Every gram of metal in an engine part is there to carry a load — and a generative-design algorithm can work out exactly which grams those are. This exhibit runs that algorithm live: it starts with a solid billet of metal filling the piston's envelope and, over and over, asks where the stress actually flows, then carves away the metal that is not pulling its weight. What is left is a bionic skeleton of load-bearing struts — the same process that produced the 3D-printed, topology-optimized pistons now appearing in high-performance engines.
Nothing here is sculpted by hand. Each step is a real finite-element stress solve followed by the SIMP density update that structural engineers actually use: the piston is divided into a grid of little elements, each given a "density" between empty and solid; the solver computes how the whole structure flexes under the combustion load pressing on the crown while the wrist-pin bosses hold it; and every element learns whether it is earning its mass. The voxels glow from blue — wasted, low stress — to red, a critical load path, and dissolve where they are not needed. It is exactly the four-beat rhythm of training a neural network: a forward pass (the stress solve), a loss (the structure's compliance), a gradient (which grams matter), and a step that nudges the design — which is why this room sits next to the Neuron Lab. Push the material budget too low and a strut overloads and fractures: the structural version of overfitting.
Press "Reveal the ideal" and the blocky live computation resolves into the smooth, hollow, finished part the optimizer is reaching toward — a lightweighted piston with its cooling gallery and gudgeon-pin bores, every surface following the flow of force. Generative design, gradient descent, and the logic you can wire up gate by gate in DigiSim's Logic Lab are the same idea: let a simple rule, applied relentlessly, discover a structure no one drew by hand.
Specifications
- Method
- SIMP density-based topology optimization
- Physics
- Live finite-element stress solve in the loop
- Analogy
- Topology optimization as gradient descent
- Simulated
- Voxel density field + von Mises stress heatmap
More from Generative design
