Automating constitutive modeling with LLMs
Please login to view abstract download link
State-of-the-art approaches in constitutive modeling, such as constitutive artificial neural networks (CANNs), achieve high accuracy, strong generalization, and guaranteed adherence to physical constraints [1]. However, when modeling new materials, these methods often require deep expertise: Choosing the right approach, configuring the model, and performing effective hyperparameter optimization can be tedious and time-consuming for experts, and a significant barrier for non-experts. This talk addresses this challenge by exploring the integration of large language models (LLMs) with CANNs. We present results on real-world datasets that capture strain–stress relationships for multiple hyperelastic materials under diverse loading scenarios [2]. These results demonstrate that CANNs generated and optimized autonomously by LLMs can match human-designed models in accuracy, generalizability, and physical consistency. Our approach not only accelerates constitutive modeling but also makes it accessible to a wider range of users. References: [1] Linka, K., Hillgärtner, M., Abdolazizi, K. P., Aydin, R. C., Itskov, M., Cyron, C. J. (2021). Constitutive artificial neural networks: A fast and general approach to predictive data-driven constitutive modeling by deep learning. Journal of Computational Physics, 429, 110010. [2] Tacke, M., Busch, M., Abdolazizi, K. P., Eichinger, J. F., Linka, K., Cyron, C. J., Aydin, R. C. (2025). Automating modeling in mechanics: LLMs as designers of physics-constrained neural networks for constitutive modeling of materials. Under review. doi: https://doi.org/10.48550/arXiv.2512.01735
