融合图像与频域特征激光切割挂渣量化预测

翟杰; 芦宇; 王鑫鑫; 夏元钦

doi:10.37188/CO.2025-0125

融合图像与频域特征激光切割挂渣量化预测

doi: 10.37188/CO.2025-0125

cstr: 32171.14.CO.2025-0125

翟杰¹,
芦宇^1, ,,
王鑫鑫¹,
夏元钦²

1.
天津职业技术师范大学电子工程学院, 天津 300222
2.
河北工业大学河北省先进激光技术与装备重点实验室, 天津 300401

基金项目: 天津市科技计划项目(No. 24YDTPJC00510)，河北省先进激光技术与装备重点实验室基金（No. HBKL-ALTE2025003）

详细信息

作者简介:
芦　宇（1982—），男，黑龙江齐齐哈尔人，博士，讲师，硕士生导师，2007年哈尔滨工业大学获得硕士学位， 2013年哈尔滨工业大学获得博士学位。主要从事激光加工技术、激光检测技术、非成像光学系统设计等方面的研究。E-mail：luyu@tute.edu.cn

中图分类号: TG485
计量
- 文章访问数: 32
- HTML全文浏览量: 14
- PDF下载量: 3
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-29
- 录用日期: 2025-11-25
- 网络出版日期: 2025-12-30

Quantitative prediction of laser-cut slag adhesion by integrating image and frequency domain features

1.
School of Electronic Engineering, Tianjin University of Technology and Education, Tianjin 300222, China
2.
Hebei Key Laboratory of Advanced Laser Technology and Equipment, Hebei University of Technology, Tianjin 300401, China

Funds: Supported by Tianjin Science and Technology Program Project (No.24YDTPJC00510); Hebei Key Laboratory of Advanced Laser Technology and Equipment(No. HBKL-ALTE2025003)

More Information

Corresponding author: luyu@tute.edu.cn

摘要

摘要:
为实现激光切割熔渣附着精准量化与工艺优化，本研究探索一种基于图像与频域特征的卷积神经网络（CNN）预测方法。构建包含2160张1 mm厚304不锈钢切割端面图像数据集。基于该数据集，采用高斯模糊、自适应阈值及形态学闭运算等图像处理算法，精确提取了挂渣的面积、高度及周长作为量化特征。为评估不同特征预测潜力，采用了RGB图像及其经二值化处理小波包分解（WPD）频域图像作为输入，并系统对比了VGG16、ResNet50和DenseNet121三种CNN架构回归性能。结果表明，在RGB图像输入路径下，VGG16网络对挂渣面积和高度预测最为精准，其平均绝对误差（MAE）分别达到0.019 mm²和0.044 mm。而对于更能反映动态过程状态轮廓周长特征，WPD频域输入路径预测效果显著提升，MAE降至0.094 mm，归一化平均误差（nMAE）为5.25%，且其预测值与真实值间拟合斜率与决定系数R²分别为0.83与0.86，呈现强线性关系。本研究证实，VGG16网络在熔渣特征预测中具备良好适用性，且WPD频域特征更有效地捕捉激光切割过程瞬态信息，所提出方法为工艺智能评估与闭环优化提供了可靠量化工具。
- 挂渣特征 /
- 卷积神经网络 /
- 小波包分解 /
- 激光切割工艺
Abstract:
To achieve precise quantification of slag adhesion and process optimization in laser cutting, this study investigates a convolutional neural network (CNN)-based prediction method that integrates both image and frequency-domain features. A dataset of 2160 cross-sectional images of 1 mm thick 304 stainless steel was constructed. From these images, key dross characteristics-area, height, and perimeter were accurately extracted using a combination of image processing techniques including Gaussian blur, adaptive thresholding, and morphological closing operations. To evaluate the predictive potential of different input representations, both RGB images and binarized images transformed via wavelet packet decomposition (WPD) were used as model inputs. The regression performance of three CNN architectures-VGG16, ResNet50, and DenseNet121 was systematically compared. Experimental results demonstrate that VGG16 achieved the highest prediction accuracy for dross area and height using RGB images, with mean absolute errors (MAE) of 0.019 mm² and 0.044 mm, respectively. For predicting the perimeter, which better reflects dynamic process behavior, the WPD frequency-domain input path yielded a significantly improved MAE of 0.094 mm and a normalized MAE (nMAE) of 5.25%. The regression fit between predicted and actual values showed a slope of 0.83 and a coefficient of determination (R²) of 0.86, indicating a strong linear correlation. This study confirms the effectiveness of VGG16 in predicting dross-related features and demonstrates the capability of WPD-derived frequency-domain features in capturing transient process information during laser cutting. The proposed methodology offers a reliable quantitative tool for intelligent process evaluation and closed-loop optimization.
- dross features /
- convolutional neural networks /
- wavelet packet decomposition /
- laser cutting process

