Fraunhofer IMS and German Industrial Partners Launch Projekt KIMIPA to Combat Edge AI Sorting Errors
Camera-based sorting systems in manufacturing mistakenly reject up to thirty percent of flawless parts. A research initiative named Projekt KIMIPA seeks to reduce this waste using retrofittable edge artificial intelligence. The project officially launched on June 1, 2026, and runs through May 31, 2028. It targets optical quality assurance in series production and contract sorting facilities, with the broader aim of achieving a productivity increase of up to 25 percent.
Hardware Architecture and Edge Learning Mechanics Without Cloud Data Transfers
The system classifies components into three distinct categories: good, not good, and pseudo-scrap, while simultaneously identifying the root cause of any misclassification, such as dirt, unfavorable lighting, or process deviations. Dennis Freudenreich, managing director of SDC Control, stated that sorting the exact same parts repeatedly and performing manual recalibration mid-run reduces the efficiency of entire production lines. Fraunhofer IMS supplies the technical foundation through its open-source framework AIfES. This framework enables neural networks to execute, train, and adapt directly on resource-constrained microcontrollers. The platform aims for a classification accuracy of at least ninety percent while operating entirely without a cloud backend, which eliminates ongoing operational costs for external computing power and data transmission.
Industrial Standards, Retrofitting, and Funding Structure of the Consortium
The platform relies on open standards and industrial interfaces such as OPC UA and MQTT to integrate into existing production lines. Funding comes from the Federal Ministry for Economic Affairs and Energy under the Central Innovation Program for Medium-Sized Businesses under funding reference 16KN126526. Fraunhofer IMS receives a grant of 279,720 euros for its role in the initiative. Alongside the research institute, the project involves SDC Control GmbH based in Neuenrade and Scemtec Hard- und Software für Mess- und Steuerungstechnik GmbH located in Reichshof-Wehnrath.
Production Testing Phases
Consortium partners plan to test the developed procedures under real manufacturing conditions through structured measurement series on physical test setups. The primary objective centers on transferring the artificial intelligence models across diverse components, inspection features, and machine types. Through a digital user interface, plant mechanics can flag incorrectly sorted parts to feed their empirical knowledge directly into the AI training, while the system provides concrete action recommendations for readjustment such as lighting, feed speed, or cleaning of system components.
