The purpose of this study is to integrate lean production and the six sigma technique into actual lean six sigma projects to solve the problem of mold surface roughness. This process is defined by Value Stream Mapping, current performance is measured, and hexahedron processing, forming, and size processing workstations are merged, in order to stabilize production flow time. Meanwhile, to solve the practical problems of stabilization improvement, the Taguchi method is used to enhance product process capability, and thus, attain the goal of improvement. To meet the multiple surface roughness requirements of different customers for different products in practice, the Artificial Neural Network prediction model is used to save the time and costs of repeatedly starting Taguchi experiment calculations.