Learning-based 3D {CAD} model generation in mechanical engineering: A survey

Computer-aided design (Cad) solid modeling lies at the core of product design in mechanical engineering, enabling designers to transform conceptual design ideas into precise, parametric, 3D Cad models. Automating solid modeling has the potential to accelerate engineering workflows, foster innovation, and lower barriers to design for non-expert users. Recently, increasing efforts are being devoted to leveraging learning-based methods to address this challenge. To the best of our knowledge, no existing survey provides a comprehensive overview of learning-based methods for 3D Cad model generation, nor addresses their applications in mechanical engineering. For that reason, we survey literature comprising state-of-the-art, learning-based methods for 3D Cad model generation, with particular emphasis on their relevance to mechanical engineering. Based on the type of input to the respective methods, we define three tasks and classify the literature accordingly: Cad Generation, Cad Reconstruction, and Cad Reverse Engineering. To provide a structured and comprehensive overview, we further organize the literature according to its types of output and the methodologies applied. Additionally, this survey highlights current research gaps, including limitations in fine-grained user control and geometric detail preservation, and discusses potential directions for future research. Finally, we present one potential use case for each of the aforementioned tasks in the mechanical engineering workflow, aiming to bridge the gap between existing methods and real-world applications. These use cases include design space exploration, quality control, and data enhancement.