TLDR: This paper presents an integrated framework for modeling and optimizing complex system-of-systems, particularly in aerospace. It addresses the challenges of computationally expensive simulations by using surrogate-based optimization techniques like Bayesian optimization, leveraging software from ONERA (SEGOMOE, SMT) and DLR (SBArchOpt, OpenTurbofanArchitecting). The approach is demonstrated effectively on jet engine architecture design, showing significant efficiency gains over traditional methods, and is part of the EU-funded COLOSSUS project for sustainable intermodal mobility.
Designing and optimizing complex systems, especially in fields like aerospace, presents significant challenges. Traditional methods often rely on detailed, physics-based simulations that are computationally expensive, time-consuming, and prone to failures. This can severely limit the exploration of innovative designs and architectures.
A recent research paper, titled “System-of-systems Modeling and Optimization: An Integrated Framework for Intermodal Mobility,” addresses these challenges by proposing an integrated framework that combines advanced modeling and optimization techniques. Authored by Paul Saves, Jasper H Bussemaker, Rémi Lafage, Thierry Lefebvre, Nathalie Bartoli, Youssef Diouane, and Joseph Morlier, the paper highlights a more efficient approach to system-of-systems (SoS) design.
Smart Optimization with Surrogate Models
The core of the proposed solution lies in surrogate-based optimization algorithms, particularly Bayesian optimization utilizing Gaussian process models. Instead of running full, expensive simulations repeatedly, surrogate models create a faster, approximate representation of the system’s behavior. This allows optimization algorithms to efficiently determine the most promising design points to evaluate, significantly reducing computational costs.
The framework leverages specialized software developed by leading aerospace research institutions. ONERA’s SEGOMOE (Super-Efficient Global Optimization with Mixture of Experts) software, which is based on surrogate models from SMT (Surrogate Modeling Toolbox), plays a crucial role in handling complex hierarchical variables. Similarly, DLR’s SBArchOpt software is used to interface optimizers with architecture models, such as those defined in OpenTurbofanArchitecting.
Also Read:
- AI Breakthrough: ChatGPT Secures Second Place in Advanced Spacecraft Piloting Simulation
- Generative AI: A Transformative Force in Australian Defence and Global Military Strategy
Real-World Applications and Collaborative Efforts
This integrated approach has been successfully applied to realistic aircraft engine test problems, specifically optimizing jet engine architecture design under real-world constraints, including hidden constraints that can cause simulation failures. The paper demonstrates SEGOMOE’s capabilities in solving these intricate problems within the DLR OpenTurbofanArchitecting software.
The research is part of the European Union (EU) funded COLOSSUS project, a collaborative effort led by DLR with significant interaction from ONERA. The project aims to explore aviation products, services, and business models, focusing on sustainable 4D-intermodal mobility. This includes investigating new aircraft configurations with lower environmental footprints, such as all-electric vertical take-off and landing (VTOL) aircraft and multi-role seaplanes, for short-range passenger air transport and even aerial wildfire fighting.
Empirical evaluations show the effectiveness of this integrated framework. For instance, in a jet engine optimization problem, the SEGOMOE algorithm achieved highly improved results compared to traditional evolutionary algorithms like NSGA-II, requiring significantly fewer computational evaluations to reach optimal values. This underscores the power of Bayesian optimization for complex engineering problems.
The collaboration between DLR and ONERA, utilizing their respective software platforms, provides a comprehensive solution for intricate design problems, encompassing both modeling and optimization. Future work aims to expand this collaboration further, integrating concepts for wildfire fighting and on-demand mobility within online platforms like DLR’s RCE. For more detailed information, you can refer to the full research paper here.
