Quantifying Steam Dropwise Condensation Heat Transfer via Experiment, Computer Vision and Machine Learning Algorithms

Abstract

Condensation of water vapor on metal surfaces has a plethora of engineering applications in the energy sector. This study demonstrates an approach of estimating the condensation heat transfer performance of a hydrophobic copper plate utilizing Artificial Intelligence (AI) techniques. The situation under study is an isothermal vertical cooled plate with an attached thin copper sheet exposed to a water-vapor environment devoid of non-condensable gases. Water was chosen due to its high latent heat of vaporization and its extensive industrial use, while copper was selected because of its high thermal conductivity (385 W/m-K). A thin layer of Teflon coating was applied to the copper surface to render it hydrophobic, allowing dropwise condensation (DwC) to be investigated. The experimental setup featured a glass condensation chamber, which allowed visualization of the condensation phenomena on the condenser; a cold plate attached to a chiller, to hold and cool down the condenser; a vacuum pump to maintain low-pressure conditions; a heat source along with pressure and temperature measurement instrumentation. The condensation setup allows to directly calculate the heat transfer coefficient of dropwise condensation and, at the same time, offers visual access to the condenser. The overall efficiency of the dropwise condensation mechanism depends on many key factors, such as nucleation density and rate, maximum droplet size and efficient condensate drainage. The surface-condensation phenomena can be video-recorded while using the instrumentation to measure the heat-transfer efficiency in real time. Computer Vision algorithms were employed to process the raw video files and generate data on the basic aspects of condensing droplet dynamics (number, diameter, area coverage, etc.). The data was evaluated and cross-compared with the heat-transfer coefficients determined from instrumentation, allowing to consider how the droplet characteristics change with experimental operating conditions. Machine learning algorithms, such as Artificial Neural Network (ANN), Multi-Layer Perceptron (MLP), General Regression Neural Network (GRNN), and one ensemble model, Gradient Boosting Regression (GBR) can help to directly determine the heat transfer performance of a specific surface (with known wettability characteristics) from its condensation characteristics. The hypothesis is explored that the visual characteristics of the condensate droplet dynamics can be used in conjunction with trained machine learning models to forecast the DwC heat transfer efficiency of this system without directly measuring the corresponding variables (e.g., temperatures, flow rate, etc.).