EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

PDFP-enhanced physics-informed neural networks for solving higher-order PDEs: Application to engineering beam problems

Shahbaz Ahmad, Neha Zaman, Muhammad Asif, Muhammad Qasim, Aaqib Hussain Shah, Anam Shahzadi, Faizan Arshid, Muhammad Israr

Mechanical Engineering Advances · 27.01.2026 · 4 (1)

DOI: 10.59400/mea4125 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mechanical Engineering Advances, online erschienen am 27.01.2026. Verfasst von Shahbaz Ahmad und 7 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study introduces a modern technique for solving beam equations on elastic foundations, applied to Euler–Bernoulli and Timoshenko beams based on the Winkler foundation. Conventional Physics-Informed Neural Networks (PINNs) have faced major challenges in handling large space-time domains, even when closed-form solutions exist. To address these challenges, we introduced a preconditioned PINNs loss function that incorporates prior knowledge from similar problems, thereby enhancing both productivity and accuracy. This novel method improves the generalization capability of PINNs in structural engineering, specifically for beam dynamics on elastic foundations. The effectiveness of the method is shown through numerical simulations on Euler–Bernoulli beams and extended domains for Timoshenko beams. Comparing with state-of-the-art PINNs shows that our method speeds up the convergence and precisely describes the behavior of the system, surpassing current strategies under the L2-norm metric. Besides, we examine the weight and weight loss plots, and present 3D visualizations of the best weight configurations. Further, we describe the superior performance of the proposed method. The present study explores the application of Preconditioned Devidon–Fletcher–Powell (PDFP)-enhanced Physics-Informed Neural Networks (PINNs) for solving higher-order partial differential equations, with particular emphasis on their implementation in engineering beam problems. It provides an overview of the new approach for addressing complex beam dynamics, including visualizations of the neural network architectures, PINNs convergence behavior, and representative solutions for the beam problems.

Zitieren

APA

Ahmad, S., Zaman, N., Asif, M., Qasim, M., Shah, A. H., Shahzadi, A., Arshid, F., & Israr, M. (2026). PDFP-enhanced physics-informed neural networks for solving higher-order PDEs: Application to engineering beam problems. Mechanical Engineering Advances, 4(1). https://doi.org/10.59400/mea4125

Vancouver

Ahmad S, Zaman N, Asif M, Qasim M, Shah AH, Shahzadi A, et al. PDFP-enhanced physics-informed neural networks for solving higher-order PDEs: Application to engineering beam problems. Mechanical Engineering Advances. 2026;4(1). doi:10.59400/mea4125

BibTeX
@article{Ahmad2026,
  author = {Ahmad, Shahbaz and Zaman, Neha and Asif, Muhammad and Qasim, Muhammad and Shah, Aaqib Hussain and Shahzadi, Anam and Arshid, Faizan and Israr, Muhammad},
  title = {PDFP-enhanced physics-informed neural networks for solving higher-order PDEs: Application to engineering beam problems},
  journal = {Mechanical Engineering Advances},
  year = {2026},
  volume = {4},
  number = {1},
  doi = {10.59400/mea4125}
}
Kostenlos starten →