Study On Damage Recognition Of Bridge Based On Support Vector Machine

Bridge structures are generally in a harsh environment,and will be affected by natural factors such as rainstorm,flood,gale,snowfall,temperature difference,freezing and thawing,earthquake,and suffered repeated impact,heavy load and even overload sometimes.Timely health detection and accurate damage recognition are the common topics at home and aborad as well as prevent the occurrence of major safety accidentsThis paper mainly studies the related content of bridge structure damage recognition,summarizes the main methods used in bridge damage recognition in recent years,such as static load damage detection,dynamic characteristic damage detection and signal processing damage detection,and focuses on frequency,modal change and curvature modal,flexibility or stiffness matrix and frequency response function in dynamic characteristic damage detection.At the same time,the related theories of Support Vector Classification（SVC）,Support Vector Regression（SVR）and Kernel Function are sorted out.And it is used in a more comprehensive and accurate multi-damage recognition of bridges.In multi-damage location recognition of bridges,aiming at the problem that the value of non-damage location is large when curvature modal difference is used to identify damage,the concept of curvature modal difference change rate is proposed,and the 1^st and the 2ⁿdd normalized curvature modal difference change rates are input into SVC for training.Taking Linxian Yellow River Bridge as an example,multi-position damage location recognition is carried out.The results show that the accuracy rate of the recognition is 99.68%in all locations,and only the non-damaged nodes near the damage locations have some misjudgments,so the damage recognition effect is good.Based on the recognition of multi-damage location of bridges,it is further proved from theoretical analysis that the change of curvature modal difference is（1-k）/k times proportional to the original stiffness,where k is the ratio of the damaged stiffness to the original stiffness,and the change of curvature modal difference of each location does not affect each other,but is only related to whether the damage occurs at this location.The relationship is validated by simply supported beam model.The ratio of the curvature modal difference to the basic value of Linxian Yellow River Bridge is used as the eigenvector to input into the support vector regression（SVR）to identify the damage degree of the bridge accurately.The difference of the damage degree recognition is basically controlled within 3%.According to the damage detection results of steel beams under artificial excitation method,the proposed SVC and SVR methods are used to discriminate the damage location and damage degree.The damage location is basically correct,and the damage degree identification is also less different from the actual situation,which proves the feasibility and correctness of the damage identification method proposed in this paper.