BIGcad: Assisting 3D CAD Modeling with Workflow Graph-Driven Bayesian Command Inferences

Abstract

Background: Recent advancements in 3D generative artificial intelligence (AI) have streamlined design processes by enabling rapid model creation. However, these tools frequently lack the accuracy and comprehensive support needed for intricate real-world applications. Consequently, designers continue to depend on command-based computer-aided (CAD) tools such as Rhino, which provide the necessary accuracy but can impose high cognitive loads. To address these challenges, we introduce BIGcad:a workflow graph-driven system that leverages Bayesian inference to optimize 3D CAD modeling, thereby enhancing both efficiency and precision. Methods: BIGcad encodes 3D modeling sequences using a Workflow graph (W-graph) and integrates a Bayesian information gain (BIG) framework to infer user intentions. By analyzing user interactions and modeling data, the system predicts subsequent steps in the modeling workflow. Implemented as a Rhino plugin, BIGcad captures command logs and snapshots in real time, providing guidance that reduces cognitive load and improves overall design efficiency. Results: The implementation of BIGcad yielded promising results, particularly in improving workflow efficiency and lowering cognitive demands. Participants reported streamlined processes, particularly during complex modeling tasks, while the visualized W-graphs provided valuable insights into alternative workflows. This approach not only reduced errors but also fostered the exploration of creative modeling strategies, underscoring the system’s potential to advance design processes. Conclusions: BIGcad introduces a new approach to improving 3D CAD modeling by integrating workflow visualization and command recommendations based on user behavior. The system not only enhances modeling efficiency and accuracy but also provides opportunities for creative exploration and learning. Future efforts will focus on expanding dataset diversity and enhancing personalized features to further optimize their utility in design processes.