EURO 2025 Leeds
Abstract Submission

620. Learn to Rank: Efficient Ranking and Visualisation of 4K Build Orientations for Surface Roughness Reduction in Additive Manufacturing

Invited abstract in session MB-4: Interpretable Optimization Methods and Applications, stream Data Science meets Optimization.

Monday, 10:30-12:00
Room: Rupert Beckett LT

Authors (first author is the speaker)

1. Peizhi Shi
University of Leeds, Leeds University Business School
2. Yuchu Qin
EPSRC Future Advanced Metrology Hub for Sustainable Manufacturing
3. Fanlin Meng
University of Exeter Business School
4. Ming Luo
University of Bristol Business School
5. Sajid Siraj
Leeds University Business School, University of Leeds

Abstract

Additive manufacturing is a promising manufacturing technique that is capable of building complex geometries and customised products. Due to its layer-by-layer printing nature, the build orientation of a part has important effect on the surface roughness of the part. In this research area, search algorithms are typically adopted to find optimal build orientations that could minimise the surface roughness. On the one hand, a search algorithm typically needs to explore a huge space. This time-consuming searching process often results in sub-optimal solutions. On the other hand, these heuristic search methods only cover a small proportion of the search space and fail to provide visualisation for the whole space. To address these two problems, this paper proposes a novel ranking network that predicts the surface roughness ranking of 4,096 orientations based on a given solid model. By adopting this approach, the time-consuming O(N) search process could potentially be converted into one forward pass with O(1) complexity, which greatly increases the search efficiency. Additionally, surface roughness of 4K orientations could be visualised based on the ranking outcome. Experimental results show that the proposed learning paradigm with the new loss function produces superior results compared to search-, unsupervised learning-, and supervised learning-based approaches.

Keywords

Status: accepted

Back to the list of papers