METHOD FOR COMPUTING STATICALLY EQUIVALENT SEISMIC LOADS USING MACHINE LEARNING FOR SINGLE-LAYER CYLINDRICAL RETICULATED SHELLS WITH VARIOUS SUPPORT CONDITIONS

This paper presents a method for predicting dominant vibration modes using neural networks to calculate the static seismic loads of single-layer cylindrical reticulated shells with various support conditions. In order to represent the structural features of these shells, such as support conditions and aspect ratios, in neural networks, this study introduces a method of expressing the displacement distribution in a fixed-length three-dimensional array, similar to image data. The three-dimensional array is fed into a neural network with convolutional layers and is used to predict the dominant vibration modes. Machine learning was applied for training, validation, and testing based on the vibration analysis results from 37,293 shells. The results demonstrate that the proposed method can accurately predict the dominant vibration modes.