Application of Developed Neural Network for Improving Forming Accuracy of Polygonal Pyramids and Spherical Segment in Incremental Forming

Abstract

This paper presents an application of a developed convolutional neural network (CNN) to improve the forming accuracy of polygonal pyramids and spherical segments in incremental forming. The developed CNN was trained using height data converted into CSV files, generating new tool paths for forming. Three models were created with resolutions of 50×50 pixels, 100×100 pixels and 200×200 pixels for training data. The research method involved comparing the traditional and CNN-generated forming processes, focusing on parameters which include wall angle, forming height, initial diameter, gap volume, and shear droop height. Among the models tested, Model100 generally provided the best results, outperforming traditional methods and other CNN models. Results showed that the CNN approach significantly reduced free surface deformation and springback, leading to more precise shapes. This study highlights the potential of CNNs in improving the incremental forming process, suggesting further application to more complex shapes and the potential for broader industrial adoption.