Cost-Optimised Cutting and Packing for Nuclear Decommissioning

The cutting and packing of large, solid structures presents a significant challenge in nuclear decommissioning as these two objectives exist in a trade-off. Segmenting a structure into smaller parts can improve packing efficiency and reduce the number of containers required, but at the cost of increased cutting effort. Conversely, minimising cuts reduces cutting cost but often leads to poor packing, raising overall disposal costs. Determining the optimal balance between cutting and packing is difficult, as the trade-off is highly context-dependent, influenced by the geometry of the structure itself as well as the choice of cutting tools and container types used.
This thesis addresses the lack of integrated approaches capable of resolving this trade-off by proposing a novel computational framework that jointly optimises both cutting and packing to minimise total cost. A new feedback-driven optimisation strategy is proposed, in which a genetic algorithm (GA) iteratively refines cutting plans based on both the cutting cost and packing cost, allowing for dynamic exploration of trade-offs between conflicting objectives.
A simplified 2D prototype is developed to evaluate the proposed methodology for jointly optimising cutting and packing. Additional 2D studies are also undertaken to investigate ways to enhance the underlying cutting and packing processes, including hyper-heuristics for improving container utilisation, comparing multi-container packing strategies, and feasible disassembly sequencing for cut structures.
Building on insights gained from the 2D studies, the methodology is extended to 3D and applied to three representative nuclear decommissioning scenarios. Results show that the integrated algorithm can consistently outperform separate cutting/packing approaches (with cost reductions of up to 17%) and achieve comparable performance to manually designed strategies.
The thesis concludes with a detailed set of future work recommendations to enhance algorithm performance and real-world applicability, including the incorporation of feature-based segmentation, post-processing for packing refinement, and integration with robotic disassembly systems.