Learning Biases in Plasticity Modelling
Please login to view abstract download link
Plasticity, characterise by high non-linearities and energy dissipation through permanent deformations, presents significant challenges for accurate and efficient modelling. Traditional methods often fail to capture these complex behaviours without incurring high computational costs. In this work, we address these challenges by integrating deep learning techniques with physical laws as inductive biases [1]. This hybrid approach leverages the strengths of deep learning to improve both accuracy and computational efficiency, learning material behaviour from data while enforcing physical consistency. By embedding fundamental principles of plasticity into the neural network architecture, our method not only captures non-linear material responses but also significantly reduces time cost, yielding faster and more scalable results.
