Scholarlink

<p>The remaining useful life prediction of rotating machinery under complex operational conditions remains a critical challenge. Existing methods are limited in their ability to jointly model long-term degradation trends and dynamic spatial dependencies in multi-sensor data, particularly under scenarios involving multiple fault modes or varying operational conditions. This study proposes an attention-based TCN-GCN prediction model that synergistically captures temporal-spatial features to address these limitations. The depth-wise separable temporal convolutional network enables efficient parallel computation to model long-span temporal dependencies. Meanwhile, the graph convolutional network dynamically constructs sensor correlation graphs to represent non-Euclidean spatial interactions. The proposed model, integrated with a channel attention mechanism to suppress noise interference, improves prediction accuracy and generalization capability on two different fault datasets, particularly on the FD002 and FD004 subsets of the C-MAPSS dataset characterized by multi-fault scenarios. The experiments validate the rationality and feasibility of the model, offering a robust solution for real-world predictive maintenance challenges.</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

TCN-GCN attention convolutional network for the remaining useful life prediction of rotating machinery

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Qi Sun,
  • Lili He,
  • YiChi Zhang

摘要

The remaining useful life prediction of rotating machinery under complex operational conditions remains a critical challenge. Existing methods are limited in their ability to jointly model long-term degradation trends and dynamic spatial dependencies in multi-sensor data, particularly under scenarios involving multiple fault modes or varying operational conditions. This study proposes an attention-based TCN-GCN prediction model that synergistically captures temporal-spatial features to address these limitations. The depth-wise separable temporal convolutional network enables efficient parallel computation to model long-span temporal dependencies. Meanwhile, the graph convolutional network dynamically constructs sensor correlation graphs to represent non-Euclidean spatial interactions. The proposed model, integrated with a channel attention mechanism to suppress noise interference, improves prediction accuracy and generalization capability on two different fault datasets, particularly on the FD002 and FD004 subsets of the C-MAPSS dataset characterized by multi-fault scenarios. The experiments validate the rationality and feasibility of the model, offering a robust solution for real-world predictive maintenance challenges.