The flow field inside the heat exchangers is associated with maximum heat transfer and minimum pressure drop. Designing a compact heat exchanger and employing various techniques to enhance its overall performance has been widely investigated and still an active research field. However, few researches deal with thermal optimization. The application of elliptic tube is an effective alternative to circular tube which can reduce the pressure drop significantly. In this study, numerical simulation and optimization of variable tube ellipticity is studied at low Reynolds numbers. The three-dimensional numerical analysis and a multi-objective genetic algorithm (MOGA) with surrogate modelling is performed. Two row tubes in staggered arrangement in fin-and-tube heat exchanger is investigated for combination of various elliptic ratio (e = minor axis/major axis) and Reynolds number. Tube elliptic ratio ranges from 0.2 to 1 and Reynolds number ranges from 150 to 750. The tube perimeters are kept constant while changing the elliptic ratio.
The numerical model is derived based on continuum flow approach and steady-state conservation equations of mass, momentum and energy. The flow is assumed as incompressible and laminar due to low inlet velocity. Results are presented in the form of Colburn factor, friction factor, temperature contours and streamline contours. Results show that increasing elliptic ratio increases the friction factor due increased flow blocking area, however, the effect on the Colburn factor is not significant. Moreover, tube with lower elliptic ratio followed by higher elliptic ratio tube has better thermal-hydraulic performance.
To achieve maximum heat transfer enhancement and minimum pressure drop, the Pareto optimal strategy is adopted for which the CFD results, Artificial neural network (ANN) and MOGA are combined. The tubes elliptic ratio (0.2 ⩽ e ⩽ 1.0) and Reynolds number (150 ⩽ Re ⩽ 750) are the design variables. The objective functions include Colburn factor (j) and friction factor (f). The CFD results are input into ANN model. Once the ANN is computed and its accuracy is checked, it is then used to estimate the model responses as a function of inputs. The final trained ANN is then used to drive the MOGA to obtain the Pareto optimal solution. The optimal values of these parameters are finally presented.