Damping springs and bolster springs (hereinafter referred to as bolster springs) are important components of a railway freight car bogie, which carry the main load when the vehicle is running. According to the relevant regulations, bolster springs need to be repaired and replaced regularly. Moreover, the driving safety of the vehicle is directly affected by the dimensions of each group of bolster springs. Thus, it is necessary to disassemble the bolster spring and perform maintenance on each part separately, and the qualified and unqualified products need to be distinguished and classified based on their dimensions. Under the traditional operation mode, the model and type of bolster spring are determined by measuring the height of the bolster spring using a free height ruler, measuring the diameter of the round steel of the bolster spring using a spring diameter gauge, and through manual visual inspection. Typically, several bolster springs of different types are used in railway freight cars . Thus, the use of the manual detection method is inefficient, labor intensive, and vulnerable to human interference, making it difficult to meet the current maintenance needs of bolster springs. To develop an intelligent maintenance system for bolster springs that can replace manual operation, a key technical problem, namely, the design of a stable and efficient detection algorithm for bolster springs, needs to be urgently solved. Through investigations and analyses, researchers have made certain advancements in the field of bolster spring detection; however, most of them include patents and very few are related to detection algorithms. Therefore, a point cloud detection method for the bolster spring size is proposed in this paper based on 3D laser point cloud technology to achieve the stable, high-precision, and efficient detection of the bolster spring size.
In this study, the bolster spring point cloud data were collected using a line laser sensor and the size of the bolster spring was measured using point cloud processing technology. First, the bolster spring measurement test platform was set up, and the three-dimensional point cloud data of the bolster spring were obtained via linear laser sensor scanning and uploaded into an industrial computer for processing. For the bolster spring point cloud data, the point cloud density was reduced and the noise was removed using the point cloud down sampling algorithm and k-means clustering algorithm. Subsequently, the free height of the bolster spring was measured using the plane fitting algorithm and point projection height calculation method based on the processed point cloud. Finally, the inner diameter and outer diameter of a group of internal and external bolster springs were calculated using the point cloud dimension reduction algorithm combined with the edge extraction, circle segmentation, and circle fitting. Through experimental research, the accuracy and efficiency of the bolster spring measurement algorithm based on point cloud technology were discussed, and the reliability and stability of the method were verified.
The manual measurement results and bolster spring point cloud data were obtained via manual measurement and the linear laser sensor scanning for 10 groups of bolster springs, respectively (Fig. 9), and the sparse point cloud data with reserved features were obtained by processing the bolster spring data using a point cloud preprocessing algorithm (Fig. 10). Furthermore, the reliability test for the height and diameter measurement was conducted on 10 groups of bolster spring data. The results show that the maximum measurement error of the point cloud measurement method is in the range of ±0.35 mm compared with that of the manual measurement, and the average measurement time is 2.7 s, which is much shorter than the manual measurement time. Additionally, the measurement accuracy and measurement efficiency are both reliable and meet the measurement requirements (Table 2 and Fig. 11). The four groups of bolster springs were measured 10 times. The results show that compared with that of the manual measurement, the maximum average error value of the proposed algorithm is in the range of ±0.35 mm, and the measurement repeatability is less than 3%, which proves that the proposed algorithm has good stability (Tables 3?5). Thus, the obtained reliability and repeatability test results are acceptable and meet the requirements stipulated in existing maintenance regulations.
In this study, a method for measuring the size of the bolster spring based on 3D point cloud technology was proposed. The point cloud data were obtained by controlling the driving line laser sensor of a six-axis robot when scanning the bolster spring. To reduce noise interference, the bolster spring point cloud data were preprocessed via sampling using the point cloud sampling algorithm and k-means clustering algorithm. After preprocessing, the point cloud data were used to solve the end and bottom plane equations of the bolster spring using the least square plane fitting algorithm. Based on the plane equation and point projection height calculation method, the free height measurements of the bolster spring were obtained. Subsequently, the bolster spring point cloud height information was removed using the dimension reduction algorithm. The reduced dimension point cloud was then circular-segmented using the maximum included angle edge extraction algorithm and threshold segmentation algorithm. Subsequently, the noise points of the segmented point cloud were filtered using the RANSAC method, and the diameter of the bolster spring was calculated using the circle fitting algorithm. To improve the diameter measurement accuracy, the gradient descent iterative algorithm was used to iterate the optimal diameter solution using the circle fitting results as the initial values, marking the end of the calculation of the inner and outer spring diameters of the bolster spring. The bolster spring measurement test platform was then set up for data collection, conducting reliability and repeatability tests on the industrial computer, and conducting reliability tests on 10 groups of bolster springs. The results show that the maximum measurement error of the point cloud measurement method is in the range of ±0.35 mm compared with that of the manual measurement, and the average measurement time is 2.7 s. Moreover, the measurement accuracy and measurement efficiency were reliable and met the measurement requirements. The measurement tests on the 4 groups of bolster springs were repeated 10 times, and the results show that compared with that of manual measurement, the maximum average measurement error value is in the range of ±0.35 mm, and the measurement repeatability is less than 3%, which proves that the point cloud measurement method has good stability. Follow-up work should be conducted to optimize the measurement system and point cloud measurement algorithm, to further improve the measurement accuracy, measurement efficiency, and measurement stability, and to apply the point cloud measurement method to the research on the development of automatic measurement and sorting equipment for bogie bolster springs and wedges.
