Research On Multi-view Vision-based Automatic Measurement Method For Complex Bend Tubes

Complex spatial bend tubes are widely used in products ranging from aeroplane to aircraft as a transformission supporter of materials and energies.The function of these bend tubes in aeroplanes are as vessels in human body.The assembly reliability of the bend tubes directly affect the performance and life cycle of products.Before the tube assembly,inspecting its 3D shape is necessary to achieve the goal of precise and stress-free assembly.However,currently,bend tubes with multi-brances,non-circular cross-sections or multiradius are still lack of accurate and rapid inspection methods.In this paper,multi-view visionbased methods are developed to solve measure problem for complex bend tubes in aircrafts,and a mesure system is also developed and verified.The main researches of the paper include the following aspects:(1)The application background and importance of bend tube measurement in aeroplane and aircraft was introduced in this paper.The current methods have disadvantages with depending on labor interactions and cannot measuring the complex bend tubes.Aiming at overcome these problems,we investigated and summaraized relate researches about machine vision based measurement technologies in aspects of close-range photogrammetry in industry,contour based 3D reconstruction,industurial parts detection technology and machine learning based shape feature recognition.The contributions of our method are also explained.(2)A multi-view vision based bend tube measurement the fundamental technologies and framework are propsed.In view of difficulties in vision-based tube measurement,the overall framework and process of complex bend tube measurement are clarified.According to the overall framework,the fundamental technology of multi-view vision-based tube measurement is based on technologies of camera calibration and sub-pixel edge extraction.The camera is calibrated by the maximum likelihood estimation.A sub-pixel edge extraction method based on clustering analysis was proposed to obtain tube’s accurate edge.(3)For the multi-branchs bend tubes,a neural network based is proposed to recognize tube’s structures.According to the shape descriptor,the different structures of bend tube are transformed into signature.Then,a neural network is trained by the signatures to recognize different tube structures as straight part,curved part,joint and othter connected components with an accuracy as high as 97%.Combined with inpection of tube and joint,the multibraches bend tubes can be measured automatically.(4)For the non-circular cross-section tubes,an extraction method for 3D skeleton of tube based on multi-view vision was proposed in this paper.The method has no limitation of the shape of tube’s cross-section to reconstruct the 3D skeleton of bend tubes,which consted by points and corresponding tangent vector of center line.Besides,during the reconstruction,the method can also excludes the unavailable images with self occultion.Based on the extracted 3D skeleton,the Plücker coordinate of skeleton can be calculated and applied to distinguish the tube’s straight parts from curved parts.Thus,the non-circular cross-section tube can be measured rapidly and accurately.(5)To measure tube’s bend radius,a photo-consistency based method is proposed.The representation for tube center line in curved part was formulated,and the mapping from tube’s center line to edge point in the image was calculated.Then,the photo-consistency was used to reconstruct 3D center line of the bend tube curved part.The measurement of tube bend radius is converted to a constrained optimization problem.Thus,the multi-radius bend tube can be measured.(6)A virtual fixture method is proposed to evaluate the tube’s processing qualification.Aiming at precise assembly,the virtual fixtrures are set to find an optimal postion for tube assembly.The optimization is calculated by a weighted least square method.Once achieving the optimal positon,the sheath deviations of evaluated tube are calculated to evaluate the tube’s processing qualification.(7)Based on the above researches and methods,a multi-view vision based bend tube measurement system is developed.To show the feasibility and accuracy of our method,experiments are conducted.The results shows that the bend tubes with multi-branches,noncircular cross-sections or multi-radius are all measured rapidly,accurately and reliably.The average deviation of tube measurment is not over 0.3m and the measurement accuracy of bend radius can reach to ±0.5 mm.The accuracy rate of our methods can reach 99.73%.The method is rapid,each single tube can be measured in 6 seconds,and whole multi-branch tubes measurement is less in 1.5 minutes without labor interactions.