![]() | Only 14 pages are availabe for public view |
Abstract The Cairo Fourier Diffractometer Facility (CFDF) has been installed at one of the ET-RR-l reactor horizontal channels and is based on the reverse time of flight (RTOF) concept. The CFOF is a unique neutron diffractometer and there is only four of such type around the world. Unfortunately, the CFDF does not contain an integral control panel to include all the functional indicators of the system. Moreover, some devices of the system have no indicators at all. Since constructing of a such control panel is out of the scope of this thesis; therefore it is assumed that there is a central control panel which contains a selected number of the most important indicators. The diagnosis of such type of equipment is a very complex task. It depends on the conditions of the indicators of the control panel of the device. The accurate diagnosis can lead to a high system reliability and save much of the maintenance costs. So an expert system for the CFDF fault diagnosis is needed to be built. In this thesis, a hybrid expert system is developed for the fault diagnosis of the CFDF. Two intelligent techniques, the multi layer feed forward-back propagation neural network and the rule based expert system, are integrated as a pre-processor loosely coupled model to build the proposed hybrid expert system. The architecture of the developed hybrid expert system consists of two levels. The first level is a feed forward back propagation neural network, and is used for isolating the faulty part of the CFDF. The second level is the rule based expert system, used for troubleshooting the faults inside the isolated faulty part. The inputs of the neural network level are the indicators conditions from the assumed CFOF control panel (symptoms). The outputs correspond to the status of the main parts of the CFDF. The back propagation training algorithm is used for the neural networks training. Learning was performed using different values of the learning factor 11,the momentum factor a and the different number of hidden layers neurons are also investigated. The rule based expert system takes the inputs and outputs of the neural networks and also information from the user by questions and answers, to define precisely the faults in the faulty part. The rule based expert system has the features of being flexible enough to build or update the knowledge base. The rule based expert system uses five independent knowledge bases related to each of the CFDF main parts. It has been found that the developed diagnostic system is both adequate and flexible for the CFDF . . The present work is organized to fall in five chapters III Chapter (1) presents the artificial intelligence strategies. The neural networks, the fundamental components of the artificial neural network, the artificial neural network learning, the activation functions and the types of learning are investigated. Also a description for the rule based expert system structure, the knowledge engineering, the rules as the knowledge engineering, the rules as the knowledge representation scheme for the knowledge base, the inference methods in forward and backward chaining, the methods of fntegration of ;’heural networks and expert systems, and the comparison between the neural networks and the expert systems are included. Chapter (2) reviews some of the previous works on the applications of neural networks and expert systems for diagnosing physical devices. Chapter (3) presents a description of the CFDF parts, the symptoms used for the fault diagnosis and its effect on the CFDF operation; and the • conventional method for fault diagnosis and its disadvantage. Chapter (4) deals with the design and implementation of the back propagation algorithm and the experimental result for the network training, the effect of changing the parameters learning factor, momentum factor, hidden nodes number on the number of iterations used for complete training. Also given are the steps for implementing the rule based expert system. Chapter (5) gives the conclusion and the future work for the present work The basic results of the present thesis were presented in a paper entitled: Fault Diagnosis Using Artificial Neural Network By: R. M. A. Maayouf, N. M. H. ElMinyawi, N. M. Ayad and S. M. Abdelwahed. The paper was published in : Arab J. of Nucl. Scien and Appl. 37(2), May (2004), (233-244). • |