HTML全文

图 1 激光切割挂渣轮廓算法流程图

Figure 1. Flowchart of the algorithm for extracting dross contours in laser cutting

下载: 全尺寸图片幻灯片

图 2 挂渣轮廓提取过程：(a) 原始图像, (b) 高斯模糊, (c) 自适应阈值分割, (d) 形态学闭运算, (e) 二值化轮廓图

Figure 2. Steps in slag adhesion contour extraction: (a) original image, (b) Gaussian blur, (c) adaptive threshold segmentation, (d) morphological closing operation, (e) binary contour map

下载: 全尺寸图片幻灯片

图 3 曝光挂渣轮廓提取过程：(a) 原始图像, (b) 高斯模糊, (c) 自适应阈值分割, (d) 形态学闭运算, (e) 二值化轮廓图

Figure 3. Contour extraction under varying exposure conditions: (a) original image, (b) Gaussian blur, (c) adaptive threshold segmentation, (d) morphological closing operation, (e) binary contour map

下载: 全尺寸图片幻灯片

图 4 调整后切口挂渣图

Figure 4. Processed image of a cut edge with slag adhesion

下载: 全尺寸图片幻灯片

图 5 (a) 原始RGB轮廓 (b) 二值轮廓 (c)能量图

Figure 5. (a)Original RGB profile (b) binary contour, (c) WPD energy map.

下载: 全尺寸图片幻灯片

图 6 VGG16模型进行激光切割挂渣识别与特征分析

Figure 6. Architecture of the VGG16 model used for slag identification and feature analysis

下载: 全尺寸图片幻灯片

图 7 三种网络架构RGB模型预测-真实散点对比。（a）、（b）、（c）分别为面积、高度、周长ResNet50预测结果；（d）、（e）、（f）分别为面积、高度、周长VGG16预测结果；（g）、（h）、（i）分别为面积、高度、周长DenseNet121预测结果

Figure 7. Comparison of Predicted vs. Actual Scatter Plots for Three CNN Architectures with RGB Image Input. (a), (b), (c) Scatter plots of predicted vs. actual values for area, height, and perimeter using ResNet50；(d), (e), (f) Scatter plots of predicted vs. actual values for area, height, and perimeter using VGG16；(g), (h), (i) Scatter plots of predicted vs. actual values for area, height, and perimeter using DenseNet12

下载: 全尺寸图片幻灯片

图 8 高挂渣轮廓表征 (a) RGB图，(b)WPD图

Figure 8. Characterization of the high-hanging slag contour (a) RGB image, (b)WPD diagram.

下载: 全尺寸图片幻灯片

图 9 低挂渣轮廓表征 (a)RGB图，(b) WPD图

Figure 9. Characterization of the low-hanging slag contour (a) RGB image, (b)WPD diagram.

