A Python library for stochastic model updating with black-box models using PyUncertainNumber - a case study with the NASA UQ challenge

Model updating plays a vital part in enhancing the credibility of computational models, especially for safety-critical engineering applications subject to polymorphic uncertainties. A long-standing objective in engineering is to reduce discrepancies arising from various sources of uncertainties and approximations between the simulation model and the corresponding physical systems. This paper presents both likelihood-based and likelihood-free methods for stochastic model updating using PyUncertainNumber, with the particular focus on reducing epistemic uncertainty. The developed library allows for a user-friendly calling signature and features scalability with blackbox models. To demonstrate its efficacy and applicability, these methods are applied to the 2025 NASA and DNV Challenge on uncertainty quantification.

Keywords: Stochastic model updating, uncertain numbers, black box model, probability box, uncertainty quantification.