Study On Damage Identification Of Steel Truss Bridges Based On Genetic Algorithms Optimizing Artificial Neural Networks

Structural health monitoring technology is in the period of vigorous development, and it will always be a very active issue in the field of civil engineering for a long time. Exact and effective methods of parameters identification is one of the key issues and the ultimate representation of structure health monitoring.Discovering the damage of structure immediately,and timely detection and repair,not only can greatly reduce the cost of the maintenance and repair,but also can avoid or reduce unnecessary loss of life and property.All of these make the technology of damage identification become a hot spot of study in the field of engineering.Steel truss bridge is one of the widely used bridges.But the security of steel,truss bridge is worrying under the influence of long-term moving load,wind load,corrosion and so on.Research on faster and more efficient way to identify the damage of steel truss is particularly important.If damage of structure occurs,it will lead changes of structure dynamic behavior. Study on damage identification of structure based on dynamic properties is very popular both in the fields of academic:and engineering:Now the most dominating methods of damage identification are signature analysis approaches,model updating approaches, artificial neural networks approaches and wavelet analysis approaches,etc.This paper give a briefing on the existing damage identification methods firstly,and then introduce the new technology in damage identification,which combining the Genetic Algorithms (GA)and Artificial Neural Networks(ANN).GA is an effective global Search optimization algorithm,hardly any local optimal solution,and is highly robust.ANN has its advantages of nonlinear mapping ability,adaptive and self-learning ability, strong generalization ability and real-time computing ability,etc,but vulnerable to local optimal solution.Using GA to optimize the scalar elements of ANN,we can use the advantages of two algorithms better.The speed and precision will be improved,and it's helpful to the development of damage identification. BP network is a kind of widely used network.This paper uses a lot of words on the introduction of its structure,mechanism and training algorithms,etc,and achieved the combination of GA and ANN in Matlab.This method is successfully applied to identify one or two damage in steel truss structure,and the results are very satisfactory.The results showed that it not only reduce the BP network's initial error,but also improved the network's training speed after combining the GA.ANN gives full play to its mapping,generalization and interpolation ability,and is good at damage identification in various degrees of structure.This method can not only identify the damage at a single point,when there are multi-damage,it can also get a satisfactory result.