MDPI Webinar | AI-Powered Materials Science and Engineering, 1 December 2025

Program

Time in CET

MDPI Host
Opening

8:00–8:05 a.m.

Prof. Dr. Jian Feng Wang
Constitutive Modelling of Granular Soils Using an Integrated Approach of X-ray Microtomography, DEM Modelling and Deep Learning

In this talk, I will present our recent progress on the micro/macro-mechanical investigation of granular soils subject to triaxial shearing using an integrated approach of X-ray micro-computed tomography (CT), three-dimensional discrete element modelling, and deep learning. Particular focus will be placed on the recent development of data-driven constitutive models of granular soils. Our results show that the effects of particle morphology, confining pressure, and initial sample density on the constitutive responses of real granular soils can be well captured by the typical recurrent neural network models, such as long short-term memory neural networks (LSTM) and gate recurrent unit neural networks (GRU). The developed deep learning models can effectively learn and reflect the intrinsic physical mechanisms underlying granular material behavior, such as stress–strain, volumetric compression and dilatancy, strain hardening and softening, and shear-induced fabric evolutions. Our latest results using a deep transfer learning technique called the few-shot learning strategy will also be presented. This talk will allow the attendees to gain an overview of the latest, cutting-edge developments of deep learning methods in the CT-based constitutive modelling of granular soils.

8:05–8:40 a.m.

Prof. Dr. Stefano Mariani
Materials Informatics and a Generative Approach at the Microscale

Materials informatics is gaining popularity for predicting the overall mechanical properties of multi-phase and polycrystalline composites. Data-driven strategies can be exploited within this framework to learn microstructural features and their relationship with the resulting macroscopic properties. However, adopting such approaches to assess the load-bearing capacity and reliability of structures and devices, accounting for stochastic effects at the microscale, still requires careful consideration, especially when only limited data or computational resources are available. In this talk, a strategy is proposed to address problems characterized by strong gradients in the stress and strain fields, which hinder the use of standard homogenization techniques. A generative adversarial network (GAN) is employed to generate reliable proxies of actual microstructures and predict the overall behavior of the studied multi-phase materials.

8:40–9:15 a.m.

Q&A Session

9:15–9:40 a.m.

MDPI Host
Closing of Webinar

9:40–9:45 p.m.