A Study on Streamlining and Denoising Point Clouds for Aero-engine Blade Surface Reconstruction

Aiming at the issue of numerous redundant, noisy, and outlier points in the reconstructed point cloud of aero-engine blade surfaces, which significantly affects reconstruction accuracy and efficiency, a point cloud streamlining and denoising algorithm based on multi discriminative parameters and improved Guided Filter is proposed. The algorithm is designed to streamline the numerous redundant points generated by the high-precision acquisition of the blade point cloud. In the feature area, the curvature of each point, the sum of the normal angles between the point and its neighboring points, and the median of these values are calculated. These discriminative parameters are then combined to extract the feature points in the feature area. For the non-feature area, the k -means clustering algorithm is used to extract key points. This algorithm uniformly streamlines the point cloud while preserving the detailed features of the blade surface point cloud data. To effectively eliminate challenging noise and outliers, the geometric features of the point cloud are used as weights for the overall smoothing and denoising of the model. The results show that after processing the point cloud data with the streamlining and denoising algorithm and then triangulating it, the surface reconstruction quality significantly improves. Specifically, the accuracy increases by $38.46 \%$, and the processing efficiency improves by $\mathbf{3 6 . 0 7 \%}$. This enhancement has led to more precise path planning for the ultrasonic nondestructive flaw detection robot used on aero-engine blades, thereby improving flaw detection accuracy and efficiency.