Abstract：In light of the three-dimensional effect of split Hopkinson pressure bar(SHPB) device for large diameter rock specimen，solid simulation is conducted with three-dimensional nonlinear dynamic finite element code；and the resultant data serials are then simulated with radial basis function (RBF) neural network and the ideal striker is self-adaptively recognized with the network. Finally，the constructed striker producing perfectly half-sine wave is compared with that back designed with one-dimensional stress wave theory；and it is demonstrated that the newly constructed striker by using three-dimensional nonlinear FEM and neural network has better half-sine waveform.
