Research On Structural Local Damage Characterization Based On Strain Data Under The Influence Of Uncertain Factors

Acquiring structural modal,displacement and strain response data through structural health monitoring system,and then analyzing structural behavior,identifying structural damage and assessing structural performance have become an effective way to ensure the safe operation of engineering structures.Compared with the modal data,the strain measurement data contain a large number of useful information in the non-resonant frequency band;compared with the displacement measurement data(acceleration,velocity,displacement,etc.),the strain measurement data relate to the local behavior of the structure directly and reflect the stress redistribution in the vicinity of the sensor.It is more suitable to characterize the local damage of the structure.However,in practice,measurement data interfere with uncertain factors such asmeasurement errors and environment and operational loads,which may lead to damage misjudgement;existing engineering structures need to be monitored and identified urgently,in which uncertainties of damage quantification is exacerbated by modeling errors caused by the lack of a suitable benchmark model;many damage identification methods that are not strictly considered are no longer applicable,and the advantages of local damage characterization based on strain measurement data are not fully made use of.Therefore,it is necessary to develop more robust and efficient processing and evaluation approaches for monitoring system-oriented measurement data.This paper focuses on the study of theory and method of local damage characterization of structure based on strain measurement data under the influence of uncertain factors.On the basis of evaluating the serviceability of existing methods,novel damage early-warning,elemental updating and data fusion methods based on strain data are proposed to solve the problems encountered under the influence of uncertain factors.The details are as follows:(1)According to three typical structural damage identification methods based on strain measurement data(direct damage localization,correlation modeling and sensitivity-based model updating),case studies are performaned to evaluate the serviceability of the methods under the influence of uncertainty factors.Firstly,the effectiveness of direct damage localization by utilizing three damage-sensitive features,static strain,modal strain and wavelet packet energy strain,which can be extracted from the strain measurement data,is evaluated,and it is clear that this method is only suitable for simple girder structures,and the influence of time-varying loads is considered urgently.Then,the effect of early-warning based on the correlation modeling between long-term strain data and temperature data is investigated by using the joint damage case of truss girder bridge,and it is found that due to the complexity of time-varying load mechanism and incomplete measurement information,it is easy to cause over-fitting of correlation model,which makes it difficult to identify early damage by this way.Finally,the measurement and modeling error interferences encountered in the structural model updating based on short-term strain data are discussed by numerical examples of riser model structure.The recursive estimation method and regularization method are deemed to have significant advantages over the traditional gradient method.However,when the initial model prediction is biased,the existing methods still produce great damage misjudgment.(2)Two types of structural early-warning and localization methods based on long-term strain monitoring data are proposed.Time-varying factors,such as environment and operating load,seriously interfere with monitoring data,while the correlation modeling method relying on time-varying load observation is difficult to identify local minor damage due to its complex mechanism and incomplete data.In this study,it is proposed to use the correlation changes of data sets composed of data of different strain measurement points to characterize the tiny anomalies of one strain measurement point data,that is,extracting local damage features which are very sensitive to damage and insensitive to time-varying load effects,and forming the damage early-warning and location algorithms.Principal component analysis and robust regression analysis are utilized to characterize the correlation between data of different strain measuring points.Specifically,the former extracts the relative entropy value of the second principal component score matrix corresponding to each strain measuring point as the local damage feature based on the moving principal component analysis,and adopts the Grubbs test method for early-warning.The latter is based on the moving robust regression analysis.The residual norm of robust regression is directly extracted from the data between each strain measuring point and its remaining measuring points as the local damage feature,and the early-warning is carried out by the control chart method.Two numerical models of simplified truss girder bridge and high-speed railway continuous box girder and steel truss test model are used as the research objects,and the robustness and computational efficiency of the two damage early-warning and localization algorithms are fully validated.(3)A type of structural elemental updating method based on short-time strain test data is proposed.Model uncertainty,especially modeling error,seriously interferes with the iterative process of the model updating,and the excessive dimension of iterative variables will also cause several calculation problems.In this strudy,an approach to update local element stiffness by using local damage information of a structure is developed:the transmissibility concept,with low-band response estimation results that are independent of structural mass and damping,is used to estimate strain response;wavelet packet energy strain is extracted from the estimated and measured strain response of target element(strain measurement element);the residual norm of both is taken as the objective function and the element stiffness is updated by swarm intelligence algorithm to obtain the damage quantification index of target element.Then,an elemental updating method based on strain response estimation and swarm intelligence algorithm is proposed.The method can make the physical parameters unrelated to local damage not participate in the elemental updating process to reduce the influence of modeling errors,and descend the dimension of the iterative variables to one to improve the convergence efficiency.Two numerical models of riser and truss substructure and simply-supported steel beam test model are used as the research objects,and the robustness and computational efficiency of the proposed algorithm are fully validated.(4)Two types of structural damage identification methods integrated by strain and displacement data are proposed.Sensors sensitive to local damage in the monitoring system are distributed in several vulnerable parts.Due to the influence of uncertain factors,it is very difficult to identify the damage at a certain measuring point only through the strain data of the measuring point.In this study,given that the complementarity of strain and displacement measurement data,a local damage identification method based on strain-displacement measuring point pair is presented on the basis of moving robust regression analysis.By this approach,the feature fusion data which have damage sensitivity only to the position of the strain measuring point and insensitive to time-varying load effect is extracted.In addition,on the basis of estimating the strain response of the target element by using the response data of the displacement measuring points arranged optimally,a type of elemental updating method combining strain and displacement measurement is proposed,which can quantify the damage of a strain measurement element independently.Numerical and experimental studies are used to verify the effectiveness and robustness of the above two damage identification approaches which fuse strain and displacement data under the influence of uncertain factors.At the end of this paper,associated with the local damage characterization method based on strain data mentioned above,some suggestions are put forward to build an index system for local damage identification of engineering structures.