Stochastic Bayesian model updating to reduce epistemic uncertainty in satellite attitude propagation

Smith, Ewan and Bi, Sifeng and Feng, Jinglang and Cavallari, Irene and Vasile, Massimiliano (2024) Stochastic Bayesian model updating to reduce epistemic uncertainty in satellite attitude propagation. In: SciTech Forum 2024, 2024-01-08 - 2024-01-12.

[thumbnail of Smith-etal-SciTech-2024-Stochastic-Bayesian-model-updating-to-reduce] Text. Filename: Smith-etal-SciTech-2024-Stochastic-Bayesian-model-updating-to-reduce.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 8 January 2025.
License: Creative Commons Attribution-NonCommercial 4.0 logo

Download (2MB) | Request a copy


The tracking of space assets within increasingly populated orbits is a critical task in the pursuit of safe space operations; however, with limited observations the gaps in time series data must be filled by numerical attitude simulation models. In practice, given the nonlinear nature of spacecraft dynamics and uncertain space disturbances, such simulations can be prone to deviating from the true underlying dynamics. In this work, we follow the approach adopted by the stochastic model updating community, where the model discrepancy is accounted for by uncertain model parameters. Working from limited prior data due to sparsity and cost, such system parameters can be hard to define with a precise distribution and are left defined by epistemic uncertainty formulations. We propose an extension of the stochastic model updating scheme, which follows the basic foundational principles of likelihood and Maximum Likelihood Estimation to reduce the amount of epistemic uncertainty attached to model parameters in time-domain models. Compared to current stochastic model updating approaches, our new methodology avoids the use of Approximate Bayesian Computation while also being able to operate in an online fashion. The application of such an approach is demonstrated against a satellite attitude propagator with initially epistemic uncertain moments of inertia to highlight the reduction in parameter uncertainty as measurements arrive.