Automatic single crystal furnace welding method and system based on CCD (Charge Coupled Device) monitoring

Abstract

The invention relates to the field of CCD automatic welding methods, in particular to a single crystal furnace automatic welding method and system based on CCD monitoring, and the method comprises the following steps: S101, obtaining a to-be-detected image of a seed crystal descending process during a welding process; s102, processing the to-be-detected image based on an image target detection model to obtain real-time seed crystal position information; and S103, acquiring a sub-image for detecting whether the seed crystal is in contact with the melt or not based on the information including the positive external rectangle of the seed crystal contour acquired in the step S102, and detecting whether the seed crystal is in contact with the melt or not according to the sub-image. The interface change of the seed crystal and the melt in the welding process can be captured in real time by utilizing the CCD camera; and through an image processing technology, the position of the seed crystal can be automatically identified, and whether the seed crystal is in contact with the melt or not, whether the seed crystal is pure original seed crystal or the original seed crystal is connected with the new seed crystal, the boundary position of the original seed crystal and the new seed crystal and the distance between the boundary position and the melt can be observed.

Description

Technical Field

[0001] The present invention belongs to the technical field of CCD automatic welding methods, and specifically relates to a single crystal furnace automatic welding method and system based on CCD monitoring. Background Art

[0002] When melting materials in a single crystal furnace, it mainly relies on manual observation windows to understand the melting situation of silicon materials in the single crystal furnace and manually control the heater power of the single crystal furnace; equipment with automatic material melting function sets a sufficiently large heater power within a sufficient time to ensure that the silicon materials in the furnace are fully melted.

[0003] A patent with publication number CN115418710A discloses a single crystal full-automatic seed crystal lowering and welding method applied to a full-automatic single crystal furnace, including: responding to a user's seed crystal lowering and welding operation request; obtaining a liquid surface temperature value, and when the liquid surface temperature value is less than or equal to a preset seed crystal lowering liquid surface temperature value, configuring the position, rotation speed, and preheating time of the seed crystal based on a preset seed crystal numerical table; when the liquid surface temperature value reaches a preset welding liquid surface temperature range, configuring a first welding position of the seed crystal to perform the seed crystal lowering and welding operation, where the preset welding liquid surface temperature value is less than the preset seed crystal lowering liquid surface temperature value. This application realizes a highly automated whole process of seed crystal welding and enhances the consistency of welding effects.

[0004] Currently, in the prior art, during the crystal pulling process of a single crystal furnace, the welding speed and position of the welding process will affect production efficiency and welding quality; in the current crystal pulling process of a single crystal furnace, manual welding and the method of lowering the seed crystal at fixed points according to the SOP table given by engineers are common operating methods. Although it is widely used, it also has the following main disadvantages:

[0005] Manual operation requires a large amount of human resources. In large-scale production, a small number of operators cannot operate a large number of crystal pulling devices; manual operation depends on the experience and techniques of operators, which in turn affects the performance consistency of crystals; manual operation may increase additional wear on equipment (such as the seed crystal rod or crucible), thereby increasing maintenance costs; the speed of lowering the seed crystal at fixed points according to the SOP is slow, especially in large-scale production, and the welding process takes a long time, thus reducing the overall production efficiency.

[0006] Therefore, the present invention provides a single crystal furnace automatic welding method and system based on CCD monitoring. Summary of the Invention

[0007] To make up for the deficiencies of the prior art and solve at least one of the technical problems proposed in the background art.

