When you boil a liquid against a hot surface, it carries heat away remarkably well, far better than air or ordinary liquid cooling. That's why boiling-based cooling shows up wherever heat gets intense: data-center electronics, electric-vehicle batteries, high-energy lasers, nuclear reactors.
The catch is that boiling doesn't always behave. Under changing loads it can fall into oscillations, pressure and temperature swinging back and forth, and the system becomes unreliable or even unsafe. Today, engineers mostly design around the problem by keeping a wide, wasteful safety margin.
My research in Georgia Tech's MiNDS Lab, with Dr. Satish Kumar, takes the other route: instead of designing around the instability, control it. I study how these oscillations form in experiments, and use machine learning to anticipate them and actively keep the system in a stable, efficient operating range as conditions change. If the control system can watch and correct in real time, the hardware gets to run closer to its true limits.
Relevant publication:
- Adhikari, D.; & Kumar, S. "Mitigation of Flow Boiling Instabilities via Active Flow Control." IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (ITherm), 2026.
Technical detail
The work sits in two-phase flow boiling: as coolant boils along a heated channel, pressure-drop and density-wave oscillations can emerge and couple, and if they grow, heat transfer degrades sharply and surface temperatures swing. The research combines flow-boiling experiments with machine-learning-based active flow control, using measured system behavior to anticipate instability onset and actuate the flow so the system stays inside a stable operating window as heat load and flow conditions vary. Target applications span electronics thermal management, EV battery cooling, high-energy laser systems, and nuclear cooling.