EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A PINN Surrogate Modeling Methodology for Steady-State Integrated Thermofluid Systems Modeling

Kristina Laugksch, Pieter Rousseau, Ryno Laubscher

Mathematical and Computational Applications · 27.03.2023 · 28 (2) · 52

DOI: 10.3390/mca28020052 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mathematical and Computational Applications, online erschienen am 27.03.2023. Verfasst von Kristina Laugksch und 2 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Research Foundation, University of Cape Town.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Physics-informed neural networks (PINNs) were developed to overcome the limitations associated with the acquisition of large training data sets that are commonly encountered when using purely data-driven machine learning methods. This paper proposes a PINN surrogate modeling methodology for steady-state integrated thermofluid systems modeling based on the mass, energy, and momentum balance equations, combined with the relevant component characteristics and fluid property relationships. The methodology is applied to two thermofluid systems that encapsulate the important phenomena typically encountered, namely: (i) a heat exchanger network with two different fluid streams and components linked in series and parallel; and (ii) a recuperated closed Brayton cycle with various turbomachines and heat exchangers. The results generated with the PINN models were compared to benchmark solutions generated via conventional, physics-based thermofluid process models. The largest average relative errors are 0.17% and 0.93% for the heat exchanger network and Brayton cycle, respectively. It was shown that the use of a hybrid Adam-TNC optimizer requires between 180 and 690 fewer iterations during the training process, thus providing a significant computational advantage over a pure Adam optimization approach. The resulting PINN models can make predictions 75 to 88 times faster than their respective conventional process models. This highlights the potential for PINN surrogate models as a valuable engineering tool in component and system design and optimization, as well as in real-time simulation for anomaly detection, diagnosis, and forecasting.

Förderung

  • National Research Foundation
  • University of Cape Town

Zitieren

APA

Laugksch, K., Rousseau, P., & Laubscher, R. (2023). A PINN Surrogate Modeling Methodology for Steady-State Integrated Thermofluid Systems Modeling. Mathematical and Computational Applications, 28(2), 52. https://doi.org/10.3390/mca28020052

Vancouver

Laugksch K, Rousseau P, Laubscher R. A PINN Surrogate Modeling Methodology for Steady-State Integrated Thermofluid Systems Modeling. Mathematical and Computational Applications. 2023;28(2):52. doi:10.3390/mca28020052

BibTeX
@article{Laugksch2023,
  author = {Laugksch, Kristina and Rousseau, Pieter and Laubscher, Ryno},
  title = {A PINN Surrogate Modeling Methodology for Steady-State Integrated Thermofluid Systems Modeling},
  journal = {Mathematical and Computational Applications},
  year = {2023},
  volume = {28},
  number = {2},
  pages = {52},
  doi = {10.3390/mca28020052}
}
Kostenlos starten →