AI-driven CAD modeling can enhance efficiency, but hurdles remain

Three-dimensional (3D) computer-aided design (CAD) is a discipline traditionally dominated by mechanical engineers, architects, manufacturers and designers. However, like most current disciplines, artificial intelligence (AI) is finding a place in 3D CAD workflows.

 

CAD models are a critical tool in design and manufacturing that allow designers to create digital representations of objects of virtually any size before manufacturing them in the real world. Importantly, individual parts can be tested for fit, movement and compatibility before investing in, for example, springs, chips and custom-designed enclosures. Conversely, models can remain digital for use online.

 

 

The learning curve for 3D CAD software can be daunting, however, and even the most proficient designers and engineers can make mistakes, inevitably increasing costs and delaying product launches. Manual 3D CAD that doesn't leverage assistance software is even more time-consuming and laborious. The rise of AI and, more specifically, machine learning (ML) and large-language models (LLMs) provides a unique opportunity to integrate these technologies in 3D CAD workflows and increase efficiency, precision and creativity.

 

A group of researchers from the University of Science and Technology of China in Hefei, China recently wrote a comprehensive literature review outlining the current AI-driven 3D CAD design state of the art and technological hurdles that remain in efficiently incorporating the AI models into current design workflows.

 

The team published their review on April 30 in the journal Computational Visual Media, published by Tsinghua University Press.

 

“CAD generation algorithms, which use point clouds, images, and other modalities as input, can be applied to reverse engineering, directly restoring the geometry of CAD models or even modeling command sequences. LLM-based algorithms can understand users' design intentions, making the design more intuitive and accessible and reducing the workload of designers in various aspects of the design process,” said Kang Wu, postdoctoral fellow at the School of Mathematical Sciences at the University of Science and Technology of China in Hefei, China and an author of the review paper.

 

The team revealed that, beyond improved accuracy and efficiency in AI-derived 3D CAD modeling, AI research is better understanding how the technology can also be leveraged to assemble more complex large-scale CAD models, provide new models beyond the limitations of human experience and create physical constraints in the CAD modeling process.

 

Despite AI’s potential in 3D CAD, the application of the technology in this field is limited and a number of drawbacks exist. Importantly, 3D CAD design files are incredibly complex, can be stored in different ways and often lack data reflecting the reasons specific designs were created. Further, the variety of CAD model representations makes establishing a large unified dataset challenging, making AI model training difficult. It can also be challenging to generate the same CAD model using different CAD modeling software via modeling command sequences, reducing the practicality of AI intervention.

 

“The models generated by current methods are not only relatively simple, but also lack a guarantee of validity. Methods often struggle to generate results with the precision, manufacturability and physical correctness required in industrial settings. Moreover, interactive and user-driven design, where the system can understand and respond to evolving user intent, remains an open research frontier,” said Wu.

 

Despite these challenges, the research team views AI-driven 3D CAD modeling as an important tool for future design.

 

“Future research can focus on constructing high-quality, general-purpose CAD datasets that include not only geometric data but also complex operation sequences and user intent annotations. Additionally, integrating AI with physical simulation and manufacturing constraints, enhancing user interaction and feedback and developing systems that more accurately capture the design intent of users will be crucial to advancing AI for CAD,” said Wu.

 

Wenzheng Wu, Yifei Li, Xiao-Ming Fu, Falai Chen and Ligang Liu from the School of Mathematical Sciences at the University of Science and Technology of China in Hefei, China; and Siyuan Xing from the School of Artificial Intelligence and Data Science, University of Science and Technology of China also contributed to this research.

 

This research was supported by the National Natural Science Foundation of China (12494555, 62025207, 92570201).

 

DOI Link:

https://doi.org/10.26599/CVM.2025.9450521

---

---