Integrated bio-inspired Design by AI: Using cell structure patterns to train an AI model to explore topology design ideas
Year 2022
Author(s) Namrata Baruah
ML Tags
Variational Autoencoder
Topic Tags
Generative Design
Structural Design
> Software & Plug-ins Used
Rhinoceros, Grasshopper for modelling the patterns of the cellular solids and testing of the AI model
Python using Tensor-Flow, Keras library for writing the AI model and CNN Google Collab for coding the VAE model
Rhinoceros, Grasshopper, Karamba for the FEM analysis (structural performance analysis)
> Workflow
Thesis Report Figure 4: Diagrammatic representation of the Research Approach
> Summary
The purpose of this thesis is to investigate design concepts using AI-generated patterns in shell structures. After selecting perforation patterns found in nature, the author trains an AI model to generate new patterns based on the training data. The aim of this research is to compare natural and AI-generated patterns to verify if the AI model generates a bigger variety of patterns while maintaining good structural performance.
The AI model used in this thesis is a Variational AUtoencoder (VAE).
The steps followed in building the model are:
Creating a continuous and uniform database by defining the parametric grashopper model variables such that generated 3D models have similarities
Formulating a parametric model in Grasshopper to create 3D models of lattice patterns
Augmentation of the Grasshopper-generated data for the deep convolutional neural networks to run properly
Training a VAE model with 80% of the data in the dataset and using the rest of the data to validate the training
Following the training of the VAE model, numerous variations of 2D image patterns were generated. The author selected the most distinct patterns and extracted the sharper ones to edit.
To use the generated patterns in the conceptual design phase of shell structures, a workflow was created. It contains the following steps:
Extraction of the geometries from a 2D image which is translated to a data structure that can be used to control a parametric model
Creating the shell form where the pattern will morph onto
Morphing the pattern on the shell
Performing FEM analysis on the morphed shell to check its structural performance
LIMITATIONS: The dataset used to train the AI contains only 2D images of cellular solid structures. The dataset was limited to small number of patterns.
> Additional Visuals
Thesis Report Figure 32: A few Iterations of the Square-Chiral lattice pattern
Thesis Report Figure 42: Sampling of Generated patterns
Thesis Report Figure 49: New topology design ideas from generated patterns
> Possible Applications
For ideas on how to implement some of the above mentioned techniques, please see
‘Possible applications for students to try with Generative Deep Learning”