Research On Inverse Method Of Uncertainty For Composite Laminates

Composite laminates are widely used in civil ships,automobiles,wind power and many other fields due to their high fatigue resistance,high corrosion resistance and good thermal stability.Compared with traditional materials,one of the advantages of composite laminates is designability.This means that researchers can design composite laminates with good mechanical properties according to their use conditions.However,in the early stage of composite design,it is necessary to investigate the relationship between response and design parameters by constructing inverse problem model according to mechanical properties.For the dynamic problems of composite laminates,the relationship between load and response can be explored by constructing an inverse problem model,and then the factors caused by the damage of composite laminates can be further studied.Therefore,the research on the inverse problem of composite laminates is one of the hot topics in current research.Generally speaking,solving model response with known model parameters and boundary conditions is called positive problem.And obtaining parameters through system response is called inverse problem.In the field of inverse problems,mixed numerical method is the mainstream method in the current inverse method.The main issues of the inverse problem are illconditioned and uncertain.First of all,because most of them are ill-posed problems,it is more complex than the normal problem.Secondly,when taking into account the measurement error of inverse parameters,processing error and other factors,it is necessary to further consider the uncertainty of parameters.The main work of this paper is summarized as follows:1.Considering that the uncertainty of geometric parameters and material parameters has an important influence on the mechanical properties of variable stiffness composite laminates.In this paper,the Approximate Bayesian method is used to identify the geometric parameters and material parameters of variable stiffness composite laminates.Compared with the traditional Bayesian method,the Approximate Bayesian method can avoid the calculation of likelihood functions,which make the reverse process more complex or difficult to obtain in practical problems.The difficulty of Approximate Bayesian method is to obtain the general statistics.If the larger generalization statistics are selected,the Approximate Bayesian method will produce larger computational cost.On the contrary,if less generalized statistics are used,it will cause large information loss,resulting in inaccurate results.In this paper,the AutoEncoder(AE)is used to complete the extraction of the Approximate Bayesian method.Meanwhile,in order to ensure the high sampling efficiency of the Approximate Bayesian method,the ANSM sampling method is integrated to achieve the sampling of the Approximate Bayesian method.Furthermore,to reduce computational cost,the Neural Network(NN)used to construct the mapping between parameters and responses are utilized.For the ill-conditioned problem of identifying the parameters of variable stiffness composite laminates,the Tikhonov regularization is integrated into the Approximate Bayesian method to solve the parameter identification problem.2.For the dynamic problem of composite laminates,it is necessary to identify the corresponding excitation changes according to the displacement changes.In this paper,two AE models are combined with the Approximate Bayesian framework to realize the boundary identification of the dynamic change process of the force as the boundary condition,and this framework is compared with the GRU model framework.