1 October 2025 in Engineering Research Express
Due to the structural constraints of sprockets and the efficiency requirements of on-site measurements, conventional external parameter calibration methods are difficult to apply to the calibration of the sprocket hub end surface. Therefore, this paper proposes a novel external parameter calibration method based on the invariance of quadratic curves. A cylindrical object with concentric circular rings is inserted into the central hole of the sprocket, and the radii of the large and small circles on the rings, along with the radius of the sprocket hub hole, are used to derive the coefficients of the equations corresponding to the three circles. Based on these coefficients, the external parameters of the sprocket hub end surface are determined. According to the contour characteristics of the sprocket teeth, a point selection model is established to identify the highest and lowest points within the tip and root circle regions. The radial runout is then calculated as the difference in radial distances between the highest and lowest points. In the experiments, eleven sprockets of different sizes were selected as test objects, and the results obtained using the proposed method were compared with those measured by a coordinate measuring machine (CMM). The results indicate that the measurement error of the tooth root and tip circles is less than 42 μm.
25 June 2025 in Engineering Research Express
Sheet metal coverings and other complex surface components feature a variety of holes in different shapes and sizes used for positioning. However, the batch measurement of these thin-walled, complex surfaces is highly challenging due to their susceptibility to deformation and the difficulty in planning measurement paths for the targets. This paper leverages laser triangulation methods and a three-degree-of-freedom test bench to capture point clouds of sheet metal surfaces. A novel Downsampling method that preserves edge features by reducing data dimensionality was developed from these point clouds, laying the groundwork for subsequent high-precision measurements. A simulated dataset was created and four feature descriptors were developed to train a Support Vector Machine (SVM) model for hole type classification. The four types of features extracted from real data are linearly separable in space, achieving accurate classification results. The measurement model was finally validated for accuracy on real data and noise simulated data, demonstrating that the measurement error of the proposed system is approximately 0.05 mm, indicating its potential for industrial applications.
14 October 2025 in Engineering Research Express
The registration of spiral-shaped three-dimensional structures is critical in engineering and biomedical fields. To address the challenging problem of registering spiral point clouds with low overlap, the Revolve ICP algorithm is proposed. By employing a single-degree-of-freedom point-to-point registration framework and dynamically adjusting the nearest-neighbor search threshold, the complex matrix decomposition issue is simplified to a single-variable optimization problem, thereby enhancing robustness while reducing computational costs. A PCA-based axis-fitting method is integrated for point cloud pre-alignment. Simulation and experimental results demonstrate that the proposed algorithm achieves significantly higher accuracy in regis-tering spiral rotational bodies compared to traditional methods, successfully achieving up to 100% overlap rates in applica-tions such as worm point cloud registration.
1 March 2026 in Engineering Research Express
The camshaft is a crucial transmission component, and its key dimensions, such as cam lift, base circle radius, and cam phase angle, significantly impact the engine’s performance. Traditional cam parameter measurement methods mainly rely on three-coordinate measurement (CMM) or dedicated contact-based methods. However, existing machine vision-based cam parameter measurement systems can only capture cam point clouds from a single cross-section and cannot measure all cam parameters along the entire camshaft or the cam-to-cam phase angles. This paper proposes a method for camshaft point cloud acquisition based on line-structured light and establishes a visual measurement model for key cam dimensions, including lift, base circle radius, and cam phase angles. Firstly, the paper proposes an accurate method for controlling the camshaft’s rotational angle and the vision system’s movement direction to obtain the complete camshaft point cloud. Secondly, a coordinate system perpendicular to the camshaft’s rotational axis is established to obtain the cam projection profile. Finally, measurement algorithms for the cam base circle radius, lift, and cam phase angle are developed. In the experiment, the camshaft of an automobile engine is used as the measurement object. The visual measurement results are compared with those from CMM. The average radius error is 0.043 mm, the average lift error is 0.072 mm, and the phase angle error is 0.317°. The total measurement time for the entire camshaft is reduced by 1198 s.
4 June 2026 in Engineering Research Express
Point cloud registration plays a crucial role in guiding the assembly of automotive sheet-metal parts. However, for sheet-metal components with smooth surfaces and sparse feature points, traditional point cloud registration algorithms often fail to achieve accurate alignment, leading to unsuccessful registration results. This paper proposes a point cloud registration algorithm that integrates both curvature and boundary features. First, principal component analysis and point-cloud plane projection are applied to the original sheet-metal point cloud to extract curvature feature points and boundary feature points, respectively. These features are then fused to construct a point set with representative registration characteristics. Subsequently, a two-stage RANSAC coarse registration method is introduced to achieve precise and efficient global alignment of the fused feature point cloud. Finally, an ICP-based fine registration algorithm is employed to iteratively optimize the aligned point cloud and further improve registration accuracy. Experiments are conducted on curved surface models and automotive sheet-metal parts exhibiting curvature and boundary features. The results demonstrate that the proposed method achieves a relative registration accuracy better than 0.22 mm within a range of 1 m. Compared with ICP-type algorithms, the proposed method provides more accurate registration for smooth surfaces and sheet-metal parts with sparse features, with relatively longer computation time but improved accuracy and robustness.