A Multidimensional Physics Informed Neural Network Assisted Damage Tolerant Design of Critical Nuclear Materials

University of Derby • GB • 3d ago

The proposed project stems from European and UKRI-wide funded projects in which digital technologies for improving the reliability of additive manufactured components are sought.

Lead Supervisor: Professor Angelo R. Maligno University of Derby
Second Supervisor: Dr Suleiman Aliyu
Industry Partner: Zentech International (Confirmed)
Project Start: Oct 2026
Advert Close Date: ASAP
Target Background: Electrical or Mechanical Engineering, Physics, Computer Science
Industrial Funding: In-kind licences, support, training
Programme: 4 year Engineering Doctorate (EngD) with industry placement

Project Summary

The proposed project stems from European and UKRI-wide funded projects in which digital technologies for improving the reliability of additive manufactured components are sought.The particular task of this study is to start the implementation of a Physics Informed Neural Networks (PINN) to support the overall reliable design of critical nuclear components with particular emphasys in additvely manuafactured parts. Damage-tolerant design focuses on creating materials and structures that can withstand damage without catastrophic failure. This approach is crucial in engineering, particularly in fields like nuclear reactors, aerospace, civil engineering, and materials science.

Long Term Vision

Physics-informed neural networks (PINNs) are a type of machine learning model that integrates physical laws into their framework. This integration allows PINNs to make accurate predictions about material behaviour, particularly under stress conditions. The overall aims of these intial study is to introduce a predictive digital engineering framework integrating three complementary scientific pillars:

(i) Integrated Computational Materials Engineering (ICME) for multiscale modelling of materials and manufacturing processes;
(ii) Damage Tolerance Design;
(iii) PINN-assisted damage tolerant design for complex loading conditions, multiple cracks and life estimation.

This integrated approach will enable earlier reliability assessment, reduced reliance on large-scale physical testing and improved predictive capability for certification-relevant structural behaviour.

Aims and objectives

The ambition of the project is to demonstrate that predictive digital engineering can transform the development of nuclear parts from a predominantly empirical, test-driven process to a simulation-supported certification paradigm. The application of physics-informed neural networks in crack propagation represents a significant advancement in the field of fracture mechanics. By combining machine learning with physical laws, PINNs provide a robust framework for predicting material behaviour and ensuring structural integrity. PINN-integrated Damage Tolerant Design have the following benefits:

Physics-informed neural networks enhance damage-tolerant design by providing accurate modelling of damage and fracture, reducing computational costs, and eliminating mesh dependency. This allows for more efficient simulations and better predictions of material behaviour under stress.
Physics-informed neural networks integrate physical laws into their learning process, making them more effective with limited or noisy data compared to traditional modelling methods.
Physics-informed neural networks can be used to model crack propagation by integrating physical laws into the neural network framework, allowing for accurate predictions of material behaviour under stress. This approach enhances the modelling of multiple cracks in materials, improving the understanding of fracture mechanics

Challenges

Limitations and challenges of applying physics-informed neural networks to crack propagation include the complexity of accurately modelling multiple cracks and the need for specialised enrichment functions to capture discontinuities and singularities. Additionally, the integration of energy-based loss functions and customised schemes can complicate the implementation process. The adoption of specialised software for three-dimensional (3D) crack propagation will overcome this bottlenecks. Software such as Zencrack provides state-of-the-art capabilities for modelling and analysing 3D cracks, predicting their behaviour and growth under specific conditions. Zencrack is an advanced engineering analysis tool for 3D fracture mechanics assessment and crack growth simulation. The program uses finite element analysis to allow calculation of fracture mechanics parameters such as energy release rate and stress intensity factors. This is achieved by automatic generation of focused cracked meshes from uncracked finite element models. A mixed-mode capability allows non-planar crack growth prediction for fatigue and time-dependent load conditions via automated adaptive meshing techniques. The integration of 3D crack propagation tools with PINN represents a significant advancement in the field of fracture mechanics.

By integrating structural modelling, reliability analysis, it will be possible to enable earlier and more accurate prediction of structural performance across the entire nuclear system lifecycle. The proposed project represents a stepping stone to support the following challenges in the design of nuclear components:

reduction in the number of required physical tests in the certification test pyramid;
improved accuracy of fatigue life prediction compared with conventional design approaches;
validated coupling between working conditions simulation, structural modelling and digital twin prediction;
quantified reduction in structural mass relative to the baseline design while satisfying the same load and reliability requirements;
quantified reduction in manufacturing discard or scrap risk proxies through explicit modellingof process variability and defect populations;
quantified improvement in predicted service life, remaining useful life, or reduction in premature replacement risk.

Scientific and Technological Objectives

The main aim of this project is to capitalise the experience on materials design and damage tolerat design gained through funded projects and develop an innovative digital engineering design paradigm by coupling specialised 3D software for damage tolerant design (crack propagation) and physics informed neural networks.Coupling 3D crack propagation with Physics-Informed Neural Networks (PINN) involves using advanced neural network techniques to model and predict the behavior of cracks in materials under various conditions, enhancing the accuracy of simulations in fracture mechanics. This approach leverages the strengths of PINNs to address complex problems related to crack dynamics and material strength.

Alignment to STAND-UP Target Impact Outcomes

50% in overall build or decommissioning process time, through digital tools, automation, new processes, monitoring and control strategies
40% in maintenance time, through new materials, robotics, smart sensors, inspection during service and digital data analytics
30% in person hours on builds, by reducing rework & right-first-time operation through digital tools and sustainable smart manufacturing