下载: 全尺寸图片幻灯片

图 10 VGG16-WPD 周长预测散点图

Figure 10. Scatter plot of perimeter predictions using VGG16 trained on WPD features

下载: 全尺寸图片幻灯片

图 11 不同CNN模型的特征可视化对比（a）样本（b）VGG16可视化图（c）Resenet50可视化图（d）RGB DenseNet121可视化图

Figure 11. A comparison of feature activation maps from different CNN models: (a) Input, (b) feature maps from VGG16, (c) feature maps from ResNet50, and (d) feature maps from DenseNet121

下载: 全尺寸图片幻灯片

图 12 不同CNN模型在WPD表征上的特征可视化对比（a）样本（b）VGG16可视化图（c）Resent50可视化图（d）Desent121可视化图

Figure 12. Comparative analysis of CNN feature visualizations on a WPD representation: (a) Input (b) feature maps from VGG16, (c) feature maps from ResNet50, and (d) feature maps from DenseNet121

下载: 全尺寸图片幻灯片

图 13 不同离焦量与速度下切割端面挂渣轮廓，（a）−0.8 mm、10 m/s，（b）−1 mm、10 m/s，（c）−0.6 mm、10 m/s，（d）−1 mm、12 m/s，（e）−0.8、12

Figure 13. Dross profiles under different defocus distances and cutting speeds: (a) –0.8 mm, 10 m/s; (b) –1 mm, 10 m/s; (c) –0.6 mm, 10 m/s; (d) –1 mm, 12 m/s; (e) –0.8 mm, 12 m/s

下载: 全尺寸图片幻灯片

表 1 三种CNN模型基于RGB图像挂渣特征预测MAE与nMAE对比

Table 1. Comparison of MAE and nMAE for dross feature prediction using three CNN models on RGB images

网络架构	挂渣面积MAE (mm²)	挂渣高度MAE (mm)	挂渣周长MAE (mm)	挂渣面积nMAE (%)	挂渣高度nMAE (%)	挂渣周长nMAE (%)
ResNet50	0.023	0.046	0.106	8.48	9.72	5.91
VGG16	0.019	0.044	0.117	7	9.38	6.49
DenseNet121	0.029	0.054	0.106	10.6	11.4	5.91

下载: 导出CSV

表 2 频域训练下挂渣周长预测MAE与nMAE

Table 2. MAE and nMAE for slag perimeter in frequency-domain training

模型	频域MAE(mm)	频域nMAE（%）
ResNet50	0.145	8.11
VGG16	0.094	5.25
DenseNet121	0.131	7.31

下载: 导出CSV

表 3 三种模型RGB与WPD支路熵对比

Table 3. Comparison of entropy between RGB and WPD branches in three models

模型	RGB 支路熵	WPD 支路熵
VGG16	2.62	2.41
ResNet50	3.12	3.68
DenseNet121	2.80	2.96

下载: 导出CSV

表 4 随机森林模型三种挂渣特征预测性能

Table 4. Prediction performance of three slag sticking features in the random forest model

挂渣特征	RMSE	MAE	R²
面积	0.029	0.022	0.901
高度	0.055	0.040	0.810
周长	0.215	0.157	0.629

下载: 导出CSV

表 5 激光切割工艺参数优化结果

Table 5. Optimization results for laser cutting process parameters

离焦量 (mm)	速度(m/s)	预测挂渣面积平均值( mm²)	超标率（%）
−0.8	10	0.18	0
−1	10	0.18	0
−0.6	10	0.19	20
−0.8	12	0.19	20
−1	12	0.20	30

下载: 导出CSV

参考文献(27)

