Deep learning-enhanced investigation of R134a/R245fa flow boiling in microchannels

Zeotropic mixtures are pivotal for high-temperature heat pumps (HTHPs), yet their heat transfer performance is significantly hampered by interfacial mass transfer resistance. This study integrates HSV with a deep learning (DL) framework (YOLOv11 and enhanced SORT) to quantify the impact of mass transfer resistance on bubble dynamics during flow boiling of R134a/R245fa mixtures in a microchannel. We establish a bubble growth suppression factor (BGSF) that directly correlates the extent of mass transfer resistance with macroscopic heat transfer coefficient (HTC) deterioration. Our results demonstrates that mixture composition profoundly suppresses bubble growth and slip velocity. The 70 wt% R245fa/30 wt% R134a mixture exhibited the most significant suppression, with a mean bubble length of only 3.3 mm at 104.2 kW/m²—a 65% suppression compared to pure R134a. Crucially, a strong quantitative link is established between the BGSF and the observed HTC reduction. Building upon these insights, a modified HTC correlation incorporating bubble dynamics parameters was developed, reducing the MAE to 10.1% and improving prediction reliability (91.7% of data within ±30%). This work establishes a data-driven, mechanistic framework that bridges microscale interfacial phenomena with system-level thermal performance.