Study On Finite Element Model Validation Method Based On Gaussian Process Response Surface And Kernel Density Estimation

Modeling and simulation (M&S) based on finite element method is extensively used in theindustrial areas. However, modelers and analyzers are all continuously puzzled about credibility of thefinite element model established. Results of the finite element model are full of variation due to kindsof uncertainties during the process of M&S. The problem from Sandia Laboratory on modelvalidation for structure dynamics and one bolt-connected structure are both regarded as the mainobjects for study, while several critical technologies on finite element model validation areinvestigated.Design expert is introduced, which plays an important role for the design of traditional responsesurface. Also, N-Modal is adopted to identify the real value of modal parameters for structures.Moreover, model updating based on RSM is accomplished for one bolt-connected splice plate.Depending on the use of MATLAB toolbox, multivariate Gaussian process response surface isestablished and its feasibility is verified by one explicit mathematical function. Furthermore, Gaussianprocess response surface is build to predict the transient response of the finite element model for acantilever beam. And it is demonstrated that feasibility and accuracy are all avalible for GaussianProcess response surface.Documents quoted, concepts and contents about model verification and model validation aredetailed. Technologies about uncertainty propagation of model parameters (KDE, PCA, Monte Carlo,etc) and model validation metrics are both discussed. One virtual problem on model validation forstructure dynamics proposed by Sandia National Laboratory is taken as an example for the applicationof model validation in the field of structure dynamics.A procedure of finite element model validation is discussed. And the whole process is applied tothe real bolt-connected structure. The whole structure is devided into two sub-structures whilecalibration and validation tests are all accomplished. Model updating is carried out to calibratedeterministic parameters and the distribution of uncertain parameters is qualified by KDE. Gaussianprocess response surface of the target application model is established in order to predict thecofidence interval of response. It is concluded that the response of the target structure can beefficiently estimated under the use of finite element model validation method, which can be worthyfor the predictation and optimization of the use of performance for the target application structure.