2023 05 01 AI in Engineering Then Now HERO 1

A History of AI in Engineering from the 1970s to Today

With the transformative potential of new AI tools like ChatGPT on everyone’s mind, it’s easy to forget that artificial intelligence has been used in product design and engineering for many years.

How AI will impact engineering is on everyone's mind these days, but AI has been used in engineering for decades.

Here's a brief history. Click to skip to the section.

I hope this offers my fellow engineers helpful context for understanding AI for product design and engineering as we all engage with the sea change currently underway and ponder what's next.

(And if you want to read more about AI and product development, don't miss our other blog posts about incorporating AI into the product development process and how to use AI responsibly in innovation work.)

2023 05 01 AI in Engineering Then Now INLINE1 FEA 1
Figure 1: Finite element analysis plot of stresses in a solar panel bracket.

FEA Simulations (1970s - early 2000s)

Early in my career, I had as my mentor an engineer who described to me how, back in the 1970s, he used to run engineering simulations via punched cards, using early versions of the ANSYS software, for Westinghouse. He recalled his great fear of dropping the stack of cards and expressed his amazement at how far computing and simulations had come since.  As an interesting side note, he contributed to a superb summary of the origins of finite element analysis (FEA).

FEA simulations are straightforward math models that can calculate mechanical behaviors such as stress, strain, heat transfer, and fluid flow. Because they are 1) problems with well-defined math models, and 2) time intensive to do manually, they were good candidates for early computer applications, whether via punched cards in the 1970s, or the more advanced GUI that my mentor and I used in 2001. 

The ability to create prototype designs using FEA simulations meant fewer hardware prototypes were needed, dramatically cutting the time and expense required to take products from concept to the manufacturing stage. 

2023 05 01 AI in Engineering Then Now inline2c stadium seat
2023 05 01 AI in Engineering Then Now inline2a stadium seat
2023 05 01 AI in Engineering Then Now INLINE2b topology optimization
2023 05 01 AI in Engineering Then Now inline2c stadium seat
2023 05 01 AI in Engineering Then Now inline2a stadium seat
Figure 2: This stadium seat design by Delve in 2013 made use of topology optimization. Tap the left/right arrows to view all the images in the slideshow.

Topology Optimization (Early 2000s)

As computing power continued to increase, a new type of computational model emerged. Topology optimization was more intelligent than traditional FEA. It used dynamic algorithms to automate what was otherwise a manual and tedious "guess and check" approach to mechanical design optimization.  

For example, in 2012, Delve applied topology optimization to the design of a stadium seat using software developed by a professor at the University of Wisconsin–Madison. This not only saved engineering time, but also resulted in a more optimized design that significantly reduced product weight and cost.  

2023 05 01 AI in Engineering Then Now INLINE3 topology optimization 1
Figure 3: Hand tool weight optimization using topology optimization (original geometries outlined) Image source: Delve

Topology optimization whittles away at an initial product shape to reduce weight, while maintaining performance requirements. Impressively, it can out-perform experienced engineers at this task. 

For this reason, and because it can “create” designs the engineer may not have considered, I would consider this an early application of AI in engineering. 

It’s a fairly narrow application given the geometry limitations and its structural optimization focus. And the output is directional, not final designs, as you can see in the stadium seat example (Figure 2) and with exploration we did around hand tool weight optimization (Figure 3). But it is an impressive tool nonetheless, and is currently available within many existing computer aided drafting (CAD) ecosystems including Solidworks.

A i by philippe starck 1 bd
Figure 4: The A.I. chair, a collaboration between AI and designer Philippe Starck, was an early generative design success. (Image via Starck.com)

How Will AI Impact Medical Devices?

Ai medical device webinar youtube thumbnail

We had a great discussion about the ways AI will impact the medical device industry.

Generative Design (Today)

Generative design is a newer and smarter generation of software that further blurs the lines between CAD modeling and simulation and marks a significant step forward in AI for product design. If topology optimization was a clumsy toddler with a crayon-in-hand, generative design is a clever, creative youth.  

Early generative design software was limited in terms of capability, was computationally intensive, and the software wasn’t user-friendly. However, that has all changed in recent years. Software improvements and cloud computing which have made generative design tools accessible and capable.  

Rather than giving the software a geometry envelope to work within, as you do with topology optimization, generative design can “imagine” and build its own structure in-between defined areas to fulfill design objectives. Using artificial intelligence, it creates something new and often very different from anything designed before. And it can do so prolifically, especially when leveraging cloud computing, with the ability to create dozens, hundreds, or even thousands of different designs to choose from or iterate upon. 

Using generative design software can feel like more than simply using software. For me, it begins to feel like a clever collaboration partner. 

Early generative design software was limited in terms of capability, was computationally intensive, and the software wasn’t user-friendly. However, that has all changed in recent years.

An early generative design success, furniture company Kartell chose to design a chair that was a collaboration between a generative design AI and designer Philippe Starck (Figure 4). The chair was released in 2019 and called “AI”. 

Autodesk’s Fusion 360 generative design package was selected as the AI collaborator. Basic inputs such as loads and constraints were provided, and the software created many iterations of structurally optimized concepts. Starck reviewed and refined these concepts into a design that was also aesthetically pleasing, as the raw output wasn’t always visually appealing. 

The AI provided the structural inspiration, but it required the design inspiration of Starck to transform it into a product that humans would appreciate. It also required human effort to make a 3D model based on this inspiration that was smooth, uniform, manufacturable, and production-ready.

AI-generated designs proved superior to those of NASA’s human engineers, demonstrating the power of the AI behind generative design.

