Scholarlink

<p>Space mission architectures often feature complex interdependencies among diverse operations, making them difficult to evaluate using traditional techniques such as sensitivity analysis, optimization, or simple trade-off analysis. We address this challenge by treating space missions as systems-of-systems and then introducing an evaluation methodology that tailors the combination of surrogate models, explainable artificial intelligence (XAI) and physics-based sandbox game simulations for this setting. We demonstrate how intricate relationships between design variables in mission architectures can be illuminated by XAI, increasing interpretability and thus efficacy in design decisions. Specifically, we find that using Shapley additive explanations, a model-agnostic interpretability technique, reduces the need for heavy computational resources, while still addressing the high dimensionality and interdependencies within our space mission design setting. Our use case on on-orbit refueling for cislunar missions makes the demonstration concrete, where we find that the design of a refueling element is a significant design variable influencing the economic feasibility of on-orbit refueling architectural options. With such insights, decision-makers can assess the feasibility of deploying reusable systems, for example, and justify their costs. To our knowledge, this is the <i>first</i> study that couples an SoS simulation of cislunar logistics with surrogate models <i>and</i> XAI in a single, auditable pipeline. The proposed X-SMART framework (i) delivers real-time feature importance analysis, (ii) quantifies explanation quality through compactness, stability, and cross-method consistency diagnostics, and (iii) generates counterfactual architectural changes that translate model insight into concrete mass, design, engine-swap, or orbit-selection actions. We also set the stage for dedicated XAI frameworks for future space mission research, advancing the explainability of complex models in this field.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Explainability-Based Framework for Evaluating Space Mission Architectures Using Sandbox Games

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Rodrigo N. Schmitt,
  • Moacir F. Becker,
  • Daniel DeLaurentis,
  • Andrea Capannolo

摘要

Space mission architectures often feature complex interdependencies among diverse operations, making them difficult to evaluate using traditional techniques such as sensitivity analysis, optimization, or simple trade-off analysis. We address this challenge by treating space missions as systems-of-systems and then introducing an evaluation methodology that tailors the combination of surrogate models, explainable artificial intelligence (XAI) and physics-based sandbox game simulations for this setting. We demonstrate how intricate relationships between design variables in mission architectures can be illuminated by XAI, increasing interpretability and thus efficacy in design decisions. Specifically, we find that using Shapley additive explanations, a model-agnostic interpretability technique, reduces the need for heavy computational resources, while still addressing the high dimensionality and interdependencies within our space mission design setting. Our use case on on-orbit refueling for cislunar missions makes the demonstration concrete, where we find that the design of a refueling element is a significant design variable influencing the economic feasibility of on-orbit refueling architectural options. With such insights, decision-makers can assess the feasibility of deploying reusable systems, for example, and justify their costs. To our knowledge, this is the first study that couples an SoS simulation of cislunar logistics with surrogate models and XAI in a single, auditable pipeline. The proposed X-SMART framework (i) delivers real-time feature importance analysis, (ii) quantifies explanation quality through compactness, stability, and cross-method consistency diagnostics, and (iii) generates counterfactual architectural changes that translate model insight into concrete mass, design, engine-swap, or orbit-selection actions. We also set the stage for dedicated XAI frameworks for future space mission research, advancing the explainability of complex models in this field.