Aviation Blade Inspection Oriented Point Clouds Registration And Surface Reconstruction

As a key component of an aviation engine,the geometric dimensions of blades have crucial influence on the performance of the engine.The loss of power may happen when blades fail to meet the design requirements.So,the measurement and evaluation of a blade is an important task in the aircraft industry.Nowadays,common methods for blade measurement in the industrial community include contact measurement and non-contact optical measurement.Optical measurement is faster and more promising compared with contact measurement.Surface structured light fringe projection is an effective way of partially measuring the blade surface.When surface structured light is used to measure a blade,due to the limitation of measurement range,multiple measurements need to be carried out and different areas of measurement data need to be registered to get a complete surface measurement data.Reconstruction of a blade surface is the important foundation for its parameter assessment.Continuous surface interpolation plays an important role in the quality assessment of a blade,it reduces the difficulty in extracting blade parameters and improves the precision of the parameters extracted.This paper works on the registration and surface reconstruction problems of blade point clouds.The main work of this paper includes: 1)This paper proposes to use section curve as a descriptor to register blade point clouds and gives a method to compute the section curve descriptor.Section curve descriptor contains more geometric information of a blade than traditional point descriptors.It effectively overcome the influence of featureless and symmetry on blade point clouds registration.2)This paper gives a method to coarsely register blade point clouds using section curve signature.The method is based on the RANSAC algorithm,and invariance property of distance under rigid body transformation is used to construct the correspondences between section curve signatures so as to complete the coarse registration.Coarse registration result can further be refined using Iterative Closest Point algorithm with point to plane error function or its variants.3)This paper proposes an applicable method for the surface reconstruction of blade point clouds.The proposed method utilizes moving least squares technique to construct local coordinate systems for the space grid points.Weighted polynomial fitting under local coordinate system is used to compute the normal projection points for the grid points.Algebraic sphere fitting is used to estimate the normals for the projection points so as to get the signed distance field.Zero iso-surface extraction from the signed distance field completes the work of surface reconstruction.The work of surface reconstruction lays the foundation for the follow up work of error evaluation.4)This paper designs experiments for the proposed registration method and reconstruction method.Accuracy and effectiveness of the proposed methods are analyzed according to the experiments.