Speeding up heterogeneous binary asteroid system propagation through the physics-informed neural network

This paper proposes the application of a physics-informed neural network (PINN) to the propagation of heterogeneous binary asteroid systems. The accuracy and efficiency of such propagation are important in the study of celestial mechanics and mission analysis, where we devote to achieving a reasonable balance. The gravitational interactions, which are necessary quantities for this integration, are formulated in Taylor expansion representation that incorporates the derivatives of the primary's gravitational potential, the secondary's generalized inertia integrals, and the relative geometry. To represent the gravity field of the primary with heterogeneous mass distribution, a hybrid model combining a quadrature-based polyhedron model and a PINN-based model is developed. The derivatives of the resultant gravitational potential are obtained by superposing those from the polyhedron and PINN-based models, with calculations performed using analytical formulas and automatic differentiation, respectively. For the gravitational potential evaluations, the hybrid model offers faster computation speed and comparable precision compared to the benchmark model. Its application to binary asteroid system propagation demonstrates that the PINN component can effectively capture the effects of non-uniform mass distribution of the body. Furthermore, our mutual dynamics simulations suggest that the heterogeneous mass distribution of the primary may significantly influence the orbital period of the system.