In hot radial forging process , the value of end surface concave of axial forging reflects the axial metal flow inhomogeneity. A threedimensional model of radial forging process for steel 45 axis part was established and verified by finite element method (FEM), and the influences of geometrical parameters of hammers and process parameters such as radial reduction rate on end surface concave were analyzed, while the Taguchi method was employed for numerical experiment design. In addition, a new process was proposed with the bite ratio, adding a step of putting the forging into hammers and forging it in situ before the radial forging process. The results show that the new process is beneficial to reduce the depth of end surface concave, and the reasonable range of geometrical parameters of hammers is hammer diameter ratio  φ≤1.2 and hammer width ratio ω≥0.5， respectively. Based on the FEM results, the TakagiSugeno Fuzzy Neural Network (T-S FNN) accurately predicts the value of end surface concave and forging load. Considering forging defects, forging force of equipment and production efficiency, a reasonable process case is presented to guarantee the quality of forgings.