Scholarlink

<p>Traditional fault prediction methods for crushers have poor accuracy because of data sparsity and sample imbalance. This study proposes a temporal convolutional network (TCN)–transformer integrated framework for crusher failure prediction, leveraging data augmentation techniques and adaptive fusion of temporal and spectral features. The time-domain TCN captures the local features of time series data via causal and dilated convolutions, whereas the frequency-domain TCN extracts the periodic patterns related to faults. When the self-attention mechanism of transformer is utilised, adaptive time–frequency feature fusion is achieved, and fault features are identified at different time scales. Experiments show that the proposed dual-path TCN–transformer model outperforms the comparative models in various evaluation indicators, with an R<sup>2</sup> of 0.997, an RMSE of 0.305 and an MAE of 0.227.</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-transformer crusher fault prediction model based on data augmentation and dual-stream feature fusion

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Xingyu Qu,
  • Jianhui Zhu,
  • Peng Lu,
  • Jian Li

摘要

Traditional fault prediction methods for crushers have poor accuracy because of data sparsity and sample imbalance. This study proposes a temporal convolutional network (TCN)–transformer integrated framework for crusher failure prediction, leveraging data augmentation techniques and adaptive fusion of temporal and spectral features. The time-domain TCN captures the local features of time series data via causal and dilated convolutions, whereas the frequency-domain TCN extracts the periodic patterns related to faults. When the self-attention mechanism of transformer is utilised, adaptive time–frequency feature fusion is achieved, and fault features are identified at different time scales. Experiments show that the proposed dual-path TCN–transformer model outperforms the comparative models in various evaluation indicators, with an R2 of 0.997, an RMSE of 0.305 and an MAE of 0.227.