Research On Several Key Problems Of Bridge Safety Assessment Using Model Updating Based On Forward Analysis

With economic development, our country has built a large number of bridges.They are built on various important transport hubs across the whole country andhave made great contributions to the healthy development of the economy. However,many built bridges have been damaged to varied degrees as a result of lack ofappropriate maintenance as well as complicated environments and frequentlyshuttled traffic stream. Therefore, safety assessment is needed for these bridges inorder that maintenance in a timely manner can be provided.Although many bridge safety assessment methods have been studied, there area lot of problems to be solved in the field of bridge safety assessment. For example,because of insufficient definite conditions of inverse problems and non-uniquemodel updating results caused by reasons including the deficiency of measured dataand the insensitivity of measured indicators to damage, bridge mechanical model isunavailable to evaluate bridge property. For long-span bridges, if two-dimensionaldynamic moving on decks of many vehicular loads is taken into consideration, thereare a number of load combinations to be considered for determining the leastfavorable load position. As a result, it is difficult to effectively determine the leastfavorable load position and internal force required for safety assessment. Althoughthere are some methods that can forecast structural life span by physical variablessuch as reliability index, model libraries to fully reflect structural property variationwith time are not established for most safety assessment methods, which makes itdifficult to effectively forecast all structural safety performance indicators in thenear future.In order to solve the above problems, this paper studies a bridge safetyassessment method based on forward analysis model updating. The details areshown as follows:Firstly, this paper proposes a forward analysis model parameter updatingmethod with stepwise uniform design by taken primary and secondary factors intoconsideration. Model updating is an inverse problem that playbacks system byexcitation and response. Most available methods solve the problem of modelupdating from the angle of back analysis, therefore, non-unique model updating results caused by insufficient definite conditions cannot be avoided. From the angleof forward analysis, this paper establishes a complete model library based on priorknowledge and takes the model which is closest to the measured one as the result ofmodel updating by means of matching response. However, the premise of thesuccess of the forward analysis method is to set up a complete model library that canreflect all properties of problems to be solved, which is definitely difficult. In orderto overcome the above difficulty, this paper uses the uniform design to arrange anumber of parameter test schemes with representation to train neural networks sothat the networks will have the function of a complete model parameter library.Through our studies, we find that the uniform design can approximately meetthe requirement of completeness. However, as the uniform design seeks uniformityand abandons orthogonality, the amount of samples of the uniform design samplingscheme is insufficient so that the neural network training is inadequate. Therefore,this paper proposes the method of rearranging factors to extend the uniform designscheme. Meanwhile, this paper uses the fuzzy analytic hierarchy process to evaluatenine kinds of identification indicators. The result shows that the displacement, thefirst-order vibration mode and the fundamental frequency are three predominantidentification indicators and their weights are0.134,0.124and0.122respectively.At last, in order to update the models of large complex structures effectively, thispaper summarize a forward analysis model parameter updating method withstepwise uniform design by taken primary and secondary factors into considerationbased on the lessons and experience of the model updating of a real steel truss.Secondly, this paper studies a method that solves the least favorable loadposition from the angle of test design. Dynamic programming is the main existingapproach to solving the least favorable load position. For bridge plane models,dynamic programming, by making the vehicular loads move along the internal forceinfluence line, evaluates extremes to get the least favorable load position. However,if two-dimensional dynamic moving on decks of many vehicular loads is taken intoconsideration, there are a number of load combinations to be considered fordetermining the least favorable load positions. At this time, the computationalworkload of the dynamic programming becomes very huge. Therefore, based on theexisting researches about the internal force influence line and envelop line of planebeam models, this paper extends the above methods so that the computing problems of the internal force influence surface and envelop diagram of bridge space beammodels can be solved. On this basis, this paper proposes the way of evaluating theleast favorable load position from the angle of test design, namely, the load-factorcombination test is arranged through the uniform design, the function betweenload-factors and bridge internal force extremes is established by regression analysis,and the least favorable load position is determined by the grid-filter method. Theresult shows that, after just a few of computations arranged by uniform design, thedeviation between the result of the method in this paper and the exact solution is1.95%, which means that the method in this paper is feasible and effective.Finally, this paper proposes a bridge safety assessment method based onforward analysis model updating. As model libraries to fully reflect structuralproperty variation with time are not established in most existing studies, this paperproposes a set of methods that establish structural health files on different times byforward analysis model updating, assess the present bearing capacity of bridges andforecast their performances in the near future based on the bridge performancevariation with time and structural health files on different times. The feasibility andeffectiveness of the bridge safety assessment method in this paper are verified by thesafety assessment of a long-span arch bridge. The results show that, for the archbridge in this paper, the biggest response error of the updated models on multi-timesis4.521%and the one of the prediction model is7.6%, which means that the healthfiles that reflect structural property variation with time are established effectively byforward analysis model updating and time series forecasting based on neuralnetworks, and a logic performance evaluating result is given.