Performance evaluation of neural and fuzzy-neural fault tolerant aircraft auto-landing controllers
Date of Issue2008
School of Electrical and Electronic Engineering
This dissertation aims to do a performance evaluation between two controllers viz. Neural network aided Baseline Trajectory Following Controller and Fuzzy-Neural aided Baseline Trajectory Following Controller for the auto-landing problem of a high performance fighter aircraft (a modified model of F-l6) under the failures of stuck control surface and severe winds. There are mainly three parts in this dissertation. The first part describes the background information and the landing task along with the available aircraft model arid baseline conventional controller used for our simulation and performance evaluation. It also describes the scope and constraints, assumptions made in our evaluation of the neural and fuzzy neural controller. The Second part describes the two algorithms viz, the GAP-RBF algorithm and the SAFIS Algorithm in detail and their implementation for the aircraft problem hence setting up the stage for the evaluation process between these two neural and fuzzy- neural algorithms respectively. The Third part finally deals with the Performance evaluation and results obtained for the various fault scenarios and the comparison between GAP and SAFIS with all other parameters being set the same. The Results show that the SAFIS (Fuzzy-Neural Controller) indeed performs better than the GAP (Neural Controller) in satisfying the pillbox requirements in a wider range of failures and the Fuzzy system closes the “holes” in the fault tolerance envelope of the neural aided controller. Thus input fuzzification really improves the performance of fault tolerant auto landing controllers.
DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Control engineering