A Machine Learning-Embedded theoretical model for precise adhesive layer stress prediction in composite bonded joints

Liu, Yalong and Gu, Zewen and Ding, Xiaoxuan and Guo, Wen and Liu, Jianlin and Hou, Xiaonan (2026) A Machine Learning-Embedded theoretical model for precise adhesive layer stress prediction in composite bonded joints. Composite Structures: 119988. ISSN 0263-8223

Text (A Machine Learning-Embedded theoretical model for precise adhesive layer stress prediction in composite bonded joints-Reviesd)
A_Machine_Learning-Embedded_theoretical_model_for_precise_adhesive_layer_stress_prediction_in_composite_bonded_joints-Reviesd.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (1MB)

Abstract

Adhesive-bonded joints are widely used in engineering applications ranging from aerospace, automobile to advanced superconducting materials due to their ability to effectively join dissimilar materials. While extensive research has investigated the factors influencing their failure modes, particularly stresses within the adhesive layer, accurate stress prediction remains challenging for joints with arbitrary geometries. This difficulty arises primarily from the complex effects of eccentric loading across varying material combinations and joint configurations. This study proposed an integrated approach combining experimental testing, numerical simulation, and machine learning to predict normal and shear stresses in single-lap joints. First, tensile tests are performed on multi-material joints (Al-Al, CFRP-CFRP, and Al-CFRP) to validate the developed finite element models. Then, theoretical models are derived for asymmetric joint configurations. For including the effects of eccentric loading, a dataset of 300 simulation results is generated to train a deep neural network (DNN) model for predicting the bending moment factors (K factors) across diverse joint geometries and material pairings. The DNN-derived K factors demonstrate exceptional accuracy when integrated into theoretical stress predictions, significantly outperforming conventional methods while maintaining robust adaptability. This work addresses a key joint mechanics challenge and offers a versatile framework for optimizing adhesive joint design in engineering.