Research On Identification And Positioning Of Multi-weld Seams In Large Tubesheet And Welding Autonomous Control Method

Pressure vessels are widely used in aerospace,petrochemical and other fields,which occupied an important position in the national economy and national defense industry.Pressure vessels are mainly produced by welding,and due to the extremely high requirements for sealing and pressure-bearing performance,pressure vessels ask for high requirements for welding quality.With the rapid increase in the number of tubesheet weld seams,the welding process is affected by size errors,multi-batch manufacturing,and thermal deformation during the welding process,resulting in low consistency of weld seams.Point cloud reconstruction of weldment has the advantages of simple data and accurate scale,which is an important way to achieve "no teaching" welding.However,there are still many difficulties: the existing point cloud reconstruction methods have a narrow field of view,which is difficult to meet their reconstruction needs;the number of welds is numerous,and the welding positioning methods in the point cloud model need to be researched;the distribution of weld seams is dense,and the change in arc length during the welding process under the effect of heat seriously affects the welding quality.To this end,this paper focuses on key technologies such as identification and positioning of multi-weld seams and arc length control for large-scale tubesheet seam welding,including the following aspects.(1)A tubesheet point cloud model construction method based on a monocular camera and line laser was proposed.By combining a monocular camera and a line laser,the scale uncertainty from the monocular camera was solved,and then multi-sensor data acquisition and fusion research were carried out.An image similarity analysis method based on the vocabulary tree algorithm was proposed to obtain the similarity between tubesheet images.Feature extraction and matching of tubesheet images were operated,and a random sampling consistency algorithm was used to improve the feature matching.Through the polar geometry and triangulation method,an initial point cloud of the tubesheet was calculated,and a point cloud model of the tubesheet was constructed by Pn P(Perspective-n-Point)and MVS algorithms.(2)The method of multi-noise tubesheet point cloud weld seam recognition was studied aiming at the large noise and multi-defects in the tubesheet point cloud model.The defects of the point cloud model were analyzed,and then the voxels of the point cloud model were divided.A method of weld seam recognition based on the voxel density of the tubesheet point cloud was proposed,and the voxel density domain of the tubesheet point cloud model was established to realize circumferential weld seam detection.To address the misdetection problem in the voxel density method,a deep learning method for weld seam detection of the tubesheet point cloud model was proposed,and a voxel-based point cloud model annular weld recognition network was established.A model loss function was constructed,and the point cloud data was augmented to complete model training.(3)A multi-layer positioning strategy was designed to realize image-based positioning of welding robot in the tubesheet point cloud model.The multi-coordinate system position relationship of tubesheet welding was sorted out,then a multi-coordinate system model was constructed,and the method to solve the multi-coordinate system conversion relationship was proposed.A feature-based maximum likelihood calculation method for the tubesheet point cloud model was proposed,the feature of the annular weld seam of the tubesheet image was extracted,and the region of interest in the point cloud was selected,based on that,a likelihood calculation method between the point cloud and the image was proposed to complete the likelihood domain calculation.And the coarse positioning of the weld seam was finished by maximum likelihood estimation.Then the region of interest in the tubesheet image was selected,the circular weld seam in the image was detected,and the center of the weld seam was extracted,by which the coordinate adjustment of the welding robot was completed and the precise positioning of the welding robot was realized.(4)An adaptive arc length control method of tubesheet welding based on an adaptive Kalman filter was proposed to realize the stability control of the welding process.The relationship between arc voltage and arc length was researched,the arc voltage pre-processing method was proposed,the arc voltage collection experiment was carried out,and the relationship between arc voltage and arc length was constructed by the least square method.The arc length estimation method based on the Kalman filter was proposed,the Kalman filter model of tubesheet welding was established,and the Kalman filter parameter selection experiment was carried out.The arc length stability control method based on the adaptive Kalman filter was proposed,and the adaptive Kalman filter algorithm was used to optimize the effect of the Kalman filter algorithm.Based on the calculation of the control step length,the stability control of the tube-sheet welding was realized.