Open Access

ARTICLE

Interpretable Machine-Learning-Assisted Stochastic Buckling Assessment of Cylindrical Shells with Random Geometric Imperfections

Yan-Ping Liang¹, Zhiqiang Wan^2,*

1 Department of Civil Engineering, Hangzhou City University, Hangzhou, China
2 School of Mechanics and Transportation Engineering, Northwestern Polytechnical University, Xi’an, China

* Corresponding Author: Zhiqiang Wan. Email:

(This article belongs to the Special Issue: AI-Enhanced Computational Mechanics and Structural Optimization Methods)

Computer Modeling in Engineering & Sciences 2026, 148(3), 12 https://doi.org/10.32604/cmes.2026.088012

Received 26 June 2026; Accepted 04 September 2026; Issue published 28 September 2026

Abstract

Initial geometric imperfections critically affect cylindrical-shell buckling, yet the influence of imperfection morphology remains difficult to separate from that of amplitude. This study develops an interpretable machine-learning-assisted framework for stochastic buckling assessment of cylindrical shells with random geometric imperfections. Circumferentially continuous Gaussian imperfection fields with different normalized correlation lengths are generated under a fixed root-mean-square (RMS) amplitude. Nonlinear Riks analyses are performed to construct a finite-element database of the corresponding buckling responses. Feature diagnosis is used to identify response-relevant descriptors of imperfection morphology. Compact surrogate models are subsequently evaluated for sample-level prediction and group-level statistical trend representation. The buckling resistance first decreases and then increases with increasing correlation length, demonstrating a pronounced non-monotonic morphology effect. Among the considered descriptors, the RMS edge jump is identified as the most informative field-derived measure and provides robust predictive information across multiple regression models. The proposed framework provides a compact link between imperfection morphology and stochastic buckling response.

---

Keywords

---