Generative design is seen as the latest tool to gain market share. However, adopting it is not as it seems. Here are tips on applying generative technology.
by Brian Thompson, divisional vice president and general manager, CAD, PTC
Generative design is a means of autonomously creating optimal designs from a set of system design requirements, such as loads, constraints, preferred materials and manufacturing processes.
Why is there so much market buzz about this technology?
First, by definition, generative design (generative) promises to unleash fully autonomous design using the power of artificial intelligence. It may even help you become a better engineer. Who wouldn’t find both of those intriguing?
Second, the potential business value is so tempting. Hard-pressed manufacturers are looking for an edge in cost-of-goods-sold, time-to-market, or anything that will keep them ahead of a competitor.
There is truth to the buzz, and I believe you can and should adopt generative design tools. The adoption process isn’t as easy as it seems. Below, I offer several suggestions for applying generative technology based on my work with PTC’s customers.
• Start small. Don’t expect to redesign your entire assembly or product with the push of a button – yet.
• Be aware that you will need to codify the design requirements as fully as possible, and in a language the generative system understands.
• Set reasonable boundaries for solution exploration.
• Be ready to continue to progress your design to production readiness.
• Be prepared to temper your expectations and apply generative where and how it makes the most sense.
I’ll now consider some of these items in greater depth.
Space Requirements–Some requirements are simple and straightforward, such as space limitations. Space limitations are the areas where the system can put geometry, where it can’t, and what geometry must exist in the design (for purposes of an interface, for example). The design space definition is critical because every design solution will honor this fundamental set of requirements.
Human engineers examine the design context, almost intuitively know these things, and instinctively design around them. The generative system, however, only knows what it’s told unambiguously. That means you’ve got some modeling to do.
The best approach is to build this design space right there in the context of your design – right next to all the other parts in the assembly. Ideally, such a design space would even include critical geometric dependencies to the surrounding parts. If those parts change, the generatively designed geometry can be updated in situ.
In that sense generatively-designed parts are no different than parts people design. The parts are still are bound by the priorities and requirements of their geometric neighbors, and the design tool in which the generative technology is deployed must be able to help the engineer manage these dependencies.
Managing other requirements–Other requirements might not be so straightforward. You’ll need to prioritize instead of loading in every conceivable requirement and then overusing – and overspending on – compute resources.
This can happen easily. I think you’d agree that most engineers would probably want a design with minimal stress. It seems simple: tell the system to minimize stress. But what about minimal mass? What about minimizing deflection or maximizing stiffness? How about maximizing the first natural frequency, or keeping the temperature differential to a minimum? The generative system eventually could handle these requests, but you would be better off prioritizing these requirements.
Take an aerospace application for example. It might be better to ask the system to deliver minimum mass, while keeping stress under a value for each material you specify, while keeping the first natural frequency above a certain value, while keeping the temperature differential below a certain value. This way, the system can focus on an “extreme” answer for the highest priority requirement while not violating any other fundamental needs of the design. The result of taking this approach will be a more efficient use of compute resources – so you can explore your design choices more quickly.
You also want to be sure you can manufacture the design alternatives your system generates. With today’s generative systems, however, manufacturing constraints will only get you so far. Fortunately, the parameters that govern them are usually ‘tweakable’ to better align to your company’s preferences. That’s a great place to start. Your efforts will save you time as you progress your generatively-developed geometry to production readiness.
Temper expectations – and apply generative where it can help you most
Step back–When you run a generative design study, the system is doing something you couldn’t hope to do yourself, and it’s using time, money, and compute resources to do so. Step back and try to be judicious in your use of those resources and in the scope of your generative study.
You might want to reconsider if you find yourself in any of these situations:
• You generate design solutions you would not be able to deliver. If you don’t have the time to qualify a new vendor that specializes in titanium extrusions, consider eliminating that option from the solution space.
• You use time and compute resources to generate thousands of design solutions. Do you really need to evaluate thousands of solutions?
• You include load cases that are redundant. Be conscious of the various load cases you’re considering. See if they can be combined so the system has fewer to consider but can still deliver candidate solutions that meet all of your needs. The extra setup time will be worth it in compute time for the entire generative study.
• You allow your biases to influence how you set up the generative study. Don’t discount combinations that you think (but aren’t sure) won’t work well. Similarly, be sure to include widely-used manufacturing processes even if your company doesn’t have internal or external supply chain expertise in it.
Remember modeling fundamentals–They still matter – even for the model of the generative design spaces. Be sure that generative design spaces are defined properly in the context of the overall design that the team is working on. Generative designs, like any other designs, need to abide by interface and space requirements. Use generative within the core design system – never outside it. This will ensure that as the overall product design evolves, your generatively designed parts can evolve at the same pace, in the same context, like all other parts.
Consider materials–Apply generative technology to problems typically solved by the right kinds of construction materials. Generative systems are limited in their understanding of certain types of material characteristics. Today, for example, large, thin-walled structures like tubing or formed sheet metal parts are not ideal candidates for generative systems. Why? The generative systems won’t honor the constant wall thickness sheet metal requirement or the constant cross section requirement of tubing. Those are temporary limitations, and you can choose many other types of parts which are great candidates, such those machined, cast, extruded or forged.
Don’t forget additive manufacturing. If you have the right production volumes and the right equipment and process expertise, you can get much more creative in the application of generative technology.
Progressing to production readiness
Even after a well-conceived generative study and the selection of a final design, chances are you’ll still need to tweak the result. You should engage once again with your manufacturing engineering colleagues to progress the manufacturability of the design as the entire project moves to production readiness.
That means you’ll need a way to shift your generative result conveniently into traditional CAD brep (boundary representation) form. Ideally, this brep transformation can occur right in the same design context, with the same dependencies and relationships to the surrounding design as was defined by the original generative problem scope.
How generative design works today
One example of generative design at use is in the Volvo development process. They wanted to replace a belt-driven fan clutch with a new fully variable electric motor, which required a new mount. The preliminary design for the mount was a three-piece sub-assembly, and their goal was to simplify to a single part.
Volvo wanted to replace a belt-driven fan clutch with an electric motor. However, this required a new mount. The first design for the mount was a three-piece sub-assembly. Volvo engineers were able to simplify the assembly to a one-piece part using generative design.
Working with the generative technology, Volvo created a design space from available volume within the engine, including all mounting points and neighboring components, as well as the motor itself; applied loads and constraints, set various mass targets, aiming for around 1 kg.
The final result was lighter than the original, optimized for machining, and exceeded the target safety factor.
A final word
You needn’t be an industrial conglomerate to see results in your own work. Start small, set up your study with care, and see where it takes you.