This work proposes a novel approach for non-deterministic extension of experimental data that considers structural model inadequacy for conditions other than the calibration scenario while simultaneously resolving any significant prior-data discrepancy with information extracted from flight measurements. This functionality is achieved through methodical utilization of model error emulators and Bayesian model averaging studies with available response data. The outlined approach does not require prior flight data availability and introduces straightforward mechanisms for their assimilation in future predictions. Application of the methodology is demonstrated herein by extending material performance data captured at the HyMETS facility to the MSL scenario, where the described process yields results that exhibit significantly improved capacity for predictive uncertainty quantification studies. This work also investigates limitations associated with straightforward uncertainty propagation procedures onto calibrated model predictions for the flight scenario and manages computational requirements with sensitivity analysis and surrogate modeling techniques.

翻译：本文提出了一种非确定性扩展实验数据的新方法，该方法在考虑除校准场景外其他条件下结构模型不充分性的同时，利用飞行测量中提取的信息解决先验数据与实测数据之间的显著差异。通过系统性地运用模型误差模拟器及基于贝叶斯模型平均的可用响应数据研究，实现了上述功能。所提出的方法无需预先具备飞行数据，并引入了将这些数据整合到未来预测中的直接机制。本文通过将HyMETS设施获取的材料性能数据扩展到MSL场景，展示了该方法在实际中的应用，经此过程得到的结果在预测不确定性量化研究方面展现出显著提升的能力。此外，本文还探究了将简单不确定性传播程序直接应用于飞行场景校准模型预测的局限性，并通过敏感性分析与代理建模技术管理计算需求。