31 Oct 2021
3D Machine Learning by LEDAS and AWV Achieves 95% Success Rate
AWV Informatik, developer of cloud based CAD solutions, and LEDAS, developer of engineering software R&D services and high-tech CAD/PLM components, today presented the results from their joint project dedicated to machine learning (ML) in 3D.
The 3D ML project showed a 95% success rate in classifying machining tools from a 50,000-model testbase provided by the tooling industry. The GTC (Generic Tool Catalog) Format is a complement to ISO 13399 that enables a productive data exchange model between cutting tool vendors and applications.
Goal of the 3D ML Project
The goal of the ML project was to classify 3D models of cutting, milling, drilling, and other machining tools into predefined GTC classes automatically. AWV and LEDAS decided on a deep learning approach that uses depth images rendered from the six primary coordinate directions.
The resulting system is a combination of pre-trained state-of-the-art convolutional neural networks that were specially tuned for working with depth images of machining tools and their accessories. OpenGL and a CAD kernel were used to generate initial depth images from 3D models provided in STEP format. Scripts were developed in Python to deal with multi-layered networks, with the help of Tensorflow and Keras frameworks.
“We decided to use depth images from the six coordinate directions, because in most cases this level of data is sufficient for use in the tooling industry,” said Alexey Ershov, CEO of LEDAS. “In terms of data size, these depth images are much smaller than voxels/3D images, so the number of processing nodes is not excessive, keeping the performance high.”
AWV and LEDAS feel that many tooling vendors and their customers will gain a significant advantage in deploying the new 3D ML system. The ability to make tool classification fully automatic will in the future be in greater demand, because of the way it can support users working with tool data management solution.
How AWV and LEDAS Cooperate
AWV and LEDAS began working together in 2007 when AWV licensed the LGS geometric solver from LEDAS. The companies decided to then enter into a long term contract on software development, which was dedicated initially to the development of ClassCAD, AWV’s flagship product. The list of joint projects successfully implemented includes GPU-based tesselation, a Web-based sketcher, and a new framework for easy development of 3D Web solutions.
About AWV Informatik
AWV Informatik AG is an innovate company that concentrates on CAD cloud software research and development. Its primary product is ClassCAD, an integrated environment for CAD applications. AWV was founded in 2000 as the spin-off from the University of Applied Sciences of Buchs in Switzerland. For more information, visit buerli.io and awv-informatik.ch.
LEDAS is an R&D software service provider to the global engineering software market. For more than twenty years, its staff of highly-qualified mathematicians and computer scientists has been taking on remarkably ambitious projects involving
and digital medicine solutions. LEDAS has success in completing custom software development projects for Dassault Systѐmes, Cimatron, JETCAM, and more. For more information, please visit www.ledas.com.
Permanent link :: http://isicad.net/articles.php?article_num=22100