Research On 3D Reconstruction Method Of Shape And Dimensions Of Compressor Blades Based On Computer Vision

Aero-engine is the core of China's aviation industry.It is of great significance in military field,national defense construction and the development of national economy.The engine blade suffers from high temperature,high pressure and high speed while the aero-engine is working.The engine blade measurement is of great importance for the production and maintenance of the aero-engine on account of its easy deformation and numerous quantity.It's obvious that the measurement is complex because the surface of the engine blade is of free-form three-dimensional curved and many key parameters can not be directly measured.In order to improve the manufacturing quality and maintenance quality of the engine blades,a rapid and efficient blade measurement method is urged to developed.In this dissertation,the compressor blade is determined as the research object.A method based on computer vision is proposed to measure the key parameters of the compressor blade.The basic idea and main work of the project are as follows:The 3D reconstruction method of structure from motion is adopted to reconstruct the compressor blade model.The algorithm consists of visual odometer and pose optimization.The visual odometer refers to the process of calculating the point cloud model from a series of images.Feature extraction,feature matching and pose estimation are the core of the algorithm.A nonlinear least square method is adopted to optimize the positions and poses evaluated.By adjusting the estimated positions and poses,the error of the reprojection is minimized and the global optimal point cloud model is obtained.With the optimized positions and poses,all pixels in the image are uniformly projected into the 3D coordinate system,and dense reconstruction result is obtained.The blade surface is fitted according to the point cloud model based on non-uniform rational B-Splines.Establishing the coordinate system according to the characteristics of the blade's own so that the point cloud model obtained from each measurement can be unified in the same coordinate system.The algorithm of random sample consensus is used in point cloud segmentation.The algorithm is robust and suitable for extracting correct models from data with noise.Comparing the Bezier,B-Spine,NURBS curved surface fitting method,both theory and experiment have proved that the NURBS surface fitting method is the most applicable to solve this problem and finally the compressor blade model isevaluated from the discrete point cloud.Furthermore,the characteristics and parameters of the compressor blade itself are introduced,including the definition of the radius of the front edge,the radius of the rear edge,the chord length,the maximum thickness and the related tolerance requirements.In addition,the algorithm of calculating the important parameters of compressor blades is designed and completed through relevant point cloud computing.Finally,the experiment and analysis of the whole compressor blade measurement system are carried out.After all the preparation of experimental conditions and environment such as the designing and producing the measured compressor blade and the selection of the camera and lens of the whole measurement system,experiments have been carried out successively.Through the texture test of the compressor blade,a suitable method of blade texture strategy is obtained.The repetitive experiment shows that the repeatability of the radius of the front edge is 0.123 mm,the radius of the rear edge 0.114 mm,the precision of the maximum thickness of the blade 0.210 mm,and the chord length could reach 0.798 mm.Compared the parameters obtained using the method in this paper with the ones using coordinate measuring machine,the error lies between-0.370mm~0.420 mm,which meet the measurement requirements,and the efficiency has advantage over coordinate measuring machine.