AI is more than simply a buzzword, with many engineers already incorporating it into their practice – and reaping the rewards.
1. Anomaly detection tools
Tech firms from startups to Siemens have recognised the potential for AI to help with CAD drawings. Systems are trained to become intelligent by using model-based reasoning, producing qualitative and quantitative analysis that predict what should come next — or even exist — in a design.
Dr. Godfrey Keung, a mechanical engineer and Chief Scientific Officer at Australian 3D printing building and construction company Luyten, told create that his company uses AI to assess compliance with building codes once engineers have developed a 3D CAD mesh of a house.
“Trained on a dataset encompassing both code-compliant and non-compliant designs, the algorithm employs convolutional neural networks to discern architectural elements and their proportions,” he said. “It then compares the input design against learned standards, flagging violations and deviations.”
2. Generative AI
Software engineer and YouTube creator Marko Stancic said that front-end developers are using ChatGPT and Github Copilot to “tidy up code” and write “unit tests for an existing function and autocomplete with deep knowledge of the codebase”.
As a back-end developer, Marko is using ChatGPT to ask for recommendations to explore different solutions and technologies.
“Then I use that to inform my designs later,” he said.
Deloitte’s AI lead Dr Kellie Nutall agreed, recently telling create that generative AI will give engineers the ability to design in an entirely new way and at a previously unimagined pace.
“It won’t be long before engineers can ask generative AI to optimise a design for very specific objectives,” she said.
“Whether that might be for climate outcomes, or they want a building to look a particular way without compromising the integrity of the structure, they will simply be able to prompt by offering clear objectives.
3. Open source deep learning libraries
Deep learning networks add brain-like architectures to computers to better solve real-world problems. Tech giants including Microsoft, Google and Facebook have released tensor frameworks for Python, making it easier for engineers to learn, build and train neural networks.
TensorFlow, PyTorch and Keras are generally seen as the most popular products in this field, each with different features to recommend them:
- TensorFlow is probably the most powerful and mature library, known for its strong visualisation capabilities and options for high-level model development.
- PyTorch is younger but has a broader community base thanks to its flexibility and object orientation, and critically, is more Python-friendly.
- Keras, an interface for the TensorFlow library, has the best plug-and-play framework for engineers to use quickly.
4. Creative generation
Robotics engineer and women in STEM advocate Marita Cheng recently published a book, Smart Girl Books, featuring curated illustrations produced by image generation AI software DALL-E.
Telling the story of Cheng’s upbringing in Cairns, it is an inspirational guide for young girls looking to find their way in the world. “I thought it would be fun to try and create something that could last and be of value,” she told create. “So I wrote a book.”
AI generated images can help illustrate pitch decks, tell stories and break up long blocks of text. The technology’s three best known programs, DALL-E 2, Midjourney and Stable Diffusion, each have different benefits:
- DALL-E 2 has the easiest interface and produces images within 12 seconds. The downside to its user-friendliness is that there’s not a high degree of accuracy, and it can only create square images.
- Midjourney produces the most intricate and high-quality images, but is more difficult to use and requires users to use it through chat platform Discord. Users can also program it to custom ratios and it’s much more responsive to prompts to control the image parameters.
- Stable Diffusion varies whether you’re using web-based or app installed versions such as DreamStudio, which give much higher-quality images. Closer to Midjourney than DALL-E 2, its standout feature is the negative prompt box, which allows users to preempt things they don’t want to see.
5. 3D visualisation and simulation
Neural radiance fields, known as NeRFs, are AI-powered computer graphics rendering models which take in multiple 2D viewpoints to generate a 3D model. This allows engineers to project realistic physical effects into their models, such as fire or liquids from a natural disaster.
Similar to the way that AI can recognise features in photos, NeRFs allow multiple images to be used as point clouds. For example, a model could be trained to identify the holes in a scanned building in a complex CAD assembly.
In the longer term, it offers a lot of upside for the graphics capabilities of engineering software. While many programs already manage both mesh and point clouds, in the future they will be able to find a way for NeRFs and other different representations to coexist.