| Safety is one of the most important principles in the design and use of bridge structures.As the service life of a structure increases,due to factors such as material ageing,external environmental influences and heavy traffic,bridges can suffer from varying degrees of cumulative damage and may experience deterioration in local stiffness,which may affect the overall stability and may even lead to bridge collapse if the damage is severe,therefore,it is very important to identify damage to bridge structures in a timely and accurate manner.This paper proposes a new method for bridge damage identification based on structural modalities,information entropy and BP neural networks,with the following main research elements and conclusions:(1)Considering the dynamic consistency of structural modalities during damage identification,the modal strain energy is combined with frequency to derive the modal frequency strain energy change rate index(MFSEGI),modal frequency strain energy change index(MFSECI)and modal frequency strain energy basis index(MFSEBI),and numerical models of simply supported beams and continuous beams are established to validate the three modal frequency strain energy indexes in different The numerical models of simply supported beams and continuous beams were developed to verify the ability of the three modal frequency strain energy indices to locate single-point and multi-point damage at different levels of damage,and the advantages and disadvantages of each in the damage identification process were compared and discussed.(2)Introducing the non-linear information entropy theory,the modal frequency strain energy index(MFSEEGI)and modal frequency strain energy entropy standard deviation index(MFSEESDI)are combined with the modal frequency strain energy to analyse the damage identification capability and the resistance to interference in handling disordered information in finite element models of continuous girders and suspension line arch bridges.(3)Different levels of Gaussian white noise were set to simulate the influence of environmental noise on the damage identification process,and the noise immunity of the modal frequency strain energy index and information entropy index to identify damage was verified.(4)Two methods,curve fitting and BP neural network prediction,were used for the quantitative study of bridge damage.The sparrow search algorithm and chaotic Tent mapping sparrow search algorithm were used to optimise the BP neural network respectively,and the network structures of SSA-BP neural network and Tent-SSA-BP neural network were constructed to optimise the initial weights and thresholds,and the differences in the accuracy of the four methods for damage quantification of simply supported and continuous beams were discussed.The results of the above study show that,firstly,both the modal frequency strain energy and modal frequency strain energy entropy indicators can accurately and effectively identify the locations where damage occurs in simply supported beams,continuous beams and suspension line arches,with the MFSEESDI indicator having the highest sensitivity to damage and the best identification effect.The MFSEBI indicator is the least affected by the damage proximity effect,and can reflect the proportional relationship between the indicator and the degree of damage.Secondly,the accuracy of the modal frequency strain energy indicator damage recognition decreases under the interference of noise,while the information entropy indicator can still identify the location of damage occurrence under the interference of noise level within10%,and has better noise resistance.Finally,the curve fitting is better,but the fitting error is larger when minor damage occurs.The ability of the optimised BP neural network regression prediction was significantly improved,with the Tent-SSA-BP neural network having the best prediction quantification and the highest accuracy rate with universal applicability. |
PDF Full Text Request