Structural Damage Identification Based On Advanced Bayesian Probabilistic Inference Sampling Approach And Application In Truss Structures

Along with the rapid development of economy and technology, many large complexstructures become being under construction. These structures, however, definitelyundertake potential damages when they are in the operation condition. How to timelyand efficiently detect the impairment that occurs in the structures to avoid engineeringaccidents has raised more and more interests among the civil engineers.In this thesis, some theoretical approaches for structural damage identificationproblem are introduced at the beginning, especially, the Bayesian probabilistic statisticalmethod implemented in the structural damage identification and model updating. Then,based on the Bayesian posterior distribution therom, the probabilistic models forstructural damage indexes are established with time domain and frequency domainseparately. Instead of applying traditional sampling methods for Monte Carlo integral,which is inefficient and inapplicable for high dimensional problem, an advancedTransitional Monte Carlo Markov Chain (TMCMC) is proposed in the paper to solveidentification problem of many uncertain parameters with the condition of various noiselevels.The innovations of the approach employed here are summarized as follows:1. Revising the original TMCMC to improve its efficiency and reduce thecomputational cost, while keep the good predicted accuracy.2. Establishing the data exchange OAPI between commercial finite element softwareSAP2000and Matlab program within Matlan environment so that the finite elementanalysis run in SAP2000is able to be controlled through Matlab code.3. The importance and influence of several parameters in the TMCMC are discussedto achieve the optimization of the algorithm for different cases.4. The revised TMCMC algorithm is applied to solve the high-dimensionalparameters identification problem.5. The complicated configuration of truss joint is simulated by unique element inSAP2000and the damage occurs in the joints is also diagnosed by proposed revisedTMCMC.6. The performance of revised TMCMC with different noise levels is investigated toqualify for the field implementation.From the simulated analysis of a complex truss structure, it is concluded that thepresented approach is capable of identifying damage among many unknown parameters with measurement noise. In addition, the location and the extent of the impairment canbe predicted through the numerical characteristics of posterior PDF. In the end, thedefect in current method is discussed and future work is put forward to make theapproach more applicable.