[0008] First aspect, the technical solution adopted by the present invention to solve its technical problems is: An automatic welding method for a single crystal furnace based on CCD monitoring according to the present invention includes the following steps:

[0009] S101. Obtain the image to be detected during the process of the seed crystal descending during the welding process;

[0010] S102. Process the image to be detected based on the image target detection model to obtain real-time seed crystal position information;

[0011] The image target detection model is a model obtained by manually annotating and training images with a seed crystal throughout the whole process from the start to the end of welding, and can detect the information of the minimum bounding rectangle of the seed crystal contour in each frame of the image collected by the CCD. The information of the minimum bounding rectangle of the seed crystal contour includes the coordinates of the upper left corner point of the minimum bounding rectangle and the width and height information of the rectangle. The coordinates, height, and width are all values based on the pixel coordinate system;

[0012] S103. Based on the information of the minimum bounding rectangle of the seed crystal contour obtained in step S102, obtain a sub-image for detecting whether the seed crystal touches the melt, and detect whether the seed crystal touches the melt according to this sub-image;

[0013] S104. Obtain whether the seed crystal touches the melt in step S103 and the obtained sub-image, segment the original seed crystal and the new seed crystal based on the image segmentation model, calculate the boundary position between the two, and the distance from the boundary position to the melt;

[0014] S105. Based on the distance from the boundary position between the new seed crystal and the original seed crystal obtained in step S104 to the position where the seed crystal touches the melt, calculate the displacement, speed, and heating power of the seed crystal adjusted by the control unit.

[0015] Preferably, step S101 further includes the following steps:

[0016] Collect images of the whole process from the start to the end of the seed crystal welding through the CCD end, and use the CCD image of the seed crystal welding process collected as the image to be detected;

[0017] The image target detection model will manually annotate and train the images of the whole process of the seed crystal welding, and obtain an image data set capable of detecting the information of the minimum bounding rectangle of the seed crystal contour in each frame of the image collected by the CCD. The information of the minimum bounding rectangle of the seed crystal contour includes the coordinates of the upper left corner point of the minimum bounding rectangle and the width and height information of the rectangle.

[0018] Preferably, the annotation information of the upper left corner point, width, and height in the initial stage of the seed crystal is the first annotation information representing the seed crystal;

[0019] The annotation information in the seed crystal contacting the melt stage includes that the category annotation information of the sub-image obtained in step 103 for detecting whether the seed crystal contacts the melt is the second annotation information characterizing the seed crystal;

[0020] The annotation information in the seed crystal contacting the melt stage includes the contour annotation of the original seed crystal, the contour annotation of the new seed crystal, and the category annotation information of the original seed crystal and the new seed crystal, which is the third annotation information characterizing the seed crystal;

[0021] Based on the annotation information of each image in the image dataset and the image dataset, generate image training samples; use the image features of each image in the image dataset as training features, and use the annotation information of each image as training targets, and train the image training samples to obtain seed crystal position detection, seed crystal-melt contact detection, new seed crystal and original seed crystal segmentation models in two different stages;

[0022] Specifically, in the process of generating image training samples, the whole process images from the appearance of the seed crystal to the end of welding in the image can be collected by a CCD; for example, one frame is collected every 1 second, and a total of n whole process images of various seed crystal welding forms are collected; in order to increase image features, perform image enhancement preprocessing on each image in the image dataset to obtain a preprocessed image dataset; then divide the states of the images in the preprocessed image dataset, and perform different types of annotation extraction work on the seed crystal welding features in each state.

[0023] When the seed crystal is in the stage of not contacting the melt, the CCD camera captures the image during the seed crystal welding process and transmits it to the seed crystal position detection model for image processing. The results output by the model are the coordinates of the upper left corner point of the positive circumscribed rectangle of the seed crystal contour, and the width and height of the rectangle.

[0024] Specifically, when the seed crystal is in the stage of contacting the melt, the CCD camera captures the image during the seed crystal welding process and transmits it to the seed crystal contact detection model for image processing.

[0025] Specifically, when the seed crystal is in the stage of contacting the melt, the CCD camera captures the image during the seed crystal welding process and transmits it to the seed crystal segmentation detection model for image processing. The results output by the model are the regions of the new seed crystal and the original seed crystal, calculate the position of the boundary line between the new seed crystal region and the original seed crystal region, and calculate the distance from the boundary line to the contact surface between the seed crystal and the melt.

