Generative Topology Design for Faster Structural Alternatives
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional topology optimization methods require significant time and effort to generate alternative designs due to the need for user input and computation of new topologies for each design variation, limiting the variety of designs that can be produced.
Innovation Solution
Employing generative design using deep learning and topology optimization, which involves training a generative ML model with intermediate and final results from topology optimization, allowing for the rapid generation and ranking of numerous design alternatives that satisfy physics constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional topology optimization solver is used to generate alternative designs, then each design can be optimized for material distribution, but the process requires significant time and user input for each design variation
Solution Approach 1:
The system performs preliminary action by pre-computing and storing intermediate results from topology optimization (such as strain energy values, stress distributions, and material density maps) before they are needed for design generation. This allows the generative model to query and utilize these pre-computed results directly, eliminating the need to re-run time-consuming optimization solvers for each design variation, thereby dramatically increasing design generation speed.
Solution Approach 2:
The system creates a digital copy of the topology optimization process by training a generative model on intermediate and final results from optimization solvers. Instead of repeatedly executing the heavy computational optimization process, the system uses the trained model to generate alternative designs by copying and varying the patterns learned from previous optimizations, achieving rapid design iteration without re-computation.
2Adaptability or versatility
If traditional topology optimization is used, then material distribution can be optimized for mechanical strength, but the variety of alternative designs is limited due to requiring user input parameters
Solution Approach 1:
The system applies self-service by enabling the generative model to automatically generate diverse design alternatives without requiring extensive user input for each variation. The model uses its trained knowledge of optimal material distributions and structural patterns to autonomously create multiple design options, with users only needing to provide high-level constraints such as target material usage percentage or functional requirements, rather than detailed geometric parameters for each design.
Solution Approach 2:
The system exploits parameter changes by training the generative model on a diverse set of topology optimization results with varying material density parameters, volume fractions, and structural configurations. This allows the model to generate a wide variety of alternative designs by sampling different parameter combinations from its training distribution, providing design diversity without requiring users to manually adjust optimization parameters for each variation.
3Strength
If solid block material is used for physical structures, then good mechanical support is provided, but substantial amount of material is consumed that is not needed
Solution Approach 1:
The system applies local quality by generating designs where material density and distribution vary locally throughout the structure. Instead of using uniform solid material, the generative model creates regions with different material densities - providing high density material where mechanical strength is critical and low density or void regions where material is not needed, thereby optimizing the balance between structural integrity and material efficiency.
Solution Approach 2:
The system utilizes porous materials by generating topologies with distributed voids, cavities, and cellular structures that provide mechanical support through geometric configuration rather than continuous material. These porous structures maintain load-bearing capacity while significantly reducing material consumption compared to solid blocks, as the porous architecture distributes stresses efficiently through its geometric pattern.
Data Source
AI summary
A generative machine learning model, such as a convolutional neural network (CNN), can be trained with solutions from a topology optimization solver for a solution for a topology of a set of structures so that the generative machine learning model can generate a plurality of alternative designs for a structure that are alternative topology optimizations (for the structure) for a set of initial setup parameters. The generative model when being trained includes a generative network and a discriminator network. The generative model can be trained using outputs from a CNN autoencoder for densities and a CNN autoencoder for strain energies.