Hybrid FEM-multitask neural network approach for processing map prediction in LPBF of novel AlFeCrX alloys
Year: 2026
Type: article
Abstract: The development of reliable predictive models for additive manufacturing (AM) remains challenging due to the limited availability of experimental data and the high costs associated with the process, particularly when working with novel alloy systems. In this study, we propose a hybrid modeling approach, that integrates finite element modeling (FEM) with artificial neural networks (ANNs) and transfer learning (TL), to predict melt pool size and defect types, specifically lack-of-fusion and keyhole porosity, and to generate processing maps for three novel aluminum alloys (AlFeCrSi, AlFeCrTi, and AlFeCrSiTi) designed for Laser-Powder Bed Fusion (LPBF). To achieve this, first, a melt pool database was generated through FEM simulations incorporating laser and material parameters. This database, containing 1386 melt pools, was trained on melt pool data to generate an initial ANN model. Subsequently, by transfer learning (TL-ANN), the model was fine-tuned using experimental measurements from 108 additively manufactured samples corresponding to three new Al-based alloys. By leveraging finite element simulation data, the TL-ANN model achieved high predictive accuracy with reduced experimental input, providing a practical framework for efficient process optimization and defect mitigation in LPBF of new materials. While the trained source model is transferable to other aluminum alloys with limited experimental datasets, the proposed hybrid methodology is broadly applicable to different metallic materials, enabling the rapid and cost-efficient generation of processing maps. Additionally, the processing maps predicting defect-type regions indicate the existence of an upper laser power threshold beyond which the formation of fully dense material becomes unlikely.
Source: Additive manufacturing
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,430
OpenAlex ID: W7203617493