Generative Design and CAD

CAD isn’t static. Over the decades, we’ve seen how CAD has evolved from relatively primitive programs allowing only for simple line drawings through to today’s complex, 4D design and BIM software which accurately portrays real-life structures and objects. The evolution of CAD has extended to other technologies as well, generally in pursuit of one goal: more useful and accurate portrayals of human designs. For instance, since the dawn of computing, software has generally been human-directed. Essentially, this meant that humans told programs what to do, rather than programs having autonomy. However, computers are now changing.

With the dawn of the cloud and more powerful processing capabilities, software has access to vastly greater amounts of computational power. They’re using this power to learn how to solve problems for themselves; to learn how to be creative. Such programs no longer need humans to give them commands. Instead, a designer simply needs to describe a problem; software then works autonomously to find a solution. This newfound computer creativity has given birth to generative design. 

Generative design sees computers taking on the mantle of designer, as regards actual design work, creating unique and innovative solutions that are beyond human imagination. The potential is therefore there for a total sea change in how designs are created. In this article, we’ll explore what generative design is, the impact it’s already having, and possibilities for the future.

What is Generative Design?

Generative design refers to the automation of the design process that results in the creation of tens, hundreds, or even thousands of design alternatives that solve a design problem. It performs some of the more mundane aspects of design, leaving the higher-level design thinking to the human professionals. But generative design isn’t fully autonomous; it requires the input of a human designer or engineer. Typically, the professional stipulates input parameters, with the generative design tool then generating designs that adhere to these parameters. The input parameters include but are not limited to:

• Functional requirements
• Design and manufacturing or construction constraints, such as cost
• Performance criteria such as strength
• Material properties
• Size and weight requirements, among others

It is crucial to point out that generative design isn’t synonymous with artificial intelligence (AI). There are aspects of generative design that are AI-related and those that aren’t. In fact, as we’ve detailed below, generative design existed before the increased adoption of AI. During the early stages of its implementation, generative design tools simply relied on the input requirements to generate multiple design alternatives. Today, this is still the case, at least to some extent, meaning modern generative design tools can function without the AI or machine learning (ML) element. Where AI and ML come in, however, is in cases where the tools learn from past designs and creations to generate better and cost-effective designs.

Generative Design vs. Generative AI

It is also worth mentioning that the integration of AI and ML into generative design doesn’t make it part of generative AI. Generative AI is a separate concept altogether. Generative AI refers to AI technology that is broadly applied in multiple industries to produce content such as audio, text, and images. In contrast, generative design is confined to a narrow domain – generating designs of products and buildings – and even then, it doesn’t always rely on AI.

History of Generative Design in CAD

Some historical accounts indicate that avant-garde architects birthed the concept of generative design in the late 1950s and early 1960s. At the time, however, they were simply just aware of its potential because the supporting computational power wasn’t readily available. Expectedly, the technology then wasn’t as advanced as it is today. It wasn’t even concretized as a working technology.

Instead, it started as purely theoretical research without any practical applications. Researchers borrowed heavily from biological phenomena and natural languages to develop and model generative algorithms like cellular automata (1951) and design grammars (1972). But later, the development of computers and software provided researchers with tools to begin actualizing their research, chiefly with the intention of improving their working processes. It is perhaps for this reason that some accounts note that the history of generative design began in the 1980s. (Coincidentally, this is the period when advances in CAD and CAM – as captured in the history of CAM – really began taking off.

Initially, the researchers were interested in how to deploy generative design in architecture. But before long, however, researchers from other fields started taking notice. As a result, they started to investigate possible ways to apply generative design in their respective fields. Professionals also began applying the technology in their respective fields, including mechanical design.

