| The effect of other external excitation such as ground motion will lead to different degrees of damage to structural members,which will affect the use of the structure.Therefore,it is necessary to find an appropriate damage location identification method to identify the location of structural member damage;after the structure is damaged,the structural response will change.When the response exceeds the limit,it will have a certain impact on the use of the structure.If the response of the damaged structure is predicted,the response value of the structure at the next time can be roughly known,and then the structure can be controlled.In this paper,taking the six story frame structure model as an example,the structural damage is simulated by the reduction of elastic modulus of beam and column members,and the structural damage location is identified and the response of the damaged structure is predicted.In the aspect of damage location identification,white noise with amplitude of 0.1 as the external excitation input of various damage conditions.The time history records of node acceleration are extracted and use the EMD method to decompose it to get the first four order IMF components,and the sample entropy corresponding to each IMF component is calculated to construct the characteristic quantity of damage identification.The characteristic quantity of damage is divided into the original characteristic quantity of damage and the Z-score standardized characteristic quantity of damage.In this paper,the SVM classifier is optimized for Bayesian,and the inverse distance square weighted KNN classifier is selected to classify and identify the damage location and make a comparison,and four types of damage location recognition are discussed.When the original characteristic quantity of damage is classified and identified by inverse distance square weighted KNN method,the accuracy of the first to fourth types of damage location classification and recognition is 100%,98.8%,91.5% and98.2%,respectively;the accuracy of the first to fourth types of damage location classification and recognition is 100%,98.8%,88.6% and 98.4% respectively for the characteristic quantity of damage after Z-score standardization.whether the damage is standardized or not,the weighted KNN classifier performs better than the optimized SVM classifier in recognition.The Z-score standardization of the damage feature reduces the recognition accuracy of the optimized SVM classifier,but has little effect on the weighted KNN classifier;Moreover,compared with the method of identifying the damage location by structure frequency,stiffness change,and so on,the process is easier when using the weighted KNN classifier to identify the damage location.In the aspect of structural response prediction after damage,according to the noise characteristics of the original seismic wave,the lifting wavelet transform is selected to denoise the original seismic wave,and the denoised RSNN139 seismic wave sequence is input as the external excitation of the structure.After comparison,the main characteristics of structural response before and after seismic wave denoising are successfully retained.The time history records after denoising of displacement,velocity and acceleration of nodes 51,47 and 43 are collected for response prediction.In the single-step prediction of NAR neural network,the prediction results are ideal,but the response prediction effect of the N43 node is the worst.Then the CEEMD-NAR method is used to predict the response of N43 node,in view of the“endpoint effect”caused by CEEMD decomposition,an extreme value symmetric continuation method is selected to solve this phenomenon.Except that the MAPE value of displacement response increases slightly to 0.1378,other evaluation index values of the responses were reduced.The above results show that NAR neural network and CEEMD-NAR method can make the prediction accuracy meet the requirements,but CEEMD-NAR method has better prediction accuracy for complex sequences. |
PDF Full Text Request