Application Of Neural Network And Genetic Algorithms In Structural Damage Identification

Because of various factors, damage, which will lead to serious consequence, will inevitably appear in civil engineering structures. Therefore, it is necessary for engineering structures to be monitored for health. In recent years, more and more structural health monitoring (SHM) system has been installed in the civil infrastructure structures such as bridges, long-span space structures and ocean platforms. However, as one of the key techniques, the damage identification (DI) has never been ultimately resolved. It confines the function of the SHM system:Therefore, the further research of damage identification (DI) method adapting the need of on-line SHM system has the important theoretical significance and engineering practical value.Essentially,the issue of structure damage identification can be regarded as that of pattern recognition.In this case,what needs to do is just to identify the relationship between the index and the state of structural damage.Specifically,pattern recognition is a method through which the pattern measured on structure could be matched by those existing figures stored in damage feature database which has already been gained by theoretical analysis. In contrast,the traditional pattern recognition theoretical could hardly trade many patterns in the field of civil engineering,such as combinatorial explosion of multiple injuries and pattern distortion caused by noise.ANN algorithms have excellent advantages in pattern recognition, and have been widely used by more and more researchers in the fields of structural damage identification.Combined with some approximate modal information,a procedure for damage detection of bridges is suggested based on Neural Network because of its stronger mode clustering ability.However, because BP algorithm is a gradient dropping method of searching for, there are inherent deficiencies unavoidably, such as converging slowly, apt to fall into extremely some snack of the error function, as to the space of larger searching, many peak values and little function can search for reaching the overall situation snack very much effectively.This article carried on the analysis to advantages and shortcoming the BP neural network and the genetic algorithm, then combine the genetic algorithm which is good at overall search with BP algorithm which has much strong local optimizing ability. According to the characteristic of the cross operator, variation operator and choice operator of GA that these operators search the overall solve with great probability in the whole variable space and converge fast and accurate near the solve, intergrading advantages of these tow characteristics, combining these tow characteristics organically, adjusting the weights of neural network by genetic algorithm and using these optimized parameters as the initial weight values of neural network.we can avoid falling into local extreme minimum value and improve convergent speed of the algorithm, then we can obtain the overall optimum solve to the question quickly.By adopting the optimized GA-BP network to process damage identification on a rectangular beam, establish ANN model, simulating verification has been processed, which is to predict the damage location and severity.It has been shown that a proper ANN model has a comparatively strong ability on damage identification.