Parameter determination for ice material model based on a bidirectional long short-term memory neural network

Parameter determination is a common problem in engineering activities. For impact problems subjected to ice projectiles, however, very few researches have addressed the inverse method to determine the material parameters of ice for finite element simulations. The present study introduced a novel method based on a sequence-to-sequence bidirectional long short-term memory (LSTM) neural network to learn the relationship between the input impact force histories and output material parameters of the finite element model, which was built to reproduce the ice impact test using a hollow tube sensor. After the trained network was evaluated by testing data set, the experimental data was used to predict the parameters for the numerical model to precisely match the test results.