基于激光散斑成像的零件表面粗糙度建模

表面粗糙度是衡量机械表面加工水平的重要参数. 通过构建一套激光散斑成像采集系统, 获取了不同表面加工类型和不同粗糙度值的零件表面激光散斑图像. 应用Tamura纹理特征理论提取图像的纹理粗糙度、对比度、方向度特征, 并分析了这三个特征与表面粗糙度的关系. 发现了纹理粗糙度特征与表面粗糙度的单调关系, 推导出平磨、外磨、研磨三种表面加工工艺的粗糙度值与图像纹理粗糙度特征的数学函数关系, 实现了表面粗糙度的测量. 同时, 利用Tamura纹理特征与加工工艺的依赖关系, 建立了基于贝叶斯网络的工艺识别推理模型, 推理出了零件表面加工工艺. 通过为多种加工类型表面建立粗糙度测量模型, 为粗糙度测量提供了新思路. 实验证明所提的粗糙度测量模型能以较高的准确率识别出零件表面加工类型并测量出其表面粗糙度值.

Surface roughness modeling based on laser speckle imaging

Surface roughness is an important parameter in measuring the roughness of surface formed by laser irradiation on the workpiece. Speckle images of rough surfaces in different classes and different surface roughness values are obtained by constructing a set of laser speckle image acquisition systems. First, the texture features of speckle images including coarseness, contrast and direction are extracted using Tamura texture theory. Then, the interactions these three features with the surface roughness are analyzed. Based on the analyses of their monotonic relations, the surface roughness functions, including flat grinding, external grinding and mill grinding craftworks, are established respectively between the texture coarseness feature of the speckle image F_crs and surface roughness R_a. Through the establishment of surface roughness function for the above three classes of workpieces, the value of surface roughness can be computed directly. However, before obtaining the value of surface roughness, the classes of processing technic should be determined because of the inconsistency of function expressions for different classes. And based on the specific connection and related dependencies between Tamura texture features and workpiece class, Bayes network is proposed to describe this uncertainty relation among different classes. Through network structure learning and parameter learning, a model for reasoning is found which can be used to determine the class of workpiece after obtaining texture coarseness feature F_crs. Thus, not only can the value of surface roughness be measured, also the class of work-piece can be recognized. Experiments are conducted to confirm the feasibility of the proposed model for measurement. The detection results indicate that high precision and accuracy are achieved for both workpiece class recognition and roughness measurement.