Computer vision application for industrial Li-ion battery module disassembly

Abstract

Automated robot-assisted disassembly is essential for the flexible disassembly of Li-ion battery modules for economic and safety reasons. In such a case, a CAD model for the planning process is of immense benefit. The geometric uncertainties due to the breathing of the Li-ion cells as well as the presence of component tolerances underline the importance of a sensor-based detection approach to determine the actual state of the battery module, which is crucial to ensure an automated and reliable disassembly process. In this paper, we present a method for 3D camera-based localization of points on deformed battery modules, aiding in identifying support points for milling operations in robot-assisted disassembly cells. This separation operation planning employs a CAD model, and our introduced computer vision “data processing pipeline”—a systematic series of processing steps—bridges the gap between the CAD model and the actual battery module. This involves capturing the module using a 3D camera and subsequently registering its points with the CAD model’s points. Central to this process are two algorithms: The Bayesian Coherent Point Drift (BCPD) algorithm ensures accurate non-rigid registration, while TEASER++ aids in reducing computational time. We demonstrate the effectiveness of these combined algorithms in our pipeline through rigorous testing and metrics, evidencing that a balance between accuracy and computational speed can be attained by adjusting point density.