A fast design tool for compact heat exchangers tube geometry to enhance thermohydraulic performance using various AI models

This study develops an effective tool for the fast design of compact heat exchangers (CHEs) based on CFD simulations and various artificial intelligence (AI) technologies. Four AI models, namely Extreme Learning Machines (ELM), Gaussian Process Regression (GPR), Improved Stochastic Configuration Network (ISCN), and Long Short-Term Memory (LSTM), are developed and validated to predict heat transfer and flow behavior. Additionally, Response Surface Methodology (RSM), a conventional statistical method, is utilized for comparison. To train the AI models, three-dimensional CFD simulations are conducted, generating 645 distinct tube geometries that serve as modeling datasets. The performance of the AI models is evaluated using a comprehensive assessment system that includes exploratory data analysis such as box plots, heatmaps, scatter plots, and Bland-Altman plots, along with traditional statistical criteria. Results indicate that GPR exhibits superior performance, especially for datasets with more outliers, such as samples of friction factor f. When choosing the most suitable model, factors such as data distribution, computation time, and data volume need to be considered. The introduction of AI models reduces the design time of heat exchangers with CFD simulations from the hourly scale to the minute level. This study provides a valuable and fast tool for AI-assisted design of CHEs.