A Methodology to Identify Physical or Computational Experiment Conditions for Uncertainty Mitigation

TL;DR

Complex, multi-disciplinary engineering designs contend with epistemic uncertainty that can propagate into costly risk. The paper proposes an ontology-driven methodology that combines system-level sensitivity analysis with targeted computational and physical activities (CX/PX) to identify and reduce key uncertainties, leveraging Sobol indices (e.g., $S_T$) and a dual-fidelity M&S toolkit (e.g., FLOPS and OpenAeroStruct) in an early-stage BWB aerostructural case. It offers a structured workflow from problem definition and ontology to uncertainty-driven experimentation, including CX guiding PX and a constrained optimization pathway for sub-scale similitude. The approach aims to deliver more reliable predictions and more efficient design processes, with practical impact on risk reduction and resource use, and is adaptable to broader aerospace design challenges.

Abstract

Complex engineering systems require integration of simulation of sub-systems and calculation of metrics to drive design decisions. This paper introduces a methodology for designing computational or physical experiments for system-level uncertainty mitigation purposes. The methodology follows a previously determined problem ontology, where physical, functional and modeling architectures are decided upon. By carrying out sensitivity analysis techniques utilizing system-level tools, critical epistemic uncertainties can be identified. Afterwards, a framework is introduced to design specific computational and physical experimentation for generating new knowledge about parameters, and for uncertainty mitigation. The methodology is demonstrated through a case study on an early-stage design Blended-Wing-Body (BWB) aircraft concept, showcasing how aerostructures analyses can be leveraged for mitigating system-level uncertainty, by computer experiments or guiding physical experimentation. The proposed methodology is versatile enough to tackle uncertainty management across various design challenges, highlighting the potential for more risk-informed design processes.

Figures (7)

• Figure 1: Illustration of a generic computer simulation and how System Response Quantities are obtained. Adapted from Roy2010
• Figure 2: Framework for uncertainty identification and reduction.
• Figure 3: Schematic illustration of the reduction of epistemic uncertainty via following the proposed framework.
• Figure 6: BWB concept used in this work.
• Figure 7: Comparison of sensitivity indices calculated by three different methods: Quasi Monte Carlo, Surrogate model with full sampling, and surrogate model with 10% sampling