A soft computing approach to tunnel face stability in a probabilistic framework

Tunnel face is important for shallow tunnels to avoid collapses. In this study, tunnel face stability is studied with soft computing techniques. A database is created based on the literature which is used to train some broadly adopted soft computing techniques, ranging from linear regression to the artificial neural network. The soil dry density, cohesion, friction angle, cover depth and the tunnel diameter are used as the input parameters. The soft computing techniques state whether the face support is stable and predict the face support pressure. It is found that the artificial neural network outperforms the other techniques. The face support pressure is predicted with the artificial neural network for statistically distributed samples, and the failure probability is obtained with Monte Carlo simulations. In this way, the stability of the tunnel face can be reliably assessed and the support pressure can be estimated fairly accurately.