Scholarlink

In order to address the challenges of fault prediction and diagnosis that are inherent to the increasing level of industrial automation and intelligence, this work proposes a composite fault diagnosis model that integrates the self-coding long- and short-term memory network (AE-LSTM) and random forest (RF). The application of the AE-LSTM is effective in the extraction of non-linear and complex features within time series data, while the Random Forest exhibits stability and high accuracy in the processing of large data sets. This model effectively combines the feature extraction function of deep learning with the classification performance of machine learning. Firstly, it captures long-term dependencies and extracts features through self-coding long and short-term memory networks. These features are then input into random forests, which enable efficient fault classification. The final application validation on the Tennessee-Eastman (TE) process dataset demonstrates that the present model achieves 95% accuracy and recall in 21 kinds of fault diagnosis. This result is superior to existing techniques and evidences the model’s excellent fault diagnosis capability.

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

Fault Diagnosis of Complex Industrial Processes Based on AE-LSTM and Random Forests

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yong Shuai Ma,
  • Lei Zhang,
  • Chang Fa Ma,
  • Guo Feng Ren,
  • Jing Jing Lin

摘要

In order to address the challenges of fault prediction and diagnosis that are inherent to the increasing level of industrial automation and intelligence, this work proposes a composite fault diagnosis model that integrates the self-coding long- and short-term memory network (AE-LSTM) and random forest (RF). The application of the AE-LSTM is effective in the extraction of non-linear and complex features within time series data, while the Random Forest exhibits stability and high accuracy in the processing of large data sets. This model effectively combines the feature extraction function of deep learning with the classification performance of machine learning. Firstly, it captures long-term dependencies and extracts features through self-coding long and short-term memory networks. These features are then input into random forests, which enable efficient fault classification. The final application validation on the Tennessee-Eastman (TE) process dataset demonstrates that the present model achieves 95% accuracy and recall in 21 kinds of fault diagnosis. This result is superior to existing techniques and evidences the model’s excellent fault diagnosis capability.