Defect detection in impeller parts utilising local geometric feature analysis

Impeller blades in aero-engines, vital for both military and civilian use, significantly impact operational efficiency through their surface quality. Surface defects like scratches can lead to severe consequences, including engine failures. Efficient defect detection in these blades is crucial. Traditional two-dimensional imaging methods fall short due to the blades’ complex geometry and hidden defect areas. Currently, manual inspection methods dominate, leading to increased costs and reduced efficiency in industrial settings. Addressing these challenges, our paper introduces a three-dimensional point cloud-based defect detection method for impeller blades. This approach involves segmenting the blades using point clouds, employing normal vectors for data extraction, and reducing computational load through voxel down-sampling. It features a unique local feature extraction technique, combining normal vector aggregation with Fast Point Feature Histograms (FPFH) and fuzzy C-means clustering, to accurately identify blade defects. This method integrates both low-dimensional and high-dimensional features using a clustering algorithm to enhance scratch defect detection in complex impeller parts. The experimental results demonstrate that compared to existing leading methods, our approach achieves a defect detection rate that is more than 5% higher and the computational efficiency is increased up to 3 times compared to traditional solutions.