Reconstructed Electromechanical Admittance Signature-based Structural Damage Identification

Piezoelectric ceramic intelligent actuator / sensor based structural damage diagnosis technique has been widely utilized in structural local fine identification and monitoring,depending on its high sensitivity to minor damage,simultaneously served as actuator and sensor,and strong immunity to far-field conditions.However,existed scientific issues such as transmission,storage and processing of large amounts of admittance signatures,packet loss during data transmission,precise damage quantification and temperature effects on damage identification via admittance signature variation need to be alleviated when facilitating electromechanical impedance technique into practical application.Therefore,reconstructed electromechanical admittance signature-based structural damage identification is proposed in this thesis,the main contributions are showed as below:(1)Integrated compressive sampling based admittance data compression technique with an orthogonal matching pursuit algorithm based admittance data reconstruction approach is proposed for structural damage identification,aimed at transmission,storage and processing of large amounts of admittance signatures and data redundancy caused by repeated measurements,multi-sensor placement,multi-case arrangement.Different from conventional electromechanical impedance technique,compressed data obtained via a linear transform based on compressive sampling method is transferred and stored into the base station as a substitute of original admittance signature,with admittance signature reconstruction accomplished on the basis of orthogonal matching pursuit algorithm,the reconstructed signatures are utilized to implement damage identification for the target structure.The proposed approach is applied to a simply-supported steel beam damage identification experimental research,optimal sparsity level interval is obtained via the analysis of total error between reconstructed conductance signature and original conductance signature versus different sparsity levels.According to the experimental results,a statistical index(i.e.,root mean square deviation)calculated via reconstructed conductance signature is validated to be feasible in damage identification.In addition,suggestions for compression ratio range are given based on the analysis of reconstruction effects with different measurement numbers and various measurement matrices.(2)A compressed sensing theory based lossy admittance signature recovery approach for structural damage identification is proposed,aimed at data loss issue during the admittance collection and transmission stage.Both experimental results of a standardized concrete cube subjected to varied temperatures and a full-size concrete shield tunnel segment undergone bolt-loosened defects are utilized to validate the feasibility of the proposed approach.Two kinds of pseudo random loss patterns,i.e.,single-consecutive-segment loss and multipleconsecutive-segment loss,are arranged to verify the robustness for admittance flow recovery.Both of the experimental results show that environmental temperature recognition and structural damage identification can be achieved via three damage index,i.e.,amplitude variations of resonance peak,frequency shifts of resonance peak and root mean square deviation.(3)A deep feature learning approach for reconstructed admittance signature based structural damage identification is proposed,aimed at piezoelectric impedance technique based precise damage quantification.Convolutional neural network is introduced to conduct feature learning for reconstructed admittance signatures,and concrete structural precise damage quantification under varied-temperature environment is achieved.Merely two groups of original admittance signatures need to be collected corresponding to each damage scenario on the basis of the proposed approach,one is utilized to generate training set,the other one is utilized to generate testing set.A series of crack detection crossover tests for standardized concrete cube subjected to varied temperatures are conducted for method validation.Enough available admittance signatures are generated based on a variation of orthogonal matching pursuit algorithm,so as to establish data library for CNN training and testing.Data library is particularly divided into several subsets according to the experimental scheme,i.e.,the same crack depth subjected to varied temperatures and different crack depths subjected to the same temperature.Totally two CNN model structures are constructed for pattern recognition of subsets aimed at temperature recognition and subsets aimed at damage quantification respectively.Testing the CNN models within training completion by training sets demonstrated that experimental results turn out to be of high accuracy,and identification effects of the subsets aimed at temperature recognition performed better than those aimed at damage quantification.