and Evaluation
Generating infinite meaningful designs
The design generation stage combines the strengths of
physics-based generative design and data-driven deep learning
to automatically generate large numbers of designs that are both engineering-sound and aesthetically meaningful.
In particular, topology optimization and parametric design techniques in engineering design
can be organically integrated with deep learning to create powerful synergies.
Set the initial design geometry and reference model.
Parametric design, topology optimization, and deep learning
are combined to generate seed data and synthetic data.
Data is generated considering both aesthetics and engineering performance.
Thousands of data samples are typically required for AI training.
Yet real-world usable 3D data may number only in the dozens.
Narnia Labs' generator enables
data synthesis even with a small amount of data.
Human-defined parameterization struggles to express
diverse designs. Parameters extracted by the generator from data
can represent diverse designs, enabling exploration of a
broader design space.