Evolution of Generative Design and CAD Since the 1980s

Like other technologies, generative design evolved progressively since the 1980s aided by the equally progressive advances in supporting technology. This evolution occurred in three phases, with the third phase representing the current dispensation of generative design:

Phase One of Generative Design Evolution

As John O’Connor, the Director of Aerospace Product and Market Strategy for Siemens Digital Industries Software, explained in a podcast episode, the early implementation of generative design in mechanical engineering and manufacturing followed a largely iterative and inspirational design process. It mainly revolved around what is now known as topology optimization.

In topology optimization, designers used generative design tools to create conceptual designs that couldn’t be manufactured. These designs mainly defined the topology of a product and were intended to help optimize its geometry based on loading and stress without considering other aspects such as manufacturability and quality (geometry and surface quality). As a result, and once the generative design tools generated the optimized designs, designers had to rework them manually to enhance their manufacturability.

Then, the development of capable CAD software birthed modeling tools that allowed designers to create interactive designs from the generated designs, making them manufacturable. Simply put, CAD eliminated manual reworks.  

Phase Two of Generative Design Evolution

What followed next was the development of partial generative design. This design approach generated designs based on user-defined conditions only. Initially, the conditions were simple and then evolved to become more complex. The complex conditions leveraged real-world physics, enabling the generative design tools to generate final designs that accurately solved the design problems.

At the same time, the partial generative design tools used different types of algorithms, which enabled the creation of designs that met the conditions to a tee. This meant that the tools incorporated more intelligence than before. As a result, it was now possible for the designers to use the tools to create geometric representations of products as well as associated product and manufacturing information.

Phase Three of Generative Design Evolution

Today, modern generative design tools incorporate machine learning and artificial intelligence. Such tools can now learn from previously generated designs, which were created using constraints and conditions similar to those of the current project, and, therefore, generate better designs based on these learnings. This has meant that the tools can create higher-level designs than before, with these designs promoting savings and better performance.

At the same time, software publishing companies have introduced different kinds of generative design tools. Some have created generative design modules/add-ons for several commercial CAD software applications, while others have included generative design capabilities in their CAD products. Regardless of the approach used, one thing is constant, that modern solutions integrate generative design with CAD. This integration merges the generative design capabilities with the CAD software’s traditional design, modeling, and simulation processes. As a result, you can use the generative design tool to generate designs, have the software run simulations on all of them, and subsequently include the best design in subsequent workflows, all within the same interface. As a result, today’s implementation of generative design is of a higher level than previously.

The third phase has coincided with major modern developments in the generative design space, particularly within the past fifteen or so years. Within this period, software publishing companies like Autodesk, PTC Inc., Siemens Digital Industries Software, and Dassault Systèmes have acquired and/or launched generative design products or capabilities. Some of these releases are summarized in the table below:

Technical Overview and Capabilities of Generative Design

Generative design tools act as co-designers. They’re currently not advanced enough to replace human designers. Even so, they’ve greatly improved the traditional design process. And to understand this impact, let’s discuss this traditional design process and how it differs from generative design.

Traditional Design Process vs. Generative Design Process.

Before the adoption of generative design tools, the design process was quite slow and followed known steps. First, designers presented several designs, including previous designs that contextualized the nature of the project and documented its history. For new projects, these designs could be borrowed from other similar projects. The designers then used CAD tools to evolve the designs (by physically creating them) into new forms that met the newly defined design constraints and conditions.

Next, the designers would evaluate the new designs against performance requirements. If they didn’t satisfy the requirements, the designs would be sent back to the previous stage, creating a feedback loop and resulting in plenty of iterations. Only after the designers and engineers were satisfied that the design met performance requirements was it forwarded to the manufacturing team for fabrication. This meant that the design professionals created the geometry first before then validating it.

In contrast, generative design tools are fed with previous designs to learn about a project’s history and context. Next, the generative design algorithms consider design constraints and performance requirements to explore hundreds or thousands of design possibilities. It is worth pointing out that generative design doesn’t aim to create existing or known solutions. Instead, it is focused on uncovering novel yet functional findings, some of which are beyond the imaginations of human designers.

