Thermal and two-phase cooling analytics
Machine learning for liquid cooling that knows its own limits.
We research trustworthy machine learning for boiling heat transfer and critical heat flux. Our work focuses on honest evaluation: benchmarks that prevent data leakage, calibrated uncertainty, and clear warnings when a model is outside the conditions it was trained on.
Why it matters
Two-phase and direct liquid cooling can remove far more heat than air, but the margin to failure is narrow. When a cold plate approaches dryout or a boiling surface nears critical heat flux, temperatures can rise quickly. Models used to monitor or predict that behavior need to be evaluated as carefully as the hardware they protect.
Leakage-safe benchmarks
Random splits of boiling data can place adjacent windows from the same physical run on both sides of the train and test boundary, which overstates accuracy. We design grouped evaluations that hold out whole runs and surfaces so reported performance reflects new conditions.
Critical heat flux and dryout
We study multimodal signals such as acoustic emission, hydrophone and thermal measurements to estimate heat flux and flag approach to critical heat flux, with the goal of supporting safer operating margins for cold plates and two-phase loops.
Calibrated uncertainty
A prediction is only useful if its confidence is honest. We report calibration alongside accuracy and treat out-of-distribution detection as a first-class result, so a model can say when it should not be trusted.
How we work
Our research follows a few consistent principles.
Public data first
We build on openly released datasets so results can be checked by others.
Conservative claims
We report the evaluation protocol with every number and state limitations plainly.
Reproducible artifacts
Code is archived with a permanent DOI and a fixed commit, alongside each preprint.
Physics in the loop
Splits, features and checks are chosen with the underlying heat transfer in mind.
Recent work
Preprints and software archived on Zenodo.
Contact
We welcome conversations with operators, equipment makers and researchers working on liquid and two-phase cooling.
brad@bcptech.dev