[0026] Preferably, S105 further includes the following steps:

[0027] S201. Collect several groups of better seed crystal descent displacement sequences during the process from the start of welding to the completion of welding, and establish a seed crystal descent displacement curve library for the welding process;

[0028] An implementation method for determining a curve library based on welding data is used to obtain multiple historical seed crystal descent displacement curves during the welding of a single crystal furnace; according to the respective data in the welding data set, multiple curves matching the optimal theoretical state data are obtained from the multiple historical curves; the multiple curves are filtered to integrate the filtered multiple curves into a final curve library, thereby accurately establishing a curve library for the seed crystal descent displacement.

[0029] S202. Select the optimal seed crystal descent target curve from the seed crystal descent displacement curve library, and construct a cost function for the seed crystal length based on the real-time seed crystal descent displacement value;

[0030] S203. Calculate the seed crystal descent displacement that needs to be adjusted for the currently calculated seed crystal length based on the cost function; the displacement of the real-time seed crystal descent during welding can be controlled by the target seed crystal descent displacement.

[0031] Preferably, the cost function for the seed crystal length constraint is:

[0032]

[0033] where L pred represents the distance from the position of the boundary line between the newly generated seed crystal and the original seed crystal obtained by the calculation in step 104 to the position where the seed crystal contacts the melt, α o is the initial acceleration of the seed crystal descent after the seed crystal just contacts the melt, ω is the change rate of the seed crystal descent acceleration, t is the time of the seed crystal descent, v o is the initial velocity of the seed crystal descent, ε is the error correction coefficient (0.7 - 0.9),

[0034] L origin is the length of the original seed crystal obtained by the calculation in step 104, and L det is the height of the seed crystal obtained by the calculation in step 102.

[0035] Preferably, when obtaining the seed crystal length of L pred and the cost function is Cost(L), based on the cost function Cost(L), the calculation method of the seed crystal descent displacement value L best (t + 1) is:

[0036] L best (t + 1) ∈ [max(L(t) - ΔL), L min , min(L(t) + ΔL, L max )]

[0037] where L best (t + 1) represents the optimal seed crystal descent displacement at the (t + 1)-th moment, and ΔL is the maximum amplitude of the single seed crystal descent preset.

[0038] Preferably, step S104 further includes the following steps:

[0039] Obtain an image data set, where the image data set includes multiple images obtained by the whole-process image acquisition of the CCD from the start to the end of the welding; divide the images into two stages: the seed crystal not contacting the melt and the seed crystal contacting the melt, and perform feature labeling on the data sets of the two different stages by different methods to obtain labeling information.

[0040] Second aspect, a single crystal furnace automatic welding system based on CCD monitoring, including:

[0041] A detection module, configured to collect image data of the seed crystal in the corresponding CCD during the welding process of a large number of single crystal furnaces, process the image data, label and integrate it into a welding detection data set to form a training set; use the training set to build a seed crystal detection model based on a deep learning algorithm;

[0042] A measurement module, configured to collect CCD image data after a large number of seed crystals contact the melt, integrate and segment the data sets of the new-born seed crystal and the original seed crystal according to the results of the detection module, label the data sets, and use a deep learning algorithm to train the segmentation model of the new-born seed crystal and the original seed crystal;

[0043] A displacement adjustment module, configured to call the optimal seed crystal descent displacement curve, construct a cost function with a seed crystal length constraint; read the real-time characteristic variables within the real-time welding stage, input the real-time CCD image to calculate the length of the seed crystal, and adjust the target displacement required for the real-time target seed crystal descent position according to the inference of the cost function; loop through the steps, and roll and optimize the target displacement adjustment until the welding is completed.

[0044] The beneficial effects of the present invention are as follows:

[0045] 1. The automatic welding method and system for a single crystal furnace based on CCD monitoring according to the present invention improve the stability and consistency of the welding process. By controlling the welding process through an automatic algorithm, it can ensure that the parameters of each welding, such as displacement, time, and welding speed, are maintained within the optimal range, thereby reducing the fluctuations caused by manual operations; avoid uneven welding quality caused by insufficient experience or operation errors of workers, and thus improve the growth stability and yield of crystals.

