VP of Solver Development, Manufacturing Products Siemens
Binder Jet Additive Manufacturing is rapidly emerging as a viable route for high-throughput production of metallic components. However, sintering-induced shrinkage and distortion remain major impediments to achieving dimensional fidelity, often necessitating extensive, costly trial-and-error approaches. This paper builds upon the foundation of our prior research presented at MIM2025 and extends the investigation by incorporating a geometric deep learning model for real-time predictions. The enhanced digital workflow combines physics-based advanced sintering simulation, material-dependent shrinkage modeling, and optimization-driven shape compensation, integrated with Physics AI, to develop an accurate predictive framework. The resulting system enables real-time predictions of densification and spatially nonuniform deformation behavior in parts with similar geometrical and material characteristics. Case studies using multiple powder sizes and material combinations with industrial-scale parts demonstrate the robustness of the approach. Experimental validation confirms substantial reductions in post-processing effort and part rejection rates, resulting in significant improvements in dimensional accuracy and a reduction in qualification cycles. This work presents a scalable framework for advancing the adoption of Binder Jet Additive Manufacturing in commercial and industrial applications by integrating physics-based process modeling with a data-driven deep learning architecture.
Learning Objectives:
Understand how geometric deep learning models can be trained on physics-based simulations to predict sintering behavior in real time.
Describe how integrating physics-based modeling with AI enables accurate and scalable prediction of shrinkage and distortion in Binder Jet Additive Manufacturing.
Demonstrate how a Physics-AI framework reduces post-processing effort and accelerates qualification cycles in additive manufacturing workflows.
