AI-Based Retrieval to Encourage Reuse of CAD-Designs: A Methodological Study and Future Perspectives

In today's market, product development faces challenges such as customization demands and shortened life cycles. To address these, companies adopt product development in generations, aiming for reuse to reduce time and costs. However, escalating product portfolios lead to opacity and increased variation, hampering effective reuse. This paper proposes leveraging AI, particularly deep learning, to address unnecessary variation creation and lack of transparency. This paper investigates a deep learning methodology, UV-Net, which converts CAD solids into tensors for processing, facilitating classification and retrieval tasks. Experimental results, based on self-supervised learning with the Fusion 360 dataset, highlight AI's potential in categorizing and retrieving CAD designs. While showing promising performance, challenges persist with complex shapes and imbalanced datasets. We discuss integrating cost considerations and future research avenues for a comprehensive AI-driven design assistance system, facilitating improved decision-making in product development.