Based On The Probabilistic Finite Element Numerical Simulation Of Structural Reliability Analysis And Software Development,

In the structure reliability analysis of mechanical products, the direct integral method and square method are no longer applied to reliability analysis, because it is difficult to describe the limit status function expression of large complex structure precisely. Meanwhile, with combination of finite element and the direct Monte Carlo method for reliability analysis, there are outstanding problems that calculated amount and solution precision increase exponentially in the proportion and is time-consuming, for finite element analysis should be done in every simulation analysis. The basic starting point for the paper:Based on the limit status response surface ideology, function expression between objective variable and input variable to replace finite element calculate model, which can increase calculating speed and efficiency, would be fitted.Based on the basic theory of structure reliability analysis, the research of random variables and the parameter distribution type are conducted; the distribution function of stochastic variables are changed into normal distribution with NATAF transform. With the stress-strength interference model as the theoretical basis, the reliability calculation equation and three reliability algorithm are analyzed.With the normal distribution as the basic types, two schemes are made which are three level as low-level analysis and seven level as high-level analysis. Using three-level analysis in central composite experiment design as probability finite element experiment design method, the number of sample Monte Carlo is determined.With the ANSYS software as the determined finite element analysis tools, the APDL commanded flow files are generated. Probability finite element analysis with the PDS module in ANSYS software is run, and then, response measurement data is generated.Base on the design of probability finite element and response surface compound simulation, with twice polynomial, KRIGING model and BP neural network as the response model, reliability analysis is done by loading response data to fit the response surface model. With the simulation results of several Monte Carlo as the exact solution of reliability analysis, the accuracy of response surface models are compared. Calculation results show that the method has high precision to meet the design requirements.Based on the researches above, using the thought of software modularization, MATLAB as software development tool, structure reliability calculate software CSU-SRC is developed. Software example shows that using this software for structure reliability analysis can improve the calculating speed greatly, without compromising the accuracy of the premise.