Enhancing Generative Design with Topology Optimization

Generative design facilitates more freedom in manufacturing by automating design processes. While topology optimization can be considered a form of generative design, its distinct nature makes it well suited for incorporating into a hybrid approach combining the two processes that can be used to improve products, particularly in terms of performance.

The main similarity between generative design and topology optimization is that they both involve using computational software to generate designs based on criteria defined by the designer, as opposed to the designer manually creating each design iteration. However, a generative design process differs from topology optimization in that it places more emphasis on the designer’s requirements for the end product’s presentation. For this reason, it is often called a rule-based process; the designer defines values for the design, and the software adheres to them. Topology optimization, on the other hand, is employed more often when the focus is on product performance rather than presentation. Furthermore, it is more physics-based than rule-based, meaning the optimization adheres to the physics of the problem rather than primarily following the rules set by the designer.

With one process prioritizing presentation and the other prioritizing performance, taking a hybrid approach that combines generative design and topology optimization allows for producing an end result with sufficient performance and optimal presentation. In addition, incorporating topology optimization rather than using a purely generative approach can result in an end product that is more sustainable, as designs created with topology optimization are based on objective criteria and thus tend to be timeless.

A hybrid approach could look like using generative design and topology optimization for different components of one larger product. Most commonly, a majority of the components would be produced traditionally, and then generative design and topology optimization would be added for only a few, specific components. It would fall to the designer to decide how to balance the needs of different components, such as determining which parts could sacrifice presentation for stronger performance, and vice versa.

Taking a Closer Look at Topology Optimization

When using topology optimization, designers create a geometric virtual space (often a box-like frame) for their design and input some required parameters into the software. The software then fills out the design by iteratively removing and adding material, producing an optimized design that fits within this geometric space.

Topology optimization is typically employed early in the design process because there is more flexibility at this stage and thus more potential for performance improvements related to topological changes of the design layout. The closer a project gets to the production stage, the fewer design changes can be made, unless the designer is willing to spend more resources to revamp the design. With topology optimization, the designer can decide on an optimal design with good performance and then, as they get closer to the production stage, shift their focus to simulation accuracy. Though simulation accuracy is always important, it is especially relevant in the latter half of the design process, when designers are using the simulation to predict how the device will work once it becomes a physical prototype.

To better understand the design freedom of topology optimization, consider the following example of creating the structure of a drone. For this model, we input two load cases, the volume fraction, and a minimum length scale.

In this case, our goal is to maximize stiffness for a certain amount of material. The load cases are symmetric, which allows us to save time and computational resources by only modeling a quarter of the domain rather than the full drone. The model is then replicated during the results visualization step so we can see the full, optimized design at the end of the process (Figure 1).

When running this topology optimization example, the computation starts with only the drone battery at the center and the four motors at the corners. We see the quarter that was originally modeled mirrored in the rest of the design and watch as the software adds material to connect the motors to each other and the battery at the center. The stiffness-to-weight ratio of the material is then adjusted until the end result has a clear physical interpretation.

Suitability for Manufacturing

When it is time to produce a design, its complexity is a key factor in deciding which manufacturing process to use. The unique results of generative design and topology optimization are usually not suitable for traditional manufacturing and mass production. For this reason, designs created with these methods, or a hybrid technique, often go hand-in-hand with additive manufacturing. For instance, the drone model example is best suited for additive manufacturing.

However, this compatibility does not mean that additive manufacturing is the only option for bringing software-generated designs to life. Often, designers may want to use topology optimization to improve on their design ideas but also need to manufacture products at a lower cost or larger scale than what additive manufacturing usually permits. Preparing such a design for traditional manufacturing may mean, for example, that milling constraints need to be considered in the optimization process. Specialized functionality for topology optimization available in the COMSOL^® simulation software can be used to account for such constraints.

Consider a scenario where a topology-optimized wheel rim design needs to be produced using subtractive manufacturing. Figure 2 shows such an example modeled with the COMSOL^® software: A wheel rim design is generated with optimal stiffness, and milling constraints are added along its axis. However, adding milling constraints reduces the design freedom and thus the stiffness for a given mass constraint. In this case, adding milling constraints results in a design that is 30 percent less stiff than it would have been if generated with conventional topology optimization. This compromise is necessary in order to meet the manufacturing requirements.

If complete design freedom were allowed, the optimal design would generally have the same symmetry properties as the load cases. However, it can require many load cases to achieve symmetry — and a large number of load cases can result in a high computational cost. In this example, design symmetry is expected since our wheel needs to be able to rotate, but we face the challenge of not having symmetric load cases. For this reason, the entire wheel needs to be modeled in every optimization iteration in order to see how the load cases affect the design.

To achieve symmetry here, we can enforce sector symmetry of the design explicitly by optimizing one of the sectors and then copying the design over to the other sectors during the optimization. Similar to how using symmetry features aided in the design of the drone model, this makes the process computationally cheaper and more efficient.

In the end, we obtain an optimized design that has good performance and still meets the manufacturing requirements.

Conclusion

Like generative design, topology optimization automates the design process, enabling designers to explore options more efficiently than with manual iteration. Both generative design and topology optimization can be leveraged with the COMSOL^® software, which also provides features for customizing automated designs so that they are suitable for specific manufacturing methods. These capabilities are applicable to all fields of engineering and scientific research and for any physics area, including structural mechanics, fluid flow, heat transfer, and acoustics. For instance, the software’s topology optimization capabilities are actively being used in the automotive industry for the design of electric motors.

COMSOL and COMSOL Multiphysics are registered trademarks of COMSOL AB.

This article was written by Beth Beaudry, COMSOL, Inc. (Burlington, MA). For more information, visit here .

---