Predictions of Heat Transfer Characteristics in Air-cooled Finned Tube Bundles Using Advanced Machine Learning Models and Techniques

The paper focuses on improving and applying machine learning (ML) methods and models used to forecast the heat transfer rate of air-cooled heat exchangers made up of horizontal single-row tube bundles. The purpose of the research is to enhance the accuracy and efficiency of predictive models by integrating advanced algorithms and data analytics techniques. Using larger datasets and refining the modeling process, the study hopes to achieve more reliable predictions that can optimize the design and operation of heat exchangers. The bundles of finned tubes were surrounded by an airflow rising upward caused by the action of buoyancy forces under conditions of thermo-gravitational (when an exhaust shaft is installed above the bundle) or free convection (without a shaft). Unsteady Navier–Stokes equations were used to conduct massive multiparameter direct numerical simulations with varying geometric and regime parameters. The tabular datasets applied for training the ML models were built on estimates of the average Nusselt number and the integral skin friction coefficient for the bundles. The features of the formulation of the physical problem, the mathematical model, and computational aspects are outlined and discussed; illustrations of computed fields and local distributions at the fin surface are given. The structure of the datasets, the characteristics of the ML models involved, related technologies, and software tools are considered. The results of a comparative analysis of the quality of trained ML models, interpretation of prediction results, and forecasts of target variables, including symbolic correlations approximating the datasets, are provided. The initial experience in developing surrogate models based on active learning methods is presented and analyzed.