ROBUST FINITE ELEMENT MODEL UPDATING METHOD BASED ON THE FEATURE MAP OF FREQUENCY RESPONSE FUNCTIONS

The frequency response function (FRF) contains all the dynamic information of a structure, making FRF-based methods a hot topic in current finite element model updating research. However, traditional FRF-based model updating methods have challenges such as frequency point selection, abnormal sensitivity matrix, and poor noise immunity. To address these problems, we propose a robust model updating method based on the feature map of frequency response functions (FMFRF). This approach treats the model updating as a forward problem using a Bayesian Convolutional Neural Network (BCNN). The FRFs of multiple sensors are combined and transformed into the feature maps, which is used as the inputs to the BCNN. The output of BCNN is the vector composed of uncertain model parameters. The Bayesian framework mitigates overfitting in neural networks with small datasets, enhancing training robustness and improving the accuracy of mapping the relationship between FMFRFs and model parameters. This method retains the advantages of traditional FRF-based methods, such as no need for mode identification and mode matching, and simultaneously updating structural parameters and damping, while significantly improving efficiency and noise immunity. This paper first verifies the method's noise immunity through numerical simulation and then experimentally validates it using a titanium alloy wall plate with six reinforcing ribs, demonstrating the proposed method's effectiveness in complex engineering structures.