[0046] An automatic welding method and system for a single crystal furnace based on CCD monitoring according to the present invention can capture the interface changes between the seed crystal and the melt during the welding process in real time by using a CCD camera; through image processing technology, the position of the seed crystal can be automatically identified, whether the seed crystal is in contact with the melt can be observed, whether the seed crystal is a pure primary seed crystal or a primary seed crystal connected to a new seed crystal, the demarcation position between the primary seed crystal and the new seed crystal, and the distance from the demarcation position to the melt can be determined; the control module can automatically adjust the heating power, the descending speed and displacement of the seed crystal according to the real-time image collected by the CCD and the results of real-time image processing.

[0047] An automatic welding method and system for a single crystal furnace based on CCD monitoring according to the present invention can dynamically adjust welding parameters, such as the displacement speed, through an automated algorithm combined with sensors and real-time data feedback, so as to ensure that the crystal is in the best state during the welding process; thereby improving the welding accuracy, quickly responding to abnormal situations at the same time, and reducing the crystal rejection rate; and the automated operation can reduce the working time of personnel in high-temperature and high-pressure environments, reduce the operation risk, and thus protect the safety of operators, while reducing the effect of equipment damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below with reference to the accompanying drawings.

[0049] Figure 1 is a schematic diagram of the single crystal detection process of the present invention;

[0050] Figure 2 is a schematic diagram of the seed crystal process flow of the present invention;

[0051] Figure 3 is a schematic diagram of the welding system module process flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0053] Embodiment 1

[0054] As Figures 1 to 2 shown, an automatic welding method for a single crystal furnace based on CCD monitoring according to an embodiment of the present invention includes the following steps:

[0055] S101. Obtain an image to be detected during the descending process of the seed crystal during the welding process;

[0056] S102. Process the image to be detected based on an image target detection model to obtain real-time seed crystal position information;

[0057] The image target detection model is obtained by manually annotating and training the images with seeds during the whole process from the start to the end of the welding, and can detect the information of the minimum bounding rectangle of the seed crystal contour in each frame of the image collected by the CCD. The information of the minimum bounding rectangle of the seed crystal contour includes the coordinates of the upper left corner point of the minimum bounding rectangle and the width and height information of the rectangle. The coordinates, height, and width are all based on the values in the pixel coordinate system;

[0058] S103. Based on the information of the minimum bounding rectangle of the seed crystal contour obtained in step S102, obtain a sub-image for detecting whether the seed crystal touches the melt, and detect whether the seed crystal touches the melt according to this sub-image;

[0059] S104. Obtain whether the seed crystal touches the melt in step S103 and the obtained sub-image, segment the original seed crystal and the newly grown seed crystal based on the image segmentation model, calculate the boundary position between the two, and the distance from the boundary position to the melt;

[0060] S105. Based on the distance from the boundary position between the newly grown seed crystal and the original seed crystal obtained in step S104 to the position where the seed crystal touches the melt, calculate the displacement, speed, and heating power of the seed crystal adjusted by the control unit.

[0061] As Figures 1 to 2 shown, S101 further includes the following steps:

[0062] Collect the images of the whole process of seed crystal welding from the start to the end of the seed crystal welding through the CCD end, and use the CCD image of the seed crystal welding process collected as the image to be detected;

[0063] The image target detection model will manually annotate and train the images of the whole process of seed crystal welding, and obtain an image dataset that can detect the information of the minimum bounding rectangle of the seed crystal contour in each frame of the image collected by the CCD. The information of the minimum bounding rectangle of the seed crystal contour includes the coordinates of the upper left corner point of the minimum bounding rectangle and the width and height information of the rectangle;

