Application of neural network to back analysis of mechanical parameters of columnar joint basalt

The bedrock of Baihetan dam is a basaltic layer of Permian that has a special structure of columnar joint type. For this type of rock,the study on its physical and mechanical properties is not enough so far,and no suitable constitutive model or failure criterion has been established yet,while they are cruicial to analyze the stability of the dam,foundation and underground cavities. At present a feasible approach to obtain these properties is to simulate the columnar joint structure by discrete element method. This paper conducts a back analysis of the joint parameters kn and ks by applying the 3DEC software and neural network methods and using the field data of load-bearing plate tests at Baihetan dam site,and then input these two parameters to 3DEC for simulation. In comparison with the field data of load-bearing plate,calculation shows the feasibility of the method presented.