太阳能电池硅片缺陷自动检测分类方法研究

发布时间：2019-01-06 14:25

【摘要】：太阳能电池硅片的质量是影响电池片转换效率以及电池组件发电效率的一个关键因素,因此对太阳能电池硅片的质量检测在生产和实验中显得尤为重要。常用的太阳能电池硅片有单晶硅片和多晶硅片,硅片在生产过程中受诸多因素的影响,或多或少地存在一些缺陷。多晶硅片常见的缺陷有边缘不纯、高不纯度、位错缺陷,单晶硅片常见的缺陷有漩涡缺陷。硅片缺陷的存在会极大地降低电池片的发电效率,减少电池组件的使用寿命,甚至影响光伏发电系统的稳定性。目前在实际生产实验中,大都是采用太阳能电池片电致发光缺陷检测,以人眼观察或者自动检测的方法进行检测。由于人眼观察的方法具有很强的主观性,并且人眼容易疲劳,大大降低了检测的可靠性和效率。另外,由于电致发光缺陷检测是针对电池片进行的检测,不能够检测生产过程中硅片、扩散片等过程片的缺陷,这样就提高了生产成本,降低了生产效率；并且电致发光检测技术是接触式检测,会给电池片带来不同程度的损伤。因此,一种能在生产过程中可以针对太阳能电池硅片缺陷的非接触式高效准确的自动检测方法是非常有价值的。本文以数字图像处理技术作为基础,对太阳能电池硅片光致发光缺陷检测分类方法进行了相关研究,并且提出了硅片缺陷的自动检测分类方法。本文的工作主要包括以下部分： 1.首先对光致发光图像预处理,包括图像去噪、增强、边缘检测、直线检测、图像旋转,目标硅片自动分割。 2.然后利用高斯曲线拟合多晶硅片图像灰度曲线方法计算分割阈值并分割缺陷,提取缺陷的面积比例与分布特征；对于单晶硅片,利用高斯曲线拟合图像中抽样像素的灰度和值曲线,提取拟合标准差；通过频域滤波结合二值化方法提取高频图像中高强度部分面积比；在高频二值化图像细化后,提取霍夫变换检测圆结果；得到漩涡缺陷的三个特征。 3.最后构造出缺陷检测分类树模型,实现缺陷的检测分类,对多晶硅片的三种缺陷采用排除法依次检测。并且基于C#完成系统软件各个功能模块的设计编写与整合。在实际应用中完成系统软件的测试,结果显示缺陷的检测分类准确率可以达到95%以上,证明本文方法的正确性与系统软件设计的合理性。本文提出一种在太阳能电池片生产中,对多晶硅片和单晶硅片进行非接触式自动化缺陷检测分类的方法,并且实现了软件的设计编写。实验证明本文的方法高效准确,有着很大的应用前景。
[Abstract]:The quality of solar cell silicon wafer is a key factor that affects the conversion efficiency of solar cell and the generation efficiency of battery module. Therefore, the quality detection of solar cell silicon wafer is particularly important in production and experiment. There are single crystal silicon wafers and polycrystalline silicon wafers in common use in solar cells. The silicon wafers are affected by many factors in the process of production, and there are some defects more or less. The common defects of polysilicon wafer are edge impurity, high impurity, dislocation defect and swirl defect of single crystal silicon wafer. The existence of wafer defects will greatly reduce the generation efficiency of the battery chip, reduce the service life of the battery components, and even affect the stability of photovoltaic power generation system. At present, in the actual production experiments, most of the solar cell electroluminescent defect detection, using human eye observation or automatic detection method to detect. The method of human eye observation is very subjective and easy to fatigue, which greatly reduces the reliability and efficiency of detection. In addition, because the detection of electroluminescent defects is aimed at the battery chip, it can not detect the defects of silicon wafer and diffusion wafer in the production process, which increases the production cost and reduces the production efficiency. And electroluminescent detection technology is contact detection, which will bring different damage to the battery chip. Therefore, a non-contact, efficient and accurate automatic detection method for silicon wafer defects in solar cells is very valuable. Based on the digital image processing technology, this paper studies the photoluminescence defect detection and classification method of solar cell silicon wafer, and puts forward the automatic detection and classification method of silicon wafer defect. The work of this paper mainly includes the following parts: 1. First, the photoluminescence image preprocessing, including image denoising, enhancement, edge detection, line detection, image rotation, target wafer segmentation. 2. Then using Gao Si curve fitting polysilicon chip image gray-scale curve method to calculate the segmentation threshold and segment defects, extract the defect area ratio and distribution characteristics; For monocrystalline silicon wafer, Gao Si curve is used to fit the gray and value curves of sampling pixels in the image, and the fitting standard deviation is extracted, and the high intensity partial area ratio in high frequency image is extracted by frequency domain filtering combined with binarization method. After the high-frequency binary image thinning, the Hough transform is extracted to detect the circle, and three features of the vortex defect are obtained. 3. Finally, a defect detection tree model is constructed to realize defect detection and classification. The three defects of polysilicon wafer are detected by eliminating method in turn. And based on C # to complete the design and integration of each functional module of the system software. The system software is tested in practical application. The result shows that the accuracy of defect detection and classification can reach more than 95%, which proves the correctness of the method and the rationality of the system software design. In this paper, a non-contact automatic defect detection and classification method for polysilicon wafer and single crystal silicon wafer in solar cell production is proposed, and the software is designed and compiled. Experiments show that this method is effective and accurate, and has a great prospect of application.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM914.4;TP391.41

【参考文献】