Key Technology Of Machine Vision Measuring And Its Application In Slender Shaft

Nowadays, there are huge needs for measuring geometrical parameters of slender shafts in machinery industry and rail transit. Conventional methods, mainly contact measurement, are time-consuming and labor intensive and measuring force may cause damage to workpieces. These methods can no longer meet the new demands of the manufacturing industries. Machine vision measuring (MVM) technique is a non-contact measurement method which takes images as transfer carrier of measurement and information and it has been caused considerable attention. Therefore, with the support from National Natural Science Foundation of China (Grant No.50805023), a new vision measuring system for the slender shafts and other large-scale parts has been developed. Main contents are showing as follows:(1) A measuring method based on an external reference has been proposed for slender shafts to solve the inconsistency problem of measurement range and measuring accuracy. According to the method, the slender shaft that to be measured is divided into a number of measurement-awaiting shaft-segments. Images of each shaft-segment, alongwith a coded and calibrated reference, are taken and gathered at different object distance. The code of the reference is obtained by using the template match technique. Consequently, the coordinates of feature points of the reference are known. Thus, the position parameters of the camera are calculated accordingly. A mathematical model among the diameter, the center coordinate and the position parameters is established on the basis of the pinhole imaging principle. The problem that only one image can’t extract the center coordinate and pseudo diameter problem are solved. A coordinate transformation model for these feature points is established and the sectional measuring results are associated by using the transformation model. The measurement will eventually be realized in the direction of slender shaft axis.(2) Method on the basis of optimization algorithm for positioning of light source is established to achieve uniform illumination with which high quality images could be gathered for MVM. The optimization objective is determined and the objective function is designed according to the question characteristics. The SA (Simulated Annealing) algorithm is chosen to solve the non-convex objective function. An experiment system which uses the TES-1330A illuminometer and the GXY2020GT4-XLE bi-dimensional worktable is developed for illumination photometry. Then the light intensity is observed on a 200mmΤ200mm target plane which is meshed by squared grid with size 5mm. The maximum difference is within 4%. The results show that the optimization objective is reasonable and the method is feasible.(3) An adaptive filter which is based on the spatial generalized autoregressive (SGAR) model is implemented to process images contaminated by mixed noises in industrial locale. The SGAR model is deduced that uses only uniform expression for both the linear and non-linear autoregressive model based on Weierstrass theory and the accuracy of the model is verified by simulation data. And on that basis we check the influence of the model type, the size of prediction window and the sub-image size on the reconstruction accuracy of texture images. Filtering experiments are carried out and the results show that the new filter can effectively denoise Gaussian noise and Poisson noise as well as impulse noise and mixed noise. The edges of restored images keep still sharp and smooth which ensures the accuracy of the successional image algorithm. A noise-insensitive and edge-preserving resolution upconversion scheme for digital image based on the SGAR model is proposed to improve measuring accuracy. Firstly, the structure of an image window is learnt adaptively by using the SGAR model. Parameters of the SGAR model are estimated in a moving window in the input low-resolution (LR) image by using the robust Generalized M-estimator (GM-estimator). Then, the interpolation model is established from the learnt model and a new feedback mechanism in accordance with the residual sum of squares minimization principle. Thirdly, the gradient simulated annealing algorithm (GSAA) is used for solving the interpolation model which can fast converge to the global optimum in probability with the help of gradient information. Experiments have been done on the worldwide datasets to evaluate the performance of our method. The results demonstrate that our method is superior as compare to the recent autoregressive model based method and bicubic, especially when images are polluted by noises.(4) A novel Canny algorithm without manually setting parameters is proposed. The earlier adaptive filter utilization and automatic calculation of low and high thresholds are studied. Firstly, the gradient image of the input image is obtained and then normalization processing is carried out for both the gray image and the gradient image which are used to calculate the gray level-gradient co-occurrence matrix. Grey entropy mathematical model is established according to the matrix and the simulated annealing algorithm is used to solve the grey entropy model. The optimal gray value and the optimal gradient value for the image segmentation are obtained. The optimal gradient value is used as the high threshold and the low threshold is equal to 0.4 times of the high threshold. Experiments are done on the worldwide datasets to evaluate the performance of our method. The results demonstrate the superiority of our method when compared with the best parameter values method and standard Canny, especially when images are polluted by mixed noises containing Gaussian noise, Poisson noise and impulse noise.A MVM experimental system for slender shafts, which synthetically utilizes all above key technology, is designed. Measurement experiments are carried out on screws and the measuring results are compared with those of coordinate measuring machine (CMM). Experimental results show that the measuring uncertainty of the experimental system is 4.9 μm when the diameter is measured and the value is 7.0 μm when the straightness is measured, respectively. The results of the CMM are 2.3 μm and 6.2 μm respectively. The system error of the experimental is 0.002 mm during the diameter measuring and the value is 0.017 mm during the straightness measuring, respectively. It is proved that the proposed theories and methods are not only feasible but also of great application value.