Partager

Document IIF

Hybrid physics-machine learning approach for predicting manifold distribution non-uniformity effects on heat exchanger performance with discrete channel network models.

Numéro : 18405

Auteurs : YORUK B., ZABUN M.

Résumé

Distribution non-uniformity (maldistribution) across parallel passages in heat exchangers reduces thermal performance and complicates design. This study develops a Python implementation of a discrete channel network manifold model to resolve header-branch interactions and predict distribution induced losses. The model is validated through experimental testing, demonstrating close agreement across representative geometries and flow regimes. A large, systematically sampled dataset is then generated with the validated solver to train supervised machine learning surrogates for performance reduction across maldistribution scenarios. The learned models accurately reproduce solver predictions and reveal key nonlinear dependencies while substantially reducing evaluation cost. The resulting physics-informed, data-driven workflow enables rapid screening, sensitivity analysis, and optimization of heat exchangers, offering a practical alternative to repeated CFD or iterative network solves for design and control.

Documents disponibles

Format PDF

Pages : 11 p.

Disponible

  • Prix public

    20 €

  • Prix membre*

    15 €

* meilleur tarif applicable selon le type d'adhésion (voir le détail des avantages des adhésions individuelles et collectives)

Détails

  • Titre original : Hybrid physics-machine learning approach for predicting manifold distribution non-uniformity effects on heat exchanger performance with discrete channel network models.
  • Identifiant de la fiche : 30034732
  • Langues : Anglais
  • Sujet : Technologie
  • Source : 9th IIR International Conference on Sustainability and the Cold Chain. Proceedings: April 12-14 2026
  • Date d'édition : 04/2026
  • DOI : http://dx.doi.org/10.18462/iir.iccc.2026.8405

Liens


Voir d'autres communications du même compte rendu (52)
Voir le compte rendu de la conférence

Indexation