17033

A Machine-Learning Framework for Design for Manufacturability

Aditya Balu, Sambit Ghadai, Gavin Young, Soumik Sarkar, Adarsh Krishnamurthy
Department of Mechanical Engineering, Iowa State University, Ames, IA, USA
arXiv:1703.01499 [stat.ML], (4 Mar 2017)

@article{balu2017machinelearning,

   title={A Machine-Learning Framework for Design for Manufacturability},

   author={Balu, Aditya and Ghadai, Sambit and Young, Gavin and Sarkar, Soumik and Krishnamurthy, Adarsh},

   year={2017},

   month={mar},

   archivePrefix={"arXiv"},

   primaryClass={stat.ML}

}

Download Download (PDF)   View View   Source Source   

855

views

Computer-aided Design for Manufacturing (DFM) systems play an important role in reducing the time taken for product development by providing manufacturability feedback to the designer while a component is being designed. Traditionally, DFM rules are hand-crafted and used to accelerate the engineering product design process by integrating manufacturability analysis during design. Such a practice relies on the experience and training of the designer to create a complex component that is manufacturable. However, even after careful design, the inclusion of certain features might cause the part to be non-manufacturable. In this paper, we present a novel framework that uses machine-learning with computer-aided design (CAD) models to provide feedback about manufacturability. We use GPU-accelerated algorithms to convert standard boundary representation CAD models into volume based representations that can be directly used for machine-learning. Our framework uses 3D-Convolutional Neural Networks (3D-CNN) to learn the representative geometric characteristics that classify difficult-to-manufacture features in a CAD model of a mechanical part and determine if the part can be manufactured or not. As a proof of concept, we apply this framework to assess the manufacturability of drilled holes. CAD models with different manufacturable and non-manufacturable drilled holes are generated and then converted into volume representations using GPU-accelerated algorithms. This data is used to train a 3D-CNN for manufacturability classification. The framework has an accuracy of more than 78% in consistently classifying the manufacturable and non-manufacturable models. Finally, the framework can explain the reason for non-manufacturability in a part using a gradient-based class activation map that can identify the non-manufacturable feature, and provide feedback to the designer about possible modifications.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2017 hgpu.org

All rights belong to the respective authors

Contact us: