Share

IIR document

Physics-informed acoustic leak detection in R32 multi-split systems via evolutionary machine learning.

Author(s) : MA X., YANG Z., LIU X., WANG Y.

Type of article: IJR article

Summary

The transition to flammable R32 refrigerant requires rapid and interpretable leak detection. Traditional acoustic diagnostics frequently fail under non-stationary conditions because they overlook the effect of Oil Circulation Rate (OCR) fluctuations on acoustic source mechanisms. A decrease in OCR shifts the flow regime from oil-filmdamped two-phase flow to a high-frequency, turbulence-dominated pure gas jet. This phase transition and the resulting spectral drift compromise models trained on steady-state features. We propose a hierarchical framework that integrates multi-objective feature selection with explainable machine learning. Using an acoustic database covering the evolution from normal circulation to oil-depleted states, an improved ensemble NSGA-II algorithm identifies 15 features insensitive to oil content variations. A Triangular Topology Aggregation Optimizer then tunes Support Vector Machine hyperparameters. The model achieves 31.8% dimensionality reduction and 97.27% accuracy, mitigating recognition errors caused by OCR fluctuations. A recall rate exceeding 94% is maintained for near-field micro-leaks, enabling alerts before refrigerant accumulation reaches the Lower Flammability Limit (LFL). SHAP analysis demonstrates consistency between model decisions and the physics of acoustic attenuation and fluid phase transitions, offering a physically interpretable solution for HVAC safety monitoring.

Available documents

Format PDF

Pages: 11 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: Physics-informed acoustic leak detection in R32 multi-split systems via evolutionary machine learning.
  • Record ID : 30034912
  • Languages: English
  • Subject: Technology
  • Source: International Journal of Refrigeration - Revue Internationale du Froid - vol. 186
  • Publication date: 2026/06
  • DOI: http://dx.doi.org/https://doi.org/10.1016/j.ijrefrig.2026.02.030

Links


See other articles in this issue (27)
See the source

Indexing