SimonOster/Porosity_prediction_PBF_with_thermography

This dataset is linked to the following publication: https://doi.org/10.1016/j.addma.2024.104502

Content:

• The attached files use Tensorflow 2.15, versions above can cause troubles due to Keras 3.0. Make sure to set up your environment accordingly.
• The datasets refer to in-situ monitoring data of a PBF-LB/M process. A SWIR thermographic camera was used to monitor the manufacturing of two similar specimen, denoted as "A" and "B".
• The datasets are part of a study in which different ML models are investigated in terms of their ability to predict local porosity in the parts bulk based on the thermographic data. The study is currently in the peer review process. As soon as it is published, it will be linked within this repository for additional information on the datasets and the model usage.
• The scripts and datasets which correspond to the individual specimen are deonted as "_SpecA" and "_SpecB".
• All models which are implemented in the attached Python scripts are already optimized in terms of hyperparameters.
• "Dataset_Features_SpecA.pkl" and "Dataset_Features_SpecB.pkl": Contain feature datasets extracted from thermography image data which can be used to train different ML and DeepL regressors. The samples are stored in pandas DataFrames. In the attached script "Example_train_model_features.py" a workflow is shown how to load and process the data. Different model types can be chosen for model training.
• The features samples are organized in 2D arrays where the first axis correspond to the specific ambient layer (first row refers to layer -3, afterwards increasing) and the second axis to the feature. The feature order is: ['Cooling_speed', 'Mean_spatter_area', 'Mean_spatter_temperature', 'Melt_pool_area', 'Melt_pool_eccentricity', 'Melt_pool_length', 'Melt_pool_max_temp', 'Melt_pool_mean_temp', 'Melt_pool_perimeter', 'Melt_pool_width', 'Number_spatters', 'tot_800', 'tot_900', 'tot_1000', 'tot_1100', 'tot_1200', 'tot_1300', 'sotot_800', 'sotot_900', 'sotot_1000', 'sotot_1100', 'sotot_1200', 'sotot_1300']
• "Dataset_IMSEQR_SpecA.zip" and "Dataset_IMSEQR_SpecB.zip": Contain raw image sequence data from thermography that is used to train a many-to-one regressor. Each individual sample is stored as a .mat-file. In the attached script "Example_train_model_IMSEQR.py" a workflow is shown in which data generators are used to load and process the sample for model training.
• For further question please get in touch by writing at [email protected]