Data for: Nonintrusive heat flux quantification using acoustic emissions during pool boiling

Monitoring two-phase cooling systems is crucial to avoid thermal runaways
and device failures. Nonintrusive monitoring methods using remote sensing,
e.g., optical and acoustic sensors, are desired to avoid interfering with
bubble dynamics and ease replacement. Compared to image-based
technologies, sound-based sensors are cheaper and do not require the same
environment as cameras. Acoustic signals during pool boiling have been
used to identify boiling states, but acoustic-based quantitative
predictions have been challenging. The present work presents a machine
learning framework to determine the heat flux during pool boiling using
acoustic signals captured through a hydrophone. This framework
investigates and compares the performance and computational cost of six
machine learning models by coupling two feature extraction algorithms
(fast Fourier transform and convolution) and three different regressors
(multilayer perceptron, random forest, and Gaussian process regression).
The fast Fourier transform-Gaussian process regression model is found to
be the most promising, with high accuracy and the lowest computational
cost. A parametric study is performed to investigate the effect of the
temporal length and sampling rates on the model predictions. It is found
that the model’s performance is improved with increasing temporal lengths
of the acoustic sequences for all sampling rates. Acoustic features below
512 Hz are found to be most significant for heat flux predictions. For
sampling rates beyond 512 Hz, the model performance is dictated by the
temporal length of the acoustic sequences.