Assessing effectiveness of a dual-barrier system for mitigating granular flow hazards through DEM-DNN framework

Constructing multiple-barrier systems is efficient in mitigating flow-like geological hazards and potentially serves as an optimal solution to protect the heavily threatened infrastructures, such as the ongoing Sichuan–Tibet railway in China. However, the design of the multiple-barrier system remains essentially empirical, which hinders the development of scientific design guidelines for a robust and cost-effective barrier system. In this study, a new framework combined discrete element modelling (DEM) and deep neural network (DNN) was proposed to assist the design of a dual-barrier system. The DEM results are used as a database to construct a DNN model, where barrier spacing (L), barrier height (H) and Froude number (Fr) are input parameters to predict the energy-trapping efficiency of barriers on resisting granular flows. The energy-trapping efficiency serves as a key index to evaluate the design of a dual-barrier system. The DEM results show that when barrier spacing L ≥ 5.0 h or barrier height H ≥ 1.0 h (where h denotes the flow depth), barrier spacing has a negligible effect on the energy-trapping efficiency. Those criteria can potentially be used to optimise the configuration of dual-barrier systems. Furthermore, the energy-trapping efficiency obtained by DEM model is reliably predicted by the DNN model, with a coefficient of determination equal to 0.9997, root mean square error equal to 0.0141, mean absolute percentage error equal to 0.0125, and variance account for equal to 99.68%, which demonstrates that the DNN model has a potential to optimize the configuration of dual-barrier systems in granular flow mitigation.