由于立铣刀为圆柱形，刀具偏转是曲面生成中的主要问题之一。通过有限元分析，揭示了刀具形状参数与刚度的关系。立铣刀的刀具刚度表示为等效直径，由描述立铣刀形状的几个几何因素决定。由于当量直径与各因素之间的关系不是解析式的，因此采用了神经网络。为了获得好的结果，反向传播算法的结构由交易决定- 关通过与实验结果的比较，验证了神经网络的仿真结果。

Due to the cylindrical type of the end mill, tool deflection is one of the main problems in surface generation. Using FEM Analysis, relations between tool shape parameters and rigidity are revealed. Tool stiffness of the end mill is represented as equivalent diameter and determined by several geometrical factors that describe shape of the end mill. Because the relation between equivalent diameter and each factor is not analytic, neural network is used. For good result, the structure for back propagation algorithm is determined by trade-off. The simulation result using neural network is verified by comparing with experimental result.