Aircraft trajectory control using artificial neural networks

Abstract

The feasibility of using artificial neural network to obtain the control inputs required for a desired aircraft trajectory is studied. The objective is to provide pilot assistance in the high work load environment of Terminal Control Areas. The trajectory between any two initial and final state vectors is broken down into three simple maneuvers. The maneuvers are designed to bring an aircraft from some initial position and velocity to the outer marker during the approach procedure. An artificial feed forward neural network is trained to accept position and velocity and give the corresponding thrust, lift coefficient, and bank angle as its outputs. The trajectory parameters compare well with the nominal position and velocity. The final conditions are achieved with an accuracy of 500 feet in ground position, 1 foot in altitude, 1 foot/second in velocity and 3 degrees in heading angle.