Research On Reliability Analysis Method Based On Support Vector Machine

There are many uncertainties in the design and service period of engineering structures, including the uncertainties of external loads, structural capacity, structural dimensions and analysis models. Structural reliability theory is a powerful tool to solve structure uncertainties and evaluate structural performance. However the performance function is always high nonlinear implicit in the complex engineering structure, which results in costly computational efforts for the reliability analysis using gradient based method or Monte Carlo Simulation(MCS). So reliability analysis with implicit performance function is a general problem for the structural reliability analysis of the practical engineering. Surrogate based analysis and optimization (SBAO) has been shown to be an effective approach for the design of computationally expensive models which is an effective way to solve the implicit performance function reliability analysis. The key issue of SBAO is how to construct a high-fidelity, stable, easy-used surrogate model with fewer samples.Support Vector Machine (SVM) is a machine-learning algorithm based on Statistic Learning Theory (SLT) which was successfully used in the areas of digit recognition, computer vision, text categorization, etc for the good learning performance with small size samples. SVM provides an effective surrogate model for approximation of the implicit performance function. In this paper, the background and principle of Support Vector Machine are presented and reliability analysis method based on Support Vector Classification (SVC) and Support Vector Regression (SVR) are proposed and compared. The parameters in SVM model which affect the training performance are discussed, and the Genetic Algorithm (GA) is employed to optimize the parameters, which can improve the performance of SVM model considerably. Numerical examples demonstrate the applicability of SVM to the region of structural reliability which is finally applied to the reliability analysis of the components of the heavy-load forging manipulator.This thesis is supported by the National Basic Research Program of China (973 Program, 2006CB705403).