The discrimination model for road cut slope stability was used to assess 216 cut slope locations on the Chuo Expressway. This model ranks the failure probability of slopes by the rainfall threshold (cumulative precipitation that causes the slope to fail) to identify slopes highly prone to collapse. Because of the complexity (nonlinearity) of the relationship between the factors relating to the cut slope failure and the precipitation that triggers a failure, it has been difficult to correctly evaluate the likelihood of failure for cut slopes. The developed model has overcome that difficulty by involving the neural network as the discrimination technique. The input data included the different factors (topography, soil properties and geology, surface layer status, change in state) in the stability investigation table prepared at the time of road slope inspection, with additional information such as catchment topography. The cut slope data were prepared, referring to a variety of information encompassing the failure history for 30 years after the commencement of service, the rainfall record at the time of failure, the maximum rainfall amount ever recorded, and the data on the status of slope protection around the time of failure. As shown by the discrimination results, the model accuracy (ratio of correct answers to number of slopes evaluated) was as high as approximately 80%, which allowed accurate determination of the amounts of rainfall inducing the failure of different slopes.
Abstract