End-of-line automation with AI

In short

The aim of the project assigned to the AI Innovation Center was to deploy AI to automate the two end-of-line steps visual quality control and goods packaging at the company Witzenmann. The result increases the quality and value of the products without the need for extra staff.

In detail

In order to manufacture competitively in a high-wage country like Germany, automated processes are indispensable. Witzenmann GmbH therefore approached the AI Innovation Center with the task of assessing whether visual quality control and packing at the end of production could be automated and, if so, to implement a prototype application that could perform both tasks. Since these are the last steps before customers receive their product, they are especially important. One challenge in this regard was the fact that the products were limp and movable.

The result of the collaboration is a demonstrator that is capable of classifying defective parts and automated bin packing with intelligent packing pattern generation based on AI using a robotic system. An image processing system first visually scans the products in detail and evaluates them regarding to damages with the help of AI. In this way, they can be classified as defective or non-defective in the next step of the packaging process. The defect-free products are then laid down and packaged in an optimized manner using an AI-based positioning generator. To train the corresponding neural network, the project partners used the AI method of reinforcement learning, i.e. learning from trial and error. Accordingly, the algorithm was “rewarded” for laying down products in the right position.

After completion of the project at IPA, the demonstrator was handed over to Witzenmann, where it is kept in the Digital Lab for demonstration and training purposes. The collaboration between the company and Fraunhofer IPA is a successful example of how AI technologies can be transferred into practice.

Demonstrator project within the scope of the AI Innovation Center, carried out at Fraunhofer IPA

Funded by the State of Baden-Württemberg, Ministry of Economics, Labor and Tourism