Abstract
Current fighter aircraft are required to be maneuverable and agile at high angles of attack. Limited control power available at these extreme flight conditions prompts the need to use control augmentation techniques. The method of tangential slot blowing is found to be an effective technique. This thesis discusses a neural network based controller designed for the problem of systems with both continuous and bang-bang effectors. The technique is applied to the benchmark problem of a generic X-29A aircraft lateral/directional dynamics. A full state feedback controller is used for the continuous control effectors. The neural network designed to control the bang-bang effectors is a three-layer network with symmetric hidden layers, which optimizes a given quadratic performance index. This performance index allows the designer to specify appropriate weights for states and control effectors to satisfy given specifications. The Neural Network Controller is directly compared to previously designed Model Predictive Variable Structure and Fuzzy Logic based controllers for the same benchmark problem. Evaluation criteria consist of closed-loop system performance, activity level of the VFC nozzles, ease of controller synthesis, time required to synthesize the controller and robustness of the controller. The neural controller designed for the benchmark problem of X-29A lateral/directional motion gives better results than the MPVSC and Fuzzy Logic based controllers. MPVSC is easier to design than the Neural and Fuzzy controllers. The neural controller is more robust and reliable to variations in plant as well as controller dynamics.
Joshi, Praveen Sudhakar (1999). Direct comparison of Neural Network, Fuzzy Logic and Model Prediction Variable Structure vortex flow controllers. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1999 -THESIS -J68.