UC Irvine

Two-phase boiling is a highly effective heat transfer mechanism for high-power thermal management systems, but its performance is governed by complex, transient, and spatially distributed liquid–vapor interfacial dynamics. Conventional measurements such as temperature, pressure, and flow rate provide valuable system-level information, yet they do not directly resolve the bubble-scale and interface-scale processes that control heat transfer degradation, critical heat flux, flow instability, and regime transition. This dissertation develops a vision-based and event-based framework for characterizing, modeling, and diagnosing flow boiling heat transfer by transforming visual data streams into physically meaningful representations of interfacial behavior.First, a flow boiling experimental platform is developed to enable high-speed visualization under controlled thermal and hydrodynamic conditions. Building on this platform, a vision-based bubble digitization framework is established using image segmentation, temporal tracking, motion estimation, and feature extraction. This framework converts raw high-speed flow boiling images into quantitative descriptors of bubble morphology, motion, vapor coverage, and wall wetting behavior. These descriptors provide a structured representation of liquid–vapor dynamics that can be analyzed across operating conditions and connected to macroscopic heat transfer behavior.The extracted interfacial features are then used to improve the physical interpretation of microgravity flow boiling heat transfer. Under microgravity conditions, suppressed buoyancy leads to prolonged vapor residence, bubble coalescence, and reduced liquid replenishment near the heated surface. To account for these effects, image-derived wetting and dryout descriptors are incorporated into heat transfer coefficient analysis. Similarly, vision-derived interfacial parameters are integrated into mechanistic critical heat flux models, providing a more direct connection between vapor–liquid structure and macroscopic thermal performance. These results demonstrate that visual features can serve not only as image-based predictors, but also as physically interpretable descriptors of boiling heat transfer mechanisms.In addition to frame-based imaging, this dissertation investigates neuromorphic event-camera sensing for flow boiling analysis. Unlike conventional cameras, event cameras asynchronously record pixel-level brightness changes generated by moving interfaces and transient vapor structures. Event-derived temporal, spectral, and spatial descriptors are developed to quantify interfacial activity, temporal unsteadiness, and spatial redistribution in flow boiling. The relationship between these descriptors and pressure drop behavior further demonstrates that event streams contain thermofluidically relevant information beyond qualitative visualization.Finally, event-driven learning architectures are evaluated for real-time flow regime classification. Among the tested models, the Event LSTM framework provides an effective balance between accuracy, latency, and computational efficiency by directly processing raw event sequences without event-frame reconstruction. The Event LSTM achieves 97.6% classification accuracy with a processing time of 0.28 ms, demonstrating the potential of sparse event streams for low latency monitoring of dynamic two-phase flow regimes.Overall, this dissertation demonstrates that visual and event-based sensing can move flow boiling analysis beyond qualitative observation toward quantitative feature extraction, physically interpretable modeling, and real-time diagnostic prediction. By connecting high-speed imaging, computer vision, neuromorphic event sensing, machine learning, and heat transfer analysis, this work provides a foundation for more interpretable, efficient, and intelligent thermal-fluid monitoring systems.