Hybrid deep learning model for prediction of monotonic and cyclic responses of sand

The monotonic and cyclic behaviors of sand are of practical significance for various problems in geotechnical engineering. In this paper, a novel hybrid deep learning model was proposed for the prediction of the responses of sand undergoing monotonic and cyclic loading conditions; two popular neural networks, the long short-term memory network (LSTM) and temporal convolutional network (TCN) were employed. A synthetic dataset generated by a constitutive model was initially used to analyze the appropriate arrangement of the LSTM and TCN layers and compare the predictive performance of the hybrid model with those of the LSTM and TCN alone models. An experimental dataset based on available laboratory tests on Karlsruhe fine sand was also used to further demonstrate the model's capacity. The predictive results showed the superiority of the proposed hybrid model and reproduction of the constitutive responses of sand with a higher accuracy under both monotonic and cyclic loading conditions.