基于DGLCM BO Stacking的工件表面粗糙度检测  

Workpiece Surface Roughness Detection Based on DGLCM-BO-Stacking

作　　者：张成辉 方灶军 Zhang Chenghui;Fang Zaojun

机构地区：[1]宁波大学机械工程与力学学院,浙江宁波315211 [2]中国科学院宁波材料技术与工程研究所,浙江宁波315201

出　　处：《机械制造》2024年第6期71-77,16,共8页Machinery

基　　金：国家重点研发计划项目(编号:2022YFB4702500);浙江省重点研发计划项目(编号:2023C01176,LD22E050007)。

摘　　要：表面粗糙度对工件使用性能和寿命有很大影响,为了实现高精度、鲁棒工件表面粗糙度检测,提出基于方向灰度共生矩阵和贝叶斯优化堆叠模型的工件表面粗糙度检测方法。采用互补金属氧化物半导体相机、同轴光源、远心镜头搭建视觉检测系统,采集不同表面粗糙度的工件纹理图像,并对采集到的纹理图像进行数据增强。使用方向灰度共生矩阵提取图像的纹理特征,构建特征描述符。使用经过贝叶斯优化的轻量梯度提升机、极限梯度提升树、随机森林作为基学习器,构建堆叠模型。使用所采集的不同表面粗糙度的纹理图像数据集进行性能试验。试验结果表明,所提出的检测方法在多个性能评价指标上都优于其它方法,具有较高的检测精度和较强的泛化能力,可以有效实现表面粗糙度的高精度、鲁棒检测。The Surface roughness has a great influence on usage performance and life of workpiece.In order to achieve high precision and robust surface roughness detection of workpiece,a workpiece surface roughness detection method based on DGLCM and BO Stacking was proposed.A visual inspection system was built by using a CMOS camera,a coaxial light source and a telecentric lens to collect texture image of workpiece with different surface roughness,and to enhance the data of the collected texture image.The DGLCM was used to extract the texture feature of the image and construct feature descriptor.Bayesian optimized LightGBM,Xgboost,and RF were used as base learners to construct the Stacking.Performance test was performed by using the acquired texture image dataset with different surface roughness.The test result shows that the proposed detection method is superior to other methods in multiple performance evaluation indexes,and has high detection accuracy and generalization ability,which can effectively achieve high precision and robust detection of surface roughness.

关 键 词：表面粗糙度 检测 方向灰度共生矩阵 贝叶斯优化 堆叠模型 

分 类 号：TH161.14[机械工程—机械制造及自动化]