Physics-Informed Machine Learning Framework for Real-Time Process Optimization and Defect Mitigation in Laser Powder Bed Fusion Additive Manufacturing

Almahdi, Mohamed Abd

doi:10.63463/kjes1230

	Physics-Informed Machine Learning Framework for Real-Time Process Optimization and Defect Mitigation in Laser Powder Bed Fusion Additive Manufacturing
Kerbala Journal for Engineering Sciences (KJES)
Volume 6, Issue 2, June 2026, Pages 196-217 PDF (2.06 M)
Document Type: Research Article
DOI: 10.63463/kjes1230
Author
Mohamed abd Almahdi^*
Islamic Azad University, Iran.
Abstract
Over the last few years, Physics-informed (PI) machine learning methods have established themselves as ideal candidates to solve quality variability in laser powder bed fusion additive manufacturing due to complex interactions, defect formations, and building parameters. The research develops a novel physics-informed learning framework that fuses physics knowledge with further state-of-the-art ensemble or neural network-based models to predict build outcomes and dynamically optimize processing parameters in quasi-real-time settings. Key physics-informed descriptors including volumetric energy density (VED) and linear energy density (LED) were extracted and fed into multiple ensemble algorithms including XGBoost, Random Forest, LightGBM, CatBoost, and gradient boosting. Simultaneously, the physics-informed neural networks were trained with thermodynamic constraints for physical consistency of the predictions. The Harvard LPBF database provided the experimental dataset which contains 1579 samples from six commercially important alloys. The findings revealed that the physics-informed neural network yielded greater performance, with an overall coefficient of determination of 0.827, with values exceeding 0.78 for four of the six alloys. Inclusion of descriptors based on physics considerably enhanced the predictive capability of the neural network, reducing the mean absolute error by up to 4.7 percent. Bayesian optimization was used to find the best parameters for mitigating defects. Moreover, the proposed framework achieved closed-loop control using layer-wise simulation with an average optimization time of 3.8 ms per step. The results obtained show that interfacing of physical knowledge with machine learning enhances reliability, prediction capability, and real-time process control in LPBF manufacturing.
Keywords
: Laser Powder Bed Fusion; Physics-Informed Machine Learning; Relative Density Prediction; Process Optimization; Additive Manufacturing

References
[1] Blakey-Milner, P. Gradl, G. Snedden, M. Brooks, J. Pitot, E. Lopez, M. Leary, F. Berto, and A. du Plessis, “Metal additive manufacturing in aerospace: A review,” Materials & Design, vol. 209, Art. no. 110008, 2021, doi: 10.1016/j.matdes.2021.110008. [2] L. Ng, G. L. Goh, G. D. Goh, J. S. J. Ten, and W. Y. Yeong, “Progress and opportunities for machine learning in materials and processes of additive manufacturing,” Advanced Materials, vol. 36, Art. no. 2310006, 2024, doi: 10.1002/adma.202310006. [3] Tang, S. Raghavan, M. B. Gorji, B. Ramamurthy, and A. Sharma, “A critical review of the effects of process-induced porosity on the mechanical properties of alloys fabricated by laser powder bed fusion,” Journal of Materials Science, vol. 57, pp. 9818–9865, 2022, doi: 10.1007/s10853-022-06990-7. [4] V. Gordon, S. P. Narra, R. W. Cunningham, H. Liu, H. Chen, R. M. Suter, J. L. Beuth, and A. D. Rollett, “Defect structure process maps for laser powder bed fusion additive manufacturing,” Additive Manufacturing, vol. 36, Art. no. 101552, 2020, doi: 10.1016/j.addma.2020.101552. [5] Karimzadeh, D. Basvoju, A. Vakanski, I. Charit, F. Xu, and X. Zhang, “Machine learning for additive manufacturing of functionally graded materials,” Materials, vol. 17, Art. no. 3673, 2024, doi: 10.3390/ma17153673. [6] P. Deshmankar, J. S. Challa, A. R. Singh, and S. P. Regalla, “A review of the applications of machine learning for prediction and analysis of mechanical properties and microstructures in additive manufacturing,” Journal of Computing and Information Science in Engineering, vol. 24, Art. no. 120801, 2024, doi: 10.1115/1.4066575. [7] E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,” Nature Reviews Physics, vol. 3, pp. 422–440, 2021, doi: 10.1038/s42254-021-00314-5. [8] Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational Physics, vol. 378, pp. 686–707, 2019, doi: 10.1016/j.jcp.2018.10.045. [9] Sajadi, M. R. Dehaghani, Y. Tang, and G. G. Wang, “Physics-informed online learning for temperature prediction in metal AM,” Materials, vol. 17, Art. no. 3306, 2024, doi: 10.3390/ma17133306. [10] O. Barrionuevo, I. La Fé-Perdomo, and J. A. Ramos-Grez, “Laser powder bed fusion dataset for relative density prediction of commercial metallic alloys,” Scientific Data, vol. 12, Art. no. 375, 2025, doi: 10.1038/s41597-025-04576-x. [11] Khalad, G. Telasang, K. Kadali, and A. Sharma, “A generalized machine learning framework for data-driven prediction of relative density in laser powder bed fusion parts,” The International Journal of Advanced Manufacturing Technology, vol. 135, pp. 4147–4167, 2024, doi: 10.1007/s00170-024-14735-w. [12] B. Schäfle, L. L. Sun, S. Wenzel, E. Slomski-Vetter, T. Melz, and E. Kirchner, “Implementation of a physics-informed neural network to predict relative densities for copper LPBF with a green laser system,” Rapid Prototyping Journal, vol. 31, pp. 333–344, 2025, doi: 10.1108/RPJ-08-2024-0343. [13] Chen, Y. Li, Y. Liu, Y. Liu, H. Wang, and H. Lu, “Development of a universal multimodal prediction method to optimise process parameters for improving densification during laser powder bed fusion,” Virtual and Physical Prototyping, vol. 19, Art. no. e2424463, 2024, doi: 10.1080/17452759.2024.2424463. [14] Zou, W. G. Jiang, Q. H. Qin, Y. C. Liu, and M. L. Li, “Optimized XGBoost model with small dataset for predicting relative density of Ti-6Al-4V parts manufactured by selective laser melting,” Materials, vol. 15, Art. no. 5298, 2022, doi: 10.3390/ma15155298.
Statistics Article View: 282 PDF Download: 110