This paper presents an intelligent design method for 2D truss structures by integrating reinforcement learning and finite element analysis (FEA). The approach uses monte carlo tree search (MCTS) on standardized triangular unit grids, with a custom reward function on mechanical performance feedback, and then FEA module evaluates structural behavior in real time by guiding the algorithm to generate truss layouts that meet strength and stiffness requirements. It is shown that the proposed intelligent design method can demonstrate the features of efficient encoding framework and physics-guided search. Finally, some simulation results are given to shown that the system reliably produces optimized structures with controlled stress and displacement under sufficient iterations.