Improved forecasting using a novel hybrid machine learning model for flow parameters in an advanced heat exchanger

İngilizce (OpenAlex)

• Developed a novel hybrid ML model for superior thermo-hydraulic predictions in PHEs. • Achieved high accuracy in predicting Nu, f, and P using multi-stage training. • Demonstrated the superiority of SE-GPR with R 2 = 0.985 and NSE = 0.981. With the rising demand for high-performance thermal management systems in critical industries such as aerospace, nuclear energy, and advanced manufacturing, optimizing the efficiency of heat exchangers is crucial. Plate heat exchangers (PHEs) are crucial for enabling efficient fluid heat transfer; however, predicting their thermo-hydraulic properties remains challenging due to the complex fluid-thermal interactions. This research presents a new integrated machine learning ML hybrid model, augmented with advanced analytics, to strengthen the prediction of thermo-hydraulic characteristics in advanced PHEs. A unique plate design, incorporating a hyperbolic tangent function for wavy structures and heat transfer with minimal frictional losses. By using the SST k-ω turbulence model, high-resolution simulations provide a detailed analysis of complex flow. The hybrid ML model undertaken in this study passes through four stages, within which the predictions that are being developed perform subsequent iterations through some classical models that include Coarse Gaussian SVMR (CG-SVMR), Matern Regression (MR), Quadratic SVMR (Q-SVMR), Rational Quadratic GPR (RQ-GPR), Robust Linear Regression (RLR), and Squared Exponential GPR (SE-GPR). The model’s adjustment is guided by performance indicators, including R 2 , SI, NSE, MAE, and RMSE. Ultimately, the SE-GPR emerged as the best-performing model in the final stage, achieving the highest performance metrics with R 2 = 0.985, RMSE = 0.015, and NSE = 0.981. The current findings suggest that the model is capable of predicting the Nusselt number (Nu), friction factor (f), and performance (P). Additionally, findings confirmed that the proposed geometry enhances heat transfer by up to 28% while maintaining lower frictional losses than conventional chevron-type plates. The study further provides a strong framework for the optimization of PHE designs to satisfy industrial needs for higher efficiency and cost reduction. Integrating machine learning techniques in the design of thermal systems can pave the way for novel improvements in heat exchanger design.