Non-intrusive acoustic sensing for real-time heat flux, flow regime, and fault detection in liquid cooling

WHAT IT IS A bolt-on sensing system that infers the thermal state inside an operating liquid-cooled system from the sound it makes. A wideband acoustic emission (AE) sensor mounts on the outside of the wall or housing. Machine-learning models, trained on synchronized acoustic, optical, and thermal measurements, convert the raw waveform into physical quantities: heat flux, boiling or condensation regime, and early indication of thermal failure. Nothing enters the flow path and the wetted surface is not modified.

WHY IT MATTERS Thermal limits in two-phase cooling are normally tracked with thermocouples or RTDs, which report conditions at the sensor rather than where a dryout or boiling crisis initiates, and which respond only after the wall has already overheated. Optical methods need an observation window. In sealed, compact, or high-voltage hardware, e.g., power modules, immersion-cooled electronics, motor windings, neither option is practical. Acoustic emission passes through the housing, so the measurement can be retrofitted to hardware that is already built.

HOW IT WORKS A wideband AE sensor and high-rate DAQ capture elastic waves generated by bubble nucleation, growth, departure, and collapse. Features are extracted from AE hits and their spectra, then mapped to thermal quantities by regression and classification models (CNN, ConvLSTM, and hit-based regression). Unsupervised domain-translation methods extend a trained model to new surfaces, working fluids, and pressures without retraining from scratch.

PROOF TO DATE

• Heat flux quantification from acoustic emission during pool boiling (Applied Thermal Engineering, 2023; Hit2flux, AI Thermal Fluids, 2025)
• Boiling crisis and critical heat flux detection by deep learning, including generalization across surfaces (Applied Thermal Engineering, 2021 and 2026)
• Flow regime identification in microchannel flow boiling and in flow condensation (International Journal of Heat and Mass Transfer, 2026)
• Partial discharge detection in electric motors and power modules, with the University of Arkansas Power Group (NASA ARMD)
• Open synchronized multimodal datasets released (Data in Brief, 2024)
• IP: US Patent 12,591,230 issued; US 2024/0210237 A1 pending
• Supported by NSF PFI-TT, NSF I-Corps customer discovery, and NASA

NEXT STEPS Seeking a partner with an operating thermal system and a failure mode worth catching early. A typical engagement starts with a short characterization campaign on partner hardware to establish the acoustic signature, followed by co-development toward an embedded package. Licensing is available through University of Arkansas Technology Ventures.