What Lies Ahead for Generative Design?

Generative design may cause a revolution in product development — not by itself, but as part of a synergistic group of technologies, product lifecycle management (PLM) consulting firm CIMdata believes.

Generative design is sometimes presented as a kind of miracle cure: It eases CAD users’ workload; creates lighter, stronger parts; and effortlessly finds solutions to design problems that are beyond human imagining. But what is it, really — and what might it become in the future?

Keith Meintjes, CIMdata fellow and executive consultant, simulation, led a webinar this month exploring these concepts. In "Beyond Generative Design: A New Paradigm for Product Development" he argued that it is not generative design alone that will dramatically reshape product development, but a confluence of technologies that also includes simulation and analysis, big data analytics, robust design, and advanced materials.

“We and others envision an environment for generative design using or exploiting artificial intelligence that will give us a platform to do generative design — what some people now call human-assisted design,” he explained, meaning “the human is in the loop, but the computer is doing the heavy lifting.”

Meintjes defined generative design as “design space exploration with optimization, and aided by computers. By our definition, it’s any method that creates or affects a physical design: the geometry, the dimensions, the material choice.” Ultimately, it’s a process of optimization, and it’s almost always underpinned by physics-based simulations.

Meintjes finds generative design a very intriguing concept because, “for millennia, ever since the beginning of time, if someone imagined a product, it had to be designed and made and tried out. The last 60 years or so, we have also been able to use simulation instead of physical tests to do virtual evaluation of proposed designs,” he noted. “But generative design proposes to create or modify feasible product designs, including the geometry, from statements of requirements and constraints. So that means that you’re able to go through the generative design process without starting from a proposed design, and end up with a design that is feasible and meets the requirements — which is, I think, quite an astounding thing.”

Generative design encompasses a variety of tools, including rules-driven parametric CAD, shape optimization, cost and manufacturing optimization, and others. Whatever the tool, Meintjes explained, “there is always an underlying simulation application to evaluate the possibilities; so as you’re going through the generative design process, the simulation application is evaluating the improvement in the design as you’re developing it.”

“Topology optimization … is a topic of huge interest today, it’s the dominant generative design tool,” said Meintjes. However, it had low adoption when it was developed in the 1990s, and “it still is very difficult to take the results from a topology optimizer and put them into usable CAD design geometry,” he observed.

Is it 3D printing that has led to current interest levels in a still-flawed technology? That’s only part of the story, according to Meintjes. “What we’re talking about here is a new environment for generative design; it’s not just layering on additive manufacturing to an old technology. We have new materials, we have advances in computational geometry — the algorithms and software — and we have this incredible rise in computing power.” Advances in IT, artificial intelligence (AI), and statistical optimization have also contributed, “but what’s also happening is the synergies between all of these advances that many of us believe will cause a revolution in product development and manufacturing,” Meintjes predicted.

Currently, topology optimization often requires difficult and time-consuming manual intervention for translation into CAD. “If you’re really going to utilize generative design to explore thousands and thousands of possibilities in the design space, having the need for human intervention at some point in each design is really unacceptable,” Meintjes declared.

“The truth is, what we need to is after we’ve completed the design, we need to revalidate that design… all the constraints may not have been considered in the generative design, so we had to alter the design after it was generated [and] just the process of translating generative design into cad may have violated or changed the geometry enough that we need to go through this step.”

The solution lies in an environment, Meintjes explained, where the generative design tools are wrapped with artificial intelligence, machine learning, and robust design capabilities. This can be provided by process integration and design optimization (PIDO) tools such as Isight, Hyperstudy, or ModeFrontier, or within the application.

There are also organizational considerations around the technology, Meintjes pointed out. “For organizations and end users, the question that needs to be answered is, ‘Who’s going to use the software?’” There are various flavors of topology optimization and generative design tools; some are focused on simulation experts or generative design experts, others on the general CAD community. CAD operators will use generative design as they are designing parts, because the technology is embedded in their CAD software.