BP Algorithm Of Learning Simultaneously And Its Application To The Damage Detection Of Bridge

The bridges will gradually wear down under the circumstances of the natural environment, such as the atmosphere erosion, the changes of the temperature and the humidity, and applied environment, such as the loading, the fatigue of the material and the structure. This is an irreversible process. Taking the consideration of the safety and the cost of the construction, we should get immediate information about the damage and make the corresponding decision which is very significant for the urban development. The article is associated with one of the projects called An intellectual decision technology for damage identification and maintenance of highway bridge in Beijing sponsored by Beijing natural scientific fund. A popular detection method in the area of the damage detection, the neural network method, is applied to study the identification of the damage in the bridges. The artificial neural network method, which is a sort of adaptive technique for mode identification, doesn't need the function of the damage identification. This method uses its own learning mechanism to create the domain of the decision. It can make best use of the samples by training each sample individually and obtaining the balanced weights with good convergence. The weights represent the nonlinear mapping relation in the network, i.e. the information about the damage of the structure, and realize the damage detection. For the shortcomings in the application of neural network method to the damage detection in the structures, a new BP (back-propagation) algorithm of which weight and threshold learning could learn simultaneously is developed. some uncertain factors in the modeling of the new BP network are investigated and suggestions and advices are proposed. The studies in this article are as follow. 1. Based on the artificial neural network, generate the BP neural network model for the damage detection of the bridges. 2. BP method with weight and threshold learning simultaneously is proposed which...