Synthetic Dataset and Code — ML-Assisted Comparative Assessment of Additively Manufactured versus Machined Gear

Synthetic Dataset and Code — ML-Assisted Comparative Assessment of Additively Manufactured versus Machined Gear Component Mechanical Performance (Enhanced Framework) This repository accompanies the manuscript "Machine Learning-AssistedComparative Assessment of Additively Manufactured versus Machined GearComponent Mechanical Performance: An Explainable, Stacked-EnsembleFramework with Sensitivity-Validated Synthetic Data" (T. O. Akande). IMPORTANT — Nature of the data `synthetic_gear_dataset.csv` is a literature-calibrated SYNTHETIC dataset,not experimental measurement data. See Section 3.3 of the manuscript.Please do not cite figures derived from this dataset as measured materialproperties. Pipeline / run order ```bashpip install -r requirements.txt python generate_dataset.py # -> synthetic_gear_dataset.csv (Table 1, seed=42)python stage_a.py # base learners + stacked ensembles -> Tables 3-4python stage_b1_shap.py # SHAP explainability -> Figs. 10-11python stage_b2_sig_pca.py # Wilcoxon significance testing + PCA/t-SNE -> Table 5, Figs. 12-13python stage_b3_sensitivity.py # sensitivity analysis -> Table 6, Fig. 14python stage_b4_uncertainty.py # conformalized quantile regression -> Fig. 15python stage_c_original_figs.py # remaining descriptive/model figures -> Figs. 2-9python framework_diagram_v2.py # schematic framework diagram -> Fig. 1``` `stage_a.py` must run before every other `stage_*` script, since they loadits saved state (`stage_a_state.pkl`). All scripts are deterministic(NumPy `Generator` seed 42 for data generation; `random_state=13`/`21` formodel training, splitting, and calibration), so outputs arenumber-for-number identical to the values reported in the manuscript. Contents | File | Manuscript output reproduced ||---|---|| `generate_dataset.py` | Table 1 parameters; `synthetic_gear_dataset.csv` (Table 2 statistics) || `stage_a.py` | Tables 3-4 (base learners + proposed stacked ensembles) || `stage_b1_shap.py` | Figs. 10-11 (SHAP explainability) || `stage_b2_sig_pca.py` | Table 5, Fig. 12 (Wilcoxon significance testing); Fig. 13 (PCA/t-SNE) || `stage_b3_sensitivity.py` | Table 6, Fig. 14 (sensitivity analysis) || `stage_b4_uncertainty.py` | Fig. 15 (conformalized quantile regression) || `stage_c_original_figs.py` | Figs. 2-9 (descriptive statistics, ROC, confusion matrix, feature importance, predicted-vs-actual, residuals) || `framework_diagram_v2.py` | Fig. 1 (framework schematic) || `requirements.txt` | Exact Python package versions used || `LICENSE` | CC BY 4.0 | Data dictionary — `synthetic_gear_dataset.csv` | Column | Type | Description ||---|---|---|| `Process` | categorical | Manufacturing route: `AM` or `Machined`. Classification target. || `Edge_Case` | boolean | `True` for the hybrid/post-processed sub-population (Section 3.3, Section 4.4). || `Surface_Roughness_Ra_um` | float | Arithmetic mean surface roughness, µm. || `Porosity_pct` | float | Internal porosity fraction, %. || `Relative_Density_pct` | float | Relative density, %. || `Surface_Hardness_HV` | float | Vickers surface hardness. || `Residual_Stress_MPa` | float | Near-surface residual stress, MPa (positive = tensile). || `Grain_Size_um` | float | Mean grain size, µm. || `UTS_MPa` | float | Ultimate tensile strength, MPa. || `Stress_Amplitude_MPa` | float | Applied cyclic tooth-root bending stress amplitude, MPa. || `log10_Nf` | float | log₁₀(cycles to failure), via manuscript Eq. (4). Regression target. || `Fatigue_Life_Cycles` | float | `10^log10_Nf`, cycles to failure. | Citation If you use this dataset or code, please cite the manuscript (full citationto be added once the DOI/publication record is available) and thisrepository's own archive DOI (see the repository landing page). Contact Timileyin O. Akande — Department of Mechanical and MechatronicsEngineering, Abiola Ajimobi Technical University (First TechnicalUniversity), Ibadan, Nigeria.