A Learning Driven Evolutionary Algorithm with Separated‑Axis C onvolutional Localization for Integrated 3D Nesting and Scheduling in Additive Manufacturing Workshop

In selective laser sintering (SLS)-based additive manufacturing workshops, the material properties of metal powder enable support-free part fabrication. Stacking parts along the Z-axis further enhances the efficiency of batch printing. However, when dealing with numerous parts and multiple parallel SLS printers, the integrated optimization of three-dimensional nesting, part sequencing, and printer allocation poses a significant challenge. In this paper, a learning-driven evolutionary algorithm (LDEVA) is proposed, which employs an adaptive operator selection mechanism. In LDEVA, voxel‑based representation is used to efficiently handle irregular custom part geometries. For efficient nesting, a novel separated‑axis convolutional localization strategy (SACL) is developed. SACL integrates the separating axis theorem with fast Fourier convolution to achieve high-density nesting while minimizing computation time. For efficient scheduling, an adaptive operator selecting framework is implemented for local search, wherein heuristic operators are dynamically combined to explore distinct neighborhood spaces. The proposed LDEVA is evaluated on 40 instances spanning different scale levels, and it can reduce more printing time than four states‑of‑the‑art methods. Results demonstrate the outstanding efficiency advantages of LDEVA, outperforming the comparison methods in 90% of instances. Notably, in scenarios with more than 100 parts, LDEVA outperforms the baseline algorithms in 95% of instances.