A visual identification detection method for superheat welding melting point

Abstract

The application provides a superheating welding melting point visual recognition detection method, steps are as follows: S1: establishing a deep learning model; S2: capturing a welding process image through a visual system at a fixed angle; S3: transmitting the welding process image to the deep learning model; S4: performing crystal seed introduction feature detection and logical judgment on the welding process image through the deep learning model, and outputting a recognition result. The application has the beneficial effects that the visual system is developed to recognize the size of the melting point, the pixel value of the melting point is directly outputted as a crystal seed introduction signal, a processor executes the crystal seed introduction process through the visual signal to achieve the consistency of the crystal seed introduction, the work efficiency of the staff is improved, the production efficiency is further improved, and the competitiveness of the enterprise is improved; the state of the welding melting point in the superheating welding process of the crystal seed is not determined manually, automatic and industrialized production is realized, the waste of working hours and the occurrence of abnormal accidents caused by insufficient manual experience are avoided, and the safety in the operation process is higher.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic monocrystalline silicon pulling production technology, and in particular relates to a visual recognition and detection method for overheated welding melting point. Background Technology

[0002] Before entering the current crystal-leading process, it is necessary to check whether there are crystal-leading characteristics. During the overheating welding process, when the liquid surface temperature of the silicon solution reaches the welding temperature, the seed crystal is automatically lowered to the liquid surface of the silicon solution and positioned at the original seed crystal, so that the seed crystal is in contact with the liquid surface of the silicon solution.

[0003] In the current temperature-controlled process, there is no fusion point in the early stage of crystal pulling, and the temperature is too high, which easily causes crystal breakage. Furthermore, it is impossible to unify the consistency of crystal pulling. Operators cannot intuitively observe whether crystal pulling characteristics appear in the early stage of entering the crystal pulling process, nor can they intuitively identify the size of the fusion point. They cannot directly issue crystal pulling signals to prompt operation, resulting in low work efficiency. Moreover, due to the influence of human error, the consistency of entering the crystal pulling process cannot be guaranteed, which may lead to waste of resources. Summary of the Invention

[0004] The problem to be solved by the present invention is to provide a visual recognition and detection method for overheated weld melting point, which is applicable to the visual judgment of the characteristics of the crystal pulling process before entering the crystal pulling process.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a visual recognition and detection method for overheated welding melting points, the steps of which are as follows:

[0006] S1: Build a deep learning model;

[0007] S2: Capture images of the welding process at a fixed angle using a vision system;

[0008] S3: Transmit the welding process image to the deep learning model;

[0009] S4: The deep learning model is used to perform crystal-leading feature detection and logical judgment on the welding process image, and the recognition result is output.

[0010] Furthermore, in S1: basic source data on the number of fusion points on each seed crystal in each single crystal furnace during the process from the start of the overheating fusion process to the end of the crystal pulling process is obtained;

[0011] The acquired source data is processed, and the source data is used as material to construct the deep learning model. The deep learning model is then trained and iteratively optimized.

[0012] Furthermore, prior to step S2, the melting point image needs to be captured using four-point tracking by the vision system;

[0013] The melting point image is transmitted to the deep learning model;

[0014] The deep learning model is used to detect and logically judge the fusion point of the melting point image, and the recognition result is output.

[0015] Furthermore, the vision system is positioned directly in front of the seed crystal, and the vision system captures one melting point image for every 90 degrees counterclockwise rotation of the seed crystal; the vision system captures the melting point image after the seed crystal has rotated one full revolution.

[0016] Furthermore, the melting point image is sent to the deep learning model via a CCD program.

[0017] Furthermore, the deep learning model outputs results based on whether a clearly visible weld point is detected in each of the melting point images;

[0018] When there is one and the weld point exists, proceed to the crystal lead characteristic judgment process;

[0019] When the fusion point is absent or the number of fusion points is greater than one, a crystal change alarm is triggered, and seed crystal processing is performed.

