Overview of AI Tools for Structural Engineers

Engineers are paying increasing attention to advances in neural networks and considering how to apply these technologies in their professional work, especially in science and engineering. Modern machine learning approaches are gradually finding their place in CAD/CAE systems. This overview highlights current tools and trends that may be useful in engineering practice.

For simplicity, we use the term AI throughout, although most examples specifically involve machine learning.

Generative models

One of the best-known technologies in this field is topology optimization, now often referred to as generative design. Such tools are available in many modern CAD systems: 3DS Generative Design, NX Generative Design, SolidWorks Topology Study, and Autodesk Generative Design.

The technology uses finite element methods to adapt a part's shape to external loads. The resulting geometry can be used directly in manufacturing only in certain cases, such as plastics, 3D printing, or presentation models for investors. More often, it serves as a starting point for further refinement using traditional methods.

Advanced packages such as Siemens NX and Autodesk Fusion can adapt parts to specific manufacturing processes, from additive manufacturing to casting and machining on equipment with different numbers of axes. Manufacturing cost estimation is particularly valuable for customers concerned with production economics as well as weight.

Generative design also has limitations: it works with individual parts and cannot always identify the best solution for an entire structure.

Autodesk Automated Modeling

Autodesk is actively incorporating machine learning into its CAD systems. One example is Automated Modeling in Fusion. The user specifies the objects that the new geometry must connect to and the objects it should avoid. The system generates alternative 3D solids that can then be refined manually.

The approach is promising but needs further development: the resulting shapes are often too complex for practical use. Its future is closely tied to creating and training specialized datasets.

Automatic sketch and drawing generation

Modern systems increasingly offer tools for automating 2D drawings. For example, the Autoconstrain module in Autodesk Fusion suggests different constraints and dimensions, while Drawing Automation automatically generates basic drawings from 3D models.

These tools are especially relevant to companies where 2D drawings remain the primary working documents. Faster documentation and the ability to check dimensions can significantly improve engineering productivity.

Intelligent assistants

Advances in generative models naturally lead to intelligent assistants.

Autodesk Assistant AI currently functions as an enhanced help system for navigating the interface. Its potential as a full-featured CAD copilot has yet to be realized.

Leo AI is positioned as an engineering copilot, but in practice it is a RAG system with limited capabilities. It can search documentation and select parts, but its model-generation results are not yet suitable for practical use.

ChatGPT and other GPT models can already help engineers find technical information, work with documentation, and analyze images. Combined with a corporate document repository and a properly designed RAG approach, these systems can become effective assistants.

Text-To-CAD and Text-To-Mesh

Some tools generate parametric models from text descriptions (Text-To-CAD) or polygonal mesh models (Text-To-Mesh).

Text-To-CAD remains at an early stage: examples such as Adam AI and Zoo Design Studio demonstrate limited functionality. These solutions currently tend to consume more time than they save.

Text-To-Mesh is developing faster. Rodin AI stands out for its ability to generate models from text descriptions, images, or point clouds. This approach may be useful in reverse engineering and could speed up the development of replacement parts.

Conclusion

Most of the tools described above are not yet available in Russia, but they illustrate directions for developing domestic solutions. The practical value of many technologies remains debatable, leaving room for genuinely useful tools.

The most promising options today appear to be:

multimodal models (such as ChatGPT-4o) for analyzing documents and images;

Rodin AI as a tool supporting reverse engineering.

Engineering workflows have not yet undergone a revolution, but more mature AI tools could substantially change design processes over the next few years.

Based on the article author's personal opinion - https://habr.com/ru/companies/lanit/articles/934500