The tools then generate these novel solutions, providing designers with plenty of design options to choose from, even before they can start physically designing using CAD software. Thus, generative design eliminates the feedback cycles and the plenty of design iterations that characterize the traditional design process, saving time and boosting productivity, as a result.

Technologies and Processes Involved in Generative Design

Modern implementations of generative design leverage the power of machine learning, artificial intelligence, and generative design algorithms to explore multiple design combinations based on user-defined input parameters, generating hundreds or even thousands of design options. It goes without saying, therefore, that generative design involves multiple steps and infuses various technologies. To understand how generative design tackles design problems to produce high-performing, manufacture-ready designs, let’s look at what the process really entails.

1. Generative Algorithms

Like all computational problems, generative design follows a set of mathematical rules or instructions, known as algorithms, that direct the entire process to produce a certain result. Typically, the generative design algorithms take into account the user-defined input parameters (design constraints) to establish how best to develop the design solution that conforms to these requirements.

There are several generative design algorithms (rules) that can be applied, with each rule serving a unique function or takng a different approach to generating the design. As such, a generative design tool combines multiple algorithms to generate designs. Generally, these generative algorithms include:

• Generative grammars: these algorithms include L-systems, graph grammars, and shape grammars. They use transformation rules to develop a shape or object.
• Genetic algorithms: they define the process of design reproduction
• Emergent and self-organized algorithms: they include cellular automata algorithms and swarm intelligence; these algorithms organize a collection of self-organizing components or agents that, by interacting with each other and their surroundings, create the final form of the design.
• Associative generative algorithms: these algorithms define how parametric design is applied in generative design. (more on this below).

Nonetheless, modern generative design tools go beyond merely using generative design algorithms. They also rely on the CAD software’s simulation algorithms, which generate data on the behavior of the product during real-life application. Combining the simulation data with the input parameters, the generative design algorithms create functional designs that satisfy most of, if not all, the requirements that a designer or engineer stipulates.

2. Ideation and Iteration

Generative design tools can generate tens, hundreds, or even thousands of designs that satisfy requirements. This process of generating designs is known as ideation. Ideation provides designers with a panoply of options from which they can choose the appropriate design even before embarking on actual CAD design and modeling. Alternatively, they can use this vast array as the basis for coming up with better input parameters that lead to improved generative designs. To put it simply, generative design isn’t limited. It allows designers to explore alternatives through continuous iterations.

3. Optimization

Optimization comes in many forms, including topology optimization and multi-disciplinary optimization. The former analyzes the load and stress components to find the optimal way to distribute material in a design space to define the part’s or product’s form. However, while topology optimization only creates one design at a time, it supports creativity. It generates technical solutions that are hard to fathom yet are safe and save material. On the other hand, the latter optimizes the designs based on fluid dynamics, electromagnetic simulations, structural analysis, and thermal management. Combining these two, generative design tools create optimal designs that satisfy performance requirements.

4. Human Input and Design Space Exploration

Generative design tools aren’t fully autonomous. They require the input of human designers who must specify the parameters. The designers must also evaluate the designs and select the best ones. Additionally, designers can choose to refine that design or feed it back to the generative design tools. If they select the latter approach, they can input more design and performance requirements to improve the design.

Going beyond the stipulating the requirements, generative design tools also allow designers to interact with and explore the design space. This capability facilitates real-time interaction at all stages of the generative process. What this helps with is that it empowers designers with knowledge of how to come up with innovative solutions that they may not have considered before. Put simply, it supports their creativity by virtually showing in real time that certain designs with complex geometries, which may be difficult for a human to conceive using traditional design methods, are indeed possible and feasible.

5. Parametric Modeling

Generative design software applications are designed to be functionally parametric. They use parametric modeling, a design paradigm wherein the design tool defines relationships between and among parameters (such as dimensions) and then assigns different values to these parameters, generating alternative designs. Moreover, the design software also creates a history tree that stores all the sequences of changes or the steps used to create features. This history tree greatly aids in design automation. This makes parametric modeling, at large, an integral part of generative design.

