Probabilistic Machine Learning for Uncertainty Quantification of FDM-Printed PETG: Model Selection, Orientation-Specific Deployment, and Reliability-Based Process Optimization

# Dataset: Probabilistic Machine Learning for Uncertainty Quantification of FDM-Printed PETG **Authors:** Mana Saedan, Watcharapong Tachajapong **Affiliation:** Department of Mechanical Engineering, Faculty of Engineering, Chiang Mai University, Chiang Mai, Thailand This dataset supports the manuscript submitted to the *Journal of Engineering Design*: "Probabilistic Machine Learning for Uncertainty Quantification of FDM-Printed PETG: Model Selection, Orientation-Specific Deployment, and Reliability-Based Process Optimization". --- ## Files | File / Folder | Description | |---|---| | `Raw-Data-Set.xlsx` | **Main training dataset** — 327 tensile test measurements from 109 unique printing conditions × 3 replicates, across 3 build orientations | | `Inference_Data/` | **Physical validation data** — raw tensile-test force curves (CSV) and Max-Load/UTS summary workbooks from printed validation specimens (batches 20260713, 20260721, 20260722, 20260727) | --- ## Raw-Data-Set.xlsx — Column Description Material: Creality™ Hyper PETG filament (1.75 mm), printed on a Creality K1 Max FDM printer (nozzle 0.4 mm). Specimen geometry: rectangular 150 × 15 × 6 mm (ASTM D3039-based). | Column | Description | Unit | |---|---|---| | `Exp` | Experiment (printing condition) ID | – | | `UTS` | Ultimate tensile strength (target variable) | MPa | | `OR` | Build orientation: X (flat), Y (on-edge), Z (upright) | – | | `SP` | Solid pattern (raster pattern of top/bottom solid layers) | – | | `IP` | Infill pattern (internal fill topology) | – | | `NT` | Nozzle temperature | °C | | `LH` | Layer height | mm | | `LW` | Print line width | % of nozzle diameter | | `WL` | Number of wall loops | – | | `SL` | Number of surface layers | – | | `ID` | Infill density | % | | `VW` | Wall printing speed | mm/s | | `VI` | Infill printing speed | mm/s | | `VS` | Surface printing speed | mm/s | | `FL` | Filament consumption (total filament length per specimen, derived from slicing software) | m | **Notes** - 327 rows = 109 conditions × 3 replicates (per-condition counts: X ≈ 108, Y ≈ 111, Z ≈ 108 raw rows; see paper Section 2). - Categorical features (`OR`, `SP`, `IP`) are treated as such in the models; all other features are numeric process settings. - `FL` is a zero-cost derived feature from the slicer — a surrogate for part mass. --- ## Inference_Data/ — Validation Data Summary workbooks (`*_Summary.xlsx`) for specimens printed to physically validate the model-selected optimal configurations: `X-UTS_Summary.xlsx`, `Y-UTS_Summary.xlsx`, `Z-UTS_Summary.xlsx` --- ## License [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) — free to share and adapt with attribution. ## How to Cite If you use this dataset, please cite the associated article (DOI will be added upon publication): > Saedan, M., & Tachajapong, W. (2026). Probabilistic Machine Learning for Uncertainty Quantification of FDM-Printed PETG. *Journal of Engineering Design*. [DOI] ## Contact Mana Saedan — [e-mail]