基于机器视觉的加热片表面缺陷检测技术研究

发布时间：2018-11-29 08:18

【摘要】：热电池是一种重要的军用电源,广泛应用于导弹、舰艇、核武器及民用航空等领域,而加热片是热电池加热系统的重要组成部分,为热电池提供热量,其质量好坏直接影响着热电池是否正常工作。加热片的制造过程分为粉料生产、粉料混合、压制成形及外观检测等工序。现阶段,粉料生产、混合及压制均已实现自动化和国产化,但加热片的外观检测仍然依靠人工目检的方式,这种检测方式严重制约了加热片的生产效率,也难以保证加热片的质量。本文的加热片由活性铁粉与高氯酸钾粉末混合并压制而成。加热片表面缺陷检测的主要任务是对完好加热片与次品进行快速区分,同时对次品加热片的四类缺陷即掉边、毛刺、夹杂与裂纹缺陷进行识别。首先,设计了加热片视觉检测系统的硬件部分,选用了康耐视公司的insight7050视觉系统等硬件。完成硬件系统的安装及调试后,利用该系统对加热片进行图像采集,获得了有利于算法处理的图像。其次,图像的预处理中,提出了一种自适应的抗噪形态学边缘的二值化阈值。在此基础上,为避免加热片图像的背景对处理结果造成干扰,提出了一种基于抗噪膨胀腐蚀型形态学边缘的前景提取算法;另外,为实现完好加热片及次品的快速分类,提出了一种基于抗噪腐蚀型形态学边缘的疑似缺陷检测算法。再次,针对加热片图像中的阴影区,高光区,阴影高光夹杂区及表面纹理,提出了一种结合高斯高频强调滤波与本文改进的Catte_pm模型图像增强算法。再次,针对加热片图像的背景引起最小误差法失效的问题,提出了一种改进的最小误差法。另外,分析比较了基于统计特征,基于主成分分析与支持向量机,基于主成分分析与神经网络的缺陷分类方法的识别准确率,并研究了基于主成分分析的两种分类方法的降维维度与识别准确率的关系。最后,完成了软件子系统的开发,并对其进行功能及技术指标验证。实验表明,本文算法实现了完好加热片与次品的快速分类;同时针对掉边、毛刺、夹杂及裂纹四类缺陷具有极高的识别准确率。
[Abstract]:Thermal battery is an important military power source, widely used in missile, naval vessel, nuclear weapon and civil aviation, etc. The heating plate is an important part of the heating system of thermal battery, which provides heat for thermal battery. Its quality directly affects whether the thermal battery works normally. The manufacturing process of heating sheet is divided into powder production, powder mixing, pressing forming and appearance inspection. At the present stage, powder production, mixing and compaction have been realized automatically and domestically, but the appearance inspection of heating sheet still depends on manual inspection, which seriously restricts the production efficiency of heating sheet. It is also difficult to guarantee the quality of the heating sheet. The heating sheet is prepared by mixing and compacting active iron powder and potassium perchlorate powder. The main task of the surface defect detection of the heating sheet is to quickly distinguish the perfect heating sheet from the defective product, and at the same time to identify the four kinds of defects of the defective heating sheet, that is, falling edge, burr, inclusion and crack defect. Firstly, the hardware of the heating slice vision detection system is designed, and the hardware of insight7050 vision system is selected. After the hardware system is installed and debugged, the image of the heating slice is collected by the system, and the image is obtained which is propitious to the algorithm. Secondly, in image preprocessing, an adaptive binarization threshold of the edge of anti-noise morphology is proposed. On this basis, to avoid the interference of the background of the heated slice image to the processing results, a foreground extraction algorithm based on the noise-resistant swelling corrosion morphological edge is proposed. In addition, in order to realize the fast classification of perfect heating sheets and defective products, a new algorithm of suspected defect detection based on noise-resistant morphological edge is proposed. Thirdly, an improved Catte_pm model image enhancement algorithm based on Gao Si high-frequency emphasis filter and the improved Catte_pm model image enhancement algorithm is proposed for the shadow region, the highlight region, the shadow highlight area and the surface texture of the heated slice image. Thirdly, an improved minimum error method is proposed for the failure of the minimum error method caused by the background of the heated slice image. In addition, the recognition accuracy of defect classification methods based on statistical feature, principal component analysis and support vector machine, principal component analysis and neural network is analyzed and compared. The relationship between dimensionality reduction and recognition accuracy of two classification methods based on principal component analysis is studied. Finally, the software subsystem is developed, and its function and technical index are verified. The experiments show that the algorithm realizes the fast classification of perfect heating plates and defective products, and it has a high recognition accuracy for four kinds of defects, such as missing edges, burrs, inclusions and cracks.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TM915

【参考文献】