DeepONet-Based Bank Angle Profiles Design for Mars Entry with Actuator Constraints

The determination of the bank angle profile is crucial for Mars atmospheric entry guidance. A significant limitation of existing bank angle profile construction methods is their neglect of the dynamic process of the bank angle. This paper introduces a novel parameterized bank angle profile construction method using deep operator networks (Deep-ONet). Initially, a design strategy for parameterized bank angle profiles, considering the dynamic characteristics of the bank angle, is presented. The bank angle profile is then modeled as an operator dependent on specific parameters. The DeepONet, which is tailor-made for learning operators, is utilized to construct parameterized bank angle profiles of unknown analytical forms. To handle rapid changes in the bank angle, Fourier feature layers are incorporated into the network to encode the time inputs. Numerical simulations demonstrate that the proposed network can effectively approximate the desired bank angle profile. In the guidance design, using the proposed network-based parameterized bank angle profile can effectively reduce the deviation between the predicted and actual flight states.