[0064] The upper left corner point, width, and height of the annotation information in the initial stage of the seed crystal are the first annotation information characterizing the seed crystal; the annotation information in the stage where the seed crystal contacts the melt includes the category annotation information of the sub-image obtained in step 103 for detecting whether the seed crystal contacts the melt, which is the second annotation information characterizing the seed crystal; the annotation information in the stage where the seed crystal contacts the melt includes the contour annotation of the primary seed crystal, the contour annotation of the new-grown seed crystal, and the category annotation information of the primary seed crystal and the new-grown seed crystal, which is the third annotation information characterizing the seed crystal; based on the annotation information of each image in the image dataset and the image dataset, image training samples are generated; the image features of each image in the image dataset are used as training features, and the annotation information of each image is used as the training target to train the image training samples, obtaining seed crystal position detection, seed crystal-melt contact detection, new-grown seed crystal and primary seed crystal segmentation models in two different stages; during the process of generating the image training samples, the whole process images of the seed crystal from its appearance in the image captured by the CCD to the end of welding can be collected; one frame is collected every 1 second, and a total of n whole process images of various seed crystal welding forms are collected; in order to increase the image features, each image in the image dataset is preprocessed by image enhancement to obtain a preprocessed image dataset; then the images in the preprocessed image dataset are divided into states, and different types of annotation extraction work are carried out on the seed crystal welding features in each state; when the seed crystal is in the stage of not contacting the melt, the CCD camera captures the image of the seed crystal welding process and transmits it to the seed crystal position detection model for image processing, and the output result of the model is the coordinates of the upper left corner point of the positive circumscribed rectangle of the seed crystal contour, as well as the width and height of the rectangle; specifically, when the seed crystal is in the stage of contacting the melt, the CCD camera captures the image of the seed crystal welding process and transmits it to the seed crystal contact detection model for image processing; specifically, when the seed crystal is in the stage of contacting the melt, the CCD camera captures the image of the seed crystal welding process and transmits it to the seed crystal segmentation detection model for image processing, and the output result of the model is the area of the new-grown seed crystal and the area of the primary seed crystal, the position of the boundary line between the area of the new-grown seed crystal and the area of the primary seed crystal is calculated, and the distance from the boundary line to the contact surface between the seed crystal and the melt is calculated;

[0065] S105 further includes the following steps: S201, collect several groups of better seed crystal descent displacement sequences during the process from the start of welding to the completion of welding, and establish a seed crystal descent displacement curve library for the welding process; determine an implementation method of the curve library according to the welding data to obtain multiple historical seed crystal descent displacement curves during single crystal furnace welding; according to the respective data in the welding data set, obtain multiple curves that match the optimal theoretical state data from multiple historical curves; perform filtering processing on the multiple curves to integrate the filtered multiple curves into the final curve library, thereby accurately establishing the curve library of the seed crystal descent displacement; the reference range of the evaluation index is defined by the standard welding process flow and the preset control target, and the evaluation index may include but is not limited to welding efficiency, temperature stability, temperature accuracy, crystal formation situation, and this embodiment does not make specific limitations on this, welding efficiency;

[0066] S202, select the optimal seed crystal descent target curve from the seed crystal descent displacement curve library, and construct a cost function of the seed crystal length according to the real-time seed crystal descent displacement value;

[0067] S203, calculate the seed crystal descent displacement that needs to be adjusted for the currently calculated seed crystal length based on the cost function; the displacement of the real-time seed crystal descent during welding can be controlled by the target seed crystal descent displacement; the cost function of the seed crystal length constraint is:

[0068]

[0069] where L pred represents the distance from the position of the boundary between the newly generated seed crystal and the original seed crystal obtained by calculating in step 104 to the position where the seed crystal contacts the melt, α o is the initial acceleration of the seed crystal descent after the seed crystal just contacts the melt, ω is the change rate of the seed crystal descent acceleration, t is the time of the seed crystal descent, v o is the initial velocity of the seed crystal descent, ε is the error correction coefficient (0.7 - 0.9), L origin is the length of the original seed crystal obtained by calculating in step 104, and L det is the height of the seed crystal obtained by calculating in step 102.

