mlpcp-interp-dic

Machine Learning for Prediction of Constitutive Parameters - Interpolation study with DIC application.

This work is forked from mlpcp-interp-num. The previous study was performed on numerical data. For this work, DIC-levelled data is used to confirm the practical applicability of the proposed interpolation approach.

📄 Reference

This repository contains the code used in the work published as:

D. Mitreiro, J. Henriques, P. A. Prates, and A. Andrade-Campos,
"Towards DIC-Subset-Independent Machine Learning Models for Constitutive Parameter Identification in Sheet Metal Forming,"
Key Engineering Materials, vol. 1049, pp. 115–122, 2026.
DOI: 10.4028/p-hcCP0o

If you use this repository, please cite this paper.

⚙️ Setup

Clone

Open a terminal, change your current working directory to the location where you want the cloned directory, then clone this repository using the URL from the green Code button above.

For example:

git clone <repository-url>

Config

Inside your repository home folder, edit config/config.ini file to define your Telegram notification variables.

Environment

Next, install Anaconda (or any other you prefer) for managing your Python environments. You can check documentation here.

After the installation, create an empty environment using Python 3.11.10

conda create --name <your_env_name> python=3.11.10
conda activate <your_env_name>

Then, navigate to your repository home folder and install dependencies

pip install -r requirements.txt

🚀 How it works

The methodology of this work is separated into the following steps.

1. New Abaqus simulations according to the previous DOE

📌 Since Abaqus is Windows-only software, the script inside the abaqus folder is meant to be run in a Windows environment.

⚠️ Before running the script, you must copy both y_train.csv and y_test.csv parameter files generated in the previous numerical study to abaqus data folder. This allows abaqus to perform exact the same numerical simulations, but this time with a different output for later use in MatchID.

Now, just run cruciform-dic.py script within abaqus folder

abaqus cae noGUI=cruciform-dic.py

2. DIC-levelling process with MatchID

📌 Since MatchID is Windows-only software, the scripts inside the matchid folder are meant to be run in a Windows environment.

2.1. Copy the files inside matchid folder (7 files) to inside both train and test subfolders.

2.2. Inside both subfolders, replace C:\ABSOLUTE\PATH\TO\test-OR-train\_base by your real path, in the following files:

• Cruciform_DIC.m3inp, lines 13, 16, 64-84, 268
• Cruciform_FEDEF.mtind, line 71

2.3. Copy the generated Abaqus train and test samples to their respective matchid train and test folders.

⚠️ Note that each sample is associated to 1 folder.

After this step, you should end up with:

• 7 files + _base folder + 2000 folders inside train folder
• 7 files + _base folder + 260 folders inside test folder

2.4. Finally, run the cruciform-dic-matchid.py script inside train and test folders to perform the DIC-levelling with MatchID in batch mode.

Within each folder, run

python cruciform-dic-matchid.py

📌 You can monitor the process by checking the dic-logs.log file in each folder.

📌 As it is a long process, you may want to create a Telegram chat bot, define your token and chat_id inside each config.ini file, so you can be notified in case something goes wrong and when the script finishes.

3. Data processing

📌 From now on, scripts must be always executed from the home folder, so defined paths inside scripts work as expected.

As an example, let's say we want to execute some random_script.py inside src folder. Then, from the home folder, we run

python src/random_script.py

3.1. Compile each sample into 1 single file with csv_pre_compile.py

⚠️ Before running the script, you must copy all the 6755 raw samples from the previous study to the data/raw/original_samples folder, as this script will use them to get $F_{xx}$ and $F_{yy}$ force values.

⚠️ Also, you must copy the Static_0000_0_Numerical_0_0.synthetic.tif.csv file generated by MatchID to data/raw folder (just pick one file from one sample, as time-step 0 is equal across all samples).

3.2. Compile all samples into 1 single file with csv_compile.py

3.3. Dataset sort according to original cruciform dataset with dataset_sort.py

3.4. Extract DIC subset coordinates from static file with extract_coords.py

3.5. Shift subset coordinates to cruciform domain with subset_shift.py

3.6. Convert shear strain values from tensorial ($\epsilon_{xy}$) to engineering ($\gamma_{xy}$) convention with epsilon_to_gamma_xy.py

📌 This conversion is necessary so shear strain results can be compared between each other, as MatchID uses tensorial convention and Abaqus uses engineering convention. So, values need to be multiplied by 2.

4. Dataset interpolation using 30x30 grid and multiquadric method with interpolation.py

5. XGBoost train and evaluation with train.py and test.py, respectively

• The obtained results are in metrics
• XGBoost models and scalers can be found in models
• Some plots can be performed using the plot scripts in tools
• All the images generated by plot scripts can be found in images

⚖️ License

This work was developed within ReachOptimum@TEMA research group, University of Aveiro, and is licensed under the MIT License, which allows anyone to use, modify, and distribute this software for free, as long as the original copyright and license notice are included. See the LICENSE file for more details.