Tolerance optimization based on neural network and genetic algorithm

Taking into account the highly nonlinear relationship between the cost and the tolerance in the product manufacturing process, a method based on the neural network in conjunction with the genetic algorithm was proposed to solve the tolerance optimal issues. The method integrates the advantage of the genetic algorithm, which can obtain the optimal result in a largescale solution space using the probability searching strategy and the strong robustness, and the superiority of the neural network that can solve the highly nonlinear problem. In the optimization process, the neural network was trained using sample data to simulate the tolerancecost function at first and get a function relationship with the black box feature between the cost and the tolerance. And then the genetic algorithm was introduced to optimize the tolerance allocation by taking the results of the trained neural network. It takes the functional requirements and the standard tolerance grades as constraints as well as the minimum of the component cost as the objective. A tolerance optimization system based on a C+[KG-*3]+ library and the Matlab was designed. Finally, an example of the latching shaft and hook mechanism component of the aircraft cargo door demonstrated the method. The analytical result proves that the new method can produce the tolerance optimization economically and accurately, and has an advantage over traditional methods.