[0070] When obtaining the seed crystal length of L pred and the cost function is Cost(L), based on the cost function Cost(L), the calculation method of the seed crystal descent displacement value L best (t + 1) is:

[0071] L best (t + 1) ∈ [max(L(t) - ΔL), L min , min(L(t) + ΔL, L max )]

[0072] where Lbest (t + 1) represents the optimal seed crystal descent displacement at the (t + 1)-th moment, and ΔL is the maximum amplitude of a single seed crystal descent preset.

[0073] S104 further includes the following steps: obtaining an image data set, which includes multiple images obtained by the CCD collecting images of the whole process from the start to the end of welding; dividing the images into two stages: the seed crystal not in contact with the melt and the seed crystal in contact with the melt, and performing feature annotation on the data sets of the two different stages by different methods to obtain annotation information.

[0074] Embodiment 2

[0075] As Figure 3 shown, an automatic welding system for a single crystal furnace based on CCD monitoring according to an embodiment of the present invention includes:

[0076] A detection module, configured to collect image data of the seed crystal in the corresponding CCD during the welding process of a large number of single crystal furnaces, and process and annotate the image data to integrate it into a welding detection data set to form a training set; based on the training set and a deep learning algorithm, a seed crystal detection model is built.

[0077] A measurement module, configured to collect CCD image data after a large number of seed crystals come into contact with the melt, integrate and segment the data sets of the new-born seed crystal and the original seed crystal according to the results of the detection module, and perform annotation on the data sets, and train a segmentation model of the new-born seed crystal and the original seed crystal by using a deep learning algorithm.

[0078] A displacement adjustment module, configured to call the optimal seed crystal descent displacement curve, construct a cost function with a seed crystal length constraint; read real-time characteristic variables within the real-time welding stage, input the real-time CCD image to calculate the length of the seed crystal, and infer the target displacement required to adjust to the real-time target seed crystal descent position according to the cost function; loop the steps to roll and optimize the target displacement adjustment until the welding is completed.

[0079] Working principle: The CCD camera can capture the interface change between the seed crystal and the melt during the welding process in real time; through image processing technology, the position of the seed crystal can be automatically identified, whether the seed crystal is in contact with the melt can be observed, whether the seed crystal is a pure original seed crystal or the original seed crystal is connected with a new-born seed crystal, the boundary position between the original seed crystal and the new-born seed crystal, and the distance from the boundary position to the melt can be observed; the control module can automatically adjust the heating power, the speed and displacement of the seed crystal descent according to the real-time image collected by the CCD and the results of real-time image processing.

[0080] The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A single crystal furnace automatic welding method based on CCD monitoring, characterized in that: The following steps are involved: S101, obtaining an image to be detected during the seed crystal descending process during the welding process; S102, processing the image to be detected based on an image target detection model to obtain real-time seed crystal position information; S103, based on the information of the positive circumscribed rectangle of the seed crystal outline obtained in step S102, obtaining a sub-image for detecting whether the seed crystal contacts the melt, and detecting whether the seed crystal contacts the melt according to the sub-image; S104, obtaining whether the seed crystal contacts the melt and the obtained sub-image in step S103, segmenting the primary seed crystal and the newly formed seed crystal based on the image segmentation model, and calculating the boundary position between the two and the distance from the boundary position to the melt; S105 , based on the distance from the boundary position between the new seed crystal and the original seed crystal to the position where the seed crystal contacts the melt obtained in step S104 , calculate the displacement and speed of the seed crystal adjusted by the control unit, as well as the heating power.

2. The single crystal furnace automatic welding method based on CCD monitoring according to claim 1 is characterized in that: The S101 further comprises the following steps: The whole process image from the beginning to the end of seed crystal fusion is collected through the CCD end, and the collected CCD image of the seed crystal fusion process is used as the image to be detected; The image target detection model will manually annotate and train the images of the entire seed crystal welding process to obtain an image data set that can detect the positive circumscribed rectangular frame information of the seed crystal outline in each frame image collected by the CCD. The positive circumscribed rectangular frame information of the seed crystal outline includes the coordinates of the upper left corner point of the positive circumscribed rectangle and the width and height information of the rectangle.

