Data for Deep learning-based semantic segmentation of machinable volumes for cyber manufacturing service
Author(s)
Yan, Xiaoliang
Williams, Reed
Arvanitis, Elena
Melkote, Shreyes
Advisor(s)
Editor(s)
Collections
Supplementary to:
Abstract
Enabling the vision of on-demand cyber manufacturing-as-a-service requires a new set of cloud-based computational tools for design manufacturability feedback and process selection to connect designers with manufacturers. In our prior work, we demonstrated a generative modeling approach in voxel space to model the shape transformation capabilities of machining operations using unsupervised deep learning. Combining this with a deep metric learning model enabled quantitative assessment of the manufacturability of a query part. In this paper, we extend our prior work by developing a semantic segmentation approach for machinable volume decomposition using pre-trained generative process capability models, which output per-voxel manufacturability feedback and labels of candidate machining operations for a query 3D part. Using three types of complex parts as case studies, we show that the proposed method accurately identifies machinable and non-machinable volumes with an average intersection-over-union (IoU) of 0.968 for axisymmetric machining operations, and a class-average F1 score of 0.834 for volume segmentation by machining operation.
Sponsor
This work was funded by an INTERN Program Supplement to the National Science Foundation EAGER, United States, award #2113672, and the National Science Foundation Future Manufacturing Program, United States, award #2229260.
Date
2023-11
Extent
Resource Type
Dataset
Resource Subtype
Rights Statement
Unless otherwise noted, all materials are protected under U.S. Copyright Law and all rights are reserved