Integration of Condition Information in UAV Swarm Management to increase System Availability in dynamic Environments

The approach of prognostics and health management (PHM) focuses on the real-time health assessment of a system under its actual operating condition and even extending this by the prediction of the future state based on up-to-date system information. This pursues the aim to derive more advanced maintenance or asset deployment strategies in order to keep the operation of the system safe and reliable. In this context, the outcome of a PHM system is often used as a decision support. For a high fidelity system where the actual state is considered at every timestep and a decision is executed immediately based up on this information, Reinforcement Learning (RL) becomes a tool to find an optimized solution. Therefore the paper presents a methodology that integrates health and operational data into a RL approach in order to derive immediate operational strategies for lower degradation and higher safety and reliability. The approach is evaluated on the basis of a swarm of unmanned aerial vehicles (UAVs) that performs a complete-area path-coverage (CAPC) mission. It can be shown that the integration of health information as well as environmental data describing dynamic operating conditions lead to lower degradation and result in more reliable operations of the swarm while achieving a more flexible mission performance compared to pre-divided swarm-missions. Varying states are also taken into account, which emphasises this approach to be a highly dynamic PHM system application.