[0020] Furthermore, in S2: the image of the welding process is captured from the completion of overheating to the end of entering the crystal-leading state;

[0021] The vision system is positioned directly in front of the seed crystal. The vision system captures one image of the welding process per second. The vision system also captures the number of small white dots in the area of ​​the fan-shaped field of view that is directly in front of the vision system.

[0022] Furthermore, in step S3: the image of the welding process is sent to the deep learning model via a CCD program.

[0023] Furthermore, in S4:

[0024] The deep learning model converts the number of weld white dots into the number of pixels of the weld white dots within the area of ​​the region in the weld process image;

[0025] The deep learning model outputs a result based on whether the number of small white dots in the area of ​​the detected welding process image meets the threshold in the deep learning model.

[0026] Furthermore, the threshold is that the area contains 40-80 pixels. When the number of pixels of the fused white dots meets the threshold, the process of chip pulling begins.

[0027] If the number of pixels of the fused white dot does not meet the threshold, the process of chip introduction will not proceed.

[0028] The above technical solution has the following advantages:

[0029] 1. By developing a vision system to identify melting point size, the identified melting point pixel value is directly output as a lead-in signal. The processor executes the lead-in process through the vision signal to achieve consistency in lead-in, which improves the work efficiency of the staff, further improves production efficiency, and enhances the competitiveness of the enterprise.

[0030] 2. It eliminates the need for manual determination of the melting point during the overheating and welding process of the seed crystal, enabling automated industrial production and avoiding wasted time and accidents caused by insufficient human experience, thus making the operation safer.

[0031] 3. By using a machine's perspective instead of a human's, the consistency of seed crystals before crystal pulling is improved, saving manpower and time;

[0032] 4. It can realize real-time monitoring of seed crystal status, timely identification of abnormal seed crystals, and improve the efficiency of single crystal manufacturing. Attached Figure Description

[0033] Figure 1 This is a flowchart of an embodiment of the present invention;

[0034] Figure 2 This is a schematic diagram of a vision system capturing a field of view in one embodiment of the present invention.

[0035] In the picture:

[0036] 1. Seed crystal 2. Rectangular region Detailed Implementation

[0037] The present invention will be further described below with reference to embodiments and accompanying drawings:

[0038] In one embodiment of the present invention, such as Figure 1 As shown, a visual recognition detection method for overheated weld melting points is described, and the detection method is as follows:

[0039] S1: Establish a deep learning model. A deep learning model is established for the number of fusion points for each seed crystal from the start of the overheating fusion process to the end of the seeding process. This includes the following steps:

[0040] S11: Obtain basic source data on the number of fusion points on each seed crystal in each single crystal furnace from the start of the overheating fusion process to the end of the crystal pulling process;

[0041] S12: Process the source data of the obtained welding points, use the source data as material to build a deep learning model, and train the deep learning model to achieve iterative optimization;

[0042] The deep learning model is built on the processor. The acquisition of the basic source data also requires a vision system on the processor. The vision system collects and captures a large number of images of the fusion points on the seed crystal 1 to build a database of fusion points, thereby building the deep learning model. Before visually recognizing and detecting the overheated fusion points, the deep learning model needs to be simulated, trained and tested, and continuously iterated and optimized.

[0043] S2: Capture melting point images through four-point tracking of the vision system: Specifically, the vision system is set facing the seed crystal, and the vision system captures a melting point image every time the seed crystal rotates 90 degrees counterclockwise; the vision system captures the melting point image of the seed crystal after one full rotation.

[0044] The vision system is mounted on the server. Specifically, the vision system can be, but is not limited to, an industrial camera. The vision system faces the seed crystal and is positioned on one side of the seed crystal 1. In this embodiment, the vision system is positioned directly in front of it, that is, facing the position of the middle arrow in the figure. First, it captures the melting point image of the point directly in front of the vision system. For every 90 degrees that the seed crystal rotates, the vision system captures one melting point image. The seed crystal rotates counterclockwise, that is, the vision system captures a total of 4 melting point images during the process of the seed crystal 1 rotating counterclockwise for one full cycle.

