Qianxu Chen
2815866C@student.gla.systa-s.com
Research title: Physics-constrained data-driven modelling of anisotropic sand behaviour
Research summary
Thesis title: Physics-constrained data-driven modelling of anisotropic sand behaviour
Thesis
To develop a physics-constrained data-driven method for modelling the anisotropic behaviour of sand. We will use the test data by Yoshimine (1998). All the data is in Excel form. What is unique in our work is that real data will be used for the modelling. Most existing research has used data from idealized soils via DEM simulations. The data of Yoshimine includes over 120 tests, which should be enough for our purpose. Before publications, keep the data confidential. This is our secret weapon. It requires some basic knowledge of constitutive modelling of soils. This is just a new method for constitutive modelling. It will be initially challenging. But once you get sufficient knowledge of artificial neural network method, you will be able to do many things fast. The main research activities include:
- Develop an artificial neural network method for modelling – work with Eky. The main issue is to form the loss functions for machine learning. This will be essential for the modelling results. The loss function may need to consider: the critical state in the p-q and e-p plots, flow rule in the 3D stress space. Should we use the components or invariants for machine learning and training?
- Modelling of the state-dependent sand behaviour first - triaxial compression first. This will be simple and help future work on anisotropic sand behaviour.
- Modelling of the anisotropic sand behaviour with data by Yoshimine (1998).
- Application of the model in Abaqus or Plaxis simulations – see existing publications for reference.
