Machine learning regression analysis of a heat exchanger with corrugated tape turbulators, by means of second law analysis

Abstract

Although using corrugated tape turbulators to enhance heat transfer in a heat exchanger is not a new application, it is still gaining popularity. Despite the need to optimize these enhanced heat transfer systems amid the global energy crisis, in-depth studies on exergetic efficiencies, advanced exergy analysis (AEA), and Machine Learning (ML) are still lacking. Thus, this study incorporates two topics: AEA of a heat exchanger with corrugated tape turbulators, the applying ML regression models to these AEA results. The independent variables of the experimental tests include thickness (t/d), width (w/d), pitch (p/d) of the corrugated tape and Reynolds number (Re) of the fluid flow. As the AEA analysis entropy generation number (Ns), exergy destruction rate (E*), efficiency (epsilon) and NTU are obtained by using heat transfer and pressure loss calculations. The correlation between the dependent and independent values are analyzed by means of ML regression models. Four multivariate linear regression models; Lasso (Least Absolute Shrinkage and Selection Operator) model, Ridge model and Elastic Net model and the well-known multiple linear regression (MLR) model are the applied regression models to the experimental AEA results. According to the values of coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE), the results are nearly close to each other. For example the coefficient of determination (R2) is about 98% for all cases, and the errors are all close to zero, meaning all tested models give successful and reliable results. In particular, the lowest error (MAE, RMSE, MAPE) and the highest determination of regression (R2) is obtained by Lasso and ElasticNet models for all output variables. By the way, the MLR model yields the weakest performance.