Comparison Of Model Updating Methods And Application On Beam Structures

There exist various kinds of errors in finite element model (FEM), whichsometimes makes it unable to reflect the dynamic characteristics of the actualstructure. However, those errors will be effectively decreased by model updatingtechnique based on modal test data and therefore, the accuracy of FEM will beimproved.Model matching and correlation analysis, which are both closely related to modelupdating technique, were firstly introduced in this paper. Three model updatingmethods, namely Berman method, Improved Cross-model Cross-mode (ICMCM)method and Radial Basis Function (RBF) neural network method, were elaborated.Their application area and updating accuracy were studied through the modelupdating of two structures, namely simply supported beam with added mass and boltconnected cantilever beam. During the updating of each structure, two model updatingmethods were used to compare the accuracy of different methods. Meanwhile,different finite element models were considered to study their effect on modelupdating.For the simply supported beam with added mass, Berman method and ICMCMmethod were adopted. The results showed that the updated model frequency byBerman method totally equaled the test ones within the test band and the relative erroroutside the test band with the fine FEM didn’t exceed5%, while the accuracy ofICMCM method was lower. However, the matrixes resulted from Berman method losttheir physical meaning and complex matrix operations made it unsuitable for largecomplex structures, for which ICMCM methods had advantages.For the bolt connected cantilever beam, ICMCM method was mainly researched.The updated model frequency not only matched the test data well within the test band,but also gave good prediction of the frequency outside the test band and the dynamiccharacteristics of the structure. Moreover, another important conclusion was drawnthat complicated FEM is no more needed for bolt connected structure. Modelupdating of the simple rigid model can satisfy the engineering requirement, whichprovides a guide for this area.Exploratory work was conducted on the RBF nueral network model updating of the bolt connected beam. Meanwhile, the most sensitive parameters to structuredynamic features were found out. As the results showed, the relative error was within2%, thanks to the nonlinear mapping established between the input and output data inthe network. Moreover, RBF neural network method can be very suitable for largecomplex structures with a number of parameters.