Research On The Key Technology And Application Of High Temperature Monocrystalline Silicon Liquid Level And Diameter Detection By Vision

As the most important raw material of information industry and new energy industry,silicon single crystal plays an important role in national economy and national strategy.In order to produce high-quality silicon single crystal efficiently using the Czochralski(CZ)method,it is necessary to measure the melt level and the crystal diameter timely and accurately during the production process.For the silicon production,machine vision is the ideal non-contact detection method.However,due to the influence of high temperature and hash environment,irritating problems such as difficult design of optical path,poor image quality,and improper image detection algorithms have hindered the application of machine vision to the silicon production.In this paper,the key technologies of applying machine vision to high-temperature productions are studied.We designed 2 machine vision systems based on our research results.One system is used for measuring the melt level of silicon single crystal in CZ furnace and the other is used for measuring the crystal diameter.The main contents are as follows.The first chapter expounds the research background,significance and purpose.We introduce the importance of silicon single crystal and the production technology of Czochralski method.Also,the defects of existing machine vision systems for melt level detection and diameter detection of silicon single crystal are analyzed in detail.At the same time,this chapter outlines the development of machine vision and the research results both at home and abroad,and points out key technologies to be solved when applying machine vision systems to the production of silicon single crystal.Finally,aiming at these problems,the source of our research and the main contents of this paper are given.In chapter 2,the design of optical path and optimization of detection principle in high-temperature applications are studied.A principle that the high-temperature vision systems should design the optical path and optimize the detection principle according to the spatial structure characteristics of the target is put forward.Combined with practical application of melt level detection of silicon single crystal,an improved laser triangulation method is proposed.This method can calculate the melt level linearly from the distance between the incident light and the reflected light on a Mo sheet produced by a linear laser.We also propose a novel method for diameter measurement of silicon single crystal based on concentric ellipses.This method doesn't apply a complex circle/ellipse fitting algorithm to every image during diameter detection,which greatly improves the measurement efficiency.In chapter 3,in order to get high-quality images in high-temperature productions,we put forward three standards and carried out a hardware design around these three standards combing the measurement system for melt level detection of silicon single crystal.Especially,the radiation characteristics of high temperature objects are studied and a SNR model for high temperature images using linear lasers is established.Although this SNR model is not exactly accurate,it is still a good guide for hardware selection of machine vision systems using lasers.Experimental results show that the system we designed can capture melt-level images of high quality.In chapter 4,we study the ellipse detection technology and propose a novel ellipse detection algorithm based on parameter reduction and randomized Hough transform.The algorithm first proves that a middle-point line of any set of parallel chords passes through the center of the ellipse.Then,the ellipse center is defined as the intersection of two middle-point lines of two different sets of parallel chords.The rotation angle of the ellipse can be computed by the symmetry relationship between the ellipse and its concentric circle.The randomized Hough transform is applied to detect the rotation angle,the long axis and the short axis of the ellipse by two points which have the same distance to the center of the ellipse.Experimental results show that the proposed algorithm has fast detection speed and is insensitive to noise and incomplete ellipse,and can be applied to the diameter measurement of silicon single crystal.In chapter 5,the problem of inaccurate edge detection in high-temperature applications is studied.Due to the measurement condition(high temperature,intense thermal radiation,flow of protective gases,etc.),edges in high-temperature images are surrounded by auras.As a result,the border between the object and the background becomes blurred and unsharp,which hinders the proper determination of the edges and decreases the measurement accuracy.A sub-pixel edge detection algorithm based on contour and Sigmoid fitting is proposed.The algorithm first uses the traditional Sobel operator to get the contour information of the target.Then the intensity distribution of pixels on the normal line direction of the coarse edge is fitted by Sigmoid function,and the sub-pixel location of the edge is obtained by the parameters of Sigmoid function.The algorithm is tested in Chapter 6 in a practical application,and experimental results prove this algorithm can measure the diameter accurately and rapidly.Chapter 6 develops two machine vision systems based on the theoretical research results of this paper.One system is used for measuring the melt level of silicon single crystal in CZ furnace.The othe system is applied to measure the diameter of silicon single crystal in CZ furnace.We verify the theoretical results presented in this paper in these 2 systems and find that these methods have very good practical values.The experimental results of the melt level measurement system show that the measurement precision of the system is 0.068mm within the whole measurement range,and it has a good performance in measurement efficiency and stability.The experimental results of the diameter measurement system show that the system has a measurement precision of 0.09mm.Compared with traditional methods,the system has a great improvement in measurement efficiency,and the system is also robust in the long-time measurement process.Both systems can meet the demands of industrial production.In chapter 7,we summarized the major work of this thesis and briefly discussed the main contents,conclusion and innovation of our research.Furthermore,the next research work is prospected.