[1]	GABZDY J T. Effects of gases on laser cutting of stainless steels[C]. Proceedings of the Laser Materials Processing Conference, Laser Institute of America, 1996: 39-44.
[2]	SHARIFF S M, SUNDARARAJAN G, JOSHI S V. Parametric influence on cut quality attributes and generation of processing maps for laser cutting[J]. Journal of Laser Applications, 1999, 11(2): 54-63. doi: 10.2351/1.521886
[3]	陈宇翔, 高亮. 正交法分析激光切割高温合金钢工艺参数[J]. 激光与光电子学进展, 2016, 53(11): 111403. doi: 10.3788/LOP53.111403 CHEN Y X, GAO L. Analysis of processing parameters for laser cutting high-temperature alloy steel by orthogonal method[J]. Laser & Optoelectronics Progress, 2016, 53(11): 111403. (in Chinese). doi: 10.3788/LOP53.111403
[4]	PARTHIBAN A, DHANASEKARAN C, SIVAGANESAN S, et al. Modeling on surface cut quality of CO₂ laser cutting for Austenitic Stainless steel sheet[J]. Materials Today: Proceedings, 2020, 21: 823-827. doi: 10.1016/j.matpr.2019.07.428
[5]	王勇, 陈娇娇, 张鹏, 等. 基于响应面-遗传算法的不锈钢激光切割工艺参数优化与试验研究[J]. 应用激光, 2023, 43(11): 85-94. doi: 10.14128/j.cnki.al.20234311.085 WANG Y, CHEN J J, ZHANG P, et al. Process parameter optimization and experimental research on stainless steel laser cutting based on response surface-genetic method[J]. Applied Laser, 2023, 43(11): 85-94. (in Chinese). doi: 10.14128/j.cnki.al.20234311.085
[6]	SHIN J S, OH S Y, PARK H, et al. High-speed fiber laser cutting of thick stainless steel for dismantling tasks[J]. Optics & Laser Technology, 2017, 94: 244-247. doi: 10.1016/j.optlastec.2017.03.040
[7]	OH S Y, SHIN J S, KIM T S, et al. Effect of nozzle types on the laser cutting performance for 60-mm-thick stainless steel[J]. Optics & Laser Technology, 2019, 119: 105607. doi: 10.1016/j.optlastec.2019.105607
[8]	SHIN J S, SONG K H, OH S Y, et al. Laser cutting studies on 10-60 mm thick stainless steels with a short focus head for nuclear decommissioning[J]. Optics & Laser Technology, 2024, 169: 110121. doi: 10.1016/j.optlastec.2023.110121
[9]	SHIN J S. Laser cutting study on 30 mm thick stainless steel for application in decommissioning of calandria shells in heavy-water reactors[J]. Nuclear Engineering and Technology, 2025, 57(3): 103242. doi: 10.1016/j.net.2024.10.004
[10]	陈达, 章轩, 赵圣斌, 等. 基于深度学习的激光熔覆层表面缺陷识别研究[J]. 光学学报, 2025, 45(9): 0915001. doi: 10.3788/AOS250462 CHEN D, ZHANG X, ZHAO SH B, et al. Surface defect recognition of laser cladding layer based on deep learning[J]. Acta Optica Sinica, 2025, 45(9): 0915001. (in Chinese). doi: 10.3788/AOS250462
[11]	王颖, 高胜, 戴哲. 基于CNN-Transformer混合网络的焊缝激光条纹分割[J]. 中国激光, 2024, 51(24): 2402110. doi: 10.3788/CJL240710 WANG Y, GAO SH, DAI ZH. Laser stripe segmentation of weld seam based on CNN-Transformer hybrid networks[J]. Chinese Journal of Lasers, 2024, 51(24): 2402110. (in Chinese). doi: 10.3788/CJL240710
[12]	张臣, 胡佩佩, 朱新旺, 等. 基于高密度点云的激光焊接缺陷智能在线检测(特邀)[J]. 中国激光, 2024, 51(4): 0402105. doi: 10.3788/CJL231293 ZHANG CH, HU P P, ZHU X W, et al. Intelligent online detection of laser welding defects based on high density point clouds (invited)[J]. Chinese Journal of Lasers, 2024, 51(4): 0402105. (in Chinese). doi: 10.3788/CJL231293
[13]	马炜鑫, 谢冠明, 林杰凯, 等. 基于机器视觉的激光焊接气孔缺陷智能检测方法[J]. 激光与光电子学进展, 2026, 63(1): 0114012. (查阅网上资料, 未找到对应的页码信息, 请确认). MA W X, XIE G M, LIN J K, et al. Laser welding porosity defect detection method based on machine vision[J]. Laser & Optoelectronics Progress, 2026, 63(1): 0114012. (in Chinese).
