Addressing Part Variability in Disassembly using Point Cloud-based Machine Learning

Machine Learning (ML) and Artificial Intelligence (AI) are well-established technologies in object classification and detection, with applications spanning from consumer products to industrial implementations such as part identification and quality inspection. Increasingly, ML and AI are seen as key enablers of the circular economy, particularly within disassembly and remanufacturing. Both fields are faced with challenges well suited for intelligent systems, notably the variability of parts. In disassembly, a wide range of different components in diverse states of wear from numerous products need to be identified and treated accordingly. This poses difficulties for traditional image-based computer vision systems, especially when training data is limited. This study investigates the application of point-cloud (PC) classification as a means to distinguish between products, enabling the identification of correct disassembly sequences and following remanufacturing treatment based on geometric differences, using real-world data to assess its practical applicability.