基于灰度共生矩阵和神经网络的Si_3N_4陶瓷推挤加工表面纹理分析  

Analysis of Surface Texture of Push-processed Si_3N_4 Ceramics Based on Gray Level Co-occurrence Matrix and Neural Network

作　　者：田欣利[1] 王龙[1] 王望龙[1] 唐修检[1] 吴志远[1] 

机构地区：[1]装甲兵工程学院装备再制造技术国防科技重点实验室,北京100072

出　　处：《兵工学报》2015年第8期1518-1524,共7页Acta Armamentarii

基　　金：国家自然科学基金项目(51475474;51105378)

摘　　要：基于边缘破碎效应驱动裂纹软推挤加工是一项新颖的加工技术。通过采集Si3N4陶瓷的软推挤加工表面形貌,运用灰度共生矩阵(GLCM)分析了对比度、熵、相关性3个特征参数与加工表面纹理分布的内在关系。通过径向基网络和竞争层网络两类神经网络的分工协作,对不同加工参数下已加工表面的纹理特征进行预测和分类,其预测结果的相对误差能控制在5%之内。随着对比度和熵越大,相关性越小;分类等级越大,表面平整程度越差。通过系统实验探讨了各加工参数对纹理特征的影响,可靠地评估了加工质量的优劣。随着车刀进给速度或槽深的增大,加工表面质量变差;随着凸缘厚度的增大,加工表面质量先逐渐变差,但经过凸缘厚度2.5 mm分界点后却又有所改善。Soft push processing based on the edge broken effect to drive crack is a novel processing tech- nique. The surface texture of machined Si3N4 ceramic is collected, and the gray level co-occurrence ma- trix （GLCM） is used to analyze the relationship among contrast, entropy, correlation and machined sur- face texture. The radial basic network and competitive network are employed to predict and classify the texture characteristics in different processing parameters. The relative error value of the predicted results can be controlled within 5%. The larger the contrast and entropy are, the smaller the correlation is, and the greater the classification level is, the worse the surface roughness is. The processing quality is ana- lyzed by exploring the effects of different process parameters on texture feature. With the increase in feed rate or groove depth, the machined surface quality is worse. With the increase in flange thickness, the machined surface quality is gradually poor. The machined surface quality can be improved when flange thickness is over 2.5 mm boundary point.

关 键 词：材料表面与界面 SI3N4陶瓷 纹理特征 灰度共生矩阵 神经网络 

分 类 号：TH161.1[机械工程—机械制造及自动化] TQ174[化学工程—陶瓷工业]