An engagement zone based modeling and control approach to the multi agent air-to-air engagement problem

The problem of launch envelope generation has been a general application problem since the beginning of military aviation history. While the first studies in this field were carried out with the help of firing tables/charts, today, complex computational methods and advanced flight simulations are used for these problems. In air missions, having real-time weapon engagement zone (WEZ) information for guided weapons or missiles is of vital importance for mission success, especially for pilots in air-to-air missions. Real time WEZ information can be obtained through fast but low-fidelity flight simulations, or it can be modeled using previously obtained high accuracy six-degree-of-freedom flight simulation data. In this technique of using simulation data, while air-ground missions can be implemented by modeling simulated results in all possible flight conditions, the number of simulations needed to cover all conditions of air-to-air engagement increases exponentially with increasing parameters such as target altitude, Mach number, and aspect angles. In the first part of this thesis, in order to solve this problem a six-degree-of-freedom simulation model for a generic air-to-air missile with high fidelity and detail level is developed. Then, in order to reduce the need for a large number of flight simulations, a new modeling method based on modern sequential design of experiments and iterative use of neural networks is presented. In this method, flight simulations and modeling are performed at the design points iteratively and mathematical WEZ models are derived by continuing until the desired performance level is reached. In the second part of the study, the problem of autonomous engagement maneuvers is addressed by developing aircraft models for a multi-agent air-to-air engagement environment, and the introduction of the obtained WEZ models to the problem in the form of control constraints and performance indices is evaluated. In contrast to traditional swarm control methods, control with reinforcement learning, which has developed rapidly in recent years, has been applied for this highly dynamic system. For multi-agent air-to-air engagements, first an algorithm to take autonomous maneuver decisions in single engagement is prepared. Then, a target allocation algorithm and a multi-engagement architecture that considers WEZ domains are established to ensure that the agents move to the desired system states. The learning performance and engagement results obtained from the study are validated in a simulation environment for different scenarios.