EUVIMEDEUVIMED entdecken
Crossref

A two-stage lstm-based prediction methodology for lamb wave resonators

Boya Liu, Xianzheng Lu, Hao Ren

Journal of Micromechanics and Microengineering · 01.06.2026 · 36 (6) · 065006

DOI: 10.1088/1361-6439/ae772f ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Micromechanics and Microengineering, online erschienen am 30.06.2026. Verfasst von Boya Liu und 2 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben. Gefördert von: ShanghaiTech Start-up, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning-assisted design methodologies have been a promising trend across multiple engineering fields. However, challenges emerge when applying deep learning to physical systems that demand both broad coverage and high-resolution accuracy. Existing deep learning models cannot balance these two objectives, as high-precision predictions over wide ranges demand substantial computational resources and exhibit poor performance when predicting long sequences. For piezoelectric resonators, prior studies are limited by narrow prediction frequency ranges, failure to account for realistic spurious modes, and neglect of effective electromechanical coupling coefficient ( k eff 2 ) effects. In this paper, we report a novel two-stage prediction methodology that addresses these limitations. The first stage employs coarse-resolution global analysis to identify resonance characteristics and spurious modes, while the second stage applies fine-resolution predictions within selected frequency windows for accurate k eff 2 calculation. Our proposed long short-term memory encoder–decoder model achieves R 2 = 0.995 and 0.993 for coarse and fine predictions, respectively, with frequency prediction accuracy exceeding 92%. The complete two-stage workflow achieves a runtime 7947 times shorter than finite element analysis (FEA) simulations while maintaining a mean relative k eff 2 prediction error below 2.79%. Experimental validation on a 9-interdigitated transducer 20 at. % scandium-doped aluminium nitride (Al 0.8 Sc 0.2 N) Lamb wave resonator demonstrates that the model predictions are in good agreement with both FEA simulations and experimental results, with the model-predicted k eff 2 (1.958%) close to FEA (1.894%) and experimental results (1.925%). This methodology offers a solution for balancing broad coverage and accurate predictions in other physical device designs.

Förderung

  • ShanghaiTech Start-up
  • National Natural Science Foundation of China

Zitieren

APA

Liu, B., Lu, X., & Ren, H. (2026). A two-stage lstm-based prediction methodology for lamb wave resonators. Journal of Micromechanics and Microengineering, 36(6), 065006. https://doi.org/10.1088/1361-6439/ae772f

Vancouver

Liu B, Lu X, Ren H. A two-stage lstm-based prediction methodology for lamb wave resonators. Journal of Micromechanics and Microengineering. 2026;36(6):065006. doi:10.1088/1361-6439/ae772f

BibTeX
@article{Liu2026,
  author = {Liu, Boya and Lu, Xianzheng and Ren, Hao},
  title = {A two-stage lstm-based prediction methodology for lamb wave resonators},
  journal = {Journal of Micromechanics and Microengineering},
  year = {2026},
  volume = {36},
  number = {6},
  pages = {065006},
  doi = {10.1088/1361-6439/ae772f}
}
Kostenlos starten →