An Inversion Model to Obtain Spacecraft Surface Potential Based on Neural Network

A surface charging detector carried on spacecraft monitors the charging level of a specific dielectric by space plasma. It is insufficient to evaluate the charging and discharging risks on different types of dielectrics on the spacecraft surface. In this study, we propose an inversion model based on the backpropagation neural network (BPNN). It can rapidly derive the surface charging potential in a nondetection area by the detector’s potential in the same charging environment. We used the Kapton surface potential, surface current, and material parameters as input and the surface potential of other materials as output. We trained BPNN with the simulation data using a surface charging model and optimized BPNN using a genetic algorithm to reduce the training error to less than 5%. Then, we verified the inversion model using a surface charging experiment on the ground. The result showed that the relative error between the inversion and experimental values was less than 12%. The model can estimate different dielectric surface charging levels based on the detector’s data.