Evaluating the impact of lighting variations and preprocessing techniques on spot weld for automated inspection of battery packs

Project description

How can computer vision techniques, including illumination and preprocessing methods, be optimized to effectively detect battery cells on battery packs, enabling automated disassembly?”

This question addresses the need to identify the most effective lighting conditions, camera setups, and image processing techniques to ensure reliable detection of weld spots on battery cells, which is critical for automating the battery revision process. It reflects the aim to bridge the gap between engineering implementation and structured research methodologies, as identified during the course of the project

Context

This project is situated in the domain of automated battery disassembly, focusing on using computer vision to detect battery cells on battery packs. The context revolves around revising e-bike batteries, currently relying on manual selection of CAD files to guide laser machines for nickel removal. By introducing camera vision, the project aims to automate this process, addressing challenges in lighting, image acquisition, and feature extraction to streamline operations while ensuring accurate and reliable detection of weld spots.

Results

The project identified the best lighting setups and image processing techniques for detecting battery cells on battery packs using camera vision. Perpendicular lighting combined with specific preprocessing methods, like Gaussian and Median-Morphological filtering, gave the clearest results for weld spot detection. These findings help make battery disassembly faster and more efficient by automating the process.