Prediction of long-term creep modulus of thermoplastics by brief testing data and interpretable machine learning

This repository contains the code to predict the Creep Modulus of thermoplastics using public material data and machine learning. This research project, conducted at Leartiker, has been accepted for publication in International Journal of Solids and Structures, DOI: https://doi.org/10.1016/j.ijsolstr.2024.113014

All credit for the dataset to the CAMPUS Plastics material information system.

Corresponding author: Héctor Lobato ([email protected])

How to install dependencies

Create conda environment:

conda create -n ENV_NAME python=3.10.9

And install requirements:

pip install -r src/requirements.txt

Or directly from the environment.yml file (slower):

conda env create -f src/environment.yml

Then activate the environment and you are ready to go.

MLOps tools:

This repository uses Kedro for project structuring and data pipelines, and Weights & Biases for experiment tracking. Refer to the documentation of these tools for more information.

Experiment tracking:

To track the experiments with Weights & Biases, you need to create an account and set use_wandb: True in conf/base/parameters.yml. If set to False, you can inspect the model_evalution_metrics.pickle corresponding to the experiment in data/05_reporting.

How to run Kedro pipelines:

You can run all the pipelines in src/creep_prediction/pipelines with:

kedro run

Or run a specific pipeline with:

kedro run --pipeline PIPELINE_NAME

To visualize the project structure and data pipelines, run:

kedro viz

How to define experiments:

All the necessary parameters for the experiments are defined in conf/base/parameters.yml. Additional information about the parameters can be found in the same file.

How to replicate paper results:

Execute run.py