Adaptive Nonlinear Neural Controller for Aircraft Under Actuator Failures

This paper presents an adaptive backstepping neural controller design for aircraft under control surface failures. The control scheme uses radial basis function neural networks in an adaptive backstepping architecture with a full state measurement for trajectory following.13; The requirement for stability is separated from the network-learning part. This allows us to use any function approximation scheme (including neural networks) for learning. For the radial basis function neural networks, a learning scheme in which the network starts with no eurons and adds new neurons based on the trajectory error is developed. Stable tuning rules are derived for the update of the centers, widths, and weights of the radial basis function neural networks. Using Lyapunov's theory, a proof of stability in the ultimate bounded sense is presented for the resulting controller. The fault-tolerant controller design is illustrated for an unstable high performance aircraft in the terminal-landing phase subjected to multiple control surface failures of hard over type and severe winds. The design uses the full 6-degree-of-freedom nonlinear aircraft model, and the simulation studies show that the above controller is able to successfully stabilize and land the aircraft within tight touchdown dispersions.