Damage Identification Research Of Concrete Structure Based On Feature Extraction And Pattern Recognition

Concrete material has been widely used in Bridges,buildings and other civil engineering due to its strong plasticity,wide source of raw materials,strong adjustable performance and high strength and durability.However,due to the long-term impact of environmental erosion,external load,construction defects and other factors,concrete structures will appear damage during service.The accumulation of damage will aggravate the deterioration of structural performance,and even lead to structural collapse in serious cases.Therefore,this paper takes prestressed concrete beams as the research object and focuses on the damage development law of prestressed concrete members,mainly including the following contents:(1)The damage test of prestressed concrete member based on piezoconductance wave is carried out.Three pretensioning prestressed concrete beams were designed and manufactured.By collecting intelligent aggregate acquisition signals at all levels of damage,the differences of time domain and frequency domain signals and wavelet packet energy in health state and damage state at all levels were studied,which laid a foundation for qualitative analysis of the damage level of concrete structures in the following paper.The results show that piezoelectric intelligent aggregate has a good sensitivity to the damage detection of concrete structures.The whole and local damage of concrete structure can be judged by the signal changes received by SA sensor.(2)Guided wave signals are analyzed by time-frequency analysis,wavelet packet transform and singular value decomposition respectively in order to conduct qualitative analysis and feature extraction of concrete structure damage state.Using the measured guided wave signals of piezoelectric sensors attached to the surface of the structure,the time-frequency matrix under different damage states was constructed based on the multi-resolution characteristics of the signals analyzed by WPT,and the singular value decomposition was carried out to excavate the essential time-frequency characteristics of guided wave signals after the cracking damage of the structure.The results show that the damage of concrete structure will cause the amplitude of SA sensor to decrease.The singular value vector constructed by combining WPT and S VD is an effective characteristic parameter of concrete structure damage.With the increase of load,the singular value of time-frequency matrix decreases gradually.The normalized singular value vector distance has a three-stage correspondence with the damage condition.(3)A damage identification method for concrete structures based on pattern recognition is proposed,and the nonlinear relationship between signal characteristics and damage is constructed.BP neural network,GABP neural network and support vector machine(SVM)are used for pattern recognition,which are characterized by singular value vectors of time-frequency matrix.The results show that the coupling of WPT and SVD can extract the damage characteristics of the piezoconductance wave signals and reflect the time-frequency matrix under multi-resolution.The singular value vector is used as the guided wave damage characteristics to establish a machine learning model to realize the damage identification of in-service concrete structures.GABP pattern recognition method can stably identify the damage location and degree of concrete structures.Compared with BP,GABP can better fit the correlation between damage signal characteristics and damage conditions,and genetic algorithm can repair the local optimal problems caused by the randomness of network structural parameters.SVM classification model can accurately identify the damage location and damage degree of concrete structures.Compared with the neural network model,it has better sensitivity in the classification of nonlinear data.