Research On Key Technologies Of Matching Based On Binocular Vision Measurement For Large Forgings

As the foundation of equipment manufacturing,forging level of large forgings is of great significance to high-end equipment manufacturing industry.To reduce energy consumption,improve processing precision and material utilization,it is necessary to measure large forgings' dimensions online during forging process.It contributes to reasonable planning of forging process and enhancement of forging level.The urgent need of major equipments in various fields for large forgings can be effectively relieved,and the industrial image and ability of high-end equipment manufacturing can be comprehensively improved.Machine vision is applied to measuring forgings because of the advantages of non contect,high precision,extensive information.During measurement based on machine vision,matching is an influential process.Therefore,research on key technologies of matching is conducted based on laser-aided binocular machine vision measurement for large forgings in this paper.To solve the problem of that the matching fearures are inconsistent,equal-matching features are constructed by analyzing the matching characteristics of laser stripes.On this basis,the features can be extracted and matched,then the dimensions can be measured.Combined with measurement requirements,the measured objects can be classified into large cylindrical-shell and general(square and long-axis)forgings,for which the corresponding matching methods are carried out.For large cylindrical-shell forgings,it is difficult to match precisely and measurement results can be influenced because of the inconsistent boundary information in left and right images.Firstly,the characteristics of laser stripes on surfaces of large cylindrical-shell parts are analyzed.Secondly,the equal-matching features are constructed based on intersection of laser stripes.Then,the high-precision feature extraction method is proposed.Eventually,the features can be matched online,precisely and robustly.For other general(square and long-axis)forgings,the measurement results can be influenced by extraction of laser stripes' centers which do not meet the one-to-one correspondence matching relationship.The adaptive threshold image segmentation method based on support vector machine(SVM)regression is proposed.Before feature extraction,the SVM model is trained with abundant experimental pictures,then coefficient of Otsu threshold is predicted to distinguish effective information of light stripes from background information.On this basis,the location information of laser stripes can be determined,and the equal-matching features are constructed based on the differences of left and right camera's fields of view.The fine feature extraction can be conducted on the basis of initial extraction for centers of laser stripes.Feature matching is achieved with epiolar constraint and gray similarity constraint.Consequently,feature points are reconstructed and dimensions are measured precisely and robustly.Accuracy verification tests are carried out in a laboratory.The relative error of diameter measurement is 0.170%,and the relative error of width measurement is 0.183%,respectively.Feasibility verification experiments are carried out in a foring workshop.The measurement for width of square forgings and diameter of long-axis forgings is conducted online.The experiments fully proves the accuracy and effectiveness of the matching method proposed in this paper.