Investigation of crack propagation within solid structures exposed to dynamic loading and impact using the truss-like discrete element method

Lead Research Organisation: University of Edinburgh
Department Name: Sch of Engineering

Abstract

Understanding how fractures and cracks develop in materials is crucial to many disciplines, e.g. aerospace engineering, materials sciences, structural engineering and geophysics. Fast and accurate computer simulation of crack propagation within realistic 3D structures would be a valuable tool to allow engineers to explore the fracture process in greater detail and yield a greater understanding of the factors that lead to the demise of a structure.

This research aims to create a new program that utilizes the truss-like Discrete Element Model (DEM) method to investigate this problem further. The program will discretize structures imported from a CAD model into an array of 2D elements that allow the user to create desired loading scenarios. It is a cutting-edge approach that aims to reduce the computational resource required by the conventional Finite Element Analysis (FEA) method. The performance of the program will be compared with an identical FEA model and aim to be validated against an equivalent experiment.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/N509644/1 01/10/2016 30/09/2021
1941431 Studentship EP/N509644/1 01/09/2017 31/07/2021 Samuel Mark Thompson
 
Description So far we have shown that machine learning algorithms can learn directly from data collected from ballistic experiments to make new predictions. In this case, a ballistic experiment refers to a blunt, cylindrical projectile perforating multi-layer armour plates with the velocity of impact and the residual velocity of the projectile being the key recorded parameters. A mutli-layer network, which is a common type of neural network used in machine learning, was trained and successfully used to make new predictions on experimental cases that the network was not trained on that were consistent with the results from real experiments given the same conditions.
Exploitation Route Current work involves using a more sophisticated type of neural network known as a Generative Adversarial Network that has the ability to generate new forms of data irrespective of data type. We aim to incorporate such algorithms and develop a bespoke network architecture that allows us to supplement data sets by using the network to generate new results that are consistent with that of real experiments. As ballistic experiments are destructive by nature, using such methods could prove to be extremely valuable by allowing researchers to obtain preliminary results without the large expense.

Furthermore, the long term application of these methods could support a future where Generative Networks are capable of suggesting complete solutions to engineering problems such as new composite materials for armour systems.
Sectors Aerospace, Defence and Marine,Digital/Communication/Information Technologies (including Software),Pharmaceuticals and Medical Biotechnology