Sign in Subscribe
Air conditioning equipment

Fault Diagnosis Strategy for Refrigerant Leakage in Internet Data Centers Presented

Strategic Insights

— 1 min read

Researchers at the Huazhong University of Science and Technology have developed a fault diagnosis strategy for refrigerant leakage in internet data centers (IDCs) using deep neural networks (DNNs). The study highlights the significant energy wastage and health risks associated with refrigerant leakage in IDCs. The researchers' approach utilizes the Gini coefficient to select important feature variables and develops a DNN-based fault diagnosis model.

Key Takeaways:

  • The IDCs air conditioning system is a significant energy consumer, and refrigerant leakage can lead to unnecessary energy waste and health risks.
  • The researchers developed a fault diagnosis strategy for refrigerant leakage using DNNs, which achieved an accuracy of 99.99% and a geometric mean accuracy (GMA) of 99.92%.
  • The proposed DNN model showed great potential for online data classification, with a small amount of online data used to update the model resulting in improved classification performance.
  • The accuracy of the model increased by 26.62%, from 73.66% to 93.27%, and the false alarm rate (FAR) decreased from 32.82% to 0%.
  • The study utilized on-the-spot experiments to collect operational data, including refrigerant charge under normal conditions and five various leakage levels.
  • The researchers demonstrated the effectiveness of the proposed DNN model in real-world scenarios, with the model being trained and tested using real IDCs air conditioning system data.

Statistics:

  • Accuracy: 99.99%
  • Geometric mean accuracy (GMA): 99.92%
  • False alarm rate (FAR): 0%
  • Missing alarm rate (MAR): 0%
  • Accuracy increase: 26.62% (from 73.66% to 93.27%)
  • False alarm rate (FAR) decrease: 32.82%

Sources:

  • A Fault Diagnosis Strategy for Refrigerant Leakage of the Air Conditioning System In High-efficiency Internet Data Centers. Energy and Buildings, 2025;346.
  • Elsevier Science Sa, PO Box 564, 1001 Lausanne, Switzerland (www.elsevier.com)
  • Information Technology Newsweekly, November 4, 2025, p 219.

Read more