Research On Reliability Analysis Method Based On Statistical Pattern Techniques

With rapid economic development and the progress of technology, the large complex structures have been applied to the practical projects. The security problems have become the hot topic. The security and reliability of structures are influenced by many factors, with features of implicit expression, high nonlinearity and high computational cost for the performance function, the traditional reliability analysis methods encounter great difficulties.Aiming at this problem, this paper presents a statistical pattern recognition based dynamic response surface method. In the method, Markov chain Monte Carlo (MCMC) is adopted to simulate initial samples, pattern recognition technique is used to generate response surface based classification as the surrogate model of performance function, besides, response surface based classification is updated by means of adding the Most Probable Point and it’s real state into the old training data set, the iterative algorithm can improve the fitting precision in the area which make a greater contribution to the failure probability. After that, the failure probability is estimated by Monte Carlo Simulation.Four pattern recognition techniques, are used to generate response surface model of the limit state surface, which include k-Nearest Neighbor, Discriminant Analysis, Support Vector Machines and Gaussian Process Classification, then the reliability analysis is conducted combining Monte Carlo Simulation based on the implementation of updating the response surface successfully. The results of examples and engineering application show that the proposed method has good applicability for solving structural reliability problems with implicit performance function, especially for problems with high calculation cost in one numeric calculation, meanwhile, the method possesses great advantage in computing efficiency and accuracy, overcoming the limitations of traditional methods in solving these problems. Moreover, the statistical pattern recognition techniques of MATLAB and commercial finite element software ANSYS are combined in the method, the connection makes full use of the convenient programming function of MATLAB and the powerful calculation ability of ANSYS, which provides much convenience for engineers. In addition, the study shows that the methodology in the proposed method have a certain universality, which also can be applied to other pattern recognition approaches, the method gives a new way to solve reliability problems of complex engineering structure efficiently.