Settlements in conventional tunneling with deep neural networks

The construction of a tunnel can induce subsidence in the ground at surface level, which can affect and cause damage to existing structures, especially in areas with high building density. This paper presents a deep neural network model (DNN) to estimate the maximum surface settlement “S_max” in a tunnel excavated with using conventional tunnelling. The structuring of the deep learning algorithm was performed using the TensorFlow library in Python 3.0. The DNN model was trained, tested, and validated using a synthetic database composed of several numerical models automated with Python in the finite element program PLAXIS2D. Variables associated with soil properties, support characteristics, surface overburden in the context of the conventional tunnelling process were considered for the modelling. To verify the estimation capacity of the DNN model, performance tests and statistical analyses were carried out, with the results showing good predictive capacity of the model for the three variables analysed.