Support Vector Machine Method Based On SRV For Damage Identification

Damage detection has been an important topic in structural health monitoring for a long time,and has drawn much attention all over the world.How to develop effective damage identification method still remains a challege.Because of the complex environment conditions,difficulty in establishment of accurate finite element models,error of measurement data,the lack of integrity as well as sensitive damage indicators,there are no common methods applicable to all cases,or at least a broad range of cases for structural damage identification.Therefore,development of new structural damage identification methods is of great significance.The structural response vector(SRV)contains enough information of structural state to reflect the structural damage state more accurately,and thereby exhibits a great perspective in playing a great role in damage identification.This dissertation is foucused on the development of novel damage indentification methods based on SRV.The studies that has been completed are detailed below.(1)According to the definition of SRV,the symmetry property of the octonion structural response vector(OSRV)was studied.The normal direction of the symmetry plane can be solved by the directly constructed SRV,so as to solve the problem that the phase trajectory was indistinguishable for damage located in symmetric positons.(2)By using the relative variation of responses as a normalization way,a new definition of SRV component was proposed.The results showed that a greater difference in phase trajectories between damges of different loacations was obtained with the new SRV,showing a better ability to identify the damage.A new OSRV phase plane method was used to locate and quantify the single damage problem with a simply-supported beam model.The results showed that,compared with the previous SRV,the newly-defined SRV has higher accuracy in damage location and quantitative identification,and also showed better performance of robustness in the presence of noise.(3)There exsit some shortcomings for the phase plane method,becase it reduces the data dimensionality with a loss of information,and is also insensitive to the structure response type.To avoid these shotcomings,the support vector mechine(SVM)method was used to locate and quantify damage with a beam structure.The influence of SRV component quantity and component type on damage detection effectiveness and computational efficiency was studied.The results showed that a greater number of SRV components does not necessarily render a better detection performance,and the selection of the appropriate number of structural responses can not only improve the recognition effectiveness,but also reduce the calculation time.It was also found that a combination of structural dynamic and static responses for SRV leads to better recognition effectiveness and calculation efficiency as compared with the method involving merely static responses.(4)By using SRV composed of dynamic and static responses as input,damage identification of truss-like structure was carried out by SVM.The recognition effectiveness was found similar to that of the previous methods.Nervertheless,as compared with the common methods involving the modal shape and high order frequencies,it is easier to measure the responses in SRV with better accuracy,.Thus,the SRV-based damage identification method has a higher feasibility for engineering applications.Considering the effect of measurement error on the recognition effectiveness,the results showed that the method is still applicable to identify damage at typical locations of truss bridges.Finally,the SRV of different damage cases were compared to find the reason that this method was more sensitive to measurement error.