A multi-level fusion similarity detection method for 3D CAD tire tread pattern models

Abstract

To address the limitations of existing similarity detection methods in comprehensively characterizing the complex topological structures and circumferential sequence features of tire tread patterns, a multi-level fusion similarity detection method for 3D CAD tire tread pattern models is proposed. The proposed framework establishes an interpretable similarity evaluation system that integrates four complementary levels—global, local, sequential, and topological. At the global level, an edge-based histogram metric quantifies differences in the overall distribution of geometric elements. At the local level, a surface feature difference matrix combined with a greedy matching algorithm captures local geometric consistency between models. At the sequential level, a common sub-chain detection approach identifies circumferential sequence features, while at the topological level, adjacency attribute graphs coupled with a simulated annealing algorithm are used to discover topological correspondences among representative substructures. Finally, a weighted fusion strategy integrates all hierarchical features to compute an overall similarity measure. Experimental validation based on similarity ranking demonstrates that the proposed method produces rank orders consistent with human evaluations across diverse tread pattern models, effectively reflecting both geometric and topological variations. The algorithm exhibits stable, monotonic ranking behavior under local incompleteness and scale variations, and maintains reasonable similarity ordering for uncalibrated samples, confirming its robustness and generalization capabilities. The proposed method achieves a comprehensive representation of 3D tire tread pattern models—from geometric and topological configurations to circumferential sequences—providing a reliable technical foundation for tread pattern design reuse, structural retrieval, and competitive analysis.