[0045] S3: Transmit the melting point image to the deep learning model. Specifically, the melting point image is sent to the deep learning model via a CCD program. The CCD program sends one melting point image to the deep learning model every second. The CCD program is used for real-time image transmission.

[0046] S4: The deep learning model is used to detect and logically judge the weld points in the melting point images and output the recognition results. Specifically, the deep learning model outputs the results based on whether there are clearly visible weld points in each melting point image.

[0047] When there is only one fusion point, proceed to the crystal lead characteristic judgment process;

[0048] When there is no fusion point or the number of fusion points is greater than one, a crystal change alarm is triggered, and seed crystal processing is performed.

[0049] The crystal change alarm device is also located on the processor. It uses four-point tracking to determine whether the seed crystal has changed. Crystal change means that the crystal structure is abnormal, which is manifested as multiple melting points or the absence of melting points. Four-point tracking is a prerequisite for determining whether to start crystal pulling. When the seed crystal 1 enters the field of view of the vision system during rotation, it is directionally tracked until it rotates to face the vision system and begins to capture. It must be ensured that there are four clearly visible fusion melting points on the seed crystal during one rotation, and the size and shape of the four fusion melting points are consistent. Four-point tracking is used to pre-filter whether the crystal pulling conditions are met. If there is an abnormality, crystal pulling is not performed, and an alarm prompt is output to process the seed crystal.

[0050] When there is only one fusion point, continue to determine the crystallization characteristics, specifically:

[0051] S5: Capture the fusion process image at a fixed angle using a vision system, specifically including the following steps: the vision system is set to face the seed crystal, the vision system captures one fusion process image per second, and the vision system captures the number of fusion white dots in the area of ​​the fan-shaped field of view facing the vision system; the process of the vision system capturing the fusion process image starts after the overheating fusion is completed and ends when the crystal is introduced.

[0052] Similarly, the position of the vision system remains unchanged. The vision system faces the seed crystal 1 and is positioned on one side of the seed crystal 1. In this embodiment, the vision system is positioned directly in front of it, that is, the vision system faces... Figure 2 The position of the middle arrow shown is configured such that a fixed angle is used to select and capture the area. The trackable field of view of the vision system is fan-shaped, as shown below. Figure 2 As shown, the fan-shaped field of view defined by the arrows on both sides is the field of view that the vision system can track. The vision system starts tracking as soon as it sees a melting point within the visible fan-shaped area, until the number of fused white dots detected in the area directly in front reaches a set threshold. Here, the fused white dots are clear and stable fused melting points. The number of fused white dots in the area is a condition for determining whether the crystal pulling process can proceed. In this embodiment, the area is a rectangular area directly facing the vision system. The size of this rectangular area can accommodate 40-80 fused white dots. The seed crystal completes one rotation in 6 seconds. The vision system continuously captures images of the fusion process during the seed crystal's rotation until the number of fused white dots detected in the area directly in front reaches the set threshold.

[0053] S6: Transmit the fusion process image to the deep learning model. Specifically, the fusion process image is sent to the deep learning model through the CCD program. The CCD program sends one fusion process image to the deep learning model every second. The CCD program is used for real-time image transmission.

[0054] S7: A deep learning model is used to perform crystal-leading feature detection and logical judgment on the welding process images, and the recognition results are output as follows:

[0055] S71: The deep learning model identifies the number of weld white dots and converts it into the number of pixels of weld white dots within the area of ​​the region in the image of the weld process;

[0056] S72: The deep learning model outputs results based on whether the number of small white dots in the area of ​​the detected welding process image meets the threshold in the deep learning model.

[0057] The threshold is set to include 40-80 pixels in the area. When the number of pixels of the fused white dot meets the threshold, the chip pulling process begins.

