Share

IIR document

A semi-supervised data-driven approach for chiller refrigerant leakage detection.

Number: 0005

Author(s) : FENG Z., WANG L., MA X., JIANG Z., CHANG B.

Summary

It’s very difficult to design a completely sealed chiller system, so refrigerant leakage is almost the most common fault in a positive pressure cycle. When a refrigerant leak occurs, chillers will have higher power consumption, even causing health and safety accidents in a closed environment. Leaking refrigerants with high global warming potential (GWP) will accelerate the greenhouse effect. This study presents a semi-supervised machine learning approach to detect refrigerant leakageand all data used for detecting leakage are from pre-installed sensors. A sophisticated experimental method was designed to collect data from a centrifugal chiller and the algorithm of anomaly detection using long short-term memory (LSTM-AD)is discussed with reconstruction error. The LSTMencoder and decoder models are trained on normal data and is used to detect leakage. It’sverified that detection sensitivity can reach 6% and the best detection coverage for leakage 6%, 11% and 16% are respectively 66%, 95% and 95%.

Available documents

Format PDF

Pages: 8 p.

Available

  • Public price

    20 €

  • Member price*

    Free

* Best rate depending on membership category (see the detailed benefits of individual and corporate memberships).

Details

  • Original title: A semi-supervised data-driven approach for chiller refrigerant leakage detection.
  • Record ID : 30031015
  • Languages: English
  • Subject: Technology
  • Source: 3rd IIR conference on HFO Refrigerants and low GWP Blends. Shanghai, China.
  • Publication date: 2023/04/05
  • DOI: http://dx.doi.org/10.18462/iir.HFO2023.0005

Links


See other articles from the proceedings (28)
See the conference proceedings

Indexing