Research On An Onsite Inspectation Of Formed Hull Plates Based On Binocular Stereovision

Although China is currently a primary shipbuilding country, it is still facing a huge gap in both manufacturing technology and process, compared with the advanced shipbuilding countries. The formation of curved plates is a key step of the manufacture of ship hull plates. However, most domestic shipbuilding enterprises still now use manual line-heating process to form a double-curved plate. The process uses wooden templates to diagnose the quality of a formed hull plate. As the wooden template is imprecise, inefficiency and costly, the inspection method based on wooden templates has become a bottleneck in the fabrication process of hull plates. Therefore, developing the digital inspection technology of formed hull plates has become a top priority of Chinese shipbuilding industry. For that, an onsite inspection technology based on binocular stereovision for formed hull plates is studied in the dissertation. According to characterizes of formed hull plates and requirements of the forming process, the development of the binocular stereovision sensor, the onsite calibration of the senor, the accurate extraction of projection features, the fast 3D reconstruction of formed surface and the self-feedback inspection are focused in this dissertation.Generally, it is very difficult to use conventional binocular stereovision to inspect a formed hull plate, because the hull plate has a large shape size, high demand of formed precision and big processing volume. To solve this problem, firstly an onsite measurement technology scheme based on binocular stereovision for formed hull plates is proposed. Secondly, the structure parameters of the binocular stereovision are further optimized and a rated working distance of 2500 mm and a baseline of 1500 mm are chosen. Finally, an onsite measurement sensor of binocular stereovision is developed.The long-term effects of vibration and temperature variation will arouse the failure of the structural parameters of the binocular stereovision, which reduces the measurement accuracy. Therefore, the structural parameters of the sensor in general require an onsite periodic calibration. However, the traditional large target is not suitable for the onsite calibration under complex conditions for inconvenient operation, custody and maintenance. In order to solve this problem, an onsite structure-parameter calibration method for large field of view(FOV) binocular stereovision based on a small-size 2D target is proposed. The proposed method uses a traditional large target to calibrate camera intrinsic parameters for the first time in the offline procedure, while uses a small 2D target to periodically calibrate the structure parameters of the sensor during the onsite procedure. The calibration results show that the measurement accuracy of the stereovision sensor calibrated by the proposed method is improved by 13% compared with the other from standard calibration method.Accurate extraction of the projection features has a direct effect on the 3D reconstruction accuracy. However, the integrity of projection spot might be damaged by a hull plate’s nonuniform optical surface, depth steps and occlusion, and then the direct ellipse-specific algebraic fitting(DESAF) method would fit an incorrect ellipse which could not accurately match the complete contour of the spot. To solve this problem, an iteration-based direct ellipse-specific algebraic fitting(IBDESAF) is proposed to remove these outlier points from the contours of incomplete spots. The experiments of computer simulated data and real data demonstrate that compared with DESAF, IBDESAF can dramatically remove the outlier points from the contour and improve the detection accuracy of center of the incomplete spot.Traditional structured light technologies projecting multiple patterns are susceptible to vibration disturbance in the workshop and have relatively low measurement efficiency. In order to solve that problem, a stereo matching method using single pattern of structured light based on neighborhood topology is proposed. The method codes each spots in the projected pattern with unique coordinates. Then, according to the coordinates of spots, the stereo correspondence can be established. It is very suitable for the measurement of formed hull plates. Experiments show that the proposed method can accurately code the spots arranged in different cases and be used to quickly obtain the 3D reconstruction of a formed hull plate, and the measurement time for a single view is less than 2 s. Finally, the performance of the developed stereovision sensor is evaluated by a test plate. The test shows that the effective working distance of the developed binocular stereovision sensor is about(2300 mm, 2900 mm), working error is no more 0.5 mm, and measurement field is no less than 2000 mm × 2500 mm.In order to evaluate the formation status of a hull plate, the measurement data and the theoretical model need be registered and compared. However, for the formation inspection procedure of a hull plate, there are some problems that should be solved. First, as the formation process of a hull plate in general is multistep and asymptotic, the measurement data and the theoretical model might have big differences in shape, position and posture, which would make conventional registration method difficult to be applied. Second, for the point cloud data, the corresponding points between the measurement data and the theoretical model are difficult to be accurately and fast determined by traditional methods based on the nearest point of point-point rule or point-plane rule, especially when there is a big difference in shape. Third, traditional methods based on comparing the measurement data and the theoretical model generally lack the onsite feedback function of formation errors. In order to solve these above problems, some methods are proposed in the dissertation. First, an automatic rough registration method based on the similar global features is proposed to solve the problem of big different shape. Second, for the formation process of hull plates, a method based on projection interpolation is proposed to find corresponding points between the measurement data and theoretical model. The proposed method is used to evaluate the surface forming accuracy, the lateral forming accuracy, the longitudinal forming accuracy and the distortion error. Third, combining the coding principle of the proposed structured light technology, a method that can project the formation error to the formed surface of a hull plate is proposed, which is useful for the workers to determine the subsequent forming process parameters. Finally, a saddle hull plate is used to test the inspection method proposed in this dissertation. The test results demonstrate that the proposed method can achieve the entire inspection process including the 3D surface reconstruction, the registration and comparison of measurement data and theoretical model and the feedback of formation error.