【作者】 李超；

【导师】 史铁林；

【作者基本信息】 华中科技大学 ， 机械工程， 2021， 硕士

【摘要】 刀具磨损状态监测在切削加工中十分重要,它直接影响着工件的尺寸精度与表面质量。通常采用光学显微镜对刀具磨损进行直接测量,这种办法需要拆卸刀具并且比较耗时。基于机器视觉的刀具磨损检测方法,可实现在机检测并提高检测效率。本文基于工件表面纹理特征对刀具磨损检测方法进行研究,通过实验采集工件表面图像数据并提取图像中的纹理特征,使用机器学习中的决策树模型完成对纹理特征数据集的分类,实现刀具磨损状态检测。本文的主要工作有:(1)建立图像采集实验并对样本图像进行预处理。首先,根据实验原理搭建了实验基本框架,确定了实验平台及其主要加工参数。然后,对于采集系统的主要部件工业相机、镜头、光源等进行对比选型,对铣削刀具磨损检测方法进行研究。开展采集试验并为采集的图像制作标签,最后对图像进行样本扩容、图像增强等预处理操作。(2)研究刀具磨损机理与纹理特征提取方法,提取Tamura纹理特征并对特征属性进行分析。首先,根据磨损机理确定各磨损阶段对应的具体磨损值范围,将对应磨损阶段标签数值化。然后,基于图像纹理特征提取方法的研究对比选择了Tamura纹理特征,编写MATLAB程序提取出6维特征属性。最后基于各个特征属性的分布图分析,剔除可分类性较差的规则度这一属性,制作特征属性数据集。(3)建立基于纹理特征的决策树分类模型,并对该模型进行参数寻优,以提高模型的分类准确率。首先,基于决策树模型原理与结构,对常见的决策树算法进行算法流程介绍与对比。然后,通过将特征数据集划分为训练集与测试集,利用训练集对模型进行初步训练得到100%的准确率,而此时测试集的准确率只有71.61%,决策树出现过拟合。之后利用网格搜索与交叉验证对决策树参数优化等操作对决策树进行优化,减小过拟合问题,分析其拟合曲线与混淆矩阵,参数优化后数据拟合效果变好,此时训练集拟合率达到100%,测试集准确率为89.20%。更多还原

【Abstract】 Tool wear State Monitoring is very important in cutting,and it directly affects the dimensional accuracy and surface quality of the workpiece.An optical microscope is usually used to directly measure tool wear.This method requires the removal of the tool and is timeconsuming.The tool wear detection method based on machine vision can realize on-machine detection and improve detection efficiency.In this paper,the tool wear detection method is researched based on the surface texture characteristics of the workpiece.Through experiments,the surface image data of the workpiece is collected and the texture features in the image are extracted.The decision tree model in machine learning is used to classify the texture feature data set to realize the tool wear status.Detection.The studies in this thesis can be summarized as follows:(1)Establish an image acquisition experiment and preprocess the sample images.First,the basic framework of the experiment was built according to the experimental principle,and the experimental platform and its main processing parameters were determined.Then,the main components of the acquisition system,industrial cameras,lenses,light sources,etc.,were compared and selected,and the milling tool wear detection methods were studied.Carry out collection experiments and make labels for the collected images,and finally perform preprocessing operations such as sample expansion and image enhancement on the images.(2)Research the tool wear mechanism and texture feature extraction method,extract the Tamura texture feature and analyze the feature attributes.First,determine the specific wear value range corresponding to each wear stage according to the wear mechanism,and digitize the corresponding wear stage tags.Then,based on the research and comparison of image texture feature extraction methods,the Tamura texture feature was selected,and the MATLAB program was written to extract the 6-dimensional feature attributes.Finally,based on the analysis of the distribution map of each characteristic attribute,the characteristic attribute data set is made by eliminating the attribute of rule degree with poor classification.(3)Establish a decision tree classification model based on texture features,and optimize the parameters of the model to improve the classification accuracy of the model.First,based on the principle and structure of the decision tree model,the algorithm flow of common decision tree algorithms is introduced and compared.Then,by dividing the feature data set into a training set and a test set,using the training set to initially train the model to obtain an accuracy of 100%.At this time,the accuracy of the test set is only 71.61%,and the decision tree is over-fitting.Then use grid search and cross-validation to optimize the decision tree parameters and other pruning operations to prun the decision tree to reduce the over-fitting problem,analyze its fitting curve and confusion matrix,and the data fitting effect becomes better after the parameter optimization.At this time,the fitting rate of the training set reaches 100%,and the accuracy of the test set is 89.20%.更多还原