MS15: Advances in modeling complex systems with applications to space engineering


  • Daoru Han, Missouri University of Science and Technology

Session A: Oct.1, 5:00pm-6:20pm, Classroom Building 118

5:00pm-5:20pm (Oct 1)
CLB 118

Bingling Huang,
University of Southern California

Understanding reinforcement learning-based agents in self-organizing systems


Complex systems can be applied to wide engineering fields, such as air force and navy systems. As the task environments become more unpredictable, it puts forwards higher requirements on the complex system design.
Self-organization refers a process of organization to an arising overall order from members' local interactions in response to changing external circumstances. One advantage of self-organization approaches is that individual can be kept relatively simple, and the emergent behaviour of the overall system can be expected to be sophisticated enough to deal with various demanding tasks. Therefore, it has become a promising strategy for the complex system design.
Reinforcement learning method has been proposed to let agents perform self-organization, where agents learn how to accomplish tasks by maximizing reward functions. However, several issues become sharper as the task complexity increases. Firstly, how to design effective reward functions to accomplish tasks with limited prior knowledge. Furthermore, how to conduct efficient trainings with predictable task performances and economic computational resources. Therefore, bridging the human’s and agents’ knowledge and understanding the mechanism inside a RL-based agent group are in big demands. Targeting at assisting the complex system design, the relevant contents and its applications in aerospace areas will be discussed in this talk.

5:20pm-5:40am (Oct 1)
CLB 118
Chen Cui,
University of Southern California

Development of grid-based Vlasov method for kinetic studies of plasma flows in space engineering
  We present the up-to-date development progress of the grid-based Vlasov method and the parallel, multi-dimensional grid-based Vlasov solver, Vlasolver for the plasma flow in space engineering. Numerical schemes and implementations are developed to solve the Vlasov-Poisson and Vlasov-Darwin systems. Two example applications are presented. The first one is a re-evaluation of one-dimensional collisionless plasma expansion into vacuum. It is shown that the grid-based method allows us to extend the self-similar solution and resolve the electron time scale perturbations. The second one is a two-dimensional simulation of plasma plume emitted from plasma thrusters. The comparison of the results from the Vlasov and PIC simulations shows that, while both models agree well on overall plume density and electric potential, the Vlasov model is able to resolve high-order moment physical properties without the effects of statistical noise. A core region and a fan region can be found in the electron heat flux. Near-Maxwellian features are found in the vx direction while top-hat shape features are found in the vy direction for electron velocity distribution function. We find that the grid-based Vlasov method, though computationally more expensive than PIC, can be advantageous in applications requiring accurate eVDF and accurate small-scale physical properties.

5:40pm-6:00pm (Oct 1)
CLB 118
David Lund,
Missouri University of Science and Technology

Kinetic Simulations of Charging of Irregularly-Shaped Dust Grains in Space Plasmas
  The objective of this study is to investigate a fully-kinetic numerical investigation of charging of irregularly-shaped dust particulates in low temperature collisionless plasmas. The recently developed Parallel Immersed-Finite-Element Particle-in-Cell (PIFE-PIC) code is utilized to self-consistently resolve the plasma environment and charging of immersed materials. This model explicitly includes the material property (dielectric constant) of dust grains. Multiple dust grain shapes/configurations will be considered and compared to find how multiple dust grains are charged in a collisionless plasmas.

6:00pm-6:20pm (Oct 1)
CLB 118
Easton Ingram,
Missouri University of Science and Technology

Kinetic Modeling of Electrostatic Transport of Lunar Regolith Particles with Applications to In-Situ Resource Utilization
  Continuous human presence on the Moon will require materials made on-site / in-situ. Many of these can be made from calcium and aluminum, elements available in the lunar regolith. However, their mineral grains (called anorthite) must first be separated from the other mineral grains in the regolith. The concept of electrostatic sieving is promising for use under the lunar environment. In this study, we will model the working process of electrostatic sieving by tracking the trajectories of charged lunar regolith particles under the electric field generated by an array of electrodes. Effects of size separation and efficacy of this concept will be presented and discussed.