EIDBO-TrICP: a coarse-to-fine algorithm for cross-source point cloud registration in aero-engine blades

Abstract

Abstract We propose a novel coarse-to-fine algorithm, named EIDBO-TrICP, for the matching and registration of aero-engine blade cross-source point clouds (CSPC) obtained from different types of sensors. This task is particularly challenging due to the presence of data inconsistencies such as missing data, noise, outliers, density variations, viewpoint changes, and scale differences, along with the complex structure and surface characteristics of the blade itself. The proposed method is composed of two main stages: In the coarse registration phase, we introduce an enhanced and improved dung beetle optimization algorithm (EIDBO), which improves the convergence speed and optimization accuracy of the standard dung beetle optimization algorithm by adopting a modified convergence factor. This approach effectively avoids local minima in the registration process by leveraging the global search capability of the improved dung beetle optimization algorithm. Additionally, singular value decomposition (SVD) is used to accelerate the local search convergence and enhance the final registration accuracy. In the fine registration phase, the Trimmed ICP (TrICP) algorithm is employed to ensure the precise alignment of the cross-source point clouds. The results from both synthetic data simulations and real aero-engine blade data obtained from a compound ray coordinate measurement system demonstrate that the proposed algorithm outperforms existing methods in both efficiency and accuracy. The significance of EIDBO-TrICP lies in its ability to effectively integrate data from multiple sensors, which is crucial for the precise measurement and analysis of aero-engine blades with complex geometries and varying structural characteristics.