28 TH INTERNATIONAL CONGRESS OF THE AERONAUTICAL SCIENCES 1 Abstract This paper presets flight demonstration of fault tolerant flight control system using a small model plane. Two types of fault tolerant controller based on neural network were developed to maintain automatic flight stability when failures suddenly occured in airframe structure. To simulate the sudden structure failure, a 1/10 scale business jet type model plane with wing-tip separation device was designed and an automatic control system installing fault tolerant controllers was developed. Flight experiments were carried out under the following scenario; the model plane flew automatically aiming four prescribed waypoints, the wing tip was separated suddenly by a command signal from the ground station, and the performance of fault tolerant controllers were investigated. The flight results could demonstrate the effectiveness of the developed controllers which adaptively regulated the serious damage of airframe structure during the flight. 1 Introduction In many future forecasts, the economic growth and demographic changes in the emerging economic nations can drive the steady increase of air traffic flow over the next 20 years. On the other hand, the fatal accident rate has remained almost constant for the past twenty years [1]. Therefore further reduction of the fatal accident rate is essential to prevent the increasing number of fatal accidents. Airline accident incident statistics [2] shows that the major accidents can be attributed to Loss of Control In-Flight (LOC-I) caused by a piloting mistake, technical malfunctions or unusual upsets due to external disturbances. It is expected that a fault tolerant flight control, or a resilient flight control improve survivability and recovery from adverse flight conditions induced faults, damaged and associated upsets [3]. While many research projects associated with the fault tolerant control have been carried out, it is very difficult to demonstrate the control performance for real airframe structure failure cases since the airworthiness should be satisfied for aircraft operated by a human pilot. One possible demonstration is the use of a scale model as unmanned aircraft. The flight test of a fighter type model plane with wing structure failure was presented [4], but flight tests for a civil type model plane with stricture failure have not been presented. The authors designed a business jet type scale model plane with wing tip separation mechanism as a model of wing structure failure. It is because there are strong indicators of an increasing market for business jet both in developed and developing counties, however it is reported that the fatal accident rate for all business jet civil operations is more than eight times that for large jet passenger aircraft in airline operation between 2000 and 2007 [5]. Therefore, a business jet type scale model was selected in our research project. Additionally, fault tolerant control system based on software algorithms is suitable for a small business jet since the use of expensive hardware devices such as redundant control surfaces is strictly limited through the viewpoint of cost or available space. Fault tolerant control algorithms based on neural networks (NNs) are investigated in our project [6]. NNs are a mathematical model that is based on biological neural networks. It is FLIGHT DEMONSTRATION OF FAULT TOLERANT FLIGHT CONTROL SYSTEM Shinji SUZUKI*, Koichi MIYAJI*, Takeshi TSUCHIYA*, Masaru NARUOKA** Tsunahiro SATO***, Naoaki ITABASHI***, and Akira YANAGIDA**** *The Univerity of Tokyo, **JAXA, ***Fuji Heavy Industry Ltd, ****The Society of Japanese Aerospace Companies tshinji@mail.ecc.u-tokyo.ac.jp