EUVIMEDEUVIMED entdecken
Crossref

INVERSE PROBLEM CHALLENGES OF MULTI-PULSE WAVE PROPAGATION IN HETEROGENEOUS MEDIA USING PHYSICS-INFORMED NEURAL NETWORKS

I. I. Kolych, R. Y. Shuvar

Ukrainian Journal of Information Technology · 2026 · 8 (1) · 71-78

DOI: 10.23939/ujit2025601.071 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Ukrainian Journal of Information Technology, 2026. Verfasst von I. I. Kolych und 1 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Developed an approach for studying inverse problems in multi-pulse wave propagation through non-uniform media, where overlapping signals make the recovery of coupled velocity profiles fundamentally ambiguous. Investigated an extension of Physics‑Informed Convolutional Neural Networks (PICNNs) to dual‑pulse scenarios to explain why ambiguity arises when two components propagate simultaneously in a heterogeneous medium. The dual‑pulse dynamics are modeled with the one‑dimensional wave equation, and the observed field is treated as the superposition of two latent wave components. PICNNs are trained to solve forward tasks (separating the two wavefields and predicting their space-time evolution) and inverse tasks (reconstructing the corresponding velocity profiles). Training data is generated with the finite element method, and absorbing boundary conditions are used to reduce spurious reflections so that the learning signal is dominated by true propagation effects. The proposed PICNN formulation performs well on forward problems: it can disentangle overlapped pulses and accurately reproduce the evolution of each component across the space-time grid. Targeted computational experiments show that separability depends strongly on velocity contrast and on the relative strength of physics‑residual constraints during training. However, the fully unconstrained inverse problem, where both velocities are treated as unknown, often fails to converge to a meaningful decomposition. Observed that optimization can become trapped in symmetric local optima, producing nearly identical outputs for the two components and preventing reliable recovery of two distinct velocity profiles. Proposed fixing one velocity profile to break symmetry between the two pulses and stabilize inversion when both velocities cannot be uniquely identified from the overlapped field alone. This study highlights the computational challenges of resolving overlapping wave signals in multi‑pulse scenarios. Concluded that multi‑component inverse problems require additional constraints or hybrid strategies that combine physics‑informed learning with source‑separation techniques to resolve ambiguity and enhance robustness. Future work will extend the approach to higher‑dimensional and multi‑component scenarios.

Zitieren

APA

Kolych, I. I., & Shuvar, R. Y. (2026). INVERSE PROBLEM CHALLENGES OF MULTI-PULSE WAVE PROPAGATION IN HETEROGENEOUS MEDIA USING PHYSICS-INFORMED NEURAL NETWORKS. Ukrainian Journal of Information Technology, 8(1), 71-78. https://doi.org/10.23939/ujit2025601.071

Vancouver

Kolych II, Shuvar RY. INVERSE PROBLEM CHALLENGES OF MULTI-PULSE WAVE PROPAGATION IN HETEROGENEOUS MEDIA USING PHYSICS-INFORMED NEURAL NETWORKS. Ukrainian Journal of Information Technology. 2026;8(1):71-78. doi:10.23939/ujit2025601.071

BibTeX
@article{Kolych2026,
  author = {Kolych, I. I. and Shuvar, R. Y.},
  title = {INVERSE PROBLEM CHALLENGES OF MULTI-PULSE WAVE PROPAGATION IN HETEROGENEOUS MEDIA USING PHYSICS-INFORMED NEURAL NETWORKS},
  journal = {Ukrainian Journal of Information Technology},
  year = {2026},
  volume = {8},
  number = {1},
  pages = {71--78},
  doi = {10.23939/ujit2025601.071}
}
Kostenlos starten →