3. The single crystal furnace automatic welding method based on CCD monitoring according to claim 1 is characterized in that: The upper left corner point, width and height of the annotation information of the seed crystal in the initial stage are the first annotation information characterizing the seed crystal; The annotation information of the seed crystal contacting the melt stage is that the category annotation information of the sub-image for detecting whether the seed crystal contacts the melt obtained in step 103 is the second annotation information characterizing the seed crystal; The annotation information of the seed crystal contacting the melt stage includes the contour annotation of the primary seed crystal, the contour annotation of the newly formed seed crystal, and the category annotation information of the primary seed crystal and the newly formed seed crystal, which is the third annotation information characterizing the seed crystal.

4. The single crystal furnace automatic welding method based on CCD monitoring according to claim 1 is characterized in that: The S105 further comprises the following steps: S201, collecting several groups of good seed crystal descending displacement sequences from the beginning of welding to the completion of welding, and establishing a seed crystal descending displacement curve library for welding process; S202, selecting an optimal seed crystal descent displacement target curve from a seed crystal descent displacement curve library, and constructing a seed crystal length cost function according to the real-time seed crystal descent displacement value; S203, calculating the seed crystal descent displacement that needs to be adjusted for the currently calculated seed crystal length based on the cost function; the seed crystal descent displacement in real time during welding can be controlled by the target seed crystal descent displacement.

5. The single crystal furnace automatic welding method based on CCD monitoring according to claim 4 is characterized in that: The cost function of the seed crystal length constraint is: Among them, L pred represents the distance from the boundary position between the new seed crystal and the original seed crystal calculated in step 104 to the contact position between the seed crystal and the melt, α o is the initial acceleration of the seed crystal just after it contacts the melt, ω is the rate of change of the seed crystal's descent acceleration, t is the descent time of the seed crystal, v o is the initial velocity of the seed crystal descent, ε is the error correction coefficient (0.7~0.9), L origin is the length of the primary seed crystal calculated in step 104, L det is the height of the seed crystal calculated in step 102 .

6. The single crystal furnace automatic welding method based on CCD monitoring according to claim 5 is characterized in that: The length of the seed crystal obtained is L pred , when the cost function is Cost(L), based on the cost function Cost(L), the seed crystal drops by a displacement value L best (t+1) is calculated as: L best (t+1)∈[max(L(i)-ΔL),L min ,min(L(t)+ΔL,L max )] Among them, L best (t+1) represents the optimal seed crystal descent displacement at the t+1th moment, and ΔL is the preset maximum amplitude of a single seed crystal descent.

7. The single crystal furnace automatic welding method based on CCD monitoring according to claim 1 is characterized in that: The S104 further comprises the following steps: An image data set is obtained, wherein the image data set includes a plurality of images obtained by collecting images of the entire process from the beginning to the end of welding by a CCD end; the images are divided into two stages, namely, a seed crystal not contacting the melt and a seed crystal contacting the melt, and feature annotations of different methods are performed on the data sets of the two different stages to obtain annotation information.

8. A single crystal furnace automatic welding system based on CCD monitoring, characterized in that: include: The detection module is used to collect the image data of the seed crystal in the CCD corresponding to the welding process of a large number of single crystal furnaces, and process and annotate the image data to form a training set for welding detection data sets; the seed crystal detection model is built based on the deep learning algorithm using the training set; The measurement module is used to collect CCD image data of a large number of seed crystals after they come into contact with the melt, integrate and segment the data sets of new seed crystals and native seed crystals according to the results of the detection module, annotate the data sets, and use the deep learning algorithm to train the segmentation model of new seed crystals and native seed crystals; The displacement adjustment module is used to call the optimal seed crystal descent displacement curve and construct a cost function with seed crystal length constraint; read the real-time characteristic variables in the real-time welding stage, input the CCD real-time image to calculate the length of the seed crystal, and adjust the target displacement required for the real-time target seed crystal descent position according to the inference of the cost function; loop the steps and roll the optimization target displacement adjustment until the welding is completed.

Images