An practical reliability method based on uniform experimental-artificial neural network

Based on the design of uniform experimental, this paper presents a new reliability calculation method, which is organically combined the with artificial neural network. The method adopt by the artificial neural network instead of FEM, which can greatly reduce calculation work. At first, according to the random variable distribution, the limited samples are extracted by uniform experimental, which can be regarded as the input parameters in finite element Analysis. Secondly the optimize artificial neural network is set up with the limited training samples based on finite element analysis result, which has a highly nonlinear mapping relationship between the efficacy and response of the structure. By making use of the generalization capability of optimize artificial neural network, sufficient response value is gained at random, and then the reliability index of structure system can be calculated. At last the method provides a new attempt for reliability analysis and prove to be feasible and effective for practical experience in complex structure, which not only makes the FEM calculation more effective based on uniform experimental, also makes full use of the merit of artificial neural network instead of the performance function.