Despite the rapid development and application of modular buildings, the structural behaviour of high-rise configurations requires further detailed investigations. In order to enable extensive studies into the nonlinear seismic time history response of such large and complex buildings, there is a need to develop computationally efficient approaches. This investigation therefore, describes the development of a machine-learning approach for predicting the time history response of modular buildings using a transformer model architecture, which is found to be particularly suitable for such sequence-to-sequence tasks. The proposed machine learning model is trained using a large database comprising the seismic time history response of a prototype high-rise modular building configuration through nonlinear time history analysis under a suite of 3000 ground motions. A special designed encoding function was applied to reflect the unique structural characteristics of modular buildings, while convolutional neural networks are used to capture both global and local features of seismic vibrations, followed by feature concatenation for the machine learning prediction. The proposed model is shown to provide a highly efficient prediction procedure that captures the time history response of such buildings. Finally, to demonstrate the applicability and effectiveness of the developed machine learning model, an illustrative example is presented in which the influence of inter-module connection properties on the seismic response of modular buildings is examined and discussed. Compared to widely used nonlinear finite element procedures, the proposed machine modelling methodology offers a fundamental approach for modular buildings to enable efficient large scale structural evaluations.
扫码关注我们
求助内容:
应助结果提醒方式:
