Artificial intelligence methodologies for the design of armour and protection systems

预测材料和结构对高能弹道冲击的机械和动态响应是设计安全、适用的装甲和防护系统的基本组成部分。英国国防科技实验室(DSTL)拥有世界领先的弹道防护专家，但防护系统的设计和优化往往依赖于昂贵而耗时的实验测试。为了提高效率和降低成本，经常使用内部建模能力来支持试验和装甲主题专家(SME)评估。为了理解和支持复杂多层装甲系统的开发，本研究将开发直接从开源弹道测试数据和有限元分析(FEA)仿真中学习的高级机器学习(ML)体系结构。有限元分析模型将被用来预测由多种材料组成的多层装甲钢板对一系列弹道冲击的弹道响应，这是真实威胁情景的代表。这项研究将开发、实施和测试可用于补充稀疏弹道数据集的最大似然方法。这将允许生成更大的数据集，这在传统方法中是不可行的。此外，这些方法可以反向预测装甲系统的关键设计参数，包括弹道极限速度(BLV)。这项研究方案旨在开发一种稳健的方法，将传统的弹道分析方法、终端弹道测试和高保真有限元数值建模和仿真结合在AI/ML框架中，以快速预测弹道性能，改进现有的基于人工智能的方法和方法。这项工作的总体目标是最终建立在现有设计能力的基础上，提出、实施、验证和测试一套可用于优化和设计基于层的装甲系统的工具，以对抗高能威胁。因此，建议研究的主要目标是：产生和/或来源相关的弹道数据以训练ML模型并自主处理用于机器学习(ML)的大量复杂数据。开发基于ML的程序和方法来辅助装甲和防护系统的设计，包括基于物理的假设、约束和概率不确定性，而不影响其效率。验证所提出的网络，确保它们提供改进和优化的防护系统(例如更高的BLV和更低的面密度)并且在计算上是可行的。提出一种更好的(即更高质量/散装效率)装甲系统的设计方法，确保与有限元分析的有效耦合

Predicting the mechanical and dynamic response of materials and structures to a high energy ballistic impact is a fundamental component of the design of safe and fit-for-purpose armour and protective systems. The UK's Defence Science and Technology Laboratory (Dstl) has world-leading experts in ballistic protection but often the design and optimisation of protection systems is reliant on expensive and time-consuming experimental testing.To increase efficiency and reduce costs in-house modelling capabilities are often used to support trials and armour Subject Matter Expert (SME) assessments. This model development, however, particularly when testing multiple configurations, can be time consuming and costly.To understand and support the development of complex multi-layered armour systems this research will develop advanced Machine Learning (ML) architectures that learn directly from open-source ballistic test data and Finite Element Analyses (FEA) simulations. FEA models will be developed to predict the ballistic response of multi layered armour plates, comprising a diverse range of materials, to a range of ballistic impacts, representative of real-life threat scenarios.This research will develop, implement and test ML approaches that can be used to supplement sparse ballistic data-sets. This will allow substantially larger data-sets to be generated, which is not feasible through conventional methods. Additionally, these methods can be reversed to predict key design parameters for armour systems, including the Ballistic Limit Velocity (BLV). This research proposal aims to develop a robust methodology of combining legacy ballistic analytical approaches, terminal ballistics testing and high-fidelity finite element numerical modelling and simulation in an AI/ML framework to rapidly predict ballistic performance, improving on existing AI-based approaches and methods.The overarching aim of this work is to ultimately build on the current design capability to propose, implement, validate and test a set of tools that can be used to optimise and design layer-based armour systems for personal and vehicle protection against high energy threats. Consequently, the main objectives of the proposed research are to:Generate and/or source relevant ballistic data to train the ML models and autonomously handle high volumes of complex data for use in Machine Learning (ML).Develop ML-based procedures and methods to aid the design of armour and protection systems, incorporating physics-based assumptions, constraints and probabilistic uncertainties, without compromising their efficiency.Validate the proposed networks, ensuring they deliver improved and optimised protection systems (e.g. higher BLV and lower areal density) and are computationally feasible.Propose a design approach for better (i.e. more mass/bulk efficient) armour systems, ensuring efficient coupling with FEA