New classification method of nonstandard parts for CNC selection based on machine learning and geometric analysis

With the continuous development of computerized numerical control (CNC) machine tool technology, selecting an appropriate machine tool is crucial for improving production efficiency and product quality. In this study, we developed a nonstandard part classification algorithm based on machine learning and geometric analysis techniques to provide a new approach for selecting CNC machine tools and sheet-metal processes. First, medial axis transform (MAT) based on the Voronoi diagram and point-cloud-filtering technology was used to obtain the 3D medial axis point cloud (MAPC) of the part. The accuracy of the optimized MAPC was superior to that of the conventional circumcenter method under the same conditions. Principal component analysis and point-cloud angle distribution histograms were used to evaluate the rotation and shape characteristics of each part. The thickness distribution of each part was analyzed using the k-means clustering algorithm. Finally, support vector machines were used to classify parts with different processes. The results showed that the classification accuracy for nonstandard parts reached 93.92%. The innovative nonstandard part classification standards and classification methods proposed in this study strongly support automation and optimization in the field of intelligent manufacturing.