Dassault Systèmes has sponsored this post.

Product designers are faced with faster design cycles and more demanding design requirements. Their designs have to be lighter, cheaper, better looking and more customized. But the tools engineers use offer little help. CAD, for example, will accurately detail geometry but cannot help engineers think of new designs.

Enter generative design. With generative design, computer-aided design can live up to its name and be used for actual design, not just detailing and documentation.

An advantage of generative design is that it can be used to reduce cost. This is done primarily by using less material, because a shape can be optimized so that no material exists where it is not needed.

Generative designs, by their nature, also create shapes that might be considered more pleasing, organic or even artistic. We are in a place where straight, flat and perfect shapes have become commonplace, even boring. Whereas a traditional designer may favor the straight and true, the novelty and rarity of generative design-produced shapes that are curvy, shapely, organic and irregular or bumpy, will draw the most attention.

Design Exploration

Letting a computer generate away, constrained only by a do-not-exceed envelope, boundary conditions and forces, is a spellbinding affair. You can start with a block of material and watch the program remove material that is not needed.

Underlying the optimization is a lot of math. If a part needs some weight loss, a finite element analysis is being done to determine where there is no stress, meaning it’s okay to remove an element there because any material there adds weight and serves no useful purpose.

Of course, you should run another stress analysis when the generative design process is done. Stresses tend to flow and change course with changes in geometry, and you want to make sure the stresses have not concentrated elsewhere.

Why Do Generative Design?

What generative design does is not conceptually difficult. Without the benefit of generative design, mechanical engineers and designers will think of a shape, or several shapes, that fit the design criteria. Maybe they will sketch them on paper. Then they will detail the best shape with their CAD program.

But have you thought of all possible shapes? Do you think you already know what works best? Why mentally iterate on a design when you can pull out a shape you designed previously—one that you know worked? What if you are hampered by a bias towards a certain shape? Maybe a bias towards the familiar?

Engineers are by nature, conservative. Why experiment when you have something that’s tried and true? An experienced engineer, therefore, may not be one to do any design exploration at all.

Now consider a generative design. Like a baby, it knows nothing of what has happened before. It is wide eyed with wonder, unbiased and will try all sorts of things. A generative design program is design exploration in the most unrestrained and unbiased sense. With the speed generative design is capable of, it can explore many variations; with the power available, it can check to make sure all performance criteria are satisfied, whether it is to be as stiff as possible or as light as possible.

It’s only with a complete lack of bias, optimization techniques, trial and error, and speed and power, that one can be sure of the possibility of finding a design that is truly wonderous, one that not only works but is elegant and aesthetically pleasing. If a camel is the result of a design requirement by a committee, could generative design have come up with an Arabian horse—an animal that is faster, lighter and more beautiful? Or, make the requirement “survivability on long trips through desert” and perhaps you generate a camel.

The power of generative design and the multiple solutions it provides may be thought of as a generous and helpful suggestion from a committee. The designer does not have to design alone, but is helped by a committee of generative designers, providing alternatives for the designer’s approval, any time of the day no matter where you are -- thanks to the cloud-based platform.

And due to the design environment afforded the designer by a platform solution, such as 3DEXPERIENCE, the flow from suggestion to incorporation occurs without model data loss, from modeling to simulation, or vice versa, without degradation or even much effort by the designer.

Performance Driven Design

Performance driven design, or generative design, has been a topic of academic research for decades. While optimization principles can be applied to any discipline, it appeared first to the design community in the field of architecture.

Performance driven design is basically design with a feedback loop that will let the design be altered. Repeat this often enough—letting the design go one way, assessing the performance and adjusting if needed—and the design will evolve. If the performance is allowed to improve, an optimal design can be eventually be reached.

The design will then be perfectly suited to the environment it will be placed in. In the case of architecture, it is a building that is optimized for the climate, whether it be the hot glare of the sun or rising water. This is in contrast to a building that is the same building as anywhere else, created for the convenience of the architect using an age-old design for a constructor that will only work with I-beams or 4-by-8 foot pieces of plywood.

In the case of product design, a performance driven design of a part will have material in the right place for the loads the part will experience, as opposed to making every part out of sheet metal simply because that is the only material you have and the only material your shop works with.

Generative Design Built on Simulation

As useful and simple as generative design appears to be in concept, the theory and application has made it difficult to use for design engineers. While optimization theory is mathematically gnarly, it can be hidden under the hood. Still, doing shape optimization entails using solvers that solve for stresses, flows, temperatures or other parameters common to simulation codes, such as FEA. They could be solving an FEA solution at each step. Almost unwittingly, a designer had become an FEA user, without any knowledge of the terms and theory of FEA, but still expected to understand loads, boundary conditions and restraints.

Biomimicry and Bone Growth

Many topology optimization codes are based on bone growth algorithms, which may seem natural and a good way to go. While biomimicry sounds good at first, most of nature’s designs are not easily manufacturable—at least not with traditional machining.

Bone growth can produce bumpy, gnarly shapes. Fractured bones heal by sending tendrils across gaps, which fuses the bones together, but with a bump. The bump will be smoothed out over time, whether it is computer time with HPC cycles or years with a broken collarbone. In the interest of time and money, true smoothness may not be achieved, which means we often end up with a shape that can only be made with a 3D printer.

