Abstract:A Bayesian probabilistic methodology to detect prestress lose of unbonded full prestressed concrete beam is presented. Free vibration of prestressing tendon with elastic rotational constraints was studied, transcendental equation of natural frequencies was derived, modeling of prestressing tendons was releasized in form of equivalent flexural rigidity in finite element model of beam. Because of noise existentence in test and randomness of material performance test, Markov chain Monte Carlo sampling technology based on adaptive Metropolis algorithm(AM—MCMC) was employed to update the finite element model in probablistic framework, the uncorrelated chain sequences were constructed for structural parameters, and distribution range of prestress lose under different prestress levels were obtained based on posterior estimates of structural parameters. The numerical results indicate that AM-MCMC sampling techonology assure the mixture capacity of chain sequences, and proper selection of sampling interval will decrease the autocorrelection effect of samplings. For different prestressing force levels, errors between the predicted stastististical means of prestress lose and test ones are less than 6% .