Generative design software: a different approach to designing

Getty Images

Users of generative design software don’t start a design process with ideas based on their own knowledge. Instead, they enter design parameters such as size, weight, manufacturing method, and cost constraints. The software then uses artificial intelligence to generate design options based on the input.

Consider how generative design software might be used to design a chair. Instead of starting out by drawing a couple of possible designs, the user tells the software that the desired chair will be made of X material, support a weight of Y, and cost Z dollars. The software then comes up with one, hundreds, or even thousands of manufacturable chair designs that meet these criteria. Most, if not all, of the generative designs wouldn’t have occurred to the software user.

Generative design programs used to be independent of CAD. The situation is different today, however. CAD companies offering generative design as an integrated tool or plug-in that can be purchased as an add-on to a CAD suite, according to Chelsea Cummings, a senior advisor for The Barnes Group Advisors (TBGA), a Pittsburgh-based firm that trains, advises, and assists clients involved with additive manufacturing.

A user of generative design software, Cummings thinks the programs available today are “pretty simple” and that new users can quickly become proficient. The learning process is made easier by features in the software itself.

For example, Cummings points out that MSC Software Corp. has integrated instructional videos into its Apex generative design software. When users hover a mouse over one of the software’s tools, a video plays that teaches them how to use it.

New users of generative design software also can find help in online videos, noted Jesse Coors-Blankenship, senior vice president of technology at Boston-based PTC Inc. In 2018 PTC acquired his startup, Frustum Inc., which developed a generative design program called Generate. Generate will be part of the Creo 7 CAD package that PTC plans to introduce later this year.

Setting Boundaries

One of the most important things users of generative design software need to master is learning to create the boundary volume for a product they want the software to design.

“How good the designer is at creating the boundary volume really determines how good the part will be,” said Kevin McClintock, a CAD/engineering consultant.

The importance of creating the boundary volume for generative design software to work is illustrated by an experience McClintock had while working as a CAD engineering consultant for the Volvo Group North America. During his stint with the automotive company, he used generative software to design truck engine parts for a project aimed at improving freight efficiency.

A member of The Barnes Group Advisors team used Generate software to redesign a robot arm base. The generative design software reduced the mass of the original base by 67% and the build time by 40%. The design of the new base, which was 3D-printed from ULTEM 9085 on an Ultimaker 3, calls for laying down material only where load paths exist between mating surfaces. TBGA

One of these engine parts was an electric fan mount. The original generative design showed that the stresses on ULTEM 1010, the material McClintock wanted to use to 3D-print the fan mount, would have been too high. So the part was CNC-machined from high-strength aluminum at a cost of $4,500.

Later, however, McClintock revisited the generated result and made design modifications that reduced the maximum stress to about 40% of that in the original design, bringing it roughly within the strength limits of ULTEM 1010. “Understanding boundary volume and how to give the generative design tool everything it needs is how I was able to reduce stresses enough to make a printed plastic part viable,” he said.

The quote to print the optimized design in ULTEM 1010 came in at just under $1,000. So had McClintock understood generative design better originally, the manufacturing plan for the part would likely have been different.

Paired With AM

Generative design software, unshackled by the constraints imposed by other design software, is being touted as a good match for additive manufacturing processes.

Generative design processes can produce parts that experience lower stresses during use, which allows the parts to be made with less material. But the resulting costs will be lower only if the generative design is paired with the right manufacturing process.

Generative designs tend to be intricate and unorthodox, which can make them difficult and expensive to produce using subtractive manufacturing processes. That’s why generative design and AM go together in many people’s minds.

Generative design and AM proved to be a winning combination for a bridge bracket McClintock designed for the Volvo Group. “For a truck design with a very aerodynamic front end, the bridge bracket holds engine components in places that they’re not [held] in a normal truck,” McClintock said.

Originally, the bridge bracket was supposed to be made of sheet metal, but the sheet metal would be too heavy and lack the strength needed to meet project requirements. So McClintock embarked on a generative-design process that called for printing the bracket from ULTEM 1010, chosen for its heat resistance and strength. When the design was complete, it took him 21 days to print the large part in pieces with his used Creality CR-10 printer.

“It was the first time anyone at Volvo had seen anything like this [part],” he said, adding that people working in traditional engineering environments are surprised by what emerges from a generative design process. “It’s a lot more like sculpture than it is engineering, though there’s a lot of engineering behind it. No one ever looks at it and says, ‘Oh, that’s nice.’ Everyone has an opinion because the parts are so radically shaped.”

Generative design software user Chelsea Cummings thinks the programs available today are “pretty simple” to use. The senior advisor at The Barnes Group Advisors added that the learning process is made easier by features in the software such as integrated instructional videos. TBGA

Finite element analysis showed that the bracket was stronger than it needed to be, so McClintock redid the generative design. The boundary volume he created included 1,500 features. “It allowed the plastic part to grow into every space it needed to get to as efficiently as possible, so I could get the stiffest and strongest part possible,” he said.

The first print of the bracket weighed 16 kg, but the final version weighs just 10 kg and looks completely different than the first design. According to McClintock, the plastic part is stiffer than the sheet metal version and weighs only a quarter as much.

Two of Stratasys Direct’s Fortus 900 FDM machines will print the plastic part. The bracket is too large to print as one piece, so it will be printed in three sections and then assembled.

A Reason to Wait

Today, Cummings estimates that half the AM industry uses generative design. Among those not using the software are companies that opt for AM to reduce lead times for existing parts. In many cases, these firms don’t want to optimize part designs for AM using generative design “because it’s expensive in engineering time and effort to change an existing design,” she explained. “I think a lot of companies are waiting to use generative design for brand new products because they haven’t invested in engineering time for those products yet.”

Coors-Blankenship sees generative design as an effective design-for-AM tool. “We hope it cuts down the amount of trial-and-error failures it takes to get to a design that will manufacture well,” he said, which translates into raw material savings for businesses.

Today, Coors-Blankenship believes generative design programs are probably better suited for 3D printing processes that don’t involve complex heat loading. With direct metal laser sintering (DMLS), for example, “you can get a lot of warping with each layer if the [part] geometry has big heat sinks in different places,” he explained, adding that current generative design programs don’t offer process simulation of DMLS. The consequences can include more trial and error to get to an acceptable design, as well as potential build failure.

Coors-Blankenship looks forward to a time in the near future when generative design programs will be able to produce designs that most if not all of the time will manufacture well on whatever machine is selected by the user. When that capability is available, he said, “we will have even greater adoption.”

Slide Show | 8 Images