Model Updating with a Neural Network Method Based on Uniform Design

In order to effectively update structural models by the neural network method based on uniform design, this paper takes beam structures as objects to study three problems that are the completeness of uniform design sampling scheme, the method of extending the uniform design sampling scheme and how to choose excellent neural network samples. The difference between the uniform design sampling scheme and the complete sampling scheme is analyzed. An idea of factor rearranging is proposed to extend the uniform design sampling scheme. A fuzzy analytical hierarchy process is used to evaluate and choose excellent neural network samples. The results show that the uniform design sampling scheme is an approximate complete sampling scheme and the idea of factor rearranging can extend the uniform design sampling scheme effectively. In the process of model updating of large complicated structures, the uniform design sampling scheme should be extended by the way of factor rearranging to improve the accuracy of model updating and physical variables with definite physical significance should be used as the samples of neural networks.