A novel model updating method based on Linear Regression and Support Vector Regression (LR-SVR) is proposed for a plate structure in thermal environments, in which the generalized eigenvalues of the mass and stiffness matrices are replaced by the natural frequencies. The proposed method demonstrates superior updating efficiency compared to traditional finite element model updating techniques and offers enhanced accuracy relative to conventional SVR-based model updating methods. The temperature-varying parameters are assumed to be polynomials w.r.t temperature. Virtual elements are introduced to facilitate boundary modeling, allowing for easier updating of the boundary conditions in the thermal environment. Both numerical simulation and experiment study on a two-side clamped plate in thermal environments are conducted to validate the efficacy of the proposed method. The experimental study accounts for a non-uniform temperature field, which can be predicted by the data obtained from both a thermal imager and thermocouples, employing a function interpolation scheme. The issue of initial deformation due to both inaccurate installation and thermal expansion is effectively resolved. The updated model exhibits high accuracy and efficiency in predicting the first five natural frequencies at different temperatures.
Read full abstract