| description abstract | The rapid proliferation of tunnel construction presents challenges to the safe operation of tunnels. Although Structural Health Monitoring Systems (SHMSs) have been widely used to prevent tunnel disasters, it is still impossible to record the mechanical behavior of the full profile of structures because of a limited number of monitoring points. Along this line, this study proposes a spatial deduction model based on a machine-learning algorithm to characterize the mechanical behavior of a tunnel structure profile driven by limited monitoring data. Strain variation is considered to reflect the mechanical behaviors of the structure, and the monitoring data obtained from the SHMS of Dinghuaimen Yangtze River tunnel are adopted for these experiments. First, the framework of the spatial deduction model, which uses a nonnegative matrix factorization (NMF) algorithm, is presented. Then, the model is formulated using the monitoring data. A numeric model is developed to reflect the geological conditions in the field to compare with the data-driven model, and the spatial deduction results are used to analyze the real-time and historical mechanical behaviors of the structure. The results indicate that the sensitive positions of the tunnel structure are the arch crown, hance, and inverted arch. The correlation between the deduction result and actual data is more than 85%, and the error is less than 2.7 με, so the presented model is reasonable. | |
