Presentation Information
[2C16]Multi-Physics Approach for Machine Learning-Based Critical Heat Flux Prediction
*Abdullah Al Mahmud1,2, Yoshihiro Tokushima1, Koji Morita1, Wei Liu1 (1. Kyushu Univ., 2. BAEC)
Keywords:
Critical heat flux (CHF),Heat transfer,Machine learning (ML),Physics-driven ML
Critical heat flux (CHF) is a complex phenomenon and a key safety metric in boiling heat transfer systems. Its prediction is challenging because of multiple mechanisms being present in flow boiling—such as departure from nucleate boiling (DNB) and dryout—. While physics-drivenmachine learning (PDML) improves over black-box machine learning (ML), conventional PDML does not possess enough physical knowledge to represent multi-mechanisms simultaneously. This work introduces a multi-physics driven ML (MPDML) framework that integrates multiple physical models within ML architecture. The database is seperated based on CHF mechanism, and ML learns from the residual in critical equlibrium quality from experimental to model results. MPDML adaptively selects the governing mechanism during prediction. Trained on a large dataset, MPDML significantly improves accuracy and physical consistency across diverse geometries and flow conditions for multi-CHF mechanisms.