Methodological Study On Concrete Structural Damage Intelligent Indentification

Artificial Neural Network (ANN) and Genetic Algorithm (GA) are two important branches of Artificial Intelligence. From the point of the combination of both NN and GA, to research the intelligent identification technology for concrete structures is conducive to the practical application of engineering. With the background of intelligent structure's damage identification, the contemporary status and trend on the research of intelligent structure's damage identification are dissertated, and furthermore the possible methods on how to identify the structure's damage are researched with the combination of NN and GA.The main contents are following: (1) From the methodology of NN, the basic theories on NN are discussed, with the emphasis on the mechanism of NN and its study rules. Then the BP NN is concerned, because it is the comparatively most widely used NN type in the field of structure's damage identification; (2) From the angle of system identification theories, the method and process of establishing the identification system model are demonstrated. Also, how to implement a structural damage identification system of concrete architecture, based on BP neural network is demonstrated in the thesis; (3) In order to find a more effective training algorithm of global approach, the way of optimizing network's weights using GA is demonstrated. The experiment shows that extensive mapping capacity of NN and rapid global convergence of GA can be obtained at the same time by combining GA and NN in some certain ways, which are researched in the thesis. And this method of combination is a relatively effective way to solve the problem of concrete structural damage identification; (4)Based on Java language, the software system used to identify the structural damage is designed and implemented. The system consists of graphic user's interface module, chief control module, samples-processing module, modal-recognition module, task-modal module. The processing is like following : Firstly, the neural network modal is set up, then the sample data are analyzed and selected. Next, the data are pre-processed to set up the neural network(NN) modal. Then, the frequency vectors are put into the NN as the input data, however, the damage position and the damage degree are treated as the desired output. The NN will be trained with the learning algorithm based on the GA till it converges. Finally, after the training, the testing on the effectiveness of the NN is carried on with the real acquired data from practice. The results can be used as the reference of judge on the status of damage.The research mentioned above shows that : the system of structural damage identification based on the combination of NN and GA has acquired relatively high accuracy and rapid convergence. As concerned as the ban with a singledamage, the beam with one, two and three damages, the system works comparatively effective. Meanwhile, the research mentioned above also shows that method based on the combination of NN and GA ,which is used to identify the structural damage, is valuable and worthy of further concern.