Physics-Informed Feature Representations and Pretrained Tabular Foundation Models for Generalizable Melt Pool Geometry Prediction
Organizations: University of South Carolina · West Virginia University
Abstract
Developing generalizable, data-efficient machine learning (ML) models for advanced manufacturing remains challenging because experimental datasets are limited and process-property relationships vary across materials and operating conditions. In laser powder bed fusion (LPBF), melt pool width, depth, and melting mode (conduction, transition, or keyhole) govern process stability and defect formation. Conventional ML models typically use dimensional process parameters, which represent laser-material interactions only implicitly and may limit cross-material transfer. This study evaluates whether two physics-informed dimensionless features, normalized enthalpy and Peclet number, combining laser processing conditions with material thermophysical properties, improve melt pool prediction over raw process parameters. We systematically benchmark seven models: four classical ML methods (Gaussian process, support vector machine, random forest, and XGBoost) and three pretrained tabular foundation models (TabPFN, TabDPT, and TabICLv2). Experiments span 316L stainless steel, IN625, IN718, and combined-material datasets, covering regression of melt pool width and depth and classification of melting modes. Dimensionless features yield comparable or superior predictive performance in most settings, with the largest gains in cross-material depth prediction (R^2 = 0.96). Tabular foundation models generally achieve stronger classification performance than classical baselines, particularly for multi-material datasets. These findings indicate that combining physically meaningful dimensionless representations with pretrained tabular models can improve data-efficient melt pool modeling and support more transferable process-window identification in LPBF.
Figures & tables
| Process Parameters (unit) | Material Properties (unit) | Dimensionless Parameters |
| Laser Power, (W) | Density, (kg m -3 ) | Normalized Enthalpy, |
| Scan Speed, (mm/s) | Specific Heat Capacity, (J kg -1 K -1 ) | Peclet Number, |
| Laser Spot Diameter, ( m) | Thermal Conductivity, (W m -1 K -1 ) | |
| Melting Temperature, (K) |
| Property (unit) | 316L | IN625 | IN718 |
| (kg m -3 ) | 7980 | 8440 | 7727 |
| (J kg -1 K -1 ) | 855 | 720 | 650 |
| (W m -1 K -1 ) | 36.7 | 28 | 29 |
| (K) | 1697 | 1563.15 | 1526 |
| Model | Explicit settings |
| Random forest | 500 trees; remaining tree controls at defaults. |
| XGBoost | 500 boosting rounds; depth 6; learning rate 0.05; squared-error objective; histogram tree method. |
| GPR | Constant RBF + white-noise kernel; input scaling. |
| SVR | RBF kernel; ; ; input and target standardization; = scale. |
| TabPFN | Pretrained regressor; 8 ensemble members. |
| TabDPT | Pretrained regressor; 8 ensembles; context size 2048; batch size 256. |
| Model | Explicit settings |
| Random forest | 500 trees; max depth unrestricted; min samples to split ; min samples per leaf . |
| XGBoost | 500 rounds; depth ; learning rate ; multi:softprob; tree method=hist. |
| GPC | StandardScaler; RBF kernel; max_iter_predict=100; one-vs-rest. |
| SVC | RBF kernel; ; = scale; |
