Research And Application Of Structural Reliability Calculation Method Based On Neural Network

Modern engineering,machines and technical equipment are becoming more and more complex.While these structures provide increasingly high-quality performance,they also put forward the higher request to its structural reliability.In the structural reliability analysis,it is difficult to calculate the structural reliability because of the complexity of structure,the incompleteness of probabilistic information,the limitation of cognition,the inadequacy of experimental sample and experimental data or highly nonlinear failure surfaces.In view of the difficulties in current reliability analysis,it has great theoretical significance and practical application value to make accurate calculation on the structural reliability by exploring a new solution approach.This paper will focus on the calculation of structural reliability under the consideration of different factors,and strive to provide new methods and ideas for structural reliability analysis.The main research contents are as follows:(1)The problem of structural reliability solution with multi-dimensional correlation variables is researched.By selecting the structure type of Copula function and solving related parameters,the joint probability density function between correlation variables is constructed,which overcomes the limitation that joint probability density function between variables is difficult to obtain directly.The direct integration method is used to construct the integral form for calculating the reliability.A dual neural network method is proposed for the calculation of multiple integrals.One network approximates integrand,the other network approximates original function.During training,only the integrand network is trained.The original function network is obtained through the relationship between the network parameters of two networks.Then the calculation of multiple integrals is achieved.This method effectively solves the difficulty of calculating multiple integrals in reliability calculation by direct integral method.Thus,under the condition of considering the correlation between variables in the structure,the complex structural reliability of multi-dimensional correlation variables is calculated with high efficiency and high accuracy.(2)The reliability of solid rocket motor grain during solidification and cooling is analyzed.The finite element ANSYS software is used to carry out three-dimensional parametric modeling of the grain.Through the analysis of transient and dynamic thermo-solid coupling under cooling conditions,the dangerous points and dangerous moments are obtained,and the maximum equivalent strain and temperature values are extracted.A dual neural network model is established based on the Copula function and the probability distribution of specific parameters.Then,the instantaneous reliability in the process of solidification and cooling of the grain is accurately calculated.Thus,the dynamic reliability analysis is realized and the practicability of the proposed method in engineering practice is verified.(3)The problem of structural reliability under fuzzy failure criterion is researched.Based on the method of Akaike Information Criterion,we measure the goodness of fitting between statistical estimated membership function and actual structure data,so as to determine the membership function of specific structure.According to fuzzy set,membership function and the probability of fuzzy random events,a mathematical model for calculating the structural fuzzy reliability is constructed.The direct integration method based on dual neural network is extended to the calculation of this mathematical model.The integrand network that composed of the fuzzy failure criterion and the joint probability density function between variables is trained.Then the original function network is simulated to obtain the structural fuzzy reliability.This method improves the calculation accuracy of structural reliability that considering the fuzzy failure criterion.Based on the mechanical properties experiment of grain and ANSYS software simulation,the fuzzy structure reliability during the grain ignition is analyzed,and the results show that the proposed method has the ability to solve practical problems.(4)For the problem with implicit performance function,a response surface method based on custom neural network is proposed to analyze the structural reliability.In this method,exponential function is used as the hidden layer activation function of neural network.Due to a multilayer neural network can be used to approximate any non-linear function with arbitrary precision,a custom neural network structure is constructed.The trained neural network not only realizes the display expression of structure performance function,but also improves the fitting precision of performance function.Compared with the traditional polynomial response surface methods,this method has better fitting effect for highly dimensional,highly nonlinear implicit performance function of structure.The response surface method based on custom neural network provides an effective modeling and analysis method for solving the reliability calculation of complex structural systems.(5)Under the condition of small sample,the performance parameters interval quantization and instantaneous reliability calculation of solid rocket motor grain structure are researched.Two important mechanical properties of material: relaxation modulus and poisson's ratio,which are obtained by experiments of grain material.Due to the data of parameters obtained is a small sample,the method of grey theory is used to mine experimental data.It realizes the uncertainty quantitative analysis of grain material performance parameters,and obtains the quantitative interval of performance parameters.Considering that the evidence theory can directly attribute probabilistic mass to the set or interval,the instantaneous reliability of grain structure is analyzed based on the evidence theory.By establishing the relationship between failure surface and the frame of discernment,and using the belief function and plausibility function to obtain the upper and lower bound probability distribution of structural reliability and failure probability,the instantaneous reliability probability interval of structure is obtained.