Bolster spring is usually installed in the inner space of the narrow bogie side frame. It is one of the important components of the vibration damping device of railway freight cars, and it requires regular repair. During the repair of the bolster spring, the inner and outer spring groups need to be removed from the side frame and reinstalled in their original positions after the repair. Presently, the bolster spring is disassembled by manually tilting and pulling out. The highly limited side frame space causes the bolster spring to easily collide with or possibly block the locating pin above the inner cavity of the side frame, thereby preventing the smooth removal of the bolster spring. Based on manual operation experience, the gap position of the upper-end surface of the outer spring is usually determined by human eyes and manually turned directly below the locating pin to prevent a collision. However, this process requires high labor intensity and owns low operational efficiency. The key technical difficulty to be solved urgently is positioning the end gap of the bolster spring in a narrow space with uncertain lighting conditions to develop a set of intelligent disassembly equipment for the bolster spring in a narrow space to replace manual operation. After analysis, it was observed that there is no relevant literature to conduct positioning research on the bolster spring object with complex geometric properties in a narrow space. This research proposes a visual indirect positioning method of bolster spring gap based on a line laser to solve the problem of automatic positioning of bolster spring gap in a narrow space, aiming at the new research object of bolster spring, based on line laser, machine vision, and image processing technologies.
In this research, a line laser vision system, an object detection algorithm based on deep learning, and an image processing algorithm were used. Firstly, the variation law between the height of the spring coil and orientation of the end gap was derived by analyzing the structural characteristics of its outer spring and collecting the height data of the first and second layers of the spring coil, aiming at the new research object of bolster spring. Then, the vision system based on a line laser was applied to the new field of bolster spring gap positioning. The adaptive solution method of spring coil height was studied with YOLOv3-tiny object detection, threshold segmentation, and bounding box fitting algorithms. Finally, the visual indirect positioning method of bolster spring gap suitable for a narrow space was analyzed using the adaptive solution method of spring coil height and mathematical model of the relationship between spring coil height and end gap. The accuracy, efficiency, and illumination resistance of bolster spring end gap positioning using this method in different orientations were analyzed experimentally.
The mathematical model based on the least square method represents the corresponding relationships between the spring coil height and gap orientation of the two types of bolster springs. The ratio Htotal/H2 of the two kinds of bolster springs shows a good piecewise linear relationship with the gap orientation, with the increased division degree (Figs. 6 and 7). The positioning experiments under different gap orientations show that the spot imaging quality under the line laser vision system is better. In this research, the adaptive threshold segmentation method is separated by the laser spot on the surface of the outer and inner springs (Figs. 14 and 15). The calculated orientation error of the bolster spring gap is kept within -5°~+ 5°, and the single positioning time is less than 0.15 s, thereby meeting the requirements of system positioning efficiency (Tables 3 and 4). The positioning experiments under different illumination intensities show that the positioning method proposed in this research has strong resistance to the change in illumination intensity (Table 5).
This research proposesd a visual indirect positioning method to bolster spring gaps based on the spot features of a linear laser. The variation law between the height of the spring coil and orientation of the end gap was derived by analyzing the structural characteristics of the outer spring and height data collection of the first and second layers of the spring coil. The collected height data were fitted, and the mathematical models of the corresponding relationships between the height of the spring coil and gap orientation of the bearing and damping springs were derived with the least square method. The YOLOv3-tiny object detection algorithm based on deep learning was used to detect the bolster spring in a complex background, and then, the region of interest segmentation, including the laser spot image, was produced. An adaptive solution method of spring coil height was proposed based on threshold segmentation and bounding box fitting algorithms, which significantly improves the flexibility and adaptability of the algorithm. Multiple sets of bolster spring gap positioning experiments have been performed on the bolster spring test platform. The results show that the proposed method can accurately and efficiently position the two types of bolster springs in a narrow space. The positioning accuracy is within -5°~+ 5° while the single positioning time is no more than 0.15 s. Meanwhile, the proposed method has strong resistance to changes in illumination intensity, thereby meeting the gap positioning requirements of the bolster spring intelligent disassembly equipment.
