Efficient damage detection for automated and safe electric vehicle Battery Recycling (eBatRe)

Purpose and goal:

The purpose of this feasibility study was to investigate the potential in using AI-based computer vision, trained on synthetic data, for electric vehicle battery pack lid damage detection, to take battery disassembly and recycling one step closer to full automation. The main goal was to create a proof-of-concept. While more time is needed to verify that the AI models can be made to perform adequately on real-world images, our results are encouraging, and we will continue to develop our solution.

Expected results and effects:

We have successfully built a highly versatile platform for synthetic data generation. The platform, which uses Blender, allows for light pattern projection a technique that potentially makes damage detection easier onto any geometry. We have also created a simple office laboratory setup for real-world pattern projection and image capturing, that mimics our software platform. Further development of this solution has the potential to open for wider adoption of computer vision in industry.

Approach and implementation:

Work on the synthetic data generation platform and the office laboratory ran in parallel during much of the fall of 2023. Several unexpected challenges were encountered, such as difficulties in isolating the projected patterns in the synthetic and real-world images, and finding a way to automatically label the damaged parts. While good working solutions were found to all these challenges, the AI modelling, which was planned to start in August, had to be postponed until the end of November.