The HDMR-hybird Network Method of Model Approximation for the Stochastic Analysis of Semi-rigid Joint

The huge economic costs with the experimental analysis and computationally expensive with the simulation are the difficulties for the stochastic analysis of engineering structures. The structural data of stochastic analysis is based on the probability of statistical results. Therefore, the focus of structural stochastic analysis is how to simulate the finite element model or the physical model by the method of approximate model with the requirements of the expected accuracy range. For the higher precision sensitivity coefficients, a large number of finite element simulations would be conducted and this process leads to intensive computation. This paper puts forward a methodology that combines the high dimensional model representation(HDMR) method and the hybrid neural network for the approximate model. The advantage of this method is the determination of coupling characteristics of the input parameters, and the complex multidimensional model could be constructed by the limited sample points. The feasibility of the method was applied to semi-rigid connection with multidimensional parameter, and the efficiency and precision were obviously superior to the traditional approximate method.