An Intelligent Plate Fin-and-tube Heat Exchanger Design System Through Integration of CFD, NSGA-II, ANN and TOPSIS.

This paper attempts to develop an intelligent plate fin-and-tube heat exchanger (PFTHE) design system, which is entirely self-programming, to achieve quickly design. The proposed design system consists of four modules: (1) formulation, (2) optimization, (3) post-processing, and (4) decision-making. The proposed design system is implemented and validated with the application of shape optimization of ellipse tubes of plate-fin heat exchanger. In the formulation module, the physical problem to be studied is mathematized and the main design variables will be determined. In the optimization module, a famous algorithm, non-dominated sorting genetic algorithm of type II (NSGA-II), is embedded in an in-house Multi-concept Heat Transfer (MHT) code to achieve call CFD simulation during optimization process. To reduce computation time, Open multi-processing (OpenMP) is employed. The optimal solutions (Pareto solutions) obtained by the optimization module will be stored in the database and also taken as the input for the post-processing module and the decision-making module. In the post-processing module, Artificial neural network (ANN) is utilized to establish the correlation between design variables and heat transfer performance indicators assisting engineers to quickly design. As a short cut for heat exchanger design, both forward and backward designs have been implemented. Finally, in the decision-making module, technique for order preference by similarity to an ideal solution (TOPSIS) is applied to determine the best compromise solution from Pareto solutions according to the actual requirements provided by users. Results show that the proposed design system could determine a best compromise solution by reducing the pressure drop (80%) of the tube bundle without sacrificing too much heat transfer performance (5%) and also save much time for designers. For the forward design, the ANNs taken six decision variables as inputs are modelled to forecast two objectives has reached acceptable precisions. This research provides a promising tool for PFTHE optimization to improve heat transfer and comprehensive performance, also for quickly design based on historical simulation or experimental results.