Multi-Objective Optimization of Shell-and-Tube Heat Exchanger Performance Using CFD-Driven Surrogate and Learning-Based Methods

Authors:

Abstract
Shell-and-tube heat exchanger performance is governed by competing thermal and hydraulic objectives that single-variable optimisation cannot resolvesimultaneously. This study introduces a unified surrogate-assisted frameworkin which baffle inclination angle θ ∈ [-25◦,25◦] and perforation diameterd ∈[2,5] mmare treated as continuous co-design variables. A twelve-memberANN ensemble trained on 35 CFD-generated samples serves as the frozenevaluation oracle for four independent optimisation paradigms: NSGA-II,Bayesian optimisation, particle swarm optimisation, and deep reinforcementlearning. Saturation diagnostics confirmed surrogate fidelity prior to everyrun. Perforation diameter emerged as the dominant hydraulic lever, withlarge perforations triggering a shift from recirculation-dominated cross-flowto jet-assisted wake suppression that reduces friction substantially withoutproportional thermal penalty. DRL-PPO located the globally superior configuration (θ = 20.88◦, d = 4.46 mm), achieving a CFD-verified thermohydraulic ratio of j/f = 0.01264: an 18.6% gain over the conventionalperpendicular unperforated baseline, a 20.0% improvement over the PECoptimal solid inclined reference, and a 35.1% friction factor reduction relative to the same inclined solid reference. All surrogate prediction errorsremained below 2.8% at verified optima, establishing physical credibility ofthe reported improvements.

This publication has 43 references indexed in Scilit: