Predicting boiling heat flux, heat transfer coefficient, and regimes Non-intrusively using external acoustics and deep learning

Abstract

Accurate monitoring of boiling heat transfer is critical for the safety and efficiency of high energy-density systems, including data center cooling, nuclear reactors, and industrial boilers. Traditional diagnostic methods relying on intrusive sensors or visual inspection become impractical in harsh industrial environments characterized by high pressures, temperatures, and radiation exposure. In this paper, we propose a non-intrusive diagnostic framework combining externally measured acoustic emission (AE) signals with advanced deep learning techniques. Pool boiling experiments were conducted from natural convection to critical heat flux (CHF), and AE signals were externally collected under various boiling conditions. Through a comprehensive evaluation of hundreds of models, a transformer-based model demonstrated optimal performance, simultaneously predicting key boiling parameters-heat flux, heat transfer coefficient (HTC), and boiling regime-with prediction errors of less than 20% for heat flux and HTC, and over 98% accuracy in boiling regime classification. Further validation on subcooled flow boiling confirmed robust generalizability. Our results reveal that frequency-domain characteristics of AE signals strongly correlate with boiling phenomena, enabling interpretable and reliable diagnostics. This method provides simultaneous prediction of critical boiling parameters without invasive instrumentation, significantly enhancing operational safety and improving reliability in thermal management systems.