Study On Acoustic Emission Identification Of Interface Slip Failure Of Concrete Filled Steel Tube Based On Improved Variational Modal Decomposition Method

Structural health monitoring based on acoustic emission is an important technology for structural health monitoring,However,how to accurately identify the damage state of the structure from the massive signals generated by the Structural health monitoring system is of great significance to the safety assessment of the structure In recent years,research on the damage mechanism of concrete-filled steel tube(CFST)based on AE signals has been widely conducted,providing an important theoretical basis for the intelligent monitoring of CFST components.This study provides a novel method for health monitoring of CFST Structural health monitoring,which combines feature selection and cluster analysis with the particle swarm optimization(PSO)optimized variational mode decomposition(VMD)method to accurately identify the structural damage from massive acoustic emission signals.This article successfully identified the patterns of slip damage and effectively improved signal quality.In addition,by constructing a slip damage classification model based on Long Short Memory(LSTM)neural network,the accuracy of damage classification reached 90.6 %,and the intelligent identification and classification of CFST were realized.The main contributions are showed as below:(1)In the aspect of damage identification of CFST structures based on acoustic emission signals: According to the characteristics of acoustic emission signals,feature selection and cluster analysis methods are adopted to analyze the acoustic emission signals from the pushout tests of CFST.The results show that there are mainly three damage modes during the sliding failure process of steel-concrete interfaces,namely,concrete damage,steel pipe damage,and steel-concrete interface damage.Among them,the peak frequency of steel pipe damage is the highest,mainly distributed around 300 k Hz,while the frequency differences between the other two damages are relatively small,but they do not affect the overall feature differences of the signals.(2)In the aspect of denoising and preprocessing of acoustic emission signals in CFST structures: To address the issue that acoustic emission signals are easily affected by noise interference and have complex components,and the traditional VMD decomposition parameter selection heavily relies on human experience,a VMD denoising method based on Particle Swarm Optimization(PSO)is proposed.Experimental results show that compared with traditional VMD,Empirical Mode Decomposition(EMD),and Empirical Wavelet Transform(EWT)algorithms,the PSO-VMD method can better improve the signal quality and has better anti-aliasing performance.(3)In the aspect of real-time classification of acoustic emission signals in CFST structures: A intelligent classification model based on LSTM neural networks is constructed,using acoustic emission signals preprocessed by the PSO-VMD method as the input of the model,and comparing the effects of signal preprocessing methods and different model hyperparameters on the recognition accuracy.The results show that the preprocessing can effectively improve the differences between different damage signals and the recognition accuracy of the model.After adjusting the model parameters,the recognition accuracy of the model reaches 90.6%.