STO-Activities: (no title)
Activity title:
Enhanced Design Processes of Military Vehicles through Machine Learning Methods
Activity Reference:
AVT-404
Panel:
AVT
Security Classification:
NATO UNCLASSIFIED
Status:
Planning
Activity type:
RTG
Start date:
2025-01-01T00:00:00Z
Actual End date:
2027-12-31T00:00:00Z
Keywords:
Artificial Intelligence, Machine Learning, Multidisciplinary Design, Simulation driven design
Background:
Military vehicle requirements continue to evolve rapidly in all AVT domains (air, space, sea, ground). To keep pace with the needs of the Alliance, the time and effort to develop next-generation military vehicles and weapon systems need to be significantly reduced, while simultaneously diminishing problems occurring late in the system development. Meeting these needs requires new and advanced capabilities for quick, accurate and thorough assessment of the design and operational spaces and reliable optimisation processes of platforms’ design. AVT recently completed several assessments of this need.
AVT-ET-054 explored the issue of affordable weapons systems and led to the formation of AVT-092, “Qualification by Analysis”, and AVT-093, “Integrated Tools and Processes for Affordable Weapons Systems”. AVT-093 focused on “the integration of tools and processes, not on the description of tools and processes”. AVT-093 also identified needs in multidisciplinary design optimisation (MDO) that could be addressed using the integration of tools and processes in a distributed parallel computing environment, that would enable feedback of information from detail to preliminary and preliminary to conceptual design. AVT-092 recognized that these capabilities described in AVT-093 are necessary to achieve the objective of rapid design and qualification of new vehicles. Both teams observed a gap between the current technology and the desired end state of rapidly developing affordable weapons systems, recognizing that developments in multidisciplinary technologies are key capabilities for closing that gap. Thus, AVT-237 focused on benchmarking the use and benefits of MDO for the development of military systems, and AVT-252 explored optimisation of aircraft and ships under uncertainty. More recently, the AVT-331 on “Goal-driven, multi-fidelity approaches for military vehicle system-level design” discussed, developed, and applied methods for accelerating vehicle design by using tools and processes that reflect different levels of fidelity throughout the MDO process.
Recognizing the interdisciplinarity inherent in MDO, AVT-331 recommended broadening the discussion in our community via cooperative events. To this aim, AVT-331 co-chairs and members organized a special session on Multi-Fidelity Methods for Vehicle Applications at the Multidisciplinary Analysis and Optimization Conference within the AIAA Aviation Forum and Exposition in 2020 and, along the same line, the AVT-354 RWS on “Multi-Fidelity Methods for Military Vehicle Design,” held in 2022 in Varna and included in the STO 2022 Highlights.
Finally, AVT-MSG-ET-232 on “Machine Learning and Artificial Intelligence for Military Vehicle Design”, a follow-on activity to AVT-331, discussed machine learning and artificial intelligence (ML/AI) as enablers to even faster and more accurate design processes. The complexity of military vehicle design problems, from the analysis of complex physical phenomena to optimal design (AVT-252) and control (AVT-351) of the vehicles, relies upon scientific computing frameworks capable of providing cost-effective and reliable solutions, beginning in the early (conceptual/preliminary) phases of the design process. In the recent years, ML/AI methods, such as supervised learning in the form of multi-fidelity methods (AVT-331, AVT-354) and data fusion approaches (AVT-351), have shown their potential in providing effective solutions to these problems. However, ML/AI techniques typically require significant amounts of training data. Additionally, the responses are often affected by lack of interpretability, and their reliability is difficult to characterize. AVT-MSG-ET-232 has discussed where and how ML can improve the design process and recommends this activity mature in the form of an RTG. In a separate proposal, the ET also proposes a companion RWS, based on the positive experience of AVT-331 (RTG) and AVT-354 (RWS) mutually benefiting from each other.
Objectives:
The goal of RTG is to identify and extend the state of the art (SoA) associated with ML-based architectures for vehicle design processes and to exploit ML methods within the context of multidisciplinary design for military vehicles. The RTG will investigate how ML can improve the vehicle design process and at what level (conceptual/preliminary design, etc.), as well aswhere in the design loop different ML approaches can be used. The expected deliverable from the RTG is a report including: a description of the current status and challenges of the SoA and the relevant different ML approaches; documentation of the method and framework technologies along with the benchmarks used in their evaluation, and an assessment of future needs and capabilities.
Topics:
The relevant scientific topics addressed in this RTG can be summarized as follows:
• Scoping review on ML-based design process for vehicles.
• Identifying where in the design loop and what kind of ML methods are best suitable for military vehicles and defining best-practice on their use.
• Assessing and comparing the utility of different ML approaches on the same benchmark problems, as well as their extension to other applications.
Contact:
Open2Partners:
Title:
Created at 24/10/2023 10:00 by System Account
Last modified at 16/05/2024 22:00 by System Account
Go back to list
Home(NATO STO)