[0058] If the number of pixels of the fused white dot does not meet the threshold, the chip pulling process will not proceed.

[0059] In this embodiment, the area is defined as a rectangular frame, and the number of fused white dots in the rectangular area 2 is set to 40-80 pixels. During the capture process, the vision system continuously captures images. If the number of pixels of fused white dots in the rectangular frame captured in each fusion process image meets the set threshold, a chip-leading signal is directly issued, and the chip-leading process is initiated. The vision system identifies the melting point size and directly outputs the chip-leading signal based on the identified melting point pixel value. The processor executes the chip-leading process through the vision signal to achieve chip-leading consistency, thereby improving work efficiency and production efficiency.

[0060] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A visual recognition and detection method for the melting point of overheated welds, characterized in that, The steps are as follows: S1: Build a deep learning model; S2: The vision system captures images of the fusion point on the seed crystal during the fusion process at a fixed angle; specifically, the vision system is set facing the seed crystal, and the vision system captures one fusion point image every time the seed crystal rotates 90 degrees counterclockwise; the vision system captures fusion point images of the seed crystal after one full rotation. S3: Transmit the image of the fusion melting point on the seed crystal during the fusion process to the deep learning model; S4: The deep learning model is used to perform crystal-leading feature detection and logical judgment on the image of the fusion point on the seed crystal during the fusion process, and the recognition result is output. The deep learning model outputs results based on whether a clearly visible weld point is detected in each of the melting point images; When there is one and the weld point exists, proceed to the crystal lead characteristic judgment process; When the fusion point is absent or the number of fusion points is greater than one, a crystal change alarm is triggered, and seed crystal processing is performed.

2. The method for visual recognition and detection of overheated weld melting points according to claim 1, characterized in that: In S1: acquire basic source data on the number of fusion points on each seed crystal in each single crystal furnace from the start of the overheating fusion process to the end of the crystal pulling process; The acquired source data is processed, and the source data is used as material to construct the deep learning model. The deep learning model is then trained and iteratively optimized.

3. The method for visual recognition and detection of overheated weld melting points according to claim 2, characterized in that: Before step S2, the melting point image needs to be captured by four-point tracking of the vision system. The melting point image is transmitted to the deep learning model; The deep learning model is used to detect and logically judge the fusion point of the melting point image, and the recognition result is output.

4. A visual recognition and detection method for overheated weld melting points according to any one of claims 1-3, characterized in that: The melting point image is sent to the deep learning model via a CCD program.

5. The method for visual recognition and detection of overheated weld melting points according to claim 1, characterized in that: In S2: the image capturing the fusion melting point on the seed crystal during the fusion process starts from the completion of overheating and ends when entering the crystal-leading state; The vision system is positioned directly in front of the seed crystal. The vision system captures one image per second of the fusion point on the seed crystal during the fusion process. The vision system also captures the number of small white dots of fusion within the area of ​​the fan-shaped field of view that is directly in front of the vision system.

6. The method for visual recognition and detection of overheated weld melting points according to claim 5, characterized in that: In step S3: The image of the fusion melting point on the seed crystal during the fusion process is sent to the deep learning model via a CCD program.

7. The method for visual recognition and detection of overheated weld melting points according to claim 6, characterized in that: In S4: The deep learning model converts the number of the weld white dots into the number of pixels of the weld white dots within the area of ​​the region in the image of the weld point on the seed crystal during the welding process. The deep learning model outputs a result based on whether the number of small white dots in the region area of ​​the image of the fusion point on the seed crystal during the fusion process meets the threshold in the deep learning model.

8. The method for visual recognition and detection of overheated weld melting points according to claim 7, characterized in that: The threshold is that the area contains 40-80 pixels. When the number of pixels of the fused white dots meets the threshold, the crystal pulling process is started. If the number of pixels of the fused white dot does not meet the threshold, the process of chip introduction will not proceed.

Images