Research On Damage Detection Method Of Grid Structure Based On EMD And Neural Network Method

With the continuous development of the domestic economy,various types of large-scale engineering construction projects represented by large-span grid structures continue to emerge throughout the country,especially in some large-scale industrial plants,large-scale equipment maintenance workshops and other industrial projects.As well as in civil projects such as large shopping centers,convention centers and libraries,the use of space grid structures has become increasingly widespread.During the use of the structures,under the joint action of internal incentives such as equipment operation and external incentives like wind,snow and earthquakes,the structure may produce some certain failures which we cannot observe with naked eyes such as bolt looseness,rod bending,and even partial micro-fractures.If these damages cannot be discovered in time and measures to be taken,they will accumulate gradually over time and suddenly erupt at some certain point,causing partial and overall collapse of the structure,threatening the people's life and property security.Therefore,it is a powerful measure to eliminate structural safety hazards by periodically performe damage identification on structures under working conditions,discover and repair the defects in time.Based on the current research results in the field,this paper proposes a method for damage identification of large-span network structures based on the combination of radial-basis function neural network(RBFNN)and empirical mode decomposition(EMD).Firstly,using the EMD method to decompose the vibration response data of the rod end node into a series of IMFs(Intrinsic Mode Functions).For the structure's original response data,after using the EMD to decompose it,the damage information and other useful contents will be distributed into various IMFs.Due to the scale of the signal containing in the IMFs is relatively simple,The damage information is difficult to be overwhelmed by interference,so the damage information in the IMFs terned to be very obvious.Using neural network's excellent feature of linear mapping,put IMFs as an input into the neural network,and compared the output with the ideal output,the damage status of the structure can be known.Meanwhile,put the original response signal into the network,and compare the results with the previous method's to reflect the superiority of the combination of EMD and RBFNN.In this paper,the numerical simulation and experimental analysis are based on the structural lab's model of Beijing University of Civil Engineering and Architecture.To simulate the random excitation of the real structure under actual working conditions,this paper uses a trolly to drive on the upper chord of the structure,and at the same time obtain the vibration response data at the joint,and use it as the damage eigenvalues.In order to improve the accuracy of the numerical simulation,the structural members are divided into three types: the upper chord,the lower chord,and the abdominal bar.At the same time,three types of damage modes such as single-bar damage,multi-bar damage and different degrees of damage were designed.In the single-damage condition,three different rods were selected for each type of the rod to identify;in the multi-damage condition,six types of damage conditions were combined: the same kind combination and the different kind combination;In degree recognition conditions,three different types of rods were selected to construct 30%,50%,and 70% of three different degrees of damage.The results using the IMFs and original data as input are comparatively analyzed in the final presentation.In the experimental stage,the steel trolley was used as the excitation source.Two steel rails were welded on the upper chord of the grid as the track for the trolley to artificially hauled reciprocate on them.Due to the limitation of the experiment,only single-rod and double-rod damage were constructed.Acceleration sensors arranged on the nodes are used to obtain vibration response data,and the different results are compared between two types of network input.Through the comparative analysis between the two stages of numerical simulation and experimental analysis,we can see that the damage identification method using EMD combined with RBFNN can be more accurately identify the damage status under various working conditions.Meanwhile,by comparing two types of input data in a neural network,it can be known that a neural network based on IMFs has higher accuracy than a network based on original data in damage identification of a long-span grid structure.Therefore,the method proposed in this paper has certain engineering practice significance.