19 Uses of AI in Product Development

19 Uses for AI in Product Development linkedin

We mapped how current AI tools might best align with and enhance our design methods and processes.

As second example, NASA has also been leveraging the power of generative design. 

There are many design considerations and constraints when designing for space, two of which are structural optimization and lightweighting. Given their low production quantities, NASA was not limited by the manufacturing constraints of traditional high-volume manufacturing processes. They were able to let the generative design AI create optimized 3D designs without manufacturing constraints, opting to fabricate parts using additive manufacturing (3D printing). 

These AI generated designs proved superior to those of their human counterparts, demonstrating the power of the AI behind generative design.

Microsoft Teams image 30
Figure 5: MJK Performance triple clamp “lightweighting” optimization via several manufacturing processes. (Image via Autodesk.com)]

As a third example, MJK Performance and Autodesk collaborated using Fusion 360 to generate custom triple clamps for motorcycles. These clamps are critical structural components as they secure several parts of the fork. 

Since they are typically bulky, the part was lightweighted and different manufacturing processes were compared, as shown in Figure 5. The “human design” was a baseline, and generative design was used to produce “lightweighted” designs optimized for CNC (both 2.5 axis and 3 axis) as well as additive and die casting that maintained a factor of safety of at least two. 

Fusion 360 can also provide cost estimations for each design and manufacturing process to consider when downselecting designs.

    However, it's worth reiterating some of the current limitations of generative design:

    • The examples above are single parts—not multi-part assemblies or complex systems.  
    • Both require significant effort from humans on both the front-end and the back-end, although this is improving.
    • Generative design doesn’t understand design intent, human factors, or aesthetics.
    • Generative design creates solid bodies, not high-quality parametric CAD models.
    2023 05 01 AI in Engineering Then Now INLINE4 ai engineering timeline 1
    Figure 6: A History of AI in Engineering from the 1970s to Today

    What’s Next: Generative AI and Beyond

    Right now, a ChatGPT-like tool does not yet exist that is capable of replicating what our Engineering team does in terms of design and 3D modeling. However, we are seeing incredible progress and have some powerful new tools at our disposal. Rapid progress is being made to solve the known shortcomings of generative design and to transform it from a specialty tool that very few use to a general-purpose tool with the potential to dramatically transform our work. 

    Rapid progress is being made to transform generative design from a specialty tool to a general-purpose tool with the potential to dramatically transform our work.

    So, what’s next for generative design that would catalyze this change? 

    • Improving generative design solvers that can tackle complex assemblies and systems and cooling, fluid flow, and multiphysics problems.
    • Supporting electrical engineers with schematic design, component selection, and PCB layout.  
    • Creating aesthetically pleasing forms and product skins alongside industrial designers.
    • Automating the process of quoting and ordering prototype or production parts.  

    And perhaps someday, generative design will be able to “understand” more of the complexities of product development such as human factors, sustainability, system-level architecture and design, and what products will face and need to survive in the real world.  

    Product Engineering Services at Delve

    Engineer working on project

    From concept to manufacturing, our engineering services can get your products to market.

    “Generative AI” is the new phrase to watch.  

    Generative AI encompasses both ChatGPT and the cascade of other AI-powered tools capable of generating content. These tools leverage huge datasets upon which to “train” the AI and can already create images, audio, and video, with more capabilities on the way.  

    I believe we are approaching an inflection point in AI that will allow designers and engineers to work far more quickly and efficiently, stay focused on higher-level thinking and work, and create products that are more sophisticated and sustainable. 

    Stay tuned for future blog posts where I will delve (pun intended) into both “Digital Twins” and Generative AI. 

    Continue this conversation on LinkedIn! Follow Aaron's LinkedIn page and share your thoughts about this blog post here.

    Let's build something together.
    Cortney uwebc event blog hero
    Fused Innovation: How Cross-Disciplinary Teams Deliver Faster, Smarter Products
    Lessons from my Wisconsin Digital Symposium keynote on building digital systems that actually work.
    2025 02 26 Microfluidics Challenges HERO
    10 Microfluidics Challenges in Diagnostics—And How to Solve Them
    Bringing breakthrough diagnostics to life takes more than lab success. It takes engineering that performs in the real world. In this article, you’ll learn how to solve ten of the toughest challenges in microfluidic device design—from fluid control and contamination prevention to scaling and usability. Want the full technical breakdown? Download the white paper.
    Ying ge D6 A Yk HWLR9o unsplash cropped smaller
    Clearing the Path to Step-Change Innovation
    Looking to move beyond incremental product tweaks? This article breaks down the three key enablers of step-change innovation—especially in complex environments shaped by new technology, regulation, and shifting organizational priorities.
    2025 03 17 Alarm Design Medical Devices HERO
    Medical Device Alarm Design Best Practices: Reducing Fatigue, Improving Clinical Response
    Are your alarms helping clinicians—or just adding to the noise? Learn how to design alarms that reduce fatigue, support clinical decision-making, and improve patient safety. This article will cover common design mistakes and actionable strategies—and includes a free quick-reference poster you can download for your team.
    2025 02 28 no shortcut medical device development HERO
    FDA Delays Could Hold Up Your Device. Are You Ready?
    In a climate of regulatory uncertainty, there’s no room for shortcuts. Rigorous risk management, human factors, and a well-structured DHF aren’t just best practices; they're critical.
    2024 12 06 Pugh Matrix
    Download Our Pugh Matrix Template to Streamline Decision-Making in Medical Device Development
    Our team relies on the Pugh Matrix to simplify complex decisions in medical device development. Learn how it works and download our customizable worksheet for free.