Research & Development

AI supports assembly specialists

Training of AI systems for the recognition of objects is still time-consuming and expensive. A new technology automates this process by using image data that is generated for all objects.

Artificial intelligence (AI) enables machines to recognize objects. For this purpose, large amounts of high-quality image data are required to manually train the algorithms. Kimoknow, a startup established at Karlsruhe Institute of Technology (KIT), Germany, has now developed a technology to automate this training.

“Training of AI systems for the recognition of objects still is time-consuming, inflexible, expensive, highly environment-dependent, and associated with a high computation expenditure,” says Lukas Kriete, one of the founders of Kimoknow. For this reason, the startup of KIT uses image data that are generated anyway for all objects in computer-aided development processes (CAD) and production data management (PDM). These data, among others, include information about the material, geometry, and position of the object. CAD and PDM data are extracted and used for automatic training of the AI. The trained object recognition system can be used for many purposes, among others, in augmented reality (AR) glasses. They capture relevant objects in the field of view of the user in real time and additionally possess the necessary context information on the object.

Application in assisted assembly

As the first use case of such AR glasses, Kimoknow developed an assistant to support specialists in the assembly of complex devices. The virtual assistant guides users through the complete assembly process, visualizes the assembly instructions step by step without an additional display, and shows in which order which part is to be assembled with which tools and assembly materials. It repeats certain steps, if errors occur, and documents the process. The mechanic has both hands free and communicates with the system via eye contact, hand signal, or voice command.

“The assembly assistant makes the process more efficient, productive, quicker, and less expensive, while quality is improved,” Kriete says. The assistant is suited for all industries producing small numbers of highly complex products. The prototype is applied for final assembly of highly specialized measurement instruments and presently being tested in cooperation with Elabo at the Center for Artificial Intelligence Talents (CAIT) of KIT’s Institute for Information Management in Engineering (IMI).

Kimoknow is a startup of IMI and was established on May 13, 2020. Apart from Lukas Kriete, Roman Wiegand, Aaron Boll, Michael Grethler, and Vesa Klumpp are members of the team of founders.

mg

www.kimoknow.de

www.imi.kit.edu

Company information
© laser-photonics.eu 2020 - All rights reserved