Merged CNN-Based Finite Element Model Update Methodology

Finite element analysis requires the construction of a model that accurately reflects the structural characteristics of a target structure. Such models must incorporate precise material and geometric properties. However, discrepancies often arise between the properties assumed during the design phase and those observed in the actual structure. To address this, the models are iteratively updated using real measurement data, minimizing the errors between measured and analytical results. This process, however, can be time-consuming, particularly for complex structures. In this study, we propose a novel approach to directly estimate the parameters of a finite element model by utilizing the modal information from the structure, including natural frequencies and mode shapes. Our algorithm integrates a deep neural network that effectively solves complex problems using a convolutional neural network that captures the relationships between the nodal data and modes. Verification using a beam model demonstrated that the proposed methodology successfully updated the finite element analysis model to accurately match the target modal information.