| In view of the existence of many small damage in the actual engineering difficult to identify the situation,taking steel pipe structure as an example,the application ability and mode of BP neural network in structural damage identification were studied.It included structural modal test,finite element model updating and neural network method.It provided a method for the identification and judgment of damage condition of actual structure by using neural network tool in practical engineering.The main research contents are as follows:(1)Modal test of pipe system without damage was carried out by acceleration sensor.Several groups of control tests were set up to calculate the sensitivity of each measuring point to structural vibration.The most sensitive points were selected as the measurement points of the non-destructive modal test and the subsequent modal test under the lossy condition.The final result of the modal test under nondestructive conditions was obtained.It provided a basis for updating finite element model.(2)Based on the experimental pipeline model,the finite element model of the pipeline system was established by using ABAQUS software.According to the modal test results,the finite element model was updated and modified to make the finite element model and the test model have enough similarity,which can accurately reflect the structural modal under the condition of lossless and the structural response under the condition of lossy.(3)BP neural network was used to identify the damage of the modified finite element model.The damage location based on natural frequency,damage degree identification based on natural frequency,damage location based on modal shape and damage degree identification based on modal shape were carried out by using neural network,ABAQUS secondary development and adding Gaussian white noise.Appropriate network training methods were selected according to the sensitivity of each modal parameter in the change of damage condition.(4)Modal acquisition tests were carried out on the test pipes under different damage conditions.Based on the experience of finite element model damage identification and training network method,the damage location and damage degree identification of the experimental pipeline structure were carried out.The ability of BP neural network to identify the damage of pipeline model depends on whether the selected training and test sample parameters are sensitive enough to the change of damage condition.For the parameters with weak sensitivity,the optimization calculation formula established in the damage identification process of finite element model can be used to optimize the data processing,so as to achieve the ability to effectively identify the damage condition. |
