Scholarlink

<p>Fouling accumulation in plate heat exchangers (PHE) significantly deteriorates thermal performance and increases operational costs. Accurate early prediction of fouling is essential for optimizing cleaning schedules and maintaining energy efficiency. This study proposes a fouling prediction framework integrating accelerated fouling experiments and time-series models to achieve effective health monitoring of heat exchangers. An experimental platform was established to simulate composite fouling involving both crystallization and particulate deposition. Throughout the experiment, temperatures, flow rates, and pressures on both the hot and cold sides were systematically monitored. The predictive performance of multilayer perceptron (MLP) model, Long Short-Term Memory (LSTM) model and CNN-BiLSTM-Attention hybrid model was comparatively evaluated. Results demonstrated that the CNN-BiLSTM-Attention hybrid model demonstrates excellent performance in predicting the cold-side outlet temperature (R<sup>2</sup> = 0.9975). This represents a significant improvement over MLP models (R<sup>2</sup> = 0.8697) and LSTM models (R<sup>2</sup> = 0.9815). Furthermore, a health condition value (HCV) based on fouling thermal resistance was introduced for dynamic assessment and early warning of performance deterioration. The findings reveal that the proposed HCV index effectively captures the temporal evolution of fouling accumulation. Alerts are triggered when the HCV reaches the predefined threshold, thereby improving operational reliability and reducing risks of unplanned downtime or safety incidents.</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

Data-driven fouling prediction and health monitoring of plate heat exchangers based on accelerated fouling experiments and deep learning models

  • 期刊论文

  • 发表日期：

    2026年7月2日

查看全文
  • HuiTing Zhang,
  • Gang Hou,
  • GuangZhong Huang,
  • JingWei Liu,
  • MeiJiao Jiang,
  • Han Jiang,
  • Zhoujian An,
  • Dong Zhang

摘要

Fouling accumulation in plate heat exchangers (PHE) significantly deteriorates thermal performance and increases operational costs. Accurate early prediction of fouling is essential for optimizing cleaning schedules and maintaining energy efficiency. This study proposes a fouling prediction framework integrating accelerated fouling experiments and time-series models to achieve effective health monitoring of heat exchangers. An experimental platform was established to simulate composite fouling involving both crystallization and particulate deposition. Throughout the experiment, temperatures, flow rates, and pressures on both the hot and cold sides were systematically monitored. The predictive performance of multilayer perceptron (MLP) model, Long Short-Term Memory (LSTM) model and CNN-BiLSTM-Attention hybrid model was comparatively evaluated. Results demonstrated that the CNN-BiLSTM-Attention hybrid model demonstrates excellent performance in predicting the cold-side outlet temperature (R2 = 0.9975). This represents a significant improvement over MLP models (R2 = 0.8697) and LSTM models (R2 = 0.9815). Furthermore, a health condition value (HCV) based on fouling thermal resistance was introduced for dynamic assessment and early warning of performance deterioration. The findings reveal that the proposed HCV index effectively captures the temporal evolution of fouling accumulation. Alerts are triggered when the HCV reaches the predefined threshold, thereby improving operational reliability and reducing risks of unplanned downtime or safety incidents.