Precision Measurement Method And Application Of Complex Parts Based On Machine Vision

The quality of key components in high-end manufacturing and intelligent equipment serves as a crucial foundation for enhancing the level of equipment in national defense and economic construction while promoting high-quality development.Measurement of large-size parts,exemplified by heavy-duty vehicle crankshafts,and high-precision measurement of micrometer-size parts,exemplified by oil pumping rod threads,is generally conducted through manual offline sampling during the manufacturing process or before leaving the factory.This approach results in slow detection efficiency,high manual labor intensity,an inability to trace detection results,and a propensity to cause missed inspections.High-precision measurement technology based on machine vision presents a significant means of improving the quality of precision products,boosting production efficiency,and reducing costs.However,the machine vision measurement system currently used online has issues with insufficient detection efficiency and accuracy of results and cannot completely replace manual inspection.This dissertation conducts in-depth research on monocular vision measurement methods and their application in precision measurement of mechanical parts,proposes a new edge detection operator to achieve rapid extraction of straight line edges in specific directions under complex image backgrounds,and solves the problem of rapid measurement of multi-edge oil pumping rod thread tooth shapes;proposes multi-view collaboration in precision measurement of large-size parts,applied to the measurement of the main shaft diameter of heavy-duty truck crankshafts.The main research contents and results are as follows:(1)The mathematical expressions of the measurement errors related to the camera parameters and pose states are theoretically derived to quantitatively analyze the measurement errors caused by the misalignment of the actual measurement plane and the system calibration plane.The measurement experiment is carried out on the rotation motion of the measurement plane around the Xw axis and the translation motion of the Zw axis with the industrial lens and telecentric lens respectively.Theoretical analysis and experimental results show that when the camera optical axis is perpendicular to the measurement plane,the measurement error is independent of the principal point coordinates of the camera’s optical axis,the translation components in the measurement plane and the rotation angle of the optical axis of the camera,and linearly related to the translation component in the direction of the optical axis of the camera and the pixel distance of the measured workpiece,and inversely proportional to the normalized focal length of the camera.When the camera optical axis is not perpendicular to the measurement plane,the measurement error is a nonlinear function of parameters of camera normalized focal length pixel sizes and rotation angles.If the measurement plane has changes in rotation and translation pose,the measurement error can be reduced by adjusting the shooting angle or position of the measured workpiece.(2)Aiming at the problem that traditional edge detection operators have poor sensitivity to image edge direction,a directional edge detection operator based on Gaussian function weighting(G-DEDA)is proposed for edge extraction of specific direction in multi-segment edge images.Unlike traditional Prewitt and Sobel algorithms,which only have corresponding edge detection convolution kernel in a specific angle,G-DEDA uses Bimodal Gaussian function of bilateral filtering to establish a general functional relationship between convolution kernel and the detection angle.Firstly,the equations of the edge line and its normal line are obtained according to the detection angle.Then,the distance from each point in the convolution kernel to the edge line and its normal line is calculated respectively to obtain the bilateral distance,which is put into the Bimodal Gaussian function to calculate the weight of the point in the convolution kernel.Finally,the weight symbol of each point is calculated according to the relative position relation between each point in the convolution kernel and the edge line,and the template of the convolution kernel is generated.The matrix sparsity increases as the size of convolution kernel increases,and it can effectively improve the generation speed of the convolution kernel,reduce the space required for storage of the convolution kernel,and improve the speed of subsequent convolution operations.The edge detection experiments with simulated images and real images show that G-DEDA has the advantages of good applicability,high detection efficiency and accuracy,and makes up the shortcomings of traditional edge detection operators in representation of edge directions.(3)In order to solve the problems of thread tooth misalignment and dust noise in the thread image,the algorithms of connected domain denoising,connected domain sorting,and connected domain merging are designed.The thread tooth angle,pitch,large diameter,medium diameter and small path are measured through accurate position of the thread teeth by the right and left edge matching algorithm,and high precision measurement.The experiment results of parameter measurement of a sucker rod thread show that the minimum and maximum error of the thread angle are 0.011° and 0.63° respectively,and the total average deviation is less than 0.08°.Compared with the universal tool microscope(UTM),the deviations of thread pitch,large diameter,medium diameter and small diameter measured by the presented method and UTM are all less than 10μm.The measurement time of all paramenters of one thread with the presented method is less than 0.13s,which can meet the real-time requirement of online thread measurement.(4)A multi-field collaborative precision measurement method is proposed to solve the problem that the measurement range and measurement accuracy cannot be balanced in the precision measurement of large-scale parts,and applied to the diameter measurement of the crankshaft shaft.The principle of collaborative measurement with dual fields of view is proposed,and a collaborative measurement system of dual fields of view is built.A customized high-precision calibration board is designed to improve the calibration accuracy of the dual camera acquisition system,and the fitting accuracy of the edge line is effectively improved through the sub-pixel processing of quadratic interpolation.The proposed method is experimented to measure the diameter of a crankshaft with a nominal diameter of 100 mm.Compared with CMM,the maximum values of μ95 measured by CMM and the proposed method are ±2.11 μm and ±1.82 μm respectively,and the mean values and mean square error values measured by the two methods are close.The results show that the proposed method achieves the same measurement accuracy as CMM.