Optimized Experimental Observation and Measurements of Pool Boiling Heat Transfer Using Computer Vision Techniques

With power dissipation rates of modern technologies on the rise, enormous efforts are going toward leveraging one of nature’s most effective thermal management mechanisms, known as boiling. In this endeavor, researchers have come to recognize the potential of visual data as it can provide real-time bubble characteristics in relation to the boiling heat flux. Key parameters to existing heat transfer performance models are nucleation site density, bubble departure diameter and bubble departure frequency. Although boiling images are richly embedded with these bubble statistics, conventional methods of extracting and analyzing such data is laborious and requires extensive user involvement. This potentially introduces errors, producing simplified and unrealistic prediction models that ultimately muddle our understanding of boiling processes. Although recent advancements in machine learning have enabled more efficient ways of processing data to classify boiling characteristics, there have been very few efforts in using image-based machine learning methods(i.e., computer vision) to correlate bubble dynamics with heat transfer performance. Such contributions of detailed data can be exploited to refine the existing, weak, theoretical models. Therefore, this study seeks to address these challenges and aid in the efforts of understanding the fundamental physics governing pool boiling through visualization techniques. We present a state of the art experimental setup and procedure that can be integrated into other laboratories as well as applied to various pool boiling experiments studying differing conditions(e.g., enhanced surfaces or fluids). Our optimized pool boiling setup is designed and fabricated to ensure excellent structural integrity, thermal insulation, and system efficiency all while allowing easy and accurate image and thermal data acquisition. This setup works in tandem with our machine learning-based computer vision model that is capable of autonomously capturing a large bandwidth of spatio-temporal bubble statistics and quantifying heat transfer performance from high-resolution boiling images. Overall, our experimental methodology and non-invasive vision-based approach to studying pool boiling dynamics, gives rise to improved heat transfer performance to progress efforts in managing thermal dissipation of high-power devices