An efficient approach for safety factor-based and reliability-based design of tunnel face support pressure using BP Neural Network

This paper develops an efficient approach for safety factor-based design and reliability-based design of tunnel face support pressure by constructing a surrogate model to predict limit support pressures. The Back Propagation Neural Network is utilised to fit a surrogate model based on a set of training samples with limit support pressures computed by a proposed numerical method. Several scenarios are used to demonstrate the applications after the prediction performance of the model has been evaluated. A safety factor-based design is implemented by predicting the limit support pressure of a scenario with strength parameters divided by the target safety factor from those of the original scenario, whereas a reliability-based design is performed by predicting the limit support pressures of a certain number of Monte Carlo simulation samples and then choosing a quantile value from the sorted limit support pressures. The design results are checked by direct numerical simulations, demonstrating that the model is valid for conventional scenarios within the specified design domain. The approach is convenient and efficient for applicable scenarios because the most complex and time-consuming process of constructing a surrogate model is no longer needed.