Design of an air to air interceptor using genetic algorithms

Murray Anderson, John E. Burkhalter, Rhonald M. Jenkins

Guidance, Navigation, and Control Conference and Exhibit · 1999 · 29 citations · 4 references

Concepts

Abstract

For this preliminary design study an apportioned pareto genetic algorithm, unique to the software package IMPROVE?“‘, was used to manipulate a solid rocket design code, an aerodynamic design code, and a three-loop autopilot to produce interceptor designs capable of accurately engaging a target. Twenty-nine design variables were required to define the optimization problem, and four primary goals were established to access the’performance of the interceptor designs. Design goals included 1) minimize miss distance, 2) minimize intercept time, 3) minimize takeoff weight, and 4) minimize maximum G-loading. In 100 generations the genetic algorithm was able to develop several basic types of external aerodynamic designs that ,performed nearly the same, with miss distances less than 10.0 feet. The solid rocket motors that propelled these external shapes shared one common characteristic: a large initial burning area. Initial combustion chamber volumes varied significantly, but the fastest interceptors generally had less fuel than would have been expected. The genetic algorithm did not prefer maximizing the amount of fuel within the rocket motor case (high fuel volume ratio). Introduction With the addition of guidance, an autopilot, and an airframe with movable control surfaces, basic rocketry expands into a more lethal and much more precise means of waging war. Rather than increasing the size of the warhead being delivered (to make-up for a loss in delivery accuracy), modem weaponeering has tended to use a small warhead coupled with an accurate control system. For ground launched systems, ideas such as “smart rocks” and “brilliant pebbles” that sprang from early Strategic Defense Initiative (SDI) research were based on the belief that the energy delivered by a small fast moving projectile without an explosive could be as lethal as a less accurate system with an explosive warhead. Similarly for air-launched ground attack weapons, Vietnam proved that delivery accuracy was paramount to defeating tough targets, and the Air Force devoted billions of dollars to the development of precise warhead delivery systems. Certainly the Persian Gulf War showed the benefit of the highly accurate weapon systems the Air Force developed. In more human terms, increased delivery accuracy means less collateral damage to nonmilitary facilities and less civilian casualties. While it is difficult to argue with .the success of the air-launched ground attack weapons that have been demonstrated in combat, these systems, and air-to-air missile systems, share a common design ‘philosophy with the less successful ground launch interceptors. When a system has a guidance system and autopilot, there is a tendency to compensate for less than stellar aerodynamic designs by shifting more and more of the delivery problems over to the autopilot. As a result, autopilots are typically very good and very robust, but the airframe and aerodynamics of the overall system are almost an afterthought. Overall system performance and system capability, therefore, suffers because of the overreliance on the autopilot to compensate for weaknesses in the aerodynamic performance of the weapon, system’. The goal of this research is to let an artificial intelligence tool, a genetic algorithm, design the aerodynamic shape while at the same time designing the propulsion system and key autopilot variables. This all-at-once approach to missile design is intended to provide a system capable of producing good aerodynamic shapes in addition to the good performance expected from an autopilot. Previous Work Rather than discuss the historical development of the autopilot, propulsion system, or’aerodynamics used for this research, it is more pertinent to discuss previous applications of genetic algorithms to autopilot and control systems. Norris and Crossleyz recently used a genetic algorithm to find gains to control a standard pedagogical two-disk torsional spring system. The approach used a very simple two-loop proportional-integral control system with velocity feedback. The two loop controller had three gains that needed to be determined as a function of variable spring stiffness. The objective functions were defined such that good gain values produce little error between the commanded and achieved disk rotation angles. Since two separate disks were being commanded in twist, a pareto genetic algorithm was used to try to minimize the rotational errors in both disks simultaneously. At the conclusion of 80 .generations, the resulting family (i.e. population of 80 members) of three controller gains were hybrid performers that would work reasonably well for both disks. Two points worth noting from this study was that the crossover * Deputy Director, Development Test Department, Senior Member, AIAA ’ Professor, Auburn University, Associate Fellow, AIAA * Associate Professor, Auburn University, Senior Member, AIAA Copyright@ 19!$lby Murray B. Anderson. Published by the American Institute of Aeronautics and Astronautics, Inc. with permission.

References

4