On-line learning neural networks for sensor validation for the flight control system of a B777 research scale model

This paper focuses on the analysis of a scheme for sensor failure, detection, identification and accommodation (SFDIA) using experimental flight data of a research aircraft model. Recent technical literature has shown the advantages of time-varying estimators and/or approximators. Conventional approaches are based on different versions of observers and Kalman filters while more recent methods are based on different approximators based on neural networks (NNs). The approach proposed in the paper is based on the use of on-line learning nonlinear neural approximators. The characteristics of three different neural architectures were compared through different sensor. failures. The first architecture is based on a multi layer perceptron (MLP) NN trained with the extended back propagation algorithm (EBPA). The second and third architectures are based on a radial basis function (RBF) NN trained with the minimal resource allocating network (MRAN) and extended-MRAN (EMRAN). The MRAN and EMRAN algorithms have recently been developed for RBF networks and have shown remarkable learning capabilities at a fraction of the memory requirements and computational effort typically associated with conventional RBF NNs. The experimental data for this study are flight data acquired from the flight-testing of a 1/24th semi-scale B777 research model designed, built, and flown at West Virginia University (WVU). Copyright (C) 2002 John Wiley Sons, Ltd.