[14]	REN X F, FAN J W, PAN R, et al. Modeling and process parameter optimization of laser cutting based on artificial neural network and intelligent optimization algorithm[J]. The International Journal of Advanced Manufacturing Technology, 2023, 127(3): 1177-1188. doi: 10.1007/s00170-023-11543-6
[15]	CHAKI S, BOSE D, BATHE R N. Multi-objective optimization of pulsed Nd: YAG laser cutting process using entropy-based ANN-PSO model[J]. Lasers in Manufacturing and Materials Processing, 2020, 7(1): 88-110. doi: 10.1007/s40516-019-00109-8
[16]	何鹏, 孙帆, 胡小方, 等. 小样本激光切割工艺参数预测模型优化[J]. 激光杂志, 2021, 42(12): 170-175. doi: 10.14016/j.cnki.jgzz.2022.12.170 HE P, SUN F, HU X F, et al. Optimization of prediction system for small sample laser cutting process parameters[J]. Laser Journal, 2021, 42(12): 170-175. (in Chinese). doi: 10.14016/j.cnki.jgzz.2022.12.170
[17]	钟毅, 栾显晔, 李晓宇, 等. 碳钢板激光切割工艺分析及优化[J]. 制造技术与机床, 2024(2): 38-45. doi: 10.19287/j.mtmt.1005-2402.2024.02.005 ZHONG Y, LUAN X Y, LI X Y, et al. Analysis and optimization of laser cutting process of carbon steel plate[J]. Manufacturing Technology & Machine Tool, 2024(2): 38-45. (in Chinese). doi: 10.19287/j.mtmt.1005-2402.2024.02.005
[18]	FRANCESCHETTI L, PACHER M, TANELLI M, et al. Dross attachment estimation in the laser-cutting process via Convolutional Neural Networks (CNN)[C]. Proceedings of the 28th Mediterranean Conference on Control and Automation, IEEE, 2020: 850-855.
[19]	ADELMANN B, SCHLEIER M, HELLMANN R. Laser cut interruption detection from small images by using convolutional neural network[J]. Sensors, 2021, 21(2): 655. doi: 10.3390/s21020655
[20]	TATZEL L, LEÓN F P. Image-based roughness estimation of laser cut edges with a convolutional neural network[J]. Procedia CIRP, 2020, 94: 469-473. doi: 10.1016/j.procir.2020.09.166
[21]	TATZEL L, TAMIMI O A, HAUEISE T, et al. Image-based modelling and visualisation of the relationship between laser-cut edge and process parameters[J]. Optics & Laser Technology, 2021, 141: 107028. doi: 10.1016/j.optlastec.2021.107028
[22]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2016: 770-778.
[23]	HE K M, ZHANG X Y, REN S Q, et al. Identity mappings in deep residual networks[C]. Proceedings of the 14th European Conference on Computer Vision, Springer, 2016: 630-645.
[24]	AL-HAIJA Q A, ADEBANJO A. Breast cancer diagnosis in histopathological images using ResNet-50 convolutional neural network[C]. Proceedings of the IEEE International IOT, Electronics and Mechatronics Conference, IEEE, 2020: 1-7.
[25]	MUKTI I Z, BISWAS D. Transfer learning based plant diseases detection using ResNet50[C]. Proceedings of the 4th International Conference on Electrical Information and Communication Technology, IEEE, 2019: 1-6.
[26]	LIU Y, ZHANG Z L, LIU X, et al. Deep learning-based image classification for online multi-coal and multi-class sorting[J]. Computers & Geosciences, 2021, 157: 104922. doi: 10.1016/j.cageo.2021.104922
[27]	MASCARENHAS S, AGARWAL M. A comparison between VGG16, VGG19 and ResNet50 architecture frameworks for Image Classification[C]. Proceedings of 2021 International Conference on Disruptive Technologies for Multi-Disciplinary Research and Applications, IEEE, 2021, 1: 96-99.