Assessment of Risk Potential due to Underground Box Structure Installation Employing ANN Model and Field Experimental Approaches

Abstract

Construction of underground box structures that underpass existing infrastructures or facilities at shallow depths is being widely conducted in urban areas in order to avoid the interference of existing structures without the rerouting of traffic. Consequently, it is crucial to ensure safety during and after construction by monitoring the ground settlement induced by underground box installations that can be influential for the existing various structures. This paper provides a method to assess the risk potential around underground box structures with artificial neural networks (ANN), taking into account input variables that can be monitored in the field. By introducing the numerical methods and ANN, the probability of failures considering the variability of design parameters such as ground conditions, structure sizes and shapes, traffic loads, and presence of existing structures could be assessed and utilized for the safe construction of underground box structures at shallow depths. Experimental programs were also performed to investigate the effect of the umbrella method, which contributes to the decrease of risk potential in a practice. A limited field test evaluation using ground-penetrating radar (GPR) along with a pneumatic dynamic cone penetrometer (PDCP) was found to be promising in the assessment of risk potential at shallow depths.