A machine-learning-based model for accurately predicting the bending behavior of sandwich beams

TY - JOUR

T1 - A machine-learning-based model for accurately predicting the bending behavior of sandwich beams

AU - Yan, Yu

AU - Duan, Yu

AU - Peng, Zhilong

AU - Zhang, Bo

AU - Yao, Yin

AU - Chen, Shaohua

N1 - Publisher Copyright: © 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

PY - 2026/7

Y1 - 2026/7

N2 - Sandwich beams always exhibit a shear-bending-coupling mechanical behavior, which is difficult to be accurately predicted by existing bending theories due to the prior assumptions, such as an overlook of the interlayer shear stress and a zigzag displacement distribution on the beam's cross-section. In this work, based on the large dataset provided by finite element simulation, an alternative model of sandwich beams is generated using machine learning (ML) methods without any assumptions, and applied to analyze the influence of modulus ratio and thickness ratio between the surface and core layers on the stress and deformation fields of sandwich beams under four point bending. It is found that an increase in modulus ratio not only causes significant shear stress in the core layer, but also results in a mixed tensile and compressive stress state in the surface layer, causing the neutral plane to shift from the core layer to the surface layers. In contrast, the thickness ratio only affects the magnitude of stress and deformation in the beam, and has little effect on the position of the neutral plane. All the predicted results are consistent with existing experimental and numerical ones, indicating that the modulus difference between the surface and core layers dominates the shear-bending-coupling behavior of sandwich beams and determines the applicability of existing bending theories. The present ML-based model should be of guiding value for the design of sandwich structures with desirable mechanical properties.

AB - Sandwich beams always exhibit a shear-bending-coupling mechanical behavior, which is difficult to be accurately predicted by existing bending theories due to the prior assumptions, such as an overlook of the interlayer shear stress and a zigzag displacement distribution on the beam's cross-section. In this work, based on the large dataset provided by finite element simulation, an alternative model of sandwich beams is generated using machine learning (ML) methods without any assumptions, and applied to analyze the influence of modulus ratio and thickness ratio between the surface and core layers on the stress and deformation fields of sandwich beams under four point bending. It is found that an increase in modulus ratio not only causes significant shear stress in the core layer, but also results in a mixed tensile and compressive stress state in the surface layer, causing the neutral plane to shift from the core layer to the surface layers. In contrast, the thickness ratio only affects the magnitude of stress and deformation in the beam, and has little effect on the position of the neutral plane. All the predicted results are consistent with existing experimental and numerical ones, indicating that the modulus difference between the surface and core layers dominates the shear-bending-coupling behavior of sandwich beams and determines the applicability of existing bending theories. The present ML-based model should be of guiding value for the design of sandwich structures with desirable mechanical properties.

KW - Bending behavior

KW - Machine learning (ML) method

KW - Modulus and thickness ratios

KW - Sandwich beam

KW - Stress and deformation fields

UR - https://www.scopus.com/pages/publications/105036253738

U2 - 10.1016/j.mechmat.2026.105701

DO - 10.1016/j.mechmat.2026.105701

M3 - Article

AN - SCOPUS:105036253738

SN - 0167-6636

VL - 218

JO - Mechanics of Materials

JF - Mechanics of Materials

M1 - 105701

ER -