A Multimodal Machine Learning Approach for Identification of Mechanical Properties in  CFRP Composites

ABSTRACT Accurate and efficient identification of the mechanical properties of carbon fiber reinforced polymer (CFRP) composites is crucial for their design and application. This paper proposes a machine learning (ML) approach based on a multimodal neural network to identify orthotropic mechanical properties including elastic modulus, Poisson's ratio, shear modulus, as well as tensile and in‐plane shear strengths. The architecture concurrently processes distinct data modalities: a multi‐layer perceptron analyzes load–displacement curves, while a convolutional neural network extracts spatial features from strain and displacement fields. The validity and reliability of the identified parameters are confirmed through a three‐point bending simulation, and an ablation study is performed to demonstrate the superiority of the proposed architecture. The results indicate that the proposed method achieves a mean relative error of 5.7% compared to experimental results, which outperforms a finite element (FE)‐based inverse method. In terms of efficiency, the model's inference time of only 0.21 s is approximately 5600 times faster than a single FE analysis. The three‐point bending simulation shows high consistency with experimental results. Finally, an ablation study confirms the superiority of the multimodal neural network. The full model achieves a coefficient of determination of 0.92, outperforming the two single‐modality models by 21% and 16%, respectively.