Inverse thermal design method for heat exchangers via digital twin model integrating deep learning surrogates with local thermal resistance evaluation

This paper introduces an inverse design method for plate-fin heat exchangers (PFHEs) integrating digital twin, deep learning and inversion algorithm, which significantly differs from the conventional design route. The local thermal resistance is calculated and regarded as the evaluation criterion for achieving longitudinal geometric parameters for PFHEs more meticulously and precisely. A complete inverse design methodology is proposed, including database construction and database expansion based on digital twin and mapping generation with a convolutional neural network (CNN). The fin pitch and staggered distance in cross-flow PFHEs are optimized. The result shows that, after 2000 training iterations, the coefficient of determination (R2) value reaches greater than 0.99 with a maximum error below 2%, indicating satisfactory convergence. Meanwhile, under the maximum flow resistance limitation, the PFHE structure obtained by the proposed algorithm displays the lowest thermal resistance value of 0.047 K·W⁻¹. As a result, compared to conventional structures optimized by mature empirical correlations, the total thermal resistance of optimized PFHE could be further reduced by 7.4% under the same flow resistance and heat transfer area. When comparing the flow resistance under the condition of the same thermal resistance, the optimized PFHE by the inverse design method exhibits a further reduction by 12.5%.