6. Computational Power

Computation is required to apply systematically the mathematical or visual properties created by the generative design algorithms. This computational power can be provided by local or cloud resources. Some tools, such as the module developed by Frustum, for example, can use both the CPU and GPU to deliver the design output faster. Others, like Autodesk’s generative design tool, leverage cloud computing, which provides near-limitless computing power.

How Companies are Incorporating Generative Design

Whilst still a comparatively new field, generative design is already turning design on its head. This is because humans and machines have entirely different approaches to how to create new designs. Humans typically find inspiration in currently existing designs. As such, what they create is likely to differ only incrementally from present designs, and is likely to be based on current ideas of what works. Machines, meanwhile, focus solely a described problem, and may take unorthodox routes to solving it, resulting in better designs. This characteristic of generative design has led to its adoption far and wide.

NASA

On the left is a high-performance antenna that NASA designed in the 1960s. On the right is a computer-generated design that performs twice as well. Image source: Autodesk Redshift

Take, for example, the two antennae shown above. NASA designed the antenna on the left in the 1960s; created by a human, its form is in keeping with what we know about antennae. However, the antenna on the right actually performs twice as well. A human would’ve been unlikely to try such a radical departure from existing norms; the machine, however, focused solely on which form would get the best results. This is just one example of how generative design could give rise to innovative and effective products.

One reason why this antenna appears so unexpected to humans is that we mostly stick to simpler forms when creating new designs. The machine-generated antenna, meanwhile, looks much more organic. Though this may surprise us, it really shouldn’t. This is because machine learning allows software to test out which design, out of countless possibilities, is best at reacting to real-world forces and stresses, and is most efficient to serve its particular purpose. As such, machines that go through this learning process are essentially experiencing “evolution”—albeit in rapid time.

Similarly, NASA has also worked with other companies to develop lighter parts for its space programs. 

Airbus

Airbus has since 2015 been using generative design to reimagine multiple structural aircraft components. In 2016, the company unveiled a bionic partition, dubbed the world’s largest 3D-printed airplane cabin component. This partition was billed as a significant piece that would help plane manufacturers minimize weight while retaining infrastructure safety and great design. Airbus has subsequently used Autodesk’s generative design tools to develop lighter parts that exceed both safety and performance standards.

Other Manufacturers

Other companies like General Motors (GM), Stanley Black & Decker, and Under Armor have also embraced generative design. For its part, GM intends to couple generative design with additive manufacturing and 3D printing to design lighter parts, including a new seat bracket (where seat belts are fastened). Using Autodesk’s generative design tool in Fusion, the company generated 150 alternative designs and ultimately selected a design that was 40% lighter and 20% stronger than the old design. It also comprised one part (shown below) instead of eight. 

Old Design vs. New Generative Design of a Seat Bracket by General Motors. Image source: Autodesk

Working with Autodesk’s generative design tools in Netfabb, Stanley Black & Decker came up with a crimping tool attachment that was 50% lighter than the old version. Moreover, in an interview with The New Atlas, Autodesk’s Australian manufacturing sales manager, Richard Elving, discussed the impact generative design is having on Autodesk’s work with footwear firm Under Armour. Autodesk used generative design to create a new sports shoe sole whose manufacture would’ve been impossible without 3D printing. This provided a glimpse into the fact that generative design and additive manufacturing go hand-in-hand.

