Development of a Deep Neural Network Model for Spacecraft Charging*

A Deep Neural Network (DNN) model using the kinetic Particle-in-Cell (PIC) simulation data for spacecraft charging has been developed. The model will be able to predict the potential structure around the spacecraft based on the system parameters of any given system with certain accuracy. DNN provides an excellent tool where we could leverage deep learning for all machine learning tasks and expect better performance with surplus data availability. The usability of DNN is unlimited if a user can train such a model with physics-based parameters. In recent studies, it has shown promising outcomes in terms of accurate prediction of physical quantities[1-2]. In the present work, we used Tensorflow[3], a deep learning library, to classify the PIC simulation datasets. Using Tensorflow, we have compared the effects of multiple activation functions on classification results and developed a library for a case study. This work shows that integrating DNN into traditional computational methods might be the new beginning of developing next-generation modeling.

1. Cheng, Chen and Zhang, Guang-Tao, Water 2021, 13(4), 423 (2021)

2. Lagaris, I. E., Likas, A., and Fotiadis, D. I., IEEE trans. on neural networks, 9(5), 987-1000 (1998)

3. Abadi M. et al., 12th USENIX (OSDI'16), pp. 265-284, (2016).