Guaranteed Stability and Improved Performance Against Actuator Failures Using Neural-aided Sliding Mode Controller for Autolanding Task Shaik Ismail*. Abhay A. Pashilkar.* Ramakalyan Ayyagari** *FMC Division, National Aerospace Laboratories, Bangalore, India (e-mail: shaik@nal.res.in, apash@nal.res.in). **Department of ICE, NIT Trichy, India (e-mail:rkalyn@nitt.edu). Abstract: This paper presents a novel Minimum Resource Allocation Neural Network (MRAN) based controller that enhances the fault tolerance capabilities of a high performance fighter aircraft during the landing phase when subjected to severe winds and failures such as stuck control surfaces. The neural controller is trained on-line to learn the inverse dynamics of the aircraft. Autolanding simulations show that the fault-tolerance envelope of the combined MRAN+SMC+BTFC controller is much wider than those of the BTFC and BTFC+SMC controllers. It is assumed that information about actuator failures is not available to the controller for use in reconfiguration, and no Fault Detection and Diagnosis (FDD) schemes are used. Keywords: autolanding; fault-tolerant control; RBF; MRAN; SMC. 1. INTRODUCTION The current trend in control law design for air vehicles is towards fault tolerant flight control focussed on ensuring and improving the required safety levels and reducing risks due to critical faults (Edwards, et. al., 2010). The most critical phase of flight of any air vehicle is landing, and both manned and unmanned air vehicles are resorting to autolanding to ensure adequate safety and flying qualities. The classical auto- landing control systems perform satisfactorily under nominal operating conditions, but generally fail to cope with faults such as the control surfaces being stuck at certain deflections and damages to aircraft structure. Therefore, there is a need for investigating novel fault-tolerant control schemes for autolanding tasks. Only actuator faults are considered in the present work. Most of the fault-tolerant control systems (FTCS) use fault detection and diagnosis (FDD) schemes for control reconfiguration/restructure. Using FDD, anticipated faults could be easily tackled by design. However, when a fault occurs, which does not belong to the class of faults designed for, it can result in a situation where that particular fault is either not detected or is incorrectly identified. Also, FDD schemes need a finite time to become effective, but in that finite time the aircraft stability and performance must be guaranteed. Difficulties also arise when some aircraft parameters need to be identified during the flight in real time and under feedback control and actuator faults and damages to aircraft structure. In view of all these problems posed by the FDD schemes, a self-learning or adaptive fault-tolerant controller, without FDD, is a plausible solution for autolanding tasks. A comprehensive survey of literature reveals three viable methods for reliable fault-tolerant control without FDD: neural-adaptive control, sliding mode control and the recently developed L 1 adaptive control. Neural networks provide a fast mechanism to achieve fault tolerant control without FDD because of their ability to learn on-line and to adapt the aircraft control systems to sudden changes in the environment as well as sensor and actuator failures. Based on a study of different architectures for neural control, a simple control architecture which uses a Minimal Resource Allocation Neural Network (MRAN) controller is used to learn the inverse dynamics of the aircraft. MRAN is a Radial Basis Function (RBF) type neural network using only online learning to represent the local inverse dynamics of the aircraft system. The ability of sliding mode control (SMC) to maintain the desired performance of the aircraft in case of faults without requiring an explicit FDD system makes it another plausible method for passive fault-tolerant control (Alwi, 2008). The recently developed L 1 adaptive control theory has also been successfully used to deal with unknown actuator failures without FDD (Patel, el. al., 2009). In this paper a novel approach is developed which combines the classical feedback control with SMC and the MRAN controllers for fault-tolerant control, without FDD, as shown in Fig. 1. This approach has guaranteed stability as well as improved fault tolerance properties under actuator failure conditions during autolanding task. The autolanding problem is discussed in Section 2. A brief description of the high performance aircraft model used is