An Under Armour shoe, created together with Autodesk using generative design. Image source: The New Atlas

Benefits of Generative Design

Companies and professionals – both junior and experienced – have a lot to gain from using generative design. This is because the generative design paradigm offers the following benefits:

1. Generative design allows designers to explore and compare numerous alternatives, with this expansive array enabling them to choose the best option.
2. By generating multiple designs, the generative design tools provide novel solutions and findings, some of which are beyond the imagination of human designers.
3. Generative design improves environmental sustainability for companies by reducing the material needed for parts. By working with PTC’s generative design tools, Cummins, for instance, reduced the material by 10-15%, according to PTC.
4. Generative design saves design time, with PTC reporting that the timesaving can be as much as 20%. In the time a designer can come up with one design, the generative design software can generate thousands.
5. The timesaving, coupled with the generation of thousands of designs, enhances productivity
6. Generative design promotes collaboration, as it needs the input of design, simulation, and optimization teams
7. Generative design can consolidate assemblies that previously were made up of multiple parts into a single part. It achieves this by designing products for additive manufacturing, which can fabricate parts with complex geometries.
8. It can create lightweight parts. For instance, when it first embraced generative design, Airbus generated a design that was 45% lighter than the traditional part yet just as strong.
9. Generative design supports better creativity by generating alternative designs from a given number of constraints, requirements, and conditions.
10. Generative design can generate designs with complex geometries, some of which may be difficult or impossible for humans to conceive. As a result, this design paradigm facilitates the creation of intricate shapes.
11. It supports customization, as it generates designs based on input parameters. So, by adjusting one or two parameters, generative design tools will simply generate the corresponding design.
12. Generative design enhances performance, especially because it generates designs that align with specific performance requirements. For instance, Autodesk reports that the design paradigm delivers 30% stronger products than those designed using traditional design methods.
13. Generative design bridges the skills gap by enabling even novice designers and engineers to generate expert-level and optimized parts.

Challenges Affecting Generative Design

While generative design offers numerous benefits, it isn’t without a few challenges, namely:

1. The early generations of generative design were mostly suitable for additive manufacturing. They couldn’t be applied to other types of manufacturing. However, modern generative design tools have solved this challenge, according to PTC. Designers can use them to generate designs for CNC machining, casting, and molding.
2. Possibility of poor designs: A great design greatly depends on the input parameters. This means that poorly defined parameters result in poor designs.

What Does the Future Hold for Generative Design?

Autodesk foresees a wholesale revolution in manufacturing; a future where we create the most effective possible designs with minimum waste. The company also predicts that generative design, coupled with generative AI, will help humans take design to new heights. The premise for this prediction lies in the fact that generative design delivers precise designs. It also delivers improved designs, with Autodesk reporting that it can generate designs for engineering products that are 30% stronger, 40% cheaper, and 40% lighter and use 40% less materials. And with generative AI tools generating data-powered yet imprecise designs faster, Autodesk predicts that the future will see the merging of the capabilities of generative design and generative AI. The result of this combination is expected to be the delivery of precise designs faster, further boosting productivity.

Conclusion

It’s clear that generative design has had – and is expected to continue having – huge implications for the CAD and manufacturing industries. In transforming computers from design tools to designers, computers are beginning to truly aid design as never before. Indeed, generative design doesn’t replace designers or engineers. Rather, it complements their work. Generative design tools use the designer-defined functional requirements, constraints, and design conditions to create a vast array of design alternatives, giving human professionals options they can subsequently use or refine. As a result, it has empowered engineers and designers with the ability to design and develop products that are stronger, cheaper, lighter, and use less materials. It also supports customization, better creativity, improved productivity, and promotes collaboration, just to mention a few benefits. Little wonder then that organizations such as Under Armor, General Motors, Stanley Black & Decker, Airbus, NASA, and more have embraced the technology.

But generative design isn’t without its challenges. For instance, the quality of designs depends on the input parameters, meaning the designer must possess the necessary skills to input the requirements. Additionally, earlier implementations of the technology generated designs that were restricted to additive manufacturing. Nonetheless, current versions support CNC machining, molding, and casting. Even so, the adoption of generative design isn’t expected to slow down. In fact, industry players expect that it will be combined with generative AI to produce precise designs faster.