Artificial Neural Networks -Driven Empirical Correlations for Enhanced Understanding of Closed-Loop Pulsating Heat Pipe Dynamics (CLPHP)

In the context of compact thermal devices producing excess heat, the effective control of thermal energy presents a complex challenge across diverse engineering domains. Closed Loop Pulsating Heat Pipes (CLPHPs), distinguished by their continuous pulsating vapor bubbles and liquid slug motions, emerge as a compelling solution for mitigating elevated heat loads. Their notable attributes, including exceptional thermal conductivity and adaptability to diverse orientations, position CLPHPs as a promising choice. Widely utilized in aerospace and electronics applications, the performance of CLPHPs is intricately tied to both design and operational considerations, mandating optimization for heightened efficiency. The primary objective of this study is to establish a semi-empirical correlation derived through the innovative application of Artificial Neural Networks (ANN). This correlation is based on a set of dimensionless numbers—Prandtl (Pr), Stephan (St), and Ohnesorge (Oh)— alongside key parameters such as the filling ratio (FR), inclination angle (hetaheta), heat flux (q") and a design characteristic factor. The resulting correlation proves instrumental in characterizing and optimizing CLPHPs across diverse applications as it contributes to predicting the thermal behavior of CLPHPs. The coolant temperature is considered as the characteristic temperature for dimensionless number calculation particularly during the early design stage when the average temperature of the evaporator and condenser is often unknown. A comprehensive database comprising varied CLPHP geometries and multiple operational points, extracted from existing literature, forms the basis of this research with 1400 experimental records. ANN models are constructed to predict CLPHP thermal resistance (K/W), and their hyperparameters are optimized for optimal accuracy. The model’s efficacy is rigorously validated against this dataset, demonstrating strong agreement between predicted and actual results. Achieving a low Mean Squared Error (MSE) of 0.058, the ANN model establishes itself as a robust and accurate tool for predicting CLPHP thermal performance. This research not only advances our understanding of CLPHPs but also establishes a bridge between empirical correlations and ANN modeling, serving as a valuable reference for the broader application of pulsating heat pipes and offering enhanced predictive capabilities in complex thermal scenarios.

Work In Progress

Contributeurs
Mira Ibrahim
Majed-Eddine Moustaid
Contact
mira.ibrahim@capgemini.com
Thématique
Transferts en milieux hétérogènes
Mots-clés
Closed Loop Pulsating Heat Pipe Dynamics
Artificial Neural Networks