Abstract : In general, it is difficult to automatically detect faults within large-scale engineering plants early during their onset. This is due to a number of factors including the large number of components typically present in such plants and the complex interactions of these components, which are poorly understood. Traditionally, fault detection within these plants has been performed through the use of status monitoring systems employing limit checking fault detection. . In this dissertation, a fault detection methodology is proposed which is better suited to the early fault detection task than the traditional limit checking. The work in this dissertation focuses on the performance of a neural networks (NNs) based fault detection system within a flight control system. This fault detection flight control system integrates sensor fault detection, identification, and accommodation (SFDIA). The SFDIA task is achieved by incorporating a main neural network (MNN) and a set of n decentralized neural networks (DNNs) for a system with n sensors assumed to be without physical redundancy. Particularly, the purpose of the MNN is to detect a wide variety of sensor faults, while the purpose of the DNNs is to identify the particular sensor that has failed and to accommodate the fault. The NNs are first trained off-line and learned from healthy data using the Extended Back-Propagation Algorithm (EBPA) and then used on-line. Because of the on-line learning, neural estimators and controllers have the capability of adapting to changes in the aircraft dynamics and/or modeling discrepancies between the actual aircraft and its mathematical model. This factor makes neural estimators and controllers an attractive option for fault _ diagnosis flight control system. The work in this dissertation addresses the problem of how to perform fault detection within the domain of large scale engineering plants when little or no analytical information is available regarding their operation. In particular, a technique is developed which allows accurate black box modeling for the plant_s components regardless of the particular types of non-linearity or dynamics, which may be present. The resulting system is also designed to permit the detection of the onset of fault conditions as early as possible; ideally, prior to the critical fault of the given component(s). Particular emphasis is placed in this work towards improving the performance of the SFDIA scheme in the presence of ramp-type soft faults, which are hard to detect without degradation of performance in terms of false alarm rates and incorrect fault identification. The SFDIA was modeled, written and tested using the MATLAB7 programming environment for a closed loop non-linear model of a De Havilland 2 _Beaver_ aircraft. Three types of faults, with six forms in each, were artificially injected as faults for each sensor separately. Then a general and most complicated case incorporating more than one sensor and one fault is taken. In each case, the SFDIA is found to be capable of accurately and timely detecting, identifying, and accommodating the faulty sensor