Machine learning for constitutive modeling

Sammanfattning: Constitutive modeling of nonlinear materials is a computationally complex and time-intensive process. To mitigate this, a neural network was used to simulate complex behaviors of non-linear materials. With supervised learning the machine learning model was able to predict the stresses in a material when given the strains. The machine learning algorithm predicted stresses in a linear elastic material with high accuracy, and in a hyperelastic material with lower accuracy. To simulate experimental conditions, artificial Gaussian noise was added to the strain data, and the model was tested with the new input.