| probability enabled; no class weighting. | |
| TabPFN | Pretrained classifier; inference batch size: package defaults. |
| MAPE (%) | NRMSE | |||
| Model | DF | PP | DF | PP |
| RF | 16.48 | 20.02 | 0.063 | 0.081 |
| XGB | 18.13 | 20.22 | 0.064 | 0.096 |
| GPR | 16.55 | 21.32 | 0.060 | 0.076 |
| SVR | 20.53 | 23.93 | 0.068 | 0.086 |
| TabPFN | 15.55 | 15.40 | 0.061 | 0.066 |
| MAPE (%) | NRMSE | |||
| Model | DF | PP+MP | DF | PP+MP |
| RF | 15.97 | 13.76 | 0.100 | 0.102 |
| XGB | 18.29 | 14.94 | 0.109 | 0.123 |
| GPR | 34.67 | 15.93 | 0.144 | 0.084 |
| SVR | 35.14 | 13.30 | 0.085 | 0.083 |
| TabPFN | 10.02 | 9.53 | 0.061 | 0.081 |
| MAPE (%) | NRMSE | |||
| Model | DL | PP+MP | DL | PP+MP |
| RF | 15.55 | 17.67 | 0.071 | 0.078 |
| XGB | 16.13 | 16.04 | 0.079 | 0.077 |
| GPR | 16.42 | 17.35 | 0.070 | 0.069 |
| SVR | 28.76 | 17.11 | 0.064 | 0.075 |
| TabPFN | 11.28 | 11.89 | 0.043 | 0.060 |
| MAPE (%) | NRMSE | |||
| Model | DL | PP | DL | PP |
| RF | 10.56 | 10.05 | 0.098 | 0.079 |
| XGB | 11.62 | 11.08 | 0.104 | 0.089 |
| GPR | 8.92 | 10.11 | 0.087 | 0.081 |
| SVR | 8.71 | 9.48 | 0.089 | 0.078 |
| TabPFN | 8.82 | 8.38 | 0.086 | 0.068 |
| MAPE (%) | NRMSE | |||
| Model | DL | PP+MP | DL | PP+MP |
| RF | 4.93 | 4.64 | 0.057 | 0.076 |
| XGB | 5.23 | 5.37 | 0.055 | 0.081 |
| GPR | 9.56 | 5.50 | 0.138 | 0.110 |
| SVR | 5.45 | 5.83 | 0.074 | 0.092 |
| TabPFN | 4.45 | 4.67 | 0.057 | 0.078 |
| MAPE (%) | NRMSE | |||
| Model | DL | PP+MP | DL | PP+MP |
| RF | 9.65 | 7.27 | 0.082 | 0.062 |
| XGB | 10.76 | 7.78 | 0.101 | 0.068 |
| GPR | 6.69 | 8.28 | 0.062 | 0.075 |
| SVR | 6.54 | 7.50 | 0.060 | 0.064 |
| TabPFN | 6.41 | 6.11 | 0.057 | 0.054 |
| Accuracy | Recall | F1 | Precision | |||||
| Model | DF | PP | DF | PP | DF | PP | DF | PP |
| RF | 0.800 | 0.781 | 0.730 | 0.700 | 0.730 | 0.692 | 0.750 | 0.720 |
| XGB | 0.814 | 0.795 | 0.743 | 0.723 | 0.738 | 0.715 | 0.750 | 0.743 |
| SVC | 0.810 | 0.752 | 0.735 | 0.649 | 0.735 | 0.649 | 0.744 | 0.755 |
| GPC | 0.795 | 0.770 | 0.714 | 0.698 | 0.715 | 0.684 | 0.723 | 0.719 |
| Accuracy | Recall | F1 | Precision | |||||
| Model | DF | PP+MP | DF | PP+MP | DF | PP+MP | DF | PP+MP |
| RF | 0.900 | 0.890 | 0.870 | 0.840 | 0.860 | 0.830 | 0.870 | 0.851 |
| XGB | 0.884 | 0.837 | 0.808 | 0.717 | 0.809 | 0.693 | 0.864 | 0.700 |
| SVC | 0.884 | 0.852 | 0.882 | 0.860 | 0.844 | 0.809 | 0.845 | 0.812 |
| GPC | 0.940 | 0.909 | 0.943 | 0.814 | 0.928 | 0.808 | 0.921 | 0.812 |
| Model | Recall | F1 | Precision |
| RF | 0.730 | 0.730 | 0.750 |
| XGB | 0.743 | 0.738 | 0.750 |
| SVC | 0.735 | 0.735 | 0.744 |
| GPC | 0.714 | 0.715 | 0.723 |
| TabPFN | 0.758 | 0.749 | 0.748 |
| TabDPT | 0.742 | 0.735 | 0.736 |
| Model | Recall | F1 | Precision |
| RF | 0.870 | 0.860 | 0.870 |
| XGB | 0.808 | 0.809 | 0.864 |
| SVC | 0.882 | 0.844 | 0.845 |
| GPC | 0.943 | 0.928 | 0.921 |
| TabPFN | 0.953 | 0.938 | 0.931 |
| TabDPT | 0.882 | 0.865 | 0.890 |
| Model | Recall | F1 | Precision |
| RF | 0.792 | 0.780 | 0.797 |
| XGB | 0.801 | 0.802 | 0.825 |
| SVC | 0.813 | 0.801 | 0.813 |
| GPC | 0.790 | 0.788 | 0.815 |
| TabPFN | 0.813 | 0.803 | 0.819 |
| TabDPT | 0.812 | 0.801 | 0.819 |
Physics-Informed ML
Physics-informed machine learning combines data-driven models with physical laws to solve, predict, and infer scientific systems.
See all 910 papers