Sign in Subscribe
Sensors

Fault Detection and Diagnosis Model for Data Center Cooling System

Strategic Insights

— 2 min read

A new research study has investigated the impacts of sensor measurement error on the reliability of data centers. The study, funded by the National Natural Science Foundation of China and Northeastern University, assessed the effects of sensor error on the convolutional neural network (CNN) based fault detection and diagnosis (FDD) model for the data center composite cooling system. The results showed that sensor errors can significantly degrade the accuracy of the FDD model, particularly in vapor compression mode, and that the accuracy threshold for fixed error is 0.2 K in the CNN model.

Key Takeaways:

  • The FDD model for the cooling system is beneficial in elevating the reliability of data centers, but its accuracy can be degraded by sensor measurement error.
  • The CNN-based FDD model is sensitive to sensor error, particularly in vapor compression mode, where a negative fixed sensor error of 1 K leads to an average 5% greater decline in accuracy compared to a positive error of the same magnitude.
  • Sensor errors have a negligible impact on model accuracy until exceeding a threshold of 0.2 K, and the evaporating temperature error is critical to FDD model accuracy.
  • In fixed bias conditions, when the error magnitude is 1 K, the accuracy of the FDD model decreases within the range of 24.8% to 45.1%.
  • The study highlights the importance of sensor accuracy and maintenance in ensuring the reliability of data center cooling systems.
  • The research has implications for the design and operation of data center cooling systems, particularly in terms of sensor placement and calibration.
  • The study's findings can inform the development of more robust and reliable FDD models for data center cooling systems.

Statistics:

  • The accuracy of the FDD model decreases by an average of 5% due to a negative fixed sensor error of 1 K in vapor compression mode.
  • The heat pipe mode is more susceptible to sensor error, where a positive fixed sensor error of 1 K causes a 6.5% higher decrease in accuracy.
  • The accuracy threshold for fixed error is 0.2 K in the CNN model.
  • In fixed bias conditions, when the error magnitude is 1 K, the accuracy of the FDD model decreases within the range of 24.8% to 45.1%.

Sources:

  • Effects of Sensor Measurement Error On Fault Detection and Diagnosis Model for Data Center Composite Cooling System. International Journal of Refrigeration, 2025;175:245-258.
  • Northeastern University, Sch Met, Sep Key Lab Ecoind, Shenyang 110819, People's Republic of China.

Read more