Scholarlink

<p>Diesel engines, characterized by high torque and economic efficiency, are widely employed as power units. However, its harsh working environment makes it prone to failures, potentially resulting in economic losses or endangering personnel safety. Traditional rule-based systems often fail to detect subtle fault features under complex conditions, leading to warnings delayed. To address this issue, we propose the C-LGA (Cascade LSTM–GRU with Data Augmentation) fault prediction model: a framework that integrates signal prediction with fault diagnosis and is adept at capturing newly emerging faults through predictive signals, thereby enabling diesel engine fault forecasting. Focusing on vibration data from diesel engines, the C-LGA model demonstrates improvements over baseline models (CNN and Transformer) in both prediction and diagnostic outcomes. Compared to the CNN, it achieves a 33% reduction in root-mean-square error (RMSE) of signal prediction and an 11% increase in fault diagnosis accuracy; relative to the Transformer, it yields a 26.08% reduction in RMSE of signal prediction and a 75% improvement in accuracy. This method can effectively predict early-stage faults in diesel engines and provide timely warning information, thus preventing fault occurrence or escalation, lowering operational costs, and improving operational safety; it can likewise be applied to other large-scale industrial equipment domains, including wind turbine systems and aircraft engine systems.</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

Research on diesel engine fault prediction method based on deep learning and vibration information

  • 期刊论文

  • 发表日期：

    2025年8月25日

查看全文
  • Zhangming Peng,
  • Weiye Zhang,
  • Rong Liu

摘要

Diesel engines, characterized by high torque and economic efficiency, are widely employed as power units. However, its harsh working environment makes it prone to failures, potentially resulting in economic losses or endangering personnel safety. Traditional rule-based systems often fail to detect subtle fault features under complex conditions, leading to warnings delayed. To address this issue, we propose the C-LGA (Cascade LSTM–GRU with Data Augmentation) fault prediction model: a framework that integrates signal prediction with fault diagnosis and is adept at capturing newly emerging faults through predictive signals, thereby enabling diesel engine fault forecasting. Focusing on vibration data from diesel engines, the C-LGA model demonstrates improvements over baseline models (CNN and Transformer) in both prediction and diagnostic outcomes. Compared to the CNN, it achieves a 33% reduction in root-mean-square error (RMSE) of signal prediction and an 11% increase in fault diagnosis accuracy; relative to the Transformer, it yields a 26.08% reduction in RMSE of signal prediction and a 75% improvement in accuracy. This method can effectively predict early-stage faults in diesel engines and provide timely warning information, thus preventing fault occurrence or escalation, lowering operational costs, and improving operational safety; it can likewise be applied to other large-scale industrial equipment domains, including wind turbine systems and aircraft engine systems.