Minimizing Structural Vibrations via Guided Diffusion Design Optimization

Structural vibrations are a source of unwanted noise in everyday products like cars, trains or airplanes. For example, the motor of a car causes the chassis to vibrate, which radiates sound into the interior of the car and is eventually perceived by the passenger as noise. Because of this, engineers try to minimize the amount of vibration to reduce noise. This work introduces a method for reducing vibrations by optimally placing beadings (indentations) in plate-like structures with a guided diffusion design optimization approach. Our approach integrates a diffusion model for realistic design generation and the gradient information from a surrogate model trained to predict the vibration patterns of a design to guide the design towards low-vibration. Results demonstrate that our method generates plates with lower vibration energy than any sample within the training dataset. To enhance broader applicability, further development is needed in incorporating constraints in the outcome plate design.