A blended physics-based and black-box identification approach for spacecraft inertia estimation

In this paper, the problem of identifying inertial characteristics of a generic space vehicle relying on the physical and structural insights of the dynamical system is presented. To this aim, we exploit a recently introduced framework for the identification of physical parameters directly feeding the measurements into a backpropagation-like learning algorithm. In particular, this paper extends this approach by introducing a recursive algorithm that combines physics-based and blackbox techniques to enhance accuracy and reliability in estimating spacecraft inertia. We demonstrate through numerical results that, relying on the derived algorithm to identify the inertia tensor of a nanosatellite, we can achieve improved estimation accuracy and robustness, by integrating physical constraints and leveraging partial knowledge of the system dynamics. In particular, we show how it is possible to enhance the convergence of the physics-based algorithm to the true values by either overparametrization or introducing a black-box term that captures the unmodelled dynamics related to the offdiagonal components.