Predicting the dynamic response of materials and structures to high ­energy ballistic impacts is a fundamental component of the development and design of fit-­for-­purpose armour and protection systems. Understanding the ballistic response of protection systems conventionally involves impacting these with a range of projectiles under various impact conditions in costly experimental campaigns. This is often combined with high-fidelity Finite Element Analyses (FEA), which contribute to a complete dynamic characterisation of the system and allows more system configurations to be assessed. The timescales for approving armour systems can be long but, in times of conflict, rapid assessment and development of protection systems is required as new threats are encountered, using materials available on operations. To support development of complex multi-layered systems and to inform rapid decision-making, this project will develop advanced Machine Learning (ML) architectures that learn directly from a combination of ballistic test data and FEA simulation results.

The main aims of the project are to: investigate the use of Multi-Layer Perceptron (MLP) models to predict the ballistic response of multi-layered armour systems, comprising a diverse range of materials, to a range of ballistic impacts; and to understand the use of Generative Adversarial Networks (GAN) to supplement sparse ballistic datasets and attempt to reverse this method to predict key design parameters for armour systems. An assessment of whether a set of conditional GAN can be used to condition the current network on additional auxiliary information or design requirements that refer to a specific property relevant to the ballistic data will also be conducted. This adaptable method will allow networks to generate not only samples specific to the class label given, but also to generate data for classes that are not present in the training set and conduct their own material characteristic campaign.