AI-assisted prediction of local damage dynamics for the design of resilient and durable network-forming materials

Figure caption: a) MD simulations of single (SN) and double polymer networks (DN) at large deformation showing different patterns of damage localization (single fracture region in SN vs nanovoids in DN). b) Propensity of bond breaking computed from the cumulated bond breaking events in a SN (represented in the initial undeformed configuration). c)The 3d simulation data is transformed into a graph input (d) which is processed by a GNN (e) which then predicts the propensity of being involved in a bond breaking event for each particle (f).

This project aims to develop an AI-driven framework to comprehensively understand, predict, and optimise the mechanical properties of network-forming systems, with a particular focus on their damage mechanisms.

Network-forming materials, such as elastomers, hydrogels, and aerogels, are widely utilised in engineering and biomedical applications due to their tunable and complex mechanical behaviors. Despite their versatility, a fundamental understanding of how their damage and failure response is influenced by synthesis parameters and the resulting mesoscopic network structure remains elusive. This project seeks to address this critical knowledge gap by combining state-of-the-art machine learning techniques, including Graph Neural Networks (GNNs), Physics-Informed Neural Networks (PINNs) and explainable AI (XAI), with high-end numerical simulations of coarse grained models for polymer networks and cutting edge experimental techniques.

By leveraging the synergy between AI-driven modeling and traditional methods, the project will uncover new insights into the relationship between material synthesis, the obtained microstructure and the micromechanical response to large deformations, ultimately enabling the design of new generations of advanced and durable materials tailored for specific applications.

Key objectives include:

1. Establishing predictive models to determine how synthesis parameters influence network failure through damage propagation.

2. Optimising network architectures to achieve durable materials able to sustain large amounts of damage accumulation to delay macroscopic failure.

3. Providing actionable insights that are scalable for various biomedical and engineering applications, including elastomer engineering, hydrogel optimisation for bio-medical applications like the coating of brain implants and shape morphing applications.

The expected breakthrough is a robust, predictive framework that eliminates reliance on trial-and-error approaches, advancing the design of network forming materials with superior mechanical performance. This innovation will set the stage for widespread adoption of AI in materials science, with implications far beyond the initial scope of polymer networks.

FUNDING

Tec 21