By Dusty Weis, Association of Equipment Manufacturers
When you’re designing the components that go into your products, why not consider thousands of different designs instead of just three or four?
Up until the era of cloud computing and artificial intelligence, that question was seldom asked because the answer was fairly straightforward. Design teams have a finite amount of time and money at their disposal, after all, and can’t afford to prototype more than a few designs—let alone thousands.
But what are these constraints costing you, in terms of the ideas that never get tested and the solutions that never get prototyped? Are there unexplored methods of building your product that might be lighter, faster or cheaper?
“There’s really not one solution for every problem; there are many,” says Bryce Heventhal, technical marketing manager for Autodesk, Inc. “This is where the power of the cloud and parallel computing is going to help us test more ideas and look at more concepts in a shorter period of time.”
Heventhal says this new approach to engineering, known as “generative design,” is going to revolutionize manufacturing over the next decade. It’s a process that enables human engineers using computer-aided design (CAD) to define an engineering problem, which is then solved over and over again by an adaptive artificial intelligence program, yielding different results each time.
Generative design was a key topic at AEM’s recent Thinking Forward event at Autodesk headquarters in San Francisco, where Heventhal presented AEM members with a working background on the technology and examples of how it’s already being used to pilot creative solutions to engineering problems.
Steve Van Zee, the engineering manager at AEM member Vermeer Corporation, says the presentation gave him a valuable grounding in technology that could have wide-ranging applications for manufacturers.
“Any time there’s a new development or a better, faster way of developing new products, it’s definitely something that we want to start talking about and understanding better,” Van Zee says.
Not surprisingly, Autodesk’s version of generative design is steeped in proprietary secrecy. But, like many modern applications of artificial intelligence, the basic notion is this: if you pool enough computing horsepower, your network is capable of completing many iterations of very complex tasks very quickly.
In Autodesk’s case, the proprietary “secret sauce” is an AI platform that has been “trained” to create solutions to engineering problems. Unlike human engineers—who first design a solution, then determine how to build it, then prototype and test its properties—generative design is capable of carrying out all three of those steps simultaneously.
The process requires human engineers to define the problem, Heventhal says, by using CAD systems to lay out basic specifications for the component that needs to be designed.
“You don’t start with the geometry,” Heventhal says. “You start where it attaches to other components and how.”
After that, engineers further refine the design parameters, specifying load requirements, deflection, rigidity, material preferences, cost of production, weight requirements and even manufacturing methods. Heventhal says attention to detail is absolutely critical during the process of defining the problem.
“If you input crap, you’re going to get back crap, essentially,” Heventhal says. “But if you actually specify the requirements early and set up to solve the right problems, that’s when you’ll see the benefits of the software.”
“So then, we can push the magic button, hit generate and send it to the cloud, and it will give you a bunch of different answers,” Heventhal says.
And by a bunch, he means a bunch. Working its way through the predefined parameters of the problem, the AI platform solves it over and over, employing a new approach each time. Like an overcaffeinated blueprint artist, the program sketches out new CAD designs, tests them in simulations and learns from its mistakes and successes.
What results is a collection of hundreds, even thousands, of computer-generated designs, catalogued by the degree to which they meet various criteria. Some designs bear similarities to the components they’re intended to replace, while others look like nothing that’s ever been manufactured before—utterly out-of-place anywhere but the set of a science fiction movie.
“You can even watch as it solves it over and over and over, right in front of you,” Heventhal laughs. “It’s actually kind of hypnotic, like watching paint dry.”
A Generative Design Success Story
Autodesk’s generative design platform is still fairly new to its portfolio, but already the company has used it to develop solutions for clients like General Motors, Toyota and Airbus.
One of the most striking examples of generative design’s power is evident in the case of a company called Lightning Motorcycles, a manufacturer that touts its electric-powered motorcycles as the fastest production bikes on the market. As with many electric vehicles, the designers at Lightning are always searching for ways to cut weight from their design, and saw in generative design a way to explore unconventional solutions to the problem.
Heventhal says Lightning specifically asked Autodesk to help cut weight from the swingarm that connects the motorcycle’s rear wheel to its frame. “They basically gave us an existing piece of geometry and said, ‘Try to make this better,’” he says. “So we did.”
With a goal of reducing that component’s mass by at least 15 percent, Autodesk sicced its generative design software on the problem. The result is hard to describe with words alone.
What was a fairly boring, dull metal part has instead been rendered as a skeletal, spindly organic thing that looks more like it was grown than manufactured. Heventhal says they not only lightweighted the previous design by 18 percent, exceeding their goals, but they improved upon the prior design’s deflection performance, too.
“The bone-like, alien geometry takes some getting used to. It would have been a nightmare manually modeling this from scratch,” Heventhal says. “Once you see it, though, it’s super cool, and there’s nothing like it out there in the market.”
It wouldn’t be possible to manufacture the unconventional swingarm design without the use of 3D printing technology, Heventhal says. But that’s just one of the many manufacturing methods that generative design takes into account when it churns out blueprints, and the AI’s human counterparts can specify which methods they have at their disposal when they input specifications.
After it has generated solutions for a problem, Autodesk’s generative design software performance tests them in simulations for buckling, fatigue and failure points. It can even account for the wear and tear of specific manufacturing processes, like the warpage and deformation that can occur in 3D printed designs, saving manufacturers time and money on dead-end plans.
“When you start to metal print a thing, if you screw up, that’s anywhere between five and 10 thousand dollars you’ve just wasted, depending on the scale,” Heventhal says.
What Generative Design Means for the Future of Engineering
There’s one thing that Heventhal wants to stress about generative design—it is not intended to replace human workers. Rather, he says artificial intelligence is a valuable mechanism for expanding, by orders of magnitude, the capabilities of engineers and designers to test new designs and incorporate them into the products they build.
“They’re still making the decisions,” Heventhal says. “We’re just giving them better tools to make those decisions. It’s not going to replace jobs, it’s going to augment jobs.”
Faced with a nearly unlimited supply of generative design solutions, engineers will have the responsibility of sorting through the performance characteristics and combining top-performing features from various designs to meet the specific needs of the product in question. Heventhal says AutoDesk also offers a virtual reality platform to assist in this process.
“Using VR, you can start to move and articulate these things, re-shaping your design in real time,” Heventhal says. “When you’re able to see how a component fits into things on a one-to-one scale in real life, it’s always better.”
Ultimately, engineering is about problem-solving, and the more data engineers have to work with, the more money they can save, the faster they can get to production and the lighter or stronger they can build a component. With generative design and the power of cloud computing, Heventhal says design teams will be able to tap into a data trove that’s deeper than any they’ve seen before.
“There’s really no one solution to any problem,” Heventhal says. “There are tons of right solutions, and we want the engineer and designer to get the information they need to make better decisions earlier in the design process.”
AEM members learned about this and other topics at a Thinking Forward event at Autodesk Headquarters in San Francisco on Aug. 23. Visit aem.org/think to learn about more of these upcoming events in a city near you, including one at Purdue University on October 16.
Dusty Weis is AEM’s strategic communications manager, covering the impact that new and emerging trends and technologies will have on the construction, agriculture and manufacturing sectors. Email him at firstname.lastname@example.org or follow him on Twitter @dustyweis.
Subscribe to the AEM Industry Advisor for regular updates about this and other cutting edge topics.