Research On Digital Twin Temperature Model Of Steel-concrete Composite Box Girder Of Cable-stayed Bridge And Its Application For Health Monitoring

At present,China is at an important moment from a large bridge country to a powerful bridge country.Under the goal of "By 2025,the infrastructure of the main framework of the national comprehensive three-dimensional transportation network has been digitalized in all elements and in all cycles." established in The 14th Five-year Plan for Digital Transportation Development of the Ministry of Transport,how to realize the interaction and integration of the physical world transportation infrastructure entity,the bridge,and the information world,has become one of the core bottlenecks in practicing the digital concept in the field of bridge health monitoring and intelligent operation and maintenance.As an important enabling technology to realize digitization,digital twin is an important way to realize digital transformation and intelligent upgrading of bridge health monitoring and intelligent operation and maintenance,however it is still in the early stage from theoretical research to practical application.How to construct a digital twin model of "precision,standardization,lightweight and visualization" and "interactive,fusible,reconfigurable and evolutionary" to realize the digital representation of the whole life cycle of the bridge remains to be further studied.Based on the temperature monitoring of steel-concrete composite box girder of a long-span cable-stayed bridge and the digital twin modeling theory of"construction-assembly-verification-fusion-validation-management",this paper focus on the perceptual access of physical entities,cleaning and maintenance of temperature monitoring twin data and digital twin temperature model of steel-concrete composite box girder based on the two.The main research contents and results are as follows:(1)An optimal arrangement method of temperature measurement points based on total least squares-improved piecewise Douglas-Peucker algorithm(TLS-IPDP)was proposed.In view of the defect that the number and spacing of temperature measurement points lacking theoretical support and often failing to capture the local extreme value of temperature gradient curve in the current temperature measurement point arrangement,TLS-IPDP was proposed by introducing the total least squares method into the traditional DP algorithm to generate the optimal measurement point arrangement scheme.Compared with the traditional temperature measuring point arrangement,the accurate perception of physical entities by digital twin temperature model was effectively realized by TLS-IPDP.(2)Based on the principle of standard Cubature Kalman Filter(CKF)and digital twin temperature model,a data cleaning method for all kinds of abnormal temperature monitoring data was developed.The improved adaptive cubature kalman filter(IACKF),anti-outliers IACKF and IPSOBP-IACKF were presented respectively for data noise,outliers suppression and drift blind alignment.For the data missing,the nearest neighbor search based on the digital twin temperature model accumulative similarity factor realized the missing filling.Some numerical examples show the feasibility of the proposed method.After data cleaning,the error can be reduced in the data processing and maintenance,so that the measured data can be used in the construction of digital twin temperature model.(3)A multi-scale temperature mechanism model of long-span cable-stayed bridge was constructed,and an off-line correction method based on improved quantum genetic algorithm(IQGA)was proposed to realize the off-line correction of the model combined with historical data.By integrating heat transfer,sub-model and sub-structure technology,the"construction-assembly" of temperature mechanism model was realized.IQGA was proposed to modify the temperature mechanism model based on off-line temperature and static load data,and realized"verification-validation" of the temperature mechanism model.The modified model completed the fine characterization of the multi-dimensional characteristics of the bridge structure from the perspective of thermal-structure coupling mechanism,and realized the iteration of verification and correction by using historical data,which significantly improved the accuracy compared with the traditional temperature mechanism model.(4)Based on ensemble learning,a temperature data-driven model of steel-concrete composite box girder with improved Stacking framework was established.The introduction of k-fold cross validation and combined weighting of accuracy and time in a traditional Stacking ensemble learning framework ensures the participation of original input features,time correlation characteristics,and the dominance of strong base models;taking a variety of machine learning algorithms as the heterogeneous base model and LASSO as the meta model,the temperature data-driven model under the improved stacking ensemble learning framework was constructed,and the "construction-assembly-verification" of the temperature data-driven model was realized from the perspective of temperature monitoring data-driven,compared with the independent basis model,the temperature data-driven model has the characteristics of more sufficient data mining,higher consistency between virtual and real,and stronger ability of reconstruction and generalization.(5)Combined with the mechanism model and data-driven model,the "global hybrid model-local hybrid model-digital twin model" method of model"fusion-validation-management" was proposed,and the "fusible,reconfigurable and evolvable" online adaptive co-fusion digital twin temperature model of steel-concrete composite box girder was completely constructed.Based on the error analysis of parallel structure,the construction strategy of global hybrid model was given;An online collaborative fusion local hybrid model establishment method based on improved just-in-time learning strategy was proposed,which realized the establishment of local model,adaptive fusion and adaptive collaboration;A complete digital twin temperature model was constructed,and the effects of specific parameters on the fusion,reconstruction and evolution performance of the model were analyzed.Compared with the prediction results of the global hybrid model for the on-line temperature monitoring process,the adaptive collaborative fusion mechanism of the digital twin temperature model effectively suppressed the divergence of the output residual sequence,made the model have the stability of the global model and the flexibility of the local model,and had the ability of real-time tracking and iterative evolution.It is suitable for the long-term tracking and prediction of structural temperature and temperature effect.(6)The specific application scenario of digital twin temperature model in bridge SHM system was discussed,and the three-dimensional visualization and interaction of twin data were tried.The statistics of temperature law in structural reliability evaluation considering historical state,temperature distribution mode calculation,compilation of temperature fatigue stress spectrum,real-time evaluation,prediction and early warning methods of temperature field and temperature effect were given respectively.Compared with the traditional model,the digital twin temperature model has the characteristics of higher accuracy,stronger adaptability and adaptive coordination,and has broad application prospects.Based on VTK library and XML language specification,the separation of storage and calculation was realized,which solved the problems of slow response speed,slow lag and huge volume of result file in traditional FEM post-processing visualization.The three-dimensional visualization and interaction of digital twin temperature model were effectively realized on the basis of lightweight and lossless.