Published: 5 April 2023| Version 1 | DOI: 10.17632/ksf6njpynh.1
Marwin Gihr


This folder contains the supplementary material for the paper "Bead geometry prediction and optimization for corner structures in WAAM using Machine Learning" Marwin Gihr*, Asif Rashid, Shreyes N. Melkote Contact: *Tel.: +49 15737922958; E-mail address: For the study three sets of experiments werde conducted. Within one experiment the travel speed and the corner angle of the beads is varied. The three experiments differ in the selected WFS. The folder Data contains the preprocessed data from the conducted experiments. For experiment 1 a WFS of 3.6 m/sec was chosen, experiment 2 was conducted with a WFS of 4.2 m/sec and experiment 3 with 5 m/sec. In the following the dataset is explained. - Each row contains the geometrical information of one bead layer cross-section and the process parameters that were used for its manufacturing - Layer contains the layer number of the particular cross-section - Group is fully correlated with the utilized TS and is used for indexing - BeadArm contains the information on which side of the corner tip the cross-section is located and is also used for indexing - Cross-section is also used for indexing. The combination of Layer, Group, BeadArm, Cross-section and experiment number creates a unique Identifier for all cross-sections - BW containts the information on the measured BW of the cross-section - OverlapRegion, CalculatedVolume, InitialTemperature, VolMeanLayerBelow, BHmax and Volume are obsolete for this study - DistanceToCorner contains the euclidean distance of the cross-section to the tip of the corner - Angle describes the corner of the bead on which the cross-section is located - WFS contains the information on the utilized WFS in manufacturing the bead on which the cross-section is located - TS contains the information on the utilized TS in manufacturing the bead on which the cross-section is located - BH0-4 contain the BH descriptors that describe the geometry of the Bead in combination with the BW 1_ExtendedBeadGeometryStudy_ModelTraining contains the code to train the Multilayer Perceptron on the collected data. 2_ExtendedBeadGeometryStudy_FeatureStudy contains the code to analyze the model through a feature study. 3_ExtendedBeadGeometryStudy_InverseModel contains the code to use the model inversively to retrieve suggestions for optimized PM and TS combinations beadwiseStudy_5points_tol20 contains the hystory of the hyperparameter optimization study conducted with the help of the Optuna library model_5Points_tol20_noOR-R2-0.71 and scaler_5Points_tol20-R2-0.71 contain the trained MLP and the scaler used to preprocess the training and testing data. The unprocessed, raw data of the study can be provided upon request.


Steps to reproduce

The raw data of the study as well as the code for data processing can be provided upon request.


Georgia Institute of Technology


Machine Learning, Advanced Manufacturing