Damage Identification Method Based On Generalized Local Curvature Modal Information Entropy And BP Neural Network

Structural health monitoring is of great significance for ensuring the safe operation of structures.Damage identification,as the core of structural health monitoring,plays an important role in assessing the health of bridges.Research es on bridge damage identification are becoming more and more mature,but there are still problems such as difficult to obtain structure information and insufficient data mining.(1)For the problem that the curvature mode index is not sensitive to the damage at the vibration mode node,the error can not be accurately quantified by the curvature mode,and the data mining is insufficient in the field of damage identification,the generalized local curvature mode information is proposed.The entropy index is derived,and its related formulas such as local probability window are deduced.The two-step method for identification is proposed.(2)The finite element model of simply supported beam bridg e is established and the single and multiple damages are simulated.The damage location is identified by the first-order generalized local curvature modal information e ntropy.The training samples of BP neural network are constructed and the damage degree of various working conditions are determined.The Gaussian white noise was simulated by MATLAB and the noise immunity of the index was studied.The simpl y supported beam model elements are refined and the relationship between the number of measuring points and the recognition effect of the damage indicators is studied.(3)In order to verify the practicability of the index,the modal test was carried out on a simply supported steel beam;the reference measuring points at different positions were set according to the characteristics of the first two modes,Multiple sets of adjacent position sensors are placed in turn and normalized by the reference point to obtain the true mode shape of the structure.The test results are verified by the acceleration waveform-spectral map and the natural frequency.The S-G filtering method is used to preprocess the measured vibration mode,and then the damage identification is studied by using the first-order generalized local curvature mode information entropy,and the curvature mode recognition result is compared.The finite element model of the test object was established by using MIDAS/Civil and ANSYS and the simulation and analysis were carried out according to the actual damage conditions.The BP neural network training samples were constructed and the damage was quantified.The effect of the indicators in the actual structural damage identification was studied.(4)The index is improved for mathematical properties of the index,and the improved second-order generalized local curvature modal information entropy is used to identify the damage of the single-point damage condition of the simply supported beam.The noise immunity of second-order generalized local curvature modal information entropy is studied by introducing Gaussian white noise.A three-span continuous beam bridge model was built by using MIDAS/Civil and the damage was simulated.Taking the midspan as the study object,the first-order generalized local curvature modal information entropy is used to identify the damage.The results show that the first-order generalized local curvature modal information entropy can better locate single-point and multi-point damage,and combined with BP neural network can accurately quantify the damage.In the identification of damage near the vibration mode node,the index proposed in this paper is better than the curvature mode.The noise imunity of the index is limited but it can obtain better damage recognition effect without intensive measuremen t points.In the test,the necessity of setting different reference points for the first two modes and the feasibility of the test method were verified.The first-order generalized local curvature modal information entropy can identify the damage location of the steel beam.The damage is quantified by the BP neural network,and the damage recognition effect is close to the vibration mode node.The improved second-order generalized local curvature modal information entropy is similar to the first-order index and can be used for damage localization.Combined with BP neural network,the damage can be quantified and the recognition accuracy of the vibration mode node is better than the curvature mode.In the continuous beam bridge,combined with the BP neural network,the first-order generalized local curvature modal information entropy can be used for damage assessment,and the damage identification near the vibration mode node is better.