Consolidated modeling and prediction of heat transfer coefficients for saturated flow boiling in mini/micro-channels using machine learning methods

Flow boiling has become a reliable mode of compensating with larger power densities and greater functions of devices because it is able to utilize both the latent and sensible heat contained within a specified coolant. There are currently very few available tools proven reliable when predicting heat transfer coefficients during flow boiling in mini/micro-channels. The most popular methods rely on semi-empirical correlations derived from experimental data. These correlations can only be applied to a very narrow subset of testing conditions. This study uses a number of data science methods and techniques to accurately predict the heat transfer coefficient during flow boiling in mini/micro-channels on a database consisting of 16,953 observations collected across 50 experiments using 12 working fluids. Exploratory data science is used to obtain confidence in the data and investigate relationships between feature variables before employing machine learning algorithms. Missing data is imputed using random forest nonparametric imputation. A variety of feature analysis techniques are employed to combine and select different optimal feature variables as input values such as principal component analysis to reduce the overall dimensionality of the dataset and the Boruta package, recursive feature elimination, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and stepwise selection to reduce the number of original variables used when modeling while preserving as much information as possible. A variety of models including linear modeling, generalized additive modeling, random forests, support vector machines, and neural networks are used to predict the heat transfer coefficient and compare the results with existing universal correlations. The support vector machine model performed best, with a Mean Absolute Percentage Error (MAPE) of 11.3%. The heat flux, vapor-only Froude number, and quality proved to be especially significant contributing variables across 90% of over 110 different models. Machine learning proved to be an extremely useful tool when predicting the heat transfer coefficient across a variety of different fluids but did struggle to predict extremely high outlier data where water was the working fluid.