A multi-modal approach to pool boiling dynamics: Acoustic and vibration data for regime classification and heat flux prediction

This paper presents a multi-scale approach to pool boiling regime classification and heat flux prediction by integrating acoustic, vibration, and imaging techniques. Pool boiling experiments are conducted on a wire heater, and data is collected using microphones, accelerometers, and high-speed cameras. Power spectral density analysis is performed on the acoustic and vibration data, revealing distinct frequency responses that correlate with bubble dynamics and heat flux variations. The spectral power distribution is then used to classify the boiling regimes, identifying transitions from nucleate boiling to critical heat flux (CHF). Proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) are applied to correlate the boiling frequencies and observe the structural dynamics of the bubbles. POD captured over 98% of the system’s energy in the first mode, while DMD identified significant frequency modes between 917.24 Hz and 9799.12 Hz, highlighting key transitions in the boiling process. Simple binary decision tree models are then employed to evaluate the performance of both acoustic and vibration datasets in predicting heat fluxes. The models achieved R${}^2$ scores consistently above 0.99. Principal component analysis (PCA) reduced the complexity of the acoustic and vibration datasets by over 97% while retaining 99% of the variance. The results demonstrate that binary decision tree models are sufficient for predicting heat fluxes with relatively low error. This study provides a framework for real-time classification of boiling regimes and heat flux prediction, offering valuable insights into the dynamics of pool boiling and contributing to the development of safe thermal management systems.