| The purpose of this thesis is to demonstrate the use of Neural Network (NN) and Fuzzy Logic(FL) in a corrective frame machine. As an application to solid mechanics, this machine restores a plastically deformed structure to its original shape via the use of an innovative control strategy. Currently, these machines are operated manually by experienced operators, and the operation of these machines require extensive experience which is often not enough, and sometimes errors will occur, costing extra time and money. This problem arises from the wrong estimates for the corrective force vector (CFV) for a given deformed shape by the operator. In this work, it is shown that a Fuzzy-Neural Control System can produce excellent results that are comparable to those obtained by experienced operators, in which the NN and the FL system determine the CFV. The controller combines NN and FL Technology to achieve this goal. The NN uses the collected and accumulated data to provide the direction of the CFV. The FL uses a Constraint Boundary to constrain the increment of the magnitude of the CFV.; Instead of using a real automobile structure, a FEM model based on nonlinear analysis is used as the simulation tool to obtain the training data for NN Plant Model. With this NN plant model, the NN and FL control strategy are evaluated in terms of three criteria. These are the probability of success of a trial, the trial numbers, and the training time with several system parameters. A study of the effects of the various; parameters on our goal is performed in this research. From this evaluation, we suggest the optimized range of the used parameters and we can make sure that this control strategy can be applied to any deformed shape by taking the parameter within this optimized range. Also we show that there is a substantial reduction in operating time. |
