Transfer learning-based artificial neural networks for hysteresis response prediction of steel braces

• Proposed a novel data-driven model to predict hysteresis response of steel braces. • Used transfer learning with pre-trained baseline LSTM networks for improved performance. • Validated the model using four case studies with experimental and synthetic data. • Applied model in a pseudo-dynamic analysis of a steel braced frame. • Model accurately estimates displacement-force relationship in steel braces. This paper proposes a novel data-driven surrogate model for predicting the hysteresis response, i.e., axial force – axial deformation, of steel braces in concentrically braced frames under seismic loading using transfer learning-based artificial neural networks. Transfer learning is utilized to leverage pre-trained baseline long short-term memory networks and transfer its knowledge to the new hysteresis surrogate model. The proposed model is validated using four case studies involving various combinations of input data obtained from laboratory tests and data generated using random earthquake-induced vibration, featuring a wide range of frequency contents, amplitudes, and durations. A pseudo-dynamic analysis is then performed on a steel braced frame system to demonstrate the application of the proposed surrogate model in system-level response evaluation while verifying the performance of the model in real-time seismic simulations. The results obtained from the validation study confirm that the proposed brace hysteresis model can properly estimate the underlying physical relationship between the input displacement and output force using the transfer learning approach. The proposed model offers an efficient method to evaluate the dynamic response of steel braced frames.