Local Scanned Point-Cloud Registration of Aeroengine Pipeline System Based on Axial Constraints

Abstract

The final assembled state is one of the crucial factors in the quality of an aeroengine pipeline system. It directly affects the safety, reliability, and thrust-to-weight ratio. Because of the limitations of aeroengine assembly and inspection technologies, ensuring the consistent assembly of pipeline systems in aeroengine production has become a challenging and key issue. We inspect the assembly of an aeroengine pipeline system with high precision and reliability using a single-frame point cloud that is acquired with a structured-light 3D scanner. To achieve this, we propose a novel registration framework that considers the structural constraints of the aeroengine pipeline system to achieve an accurate point-cloud alignment between the single-frame point-cloud data and the standard model. First, pipe axes are extracted from the single-frame point-cloud and standard models as the constraints for the coarse registration. Subsequently, a new point-cloud descriptor for curved segment features is designed to implement the rapid retrieval of structural features from both the single-frame data and relative models automatically. Finally, we use a voting mechanism to determine the best result when there are multiple matches. This allows us to achieve accurate and efficient registration with a non-uniform assembly structure. Experiments on both actual and synthetic datasets illustrate that the proposed registration method achieves higher accuracy and efficiency. Compared with the state-of-the-art algorithm, our method reduces the RMSE value by approximately 9 times during the quasi-real-time single-view raw point cloud assembly process of the aeroengine pipeline system.