Structural Reliability Analysis Based on Support Vector Machine

摘要

Structural reliability and safety play a major role in all facts of human lives.Over the past few decades significant advancements have been made in incorporating and consideration of reliability and uncertainty analysis in a wide range of engineering disciplines and practices.However,due to lack of a complete understanding and predicting the structural response under various environmental impacts,changes and variations occurring in a structure through its life time,and/or modifications and redesign of a structure’s components during its service life,it might be difficult to obtain the up to date information of structural reliability.Therefore,a non supervised approach to evaluate the reliability information is highly desired.In this work,a Support Vector Machine(SVM)based reliability analysis approach is introduced.Support Vector Machine(SVM)is a promising machine learning algorithm for data classification and regression.For the classification problem,the major feature of SVM is its capability of minimizing the training error while simultaneously maximizing the margin between two classes.This leads to the unique characteristic of its ability of generalization from the small data sets.In reliability analysis,on the basis of the simulated data sets,a hyperplane to classify the safety region and the failure region can be found by using SVM.Other testing data can be classified according to this classifier.The advantage of this approach is to classify the new data points without going through the calculation of the limit state function.In this section of the research work,the support vector machine algorithm was implemented to classify the safe region and the failure region.The failure probability for a set of test data was found based on the classifier.The reliability updating can be performed based on the learning machine function by updating the hyperplane with only a small set of new measurements.