Segmentation of peen forming patterns using k-means clustering

Shot peen forming is a prominent manufacturing process for shaping aircraft wing skins, which mostly relies on manual intervention, where operators use their expertise to select appropriate peening parameters and apply them in a specific pattern. This paper introduces a crucial advancement towards automating shot peen forming —a segmentation strategy designed to partition peen forming patterns into uniformly treated zones. Unlike existing optimization methods that necessitate predefined peening treatments, our segmentation strategy automatically identifies optimal treatment parameters for each segment. Our approach comprises a novel clustering algorithm, which divides the pattern into segments, and a noise filtering algorithm that eliminates excessively small segments. The clustering algorithm is a unique adaptation of the k-means method, which considers interconnected centroids due to the coupling of the effects of the top and bottom treatments of the part. The filtering algorithm leverages cellular automata principles. Both algorithms underwent numerical testing using 200 randomly generated test cases. The results indicate that the segmentation strategy consistently maintained forming error within an acceptable range, and remarkably, reduced the forming error in 67 out of 200 cases. This segmentation strategy can seamlessly integrate with existing shape optimization tools and a peening treatment library, leading to a fully automated shot peen forming system. The source code for our algorithms is publicly available on GitHub, fostering accessibility and further research in this domain.

Mots clés

• segmentation
• clustering
• filtering
• k-means
• peen forming