Generative Architected-Materials Design using Mutiobjective Active Learning Loop (GAD-MALL): A active learning pipeline for designing the architected materials
Architected materials that consist of multiple subelements arranged in particular orders can demonstrate a much broader range of properties than their constituent materials. However, the rational design of these materials generally relies on experts’ prior knowledge and requires painstaking effort. Here, we present a data-efficient method for the high-dimensional multi-property optimization of 3D-printed architected materials utilizing a machine learning (ML) cycle consisting of the finite element method (FEM) and 3D neural networks. Specifically, we applied our method to orthopedic implant design. Compared to expert designs, our experience-free method designed microscale heterogeneous architectures with a biocompatible elastic modulus and higher strength. Furthermore, inspired by the knowledge learned from the neural networks, we developed machine-human synergy, adapting the ML-designed architecture to fix a macroscale, irregularly shaped animal bone defect. Such adaptation exhibits 20% higher experimental load-bearing capacity than the expert design. Thus, our method opens a new paradigm for the fast and intelligent design of architected materials with tailored mechanical, physical, and chemical properties.
The following libraries are necessary for running the codes.
tensorflow-gpu == 2.5.0
keras == 2.3.1
tqdm == 4.59.0
scipy == 1.6.2
numpy == 1.19.2
pandas == 1.2.4
matplotlib==3.3.4
Please install requirements using below command.
pip install -r requirements.txt
which should install in about few minutes.
The developmental version of the package has been tested on the following systems and drivers.
- Ubuntu 18.04
- CUDA 11.4
- cuDNN 8.1
- RTX3090 Ti
To run the generative architecture design-multiobjective active learning loop (GAD-MALL) pipeline, please follow these steps:
- Train 3D-CAE model as the generative model by running the
3D_CAE.py
Python code in the folderGAD-MALL
, please run the following line in terminal:
python 3D_CAE.py
Note: The raw data 3D_CAE_train.npy
for training 3D-CAE can be downloaded here.
- Train 3D-CNN models as surrogate models of GAD-MALL by running the
3D_CNN.py
Python code in the folderGAD-MALL
, please run the following line in terminal:
python 3D_CNN.py
Note: The raw data for training 3D-CNN is in the folder GAD-MALL/GAD-MALL/data/
. One big file named Matrix60.npy
can be downloaded here.
- To search for high-performance architected materials with specific elastic modulus and high yield strength using GAD-MALL, please run the following line in terminal:
bash run_GAD_MALL.sh
- After completing the GAD-MALL process, you'll obtain porosity matrices with specific predicted elastic modulus (E=2500 MPa, E=5000 MPa) and the highest predicted yield strength. You can use these matrices to generate TPMS-Gyroid scaffolds. For more about
TPMS-Gyroid Structure Generation
, see code andREADME
in folderTPMS-Gyroid-Generate-by-MATLAB
- Conduct Finite Element Method (FEM) analysis of the TPMS-Gyroid scaffolds using ABAQUS (or other software capable of mechanical simulation). For more about the
Finite Element Method
, see code andREADME
in the folderFinite Element Method by ABAQUS
. Moreover, to automate the pipeline of FEM (Matlab for Gyroid structures generation in STL format → Hypermesh draw the mesh in .inp format → Abaqus conduct FEM analysis and output a .odb file for further calculation of mechanical properties), see code andREADME
in the folderFEM_automation
. - Once you've combined the new mechanical property data from FEM results with your initial dataset in the
Datasets
folder, return to step 2 and run the3D_CNN.py
Python code again to update your surrogate models. Continue this active learning loop until you find Gyroid porosity matrices with promising mechanical properties.
For more about Model Parameters Optimizing using Bayesian Optimization
, see code and README
in folder Model Parameter Optimize_Bayesian Optimization
For more about Other State-of-the-art Active Learning Methods
compared with our best method GAD-MALL, see cede and README
in folder Other-State-of-the-art-Active-Learning
If you find this work interesting, welcome to cite our paper!