基于ROI-RSICP算法的车轮廓形动态检测 
下载： 948次
Dynamic Inspection of Wheel Profile Based on ROI-RSICP Algorithm
1 中南大学交通运输工程学院, 湖南 长沙 410075
2 汉阳大学机械工程学院, 首尔04763, 韩国
图 & 表
图 1. 不同情况下的测量车轮廓形示意图。(a)激光平面与车轮表面的交线不过轮心;(b)畸变车轮廓形;(c)激光平面与车轮表面的交线过轮心;(d)无畸变车轮廓形
Fig. 1. Measured wheel profiles under different situations. (a) Intersection line of laser plane and wheel surface not crossing wheel axis; (b) wheel profile with distortion; (c) intersection line of laser plane and wheel surface crossing wheel axis; (d) wheel profile without distortion
下载图片 查看原文
图 2. 车轮廓形磨耗演变示意图
Fig. 2. Wear evolution of wheel profile
下载图片 查看原文
图 3. PointNet点云分割网络图
Fig. 3. Point cloud segmentation network of PoinNet
下载图片 查看原文
图 4. 车轮廓形动态检测流程图
Fig. 4. Flow chart of dynamic inspection of wheel profile
下载图片 查看原文
图 5. 总损失随训练次数的变化曲线
Fig. 5. Total loss versus training times
下载图片 查看原文
图 6. mIoU随训练次数的变化曲线
Fig. 6. mIoU versus training times
下载图片 查看原文
图 7. 基于PointNet提取的ROI部分结果图。(a)无磨耗;(b)轻度磨耗;(c)重度磨耗
Fig. 7. Partial results of extracted ROI based on PointNet. (a) Standard; (b) slightly worn; (c) badly worn
下载图片 查看原文
图 8. ICP、SICP与本文算法的配准精度对比。(a) 原始模型;(b) ICP;(c) SICP;(d) ROI-RSICP
Fig. 8. Accuraccy comparison among ICP, SICP and proposed methods. (a) Original model; (b) ICP; (c) SICP; (d) ROI-RSICP
下载图片 查看原文
图 9. 不同IoU值下的车轮廓形配准结果图。(a)轻微磨耗;(b)严重磨耗
Fig. 9. Registration results of wheel profile under different IoU values. (a) Slightly worn; (b) badly worn
下载图片 查看原文
图 10. 本文方法检测结果与第四种检查器检测结果的差值随IoU的变化
Fig. 10. Difference between inspection result by proposed method and that by the 4th kind of inspector versus IoU
下载图片 查看原文
图 11. 不同夹角下 ICP、SICP与本文算法的稳定性比较。(a) 30°;(b) 35°;(c) 40°;(d) 45°;(e) 50°;(f) 55°;(g) 60°
Fig. 11. Stability comparison of ICP, SICP and proposed methods under different included angles. (a) 30°; (b) 35°; (c) 40°; (d) 45°; (e) 50°; (f) 55°; (g) 60°
下载图片 查看原文
图 12. 各算法计算的踏面磨耗值与第四种检查器检测结果的差值
Fig. 12. Difference between tread wear calculated by each method and result measured by the 4th kind of inspector
下载图片 查看原文
表 1ICP、RSICP、ROI-RSICP算法与人工方法检测的踏面磨耗值
Table1. Tread wear detected by ICP, SICP, ROI-RSICP and artificial methods unit: mm
| Method | Data type | Position 1 | Position 2 | Position 3 | Position 4 | Position 5 | Position 6 |
|---|
| ICP | Slightly worn | 1.60 | 1.74 | 1.88 | 1.23 | 2.23 | 1.49 | | SICP | 1.13 | 1.27 | 1.14 | 1.25 | 1.56 | 1.36 | | ROI-RSICP | 1.76 | 1.94 | 1.99 | 1.75 | 2.09 | 1.84 | | Artificial method | 1.80 | 1.90 | 1.90 | 1.80 | 2.00 | 1.70 | | ICP | badly worn | 2.70 | 3.11 | 2.55 | 2.96 | 1.04 | 2.43 | | RSICP | 0.99 | 1.01 | 1.81 | 2.21 | 2.14 | 1.78 | | ROI-RSICP | 4.19 | 4.11 | 4.15 | 4.21 | 4.12 | 4.39 | | Artificial method | 4.10 | 4.00 | 4.10 | 4.30 | 4.10 | 4.50 |
|
查看原文
表 2严重磨耗车轮的重复检测结果
Table2. Repeated detection results of badly worn wheel unit: mm
| Number ofmeasurements | Flange height | Flange thickness | Tread wear |
|---|
| ROI-RSICP | Artificial method | ROI-RSICP | Artificial method | ROI-RSICP | Artificial method |
|---|
| 1 | 31.14 | 31.10 | 30.11 | 30.20 | 4.14 | 4.10 | | 2 | 31.18 | 31.10 | 30.09 | 30.10 | 4.18 | 4.10 | | 3 | 31.12 | 31.00 | 30.12 | 30.20 | 4.12 | 4.00 | | 4 | 31.13 | 31.00 | 30.07 | 30.10 | 4.13 | 4.00 | | 5 | 31.12 | 31.10 | 30.12 | 30.20 | 4.12 | 4.10 | | 6 | 31.20 | 31.10 | 30.10 | 30.10 | 4.20 | 4.10 | | Mean | 31.15 | 31.07 | 30.10 | 30.15 | 4.15 | 4.07 | | SD | 0.033 | 0.051 | 0.0194 | 0.054 | 0.033 | 0.051 |
|
查看原文
表 3车轮几何参数的动态检测结果
Table3. Dynamic inspection results of wheel geometric parameters unit: mm
| Parameter | Method | Measurement result | Maximumerror |
|---|
| 1 | 2 | 3 | 4 | 5 | 6 |
|---|
| Flange height | ROI-RSICP | 32.11 | 32.08 | 32.11 | 32.07 | 32.08 | 32.04 | | | Artificial method | 32.2 | 32.1 | 32.1 | 32.1 | 32.2 | 32.1 | 0.12 | | Flange thickness | ROI-RSICP | 31.08 | 31.12 | 31.13 | 31.12 | 31.14 | 31.14 | | | Artificial method | 31.2 | 31.2 | 31.2 | 31.1 | 31.2 | 31.2 | 0.12 | | Tread wear | ROI-RSICP | 5.11 | 5.08 | 5.11 | 5.07 | 5.08 | 5.04 | | | Artificial method | 5.1 | 5.1 | 5.1 | 5.0 | 5.1 | 5.0 | 0.07 |
|
查看原文
易倩, 钟浩宇, 刘龙, 刘文龙, 易兵. 基于ROI-RSICP算法的车轮廓形动态检测[J]. 中国激光, 2020, 47(11): 1104006. Yi Qian, Zhong Haoyu, Liu Long, Liu Wenlong, Yi Bing. Dynamic Inspection of Wheel Profile Based on ROI-RSICP Algorithm[J]. Chinese Journal of Lasers, 2020, 47(11): 1104006.
PDF全文
