Formation Flying of Satellites

Background

Nowadays the field of satellite research shifts to the research area of multiple agent system (MAS). Using multiple (small) satellites which work in cooperation for a specific mission is a more fail-safe and cheaper solution that using one large satellite. In case of a failure of one the satellites in the MAS, the remaining satellites can still perform the task (albeit with a reduction in performance). Obviously, when using one large satellite, a failure of that satellite terminates the mission altogether. Also the launch of multiple smaller satellites is cheaper than launching one large satellite. Smaller satellites can piggy-back on other launchers. Risk reduction is also significant when looking at launches. Losing several smaller satellites on a launcher is less expensive than losing one large satellite.

Figure 1: Darwin (source: ESA)

Next to the safety, economic, and risk management reasons there is the feasibility reason. Some mission cannot be fulfilled with one large satellite. Examples are SAR (Synthetic Aperture Radar) missions which require a very large aperture. Other applications like deep space exploration also benefit from using multiple satellites which can mimic one very large telescope. Mission as for instance Darwin from ESA (see figure 1) and Terrestrial Path Finder from NASA (see figure 2) are both multi-satellite mission.

Project Goal

Figure 2: Terrestrial Path Finder (source: NASA)

The main goal of the project is to develop a formation flying control system for multiple satellites. The control system must be highly autonomous and must optimize the performance of the cluster. In the project proposal the control system structure is divided into two levels: high level and low level control (see figure 3).

At the higher level, the cluster reconfiguration in terms of satellite positioning and subsequent trajectory reference planning is performed. The problem of reconfiguration is a non-linear global optimization problem. The optimization includes aspects such as mission life span, collision avoidance during cluster reconfiguration, and mission related requirements such as resolution, observation time, and satellite failures and subsequent performances. The technique which will be used to solve the global optimization problem is interval analysis. By using interval analysis we are guaranteed to find all global optima with probability one.

The low level control is performed within each satellite and uses the reference trajectory set by the high level control. Low level control performs satellite attitude control and trajectory tracking. In case of a malfunction of some part of the satellite, the performance of the low level control will be degraded. Depending on the type and magnitude of the malfunction, the satellite can still perform some tasks. To optimization the remaining capabilities of the satellite we should make the low level control adaptive. Reinforcement learning (RL) is proposed as the technique to make the control adaptive. RL is optimal learning technique which can handle with, grey and black box cases.