The Structure Defect Determination Of Large-Span Steel Structure

With the improvement of people's request to the function and capacity of stadium, Construction of large span steel structure with its advantages of speed is widely used in stadium,terminal,exhibition center and other large public buildings by engineers.But it is easily damaged becaused of defect existed in pole structure.Once damaged,it is difficult to find this pole damage and repair them in time.At present,the method which is used to test and repair damage poles of structure is to investigate them one by one.It wastes so much time and money when used those methods.Based on this,in recent years,many scholars researched this issue and found some new methods,such as wavelet analysis,frequency domain method,neural network,genetic algorithm to solve this question.Among them,the neural network damage detection method has the most potential.Artificial neural network is used to make a damage detection to a stadium,locate structural damage and determine damage degree with the dynamic characteristic difference between a prototype structure and a damage structure.The main research contents include:(1) Structural damage identification is divided into three steps under the modal parameters of structure before and after damage which got from ANSYS.The three steps include the initial positioning of damage identification,the specific positioning of damage identification and damage identification.And analysis is done to compare the effect of different networks during identification process.(2) When using neural network to make damage identification,network training which has a large sample size and vector dimensions leads to the lower accuracy of damage identification.The sub-structure is introduced to make the damage identification of large-scale complex structure into possible.The principle is that the damage sub-structure is subdivided into sub-structure so that the scope of damage is narrowed and the precision of neural network is improved.(3)The neural network is affected by hih-frequency and mode displacement.Now those factors are pulled in different neural networks(BP,RBF,PNN) to find them how affect neural network capacity.The conclusion provided a reference to the future research at improving network accuracy,anti-interference ability and the generalization ability of the network.