AI-DRIVEN PREDICTION OF BUCKLING LOADS IN IMPERFECT CYLINDRICAL SHELLS BASED ON NON-DESTRUCTIVE PROBING
Abstract
Cylindrical shell structures are widely used in engineering due to their high strength-to-weight ratio, with applications in automotive, aerospace, and marine. However, these structures are highly susceptible to buckling under axial compression, with experimental studies showing significantly lower load-bearing capabilities than analytical predictions due to their sensitivity to geometric imperfections. This study proposes an AI-driven model to predict the critical buckling loads of cylindrical shells with unknown imperfections, using non-destructive probing test responses and a model trained on simulation data. Finite Element Analysis (FEA) is performed on a model without imperfections to obtain its buckling modes. These modes’ deformations are then scaled and superimposed onto the perfect shape to introduce arbitrary imperfections in samples. Through conducting buckling simulations on a soda-can-size cylindrical shell, a data set comprising over 2,000 imperfect buckling samples was generated, and their load-bearing capacity and effective stiffness was obtained. Furthermore, a lateral probing experiment was simulated on the midpoint of the shell face to obtain the stability landscape in different imperfection conditions. The data was preprocessed, and meaningful features were extracted from it. An innovative network comprising of 2D convolutional neural networks (CNNs) and Extreme Gradient Boosting (XGBoost) models were employed to predict the buckling load and imperfection state of specimens.
Publication Title
ASME International Mechanical Engineering Congress and Exposition Proceedings Imece
Recommended Citation
Rahmati, A., & Guan, Y. (2025). AI-DRIVEN PREDICTION OF BUCKLING LOADS IN IMPERFECT CYLINDRICAL SHELLS BASED ON NON-DESTRUCTIVE PROBING. ASME International Mechanical Engineering Congress and Exposition Proceedings Imece, 8 https://doi.org/10.1115/IMECE2025-166262
