增材制件亚表熔融沉积层孔隙缺陷逐层检测技术研究

Presently, there are few researches on quality detection of additive manufacturing (AM), most of which are based on the traditional detection methods. In this research project, according to the characteristic of the layer by layer processing of AM, the radiographic testing and machine vision technologies are used to realize the defect detection and classification. The contents of the research project include: transient melting characterization of AM on the subsurface, porosity defect area identification, overlap defects segmentation and edge reconstruction, porosity defects classification. The contributions of the research project are: we proposed the dynamic differential statistical model based subsurface image reconstruction technology, the optional shape constraint coupling snake model based multi-stage defect outline fitting method, and the defect classification method based on random-triangle-distribution feature and sparse representation classifier. Based on theoretical studies, set up the experimental platform of laser 3D printer, develop the subsurface defect detection system of AM, and take the titanium alloy honeycomb component as the application object to verify the proposed method. The completion of this project has important theoretical and practical value for the realization of monitoring and ensuring the quality of the additive manufacturing.

目前专门针对增材制造质量检测的相关研究还较少，多沿用传统减材制造检测手段。本项目针对增材制造特有逐层加工特性，在加工过程中采用射线成像技术，并配合机器视觉检测方法，实现逐层缺陷检测及其定量识别。项目研究内容包括增材制造亚表二次瞬态熔凝表征、亚表单层透射影像缺陷区域辨识、多重影缺陷分割及真实轮廓重构、及沉积微区孔隙缺陷定量分类识别。在基于动态差分统计模型的增材制件亚表熔融沉积层影像重构技术、基于可选形状约束耦合Snake 模型的多阶重影缺陷轮廓拟合方法、以及基于边缘随机三角分布特征及低秩稀疏表示的孔隙缺陷分类方法方面取得创新。搭建射线检测激光3D打印实验平台，开发增材制件亚表层缺陷逐层检测系统，并以钛合金蜂窝构件为应用对象进行应用验证。本项目的完成，对实现增材制造过程监控、保障增材制件质量有着重要的理论及应用价值。

增材制件在制备过程中难免会出现孔隙、孔洞、夹杂、裂纹等缺陷。宏观缺陷如孔洞、夹杂、裂纹等相对比较容易检测，而微观缺陷如孔隙缺陷则难以检测，缺少针对性的检测技术。本项目针对这些检测难点，在增材制造在线逐层缺陷检测、增材制件三维形貌测量方面取得了一系列重要成果。（1）提出了基于像素映射误差校正的增材制造主动视觉逐层形貌测量方法，可对增材制造逐层形貌进行精确测量及重构，排除阴影及制件反光等影响，避免系统误差弱化增材制造形貌缺陷，增强增材制造形貌缺陷的可辨识度。（2）提出了基于时序逻辑边缘阴影抑制的增材制件熔融沉积层点云重构技术，采用模型及重建层配准检测熔积缺陷，针对检测单层多区域打印，可单次完成打印层全貌检测，避免重复拟合分割，显著提高检测效率。（3）提出了一种基于超体聚类的3D打印亚表区域分割方法。将三维点云划分为细小区域，并提取特征进行缺陷检测可极大提高检测鲁棒性。与传统方法对单个三维点进行缺陷检测的方法相比，区域描述与检测能利用缺陷的异常点聚集特性，检测出偏离幅值低于噪声的细微缺陷，可满足实际工业应用需求。本项目进行期间相关研究成果在国际上具有较大的影响力，共发表了权威SCI期刊论文共25篇，授权发明专利18项。对提高增材制造质量具有重要意义。