techniques encounter efficiency bottlenecks when processing long-span data and exhibit limited feature extraction capabilities within a single domain, which restricts the model’s ability to learn from long-time-series data. To address this issue, this paper designs the GDRP-Net to improve the Informer model, significantly enhancing its capacity to learn long-time-series features. The study extracts a total of 23 features from the time domain, frequency domain, and time-frequency domain, forming a comprehensive original feature set. These features are then filtered based on criteria of monotonicity, correlation, and robustness. Subsequently, Principal Component Analysis (PCA) is employed for feature fusion and dimensionality reduction to obtain key features and generate life span labels. Finally, the design-optimized PE-Informer model, which is tailored for attention mechanisms and long-time-series dependency processing, is utilized to model and predict the processed health indicators. Experiments on the XJTU-SY dataset demonstrate that the proposed method outperforms the Convlstm, Transformer models and SGACNN, with reductions in Root Mean Square Error (RMSE) by 45.54%,32.2% and 59.27%, respectively, and Mean Absolute Error (MAE) by 52.22%,29.03% and 65.78% respectively. This provides an efficient solution for predictive maintenance of industrial equipment.