Thanks to an integrated design and manufacturing platform on the cloud, the back and forth between generative design and design for manufacturing is more agile. Designers and manufacturers from different companies can share the up to date design instantaneously and collaborate, avoiding delay between designing groups as well as the manufacturing facility. A part is being analyzed while it is taking shape in generative design. It has to be studied and smoothed out to be machined, otherwise the machine shop will throw it back over the wall into your inbox with a message of, “We can’t make this!”

Keeping it all under one roof—optimization, simulation, design and manufacturing—is what Dassault Systèmes calls a digital continuity integrated into a design environment – and aims to provide a seamless integration with the 3DEXPERIENCE platform. This is made possible with a cloud infrastructure. A cloud infrastructure can “liberate” small or medium business from the overhead involved, such as installation, maintenance, license management, etc. for those who use it.

The Process

You can start in CATIA with a basic solid model or dumb geometry (no features or history). Should your solid model be smarter, you can defeature it. That will be necessary to avoid fine meshing and the subsequent drain on computer resources when every little detail is meshed. Forces and constraints are applied, and functional specifications come in with the design.

Start with a dumb model. (Picture courtesy of Dassault Systèmes.)

Just like an FEA, you apply constraints and loads. Like generative design, you specify a target parameter. You can minimize volume, maximize stiffness, or something else. You also tell it what manufacturing process will be used to create the part. (Image courtesy of Dassault Systèmes.)

Didn’t need all the material, after all. The generative design is almost done. Here you can see it is leaving only the material needed to withstand the given loading. (Image courtesy of Dassault Systèmes.)

Which One is Best?

Getting a generative design program is something a computer can do too well. Generative design can generate a lot of designs—possibly too many—all of which will meet the specifications. The program will generate one design and stop when it meets the set specifications, then it will start again. Because the paths taken towards optimization can fork one way or the other, at random and over many forks, there are too many possible combinations. How can you select the winner?

Imagine a flip-book with a thousand pages; that’s a lot to flip through. Each design is distinct, but not really all that different.

How to sort it out? If you have an app that makes a thousand designs, can you have an app to help you sort them out afterwards? Maybe you can put them in order of least weight, or classify them according to a specific performance metric.

Compare variations, sorting the results of the generative designer using KPIs (lower left), which can be mass, various stresses (like Von Mises), volume, strain energy and others, plus an overall score. (Image courtesy of Dassault Systèmes.)

Pick your process and see the stresses in each. (Image courtesy of Dassault Systèmes.)

With Dassault Systèmes’ Functional Driven Generative Design, engineers can create multiple variations for comparison and conduct trade-off studies. Varying the loads, constraints, goals (such as weight reduction) and specifying the manufacturing processes (molding, forging, machining and additive manufacturing) will each create a variation. Designers can then compare and assess the resulting design using key performance indicators, such as weight, to select the best version.

The generative design is used as a base on which smooth surfaces are overlaid and patched together using CATIA modeling tools. (Image courtesy of Dassault Systèmes.)

The generative design parts can be shown in context of the assembly. You can look at each variation one by one. Here we see the 3D printed variation and can compare it to the cast part. It is a nice visual check. (Image courtesy of Dassault Systèmes.)

Finally, A Little History

Dassault Systèmes has a few functional generative design products, which are included in the newly introduced Roles that the company suggests. Roles are like job functions. For example, one role could be Fastener Designer. Another role is Function Driven Generative Designer. The former has no use for generative design, while the latter is defined by it.

The Function Driven Generative Designer role integrates 21 Dassault Systèmes applications, including CATIA Functional Generative Design, CATIA Functional Generative Design (TOSCA) and CATIA Generative Wireframe and Surface, as well as the old classics for the designer, including wireframe, surface and solid modeling, assemblies, 2D drawings and more.

TOSCA a generative design program created by the German company FE-DESIGN, which was acquired by Dassault Systèmes in 2013. Back then, FE-DESIGN’s two main products were TOSCA Structure and TOSCA Fluid. Dassault Systèmes had already tried used FE-DESIGN technology inside it’s ABAQUS application, calling it ATOM (Abaqus Topology Optimization Module).

The TOSCA generative design can start with your initial shape and “nibble away at it,” in the words of industry analyst Monica Schnitger, until what is left is just enough to have the part do what it was supposed to – and nothing more. (Picture courtesy of video from TECHNIA UK.)

With Dassault Systèmes products, engineers can also get workflow assistance. The geometry created by generative design is available for use in detailed design and manufacturing. The shapes generated are not triangular quad meshes, but true solid models. They are parametric, as well.

With CATIA, the same environment can have a parametric solid, a surface, a subdivision (subD) model and direct modeling.

The generative design takes place knowing the manufacturing process that will be used to create it. These manufacturing processes can be checked off: molding, forging, NC, 3D printing and casting, and the design that emerges will be specific to that process.

The Generative Design Process

The conventional process to create a part with generative design is to let the software automatically generate function-driven conceptual and detailed organic shapes. The shapes are “science-based” shapes that can be refined with parametric optimization. An analysis of the resulting shape should guarantee that the refinements keep within the parameter of the design.

To learn more about 3DEXPERIENCE on the cloud, visit Dassault Systèmes.