Scholarlink

With the increasing complexity of industrial processes, fault detection has become crucial for ensuring equipment safety and production efficiency. Traditional methods fail to capture intricate relationships in nonlinear scenarios and struggle with processing large-scale complex industrial datasets. This paper proposes a fault detection model, called 3DCNN-LSTM, that combines a three-dimensional convolutional neural network (3DCNN), long short-term memory network (LSTM), and multi-head attention mechanism. The model extracts spatiotemporal features using 3DCNN, captures long-term temporal dependencies with LSTM, and enhances the perception of key feature correlations through the attention mechanism. The TE (Tennessee Eastman) dataset, a benchmark for chemical process simulation, contains multivariate time-series data with 21 predefined fault modes and is extensively utilized for evaluating the performance of process monitoring, fault detection, and diagnostic algorithms. The proposed method demonstrates significantly higher diagnostic accuracy on this dataset compared to existing approaches.

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

Dual-Channel 3DCNN-LSTM Fault Detection for Industrial Processes

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jinlong Han,
  • Ni Bu,
  • Xiaohong Yin,
  • Shaoyuan Li,
  • Wentao Liu

摘要

With the increasing complexity of industrial processes, fault detection has become crucial for ensuring equipment safety and production efficiency. Traditional methods fail to capture intricate relationships in nonlinear scenarios and struggle with processing large-scale complex industrial datasets. This paper proposes a fault detection model, called 3DCNN-LSTM, that combines a three-dimensional convolutional neural network (3DCNN), long short-term memory network (LSTM), and multi-head attention mechanism. The model extracts spatiotemporal features using 3DCNN, captures long-term temporal dependencies with LSTM, and enhances the perception of key feature correlations through the attention mechanism. The TE (Tennessee Eastman) dataset, a benchmark for chemical process simulation, contains multivariate time-series data with 21 predefined fault modes and is extensively utilized for evaluating the performance of process monitoring, fault detection, and diagnostic algorithms. The proposed method demonstrates significantly higher diagnostic accuracy on this dataset compared to existing approaches.