Deep CAD Shape Recognition for Carbon Footprint Estimation at the Design Stage

• Published in: Procedia CIRP Vol. 122; pp. 545 - 550
• Main Authors: Hasebe, Tatsuya, Katayama, Erika, Yoshiteru, Katsumura
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2024
• Subjects: carbon footprint assessment; computer aided engineering; machine learning; machine learning; computer aided engineering; carbon footprint assessment
• ISSN: 2212-8271; 2212-8271
• DOI: 10.1016/j.procir.2024.01.080

Estimating the carbon footprint of products at the early stage of design is crucial for streamlining the engineering process of sustainable products. However, the carbon footprint estimation of the products requires material and manufacturing information that is typically not available at the early design stage. In this study, a novel method is proposed for carbon footprint estimation, which can evaluate the carbon footprint through the shape recognition of computer aided design (CAD) models based on the graph deep learning. The learning model utilizes the boundary representation of CAD models for deeper understandings of the CAD model shape. The proposed method trains the deep learning model on the existing CAD models to recognize important sub-shapes and attributes, including materials and manufacturing information, such as welded parts, which are the essential data for the carbon footprint estimation. The proposed method enables the designers to estimate the carbon footprint without the laborious condition setting, which facilitates concurrent monitoring and improvement of the carbon footprint. The method is applied to actual assembly models and demonstrated that the material and welded parts, which are attributes required for emission prediction, can be recognized with an accuracy of more than 80%.