A Method Of Estimating Structural Failure Probability And Applied In Levee Pile Foundation Project

With the rapid development of the national economy, the government will increase the financial investment to infrastructure and large complicated structures have been used extensively in practical engineering. So the security problems of these structures will become hot issues which are concerned by more and more people. However, besides that there are many uncertainties affecting the security of structures, existing methods have the deficiencies of high computational cost and low calculating accuracy when solving the problems of structures reliability. Consequently, it is imperative to explore an effective and practical structures reliability analysis method aiming at the current practical engineering.GPR-MCS is new reliability computational procedure which combines the machine learning and traditional method which has advantages in solving structures reliability problems. But the result rely too much on learning samples and the choice of learning samples make much sense to calculation result., So a dynamic learning mechanism is applied to GPR-MCS. In the proposed method, GP predicted value and variance are employed to assess to the predicted results, and then new learning samples are acquired and GPR models are rebuilt repeatedly based on dynamic learning samples, finally adopting MCS to solving the problems directly. In the thesis, fundamental of GPR-MCS, select strategy and dynamic learning mechanism are stated, the results of performance test by different benchmark functions and application in practical engineering show that the proposed method is feasible. Compared with traditional typical method shows it is efficient.Given the complicated problem of tackling with horizontal bearing pile foundation, mechanical response and destruction model are analyzed under horizontal load basing on three-dimensional medol of the rock-socketed pile built by the finite difference software FLAC3D. Comparing with Single pile horizontal load experiment, FLAC3D is reasonable in solving nonlinear problems of interaction between pile and soil. Finally taking counterfort levee pile foundation is taken as an example, and FLAC3D is applied to model the grounded piles and analyze their bearing features and destruction model. Aiming at the phenomenon of excessive horizontal displacement at the top of piles and considering the variability of load and material, function of solving structure reliability is established according to the limited displacement top piles, so the piles foundation problems can be solved by GPR-MCS basing on dynamic learning mechanism. Result indicates that piles foundations will failure under condition of uncertainty and the large horizontal force, reinforcement treatment must be done.