In the never-ending quest to create better products, the latest tool is a technology called “generative design” (GD). A GD algorithm receives high-level requirements as input and generates an optimal design as output. A typical scenario is minimizing the weight of a mounting bracket subject to some structural requirements.
I can tell the GD system what stiffness I require, and an algorithm will then generate the bracket shape that minimizes weight. This is typically done using a topology-optimization algorithm, which gradually whittles away material where stresses are low, thereby reducing weight without compromising strength.
The results are often beautiful organic shapes.
This sounds fabulous. What could go wrong?
First, most part design is a lot more complex than just minimizing weight subject to strength constraints. In real designs, the requirements are enormously varied and complex. You might be concerned with the aesthetics of the part, safety, manufacturability, service and conformance to laws and standards. Use of additive manufacturing removes many design constraints (and introduces new ones), but there’s still a long list of other concerns. Typically, there’s no way to communicate these requirements into a “black box” GD algorithm. So it can’t satisfy them. We need a more open mechanism for communicating requirements to the GD algorithms.
Second, even if we’re just concerned with minimizing weight, there are many ways to accomplish this. Using traditional topology optimization to whittle away material is one way. But it might be better to replace solid material with lattice structures, or solid beams with hollow ones. Or, in thin-walled shell parts, we might vary the wall thickness or reduce it and add structural ribbing where needed. All of these are good ways to reduce weight, but they won’t be considered by typical GD algorithms. We need to be able to tell the algorithm that it should do more than just whittle away material.
Finally, topology-optimization algorithms produce radically different answers depending on what starting shape you give them. Some GD algorithms just produce a single result and pretend that it’s optimal. Others try various different starting points, produce several different answers and pick the best one by some heuristic. The more honest ones confess to not knowing what “best” means, so they ask you to choose.
The three problems arise because a typical GD algorithm is a magical black box. You have no idea what it’s doing internally and certainly no way to control it.
There is a better approach. Using nTopology software, you decide which requirements should be considered, and the techniques used to generate geometry. You create “workflows” that automatically generate designs in a deterministic way, and you control exactly how the “best” design is selected. This increased transparency, and control is the key to successful GD.
A good example is the work done recently at Yamaichi Special Steel. The firm was able to make highly complex designs (including lattices & organic shapes), but actually designing them was difficult. The optimal geometry is application-specific and had to be found by tedious trial and error. So, Yamaichi created workflows that automatically generate thousands of potential designs in hours. The firm then used a proprietary application to pick the optimal structure for a given application. The folks at Yamaichi Special Steel are succeeding because they control the GD process; it doesn’t control them.
Connect With Us
