Structural Damage Detection Based On Flexibility Curvature Difference And BP Neural Network

Structural damage detection is an effective technique to detect structural health,preventing sudden collapse of structures,avoiding or reducing casualties and heavy economic losses,it is of great scientific significance.Due to the natural environment,load action and material aging,minor damage in the structure material evolves into major damage over time,which is likely to lead to structural collapse and catastrophic consequences.The identification of structural damage is to detect the potential danger of the structure before the accident happens,so as to adopt appropriate plan to avoid disaster.Among the methods based on changes in structural mechanical properties,the flexibility method and the modal strain energy method are two good methods,which have high sensitivity to local damage and easy to be constructed.In this paper these two methods are improved,one is based on changes in structural mechanical properties,a new Flexibility Curvature Difference(FCD)method is proposed;the other is based on the signal processing method,the Modal Strain Energy Difference is used as the input of the BP Neural Network for training the network,then a new set of data is inputted to the trained network to classify damage.In this paper,a new method of comprehensive average flexibility curvature difference is proposed.Firstly,the modal shapes of three-dimensional structure in four different directions are taken,from which four different flexibility curvature differences are obtained,the absolute values of these four indicators are then averaged to obtain a new indicator.The numerical simulation results show that there are interference elements when the four indicators are taken.The new comprehensive average indicator can accurately identify the damage location.The new method is better than the general flexibility curvature difference method.Another method is to use the modal strain energy difference(the first-order mode)as the input of the BP neural network.By finite element numerical simulation analysis of the space steel frame,a total of nine networks have been trained and tested.The statistical analysis of the test results shows that the single damage indicator is used as the input of the BP neural network,and the single hidden layer is used,the trained network can make accurate damage prediction for a new set of data.This method has remarkable advantages over only using modal strain energy.It can identify the location and extent of damage at the same time,and the accuracy is greatly improved.