Generative design: algorithms that propose shapes under constraints
Generative design is a 3D CAD capability that uses AI to automatically create multiple design alternatives from a set of requirements, such as materials and manufacturing processes. Topology optimization, a related method, adds and removes material within a design space to meet performance goals like minimizing mass or staying within stress limits. Both approaches are used in industries from aerospace to consumer products, and they can produce organic shapes that are often manufacturable with additive techniques.
What generative design does
Generative design is a 3D CAD feature that relies on AI to automatically produce optimal designs from a set of system requirements. Engineers specify their goals, including preferred materials and manufacturing processes, and the generative engine generates multiple versions of manufacture-ready designs that designers can explore quickly, as described by PTC's generative design page.
This approach changes traditional design processes because engineers can work with the technology to create better designs and foster product innovation more efficiently. It allows users to explore many design alternatives in a short period, which can help them create more efficient and innovative designs more quickly.
Generative design weighs many factors, such as materials, manufacturing, and performance needs, so it can produce designs that are stronger, lighter, and more efficient for the stated goals. It also shortens the design cycle by automating the generation of many alternatives and evaluating them rapidly.
Topology optimization as a core method
Topology optimization is a method for finding the best layout of material within a given space to meet a performance target while respecting constraints like loads and boundary conditions. It works by adding or removing material and adjusting feature sizes, unlike shape optimization that only tweaks a predefined shape, as explained by Ansys's topology optimization article.
Common goals include reducing weight, keeping temperatures within limits, avoiding certain vibration frequencies, or ensuring stress and deformation stay below allowable values. The algorithm lets the object's performance dictate where material should be placed.
Topology optimization was once limited to research because the resulting shapes were hard to make with conventional methods. The rise of additive manufacturing, particularly metal 3D printing, made it possible to produce these organic forms. Adding manufacturing constraints to the optimization tools also allows production via casting, molding, or forming.
Neural networks for faster optimization
Researchers have explored using deep learning to speed up topology optimization. A paper by Ivan Sosnovik and Ivan Oseledets proposes a deep learning based approach for speeding up topology optimization methods, stating the problem as an image segmentation task, as detailed in the arXiv paper on neural networks for topology optimization.
The authors introduce a convolutional encoder-decoder architecture and demonstrate significant acceleration of the optimization process. The proposed approach has excellent generalization properties, and the paper discusses successful results as well as drawbacks of the current method.
Real-world applications at NASA
NASA has turned to AI to design mission hardware, using generative design and evolutionary algorithms to create components that meet strict requirements. For example, NASA's Space Technology 5 (ST5) mission used advanced evolutionary algorithms to automatically design antennas, as reported in NASA's article on AI-designed mission hardware.
The ST5 antenna was a challenging design problem due to the combination of wide beamwidth for a circularly-polarized wave and wide impedance bandwidth. The evolved antenna had advantages in power consumption, fabrication time, and performance, and it required approximately three person-months to design and fabricate, compared to about five for a conventional antenna, according to the NASA NTRS paper on automated antenna design.
The evolved antenna had an unusual organic looking structure that expert antenna designers would not likely produce. In addition, an S-band phased array antenna element for NASA's TDRS-C communications satellite was evolved, meeting specifications for both receive and transmit/receive elements, simplifying the antenna and reducing testing and integration costs.
| Aspect | Generative design | Topology optimization |
|---|---|---|
| Definition | AI-driven 3D CAD capability that creates optimal designs from requirements | Methodology that adds or removes material to achieve a goal under constraints |
| Input | Design requirements, materials, manufacturing processes | Loads, boundary conditions, material properties, performance goals |
| Output | Multiple manufacture-ready design alternatives | Optimized material layout within a design space |
| Manufacturing | Can consider traditional and additive methods | Often paired with additive manufacturing, but can include casting, molding |
| AI role | Uses AI and machine learning to generate and evaluate options | Uses algorithms; can be accelerated with neural networks |
Frequently asked questions
Is generative design the same as topology optimization?
No. Generative design is a broader CAD capability that uses AI to create multiple design alternatives from requirements, while topology optimization is a specific method that adds or removes material within a design space to meet performance goals. Generative design often incorporates topology optimization as one of its techniques.
Can generative design be used with traditional manufacturing?
Yes. Generative design can be applied to parts that are machined, cast, or forged, not just additive manufacturing. Engineers can specify preferred manufacturing processes, and the generative engine produces designs that are manufacturable with those methods.
What are the benefits of using generative design?
Benefits include improved efficiency by exploring many design alternatives quickly, enhanced performance by considering a wide range of factors, and reduced design time by automating the design process. It can lead to stronger, lighter, and more efficient innovations.
How does topology optimization work?
Topology optimization starts with a solid mass filling the design space. The engineer creates a finite element analysis model, defines how the object is held, what forces are applied, and material properties. The algorithm evaluates each element and removes those that do not experience loading, turning elements on or off to achieve the optimal layout.
Has NASA used generative design for real missions?
Yes. NASA used evolutionary algorithms to automatically design antennas for the ST5 mission and a phased array element for TDRS-C. The evolved antennas met mission requirements and had advantages in power consumption and fabrication time.