Answers are grounded in this paper and cite their sources.
The design of composite structures for impact and crash scenarios critically relies on understanding strain-rate effect, where mechanical properties significantly vary with the local loading rates of the material. Characterising the dynamic material responses of these materials for crashworthy design is traditionally resource-intensive. While standard data-driven machine learning (ML) models offer a convenient way to learn the underlying complex characteristics from experiments, they often lack physical consistency. Additionally, they are resource-intensive, requiring large volumes of experimental data. Circumventing these two issues, we propose a novel approach utilising Constitutive Artificial Neural Networks (CANNs) to characterise the strain-rate dependent behaviour of carbon/epoxy composites. Unlike existing black-box algorithms, this approach directly embeds constitutive equations into the neural network architecture. The model is tested for IM7/8552 under uniaxial compression at strain rates up to 200 s −1 . The model predictions demonstrate good agreement with experimental data (nRMSE = 7.3%), effectively capturing complex rate-dependent characteristics. Apart from demonstrating accurate interpolation, the model also shows robust extrapolation to out-of-distribution (OOD) strain rates. By combining the flexibility of ML with the reliability of physical laws, this approach significantly reduces the experimental burden, offering a powerful method to accelerate the design of safe and high-performance composite structures.
No takes yet. Share an insight, caveat, or question.
Yoo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: