Genetic algorithms for tuning fuzzy membership functions in flight control software

Abstract

In the past twenty years, the U.S. general aviation industry has been beset by flagging membership and substantial increases in operating costs. In an attempt to reduce costs, improve training programs, and increase safety factors, the Knowledge Based Control Systems Group at Texas A&M University has pursued research in the area of general aviation cockpit management. This research has investigated low-cost solutions which bring the sophisticated cockpit technology of commercial aircraft to their general aviation brethren. One project developed during the course of this research was a Flight Mode Interpreter (FMI). The FMI performs real-time, high-level situation recognition based on data from aircraft flight instruments. Flight modes such as "Takeoff' and "Cruise" are defined in a rulebase file, along with fuzzy membership functions (MBFs) which map the flight modes to expected partitions of the aircraft state space. The FMI uses a fuzzy logic inference engine to determine the current flight mode based on applying aircraft sensor data to the rules contained in the rulebase. During development of the FMI, it was discovered that the construction of these fuzzy rulebases was a tedious and time-consuming task. Defining the flight modes to be recognized was simple, but developing MBFs which allowed the FMI to make correct decisions was not. Because the FMI is meant to operate on numerous aircraft, it was advantageous to study methods for automating parts of the rulebase development process. This would allow the construction of independent rulebases for different aircraft to proceed in the most efficient manner possible. A solution based on a parameterized genetic algorithm (GA) was developed to address this need for automation. Known as the GAPATS Expert Builder, this tool aids in the development of rulebases for the FMI by automating the trial-and-error process used to generate first-cut MBFS. While the tool does not produce final, polished solutions, it provides a valuable head start in the development of rulebases, and also proved to be a useful mechanism by which the FMI software may be studied and functionally verified.