Silicon ingot appearance defect detection system

Abstract

The present invention discloses a silicon ingot appearance defect detection system, and relates to the technical field of defect detection, the silicon ingot appearance defect detection system comprises a laser source module, a camera module, an image processing module, an AI detection module and a control and feedback module, the laser source module uses a 200W 1300nm laser for illumination and performs oblique shooting from the side surface, the camera module adopts two rows of Sensor 1K black and white line scanning indium gallium arsenic cameras, and the image processing module is used for image processing. The image processing module is used for acquiring two images and setting two sets of exposure parameters to realize high dynamic range imaging, the image processing module is used for processing the two acquired images by using a specific algorithm, including row alternating extraction, splicing and brightness uniformity comparison, and the AI detection module is used for analyzing the selected images through a trained AI model, judging whether hidden crack defects exist or not, and if the hidden crack defects exist, judging whether the hidden crack defects exist or not. The control and feedback module controls cooperative work of all the modules, feeds back a detection result to the user interface and provides a system state and the detection result, and the system has the advantage of being high in automation degree.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a silicon ingot appearance defect detection system. Background Art

[0002] With the rapid development of the solar photovoltaic industry and the semiconductor industry, the market demand for silicon ingots, as their main raw material, is also growing. At present, domestic silicon ingot manufacturers mainly use automatic assembly lines to process silicon ingots. The cylindrical raw rods are cut horizontally by a cutter, cut into square silicon blocks of equal size by a square machine, and then ground and polished by a polisher to obtain finished silicon ingots.

[0003] Due to the physical characteristics of silicon rods and the limitations of processing machines, silicon ingots often have end cracks during processing, which seriously affects the safety performance and production efficiency of the production line. Most manufacturers use manual methods to directly remove silicon ingots with end cracks from the production line. This method increases production costs, increases dependence on manual labor, reduces safety performance, and seriously affects the production efficiency of automatic production lines. At present, how to control the quality of incoming solar cell materials from the source has become an urgent need for major silicon ingot processing companies. Summary of the invention

[0004] The object of the present invention is to provide a silicon ingot appearance defect detection system to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a silicon ingot appearance defect detection system, including a laser source module, a camera module, an image processing module, an AI detection module, and a control and feedback module. The laser source module uses a 200W 1300nm laser as illumination and shoots obliquely from the side. The camera module uses a 1K black-and-white line-scan indium gallium arsenide camera with two rows of sensors, and sets two sets of exposure parameters to achieve high dynamic range imaging. The image processing module uses a specific algorithm to process the two acquired images, including alternating row extraction, splicing, and brightness uniformity comparison. The AI ​​detection module is used to analyze the selected image through a trained AI model to determine whether there is a hidden crack defect. The control and feedback module controls the coordinated work of each module, feeds back the detection results to the user interface, and provides system status and detection results.

[0006] According to the above technical solution, the working method of the system is:

[0007] S1. After the laser source module is started, a 200W 1300nm laser is irradiated to the N-type silicon ingot, and the image of sequential laser illumination is obtained by side oblique shooting;

[0008] S2. A black dust cover is designed around the camera module. The camera module takes pictures alternately line by line through the polling function according to the laser lighting conditions, and uses two sets of exposure parameters to cope with different light transmittances to form a light and dark image.

[0009] S3, in the image processing module, extracting and splicing the images of the odd-numbered rows and the even-numbered rows respectively to obtain two complete images;

[0010] S4, comparing the brightness and wear marks of the two images through an algorithm, and determining which exposure image to use as the final target to be detected;

[0011] S5. The selected image is handed over to the AI ​​detection module, and the AI ​​formula is set. The AI ​​model is trained by using a large number of pictures with and without hidden cracks to determine whether there are hidden cracks inside. The trained AI model then feeds the results back to the control and feedback module to update the system status and user interface.

[0012] According to the above technical solution, in S3, the captured image is rotated, and through ROI processing, the part outside the detection area is cropped to brighten and magnify the defects, which provides important support for AI detection and makes the image and the real object in the same direction, which is more conducive to the operator to find the defect location. During the defect screening process, the operator can modify the algorithm through software to detect parameters such as defect area and quantity to customize the tightness of the card control to achieve flexible detection needs.

[0013] According to the above technical solution, in S5, the model parameter setting of the AI ​​detection module includes setting the target category to be detected, the network parameters of the model, the training parameters, and the augmentation parameters, selecting representative samples, and selecting pictures with various lighting, angles, and representative backgrounds. When a picture is marked, it is automatically added to the training sample set, all the targets in the picture are marked, and the ambiguous areas are covered with LMask. When you click to stop training, the tool will automatically trigger the test operation. If a new picture is imported and the recognition result of the preceding dependent node changes, after the update operation is completed, click the test button to view the recognition effect of the model, repeat the marking, training, and testing process until the recognition requirements are met, and if a hidden crack defect is found, a signal is given to the buzzer to alarm, and the picture is classified into the NG folder. If broken edges, wear marks, bright lines, and glue marks are found, they are only detected and recorded without triggering an alarm.

[0014] According to the above technical solution, the step S5 further includes determining whether the hidden crack has reached a limit size, and the specific method is as follows:

[0015] S5-1. The scanned standard image with hidden cracks is enlarged along the hidden crack area until the enlarged image with hidden cracks reaches a degree that cannot be used normally, which is recorded as 100% deformation, i.e., the worst quality. The model of the deformation degree of the silicon ingot area is the limit state, while the standard image with hidden cracks is 0% deformation, i.e., the quality for recycling;

[0016] S5-2. By collecting two different sizes of hidden crack areas of silicon ingot images and the deformation degree of the image model scanned after the silicon ingot images are scanned, the deformation degrees of the two image models are between 0-100%, and the functional relationship between the size of the hidden crack area of ​​the silicon ingot image and the image model scanned after scanning is obtained, and the formula is:

[0017]

[0018] Among them, D x is the area of ​​the hidden crack region of the recyclable silicon ingot image, D y is the area of ​​the hidden crack region of the silicon ingot image during scanning, A% is the degree of deformation of the image model of the recyclable silicon ingot image, B% is the degree of deformation of the image model of the silicon ingot image during scanning, and finally k is obtained;

[0019] S5-3. After the functional relationship is obtained, the size D0 of the hidden crack area corresponding to 100% deformation, that is, when the model of the degree of deformation of the hidden crack area is in the limit state, is obtained. This value is the maximum value allowed for the area of ​​the hidden crack area.

[0020] According to the above technical solution, in S5, the specific method for comparing the wear marks of the two images is: under the premise of ensuring that the area of ​​the hidden crack region is greater than D0, the number of wear marks scanned after scanning the silicon ingot image is counted multiple times, and the quality of the silicon ingot is judged according to the number of wear marks. The fewer the number of wear marks, the better the quality of the silicon ingot. Since the number of wear marks and the area of ​​the hidden crack region are not in a linear positive correlation, the quality of the silicon ingot corresponding to the area of ​​the hidden crack region is best when the number of wear marks decreases fastest with the change in the area of ​​the hidden crack region.

[0021] According to the above technical solution, in S3, after the camera module completes shooting, a series of image data with alternating light and dark will be obtained, each image is composed of a number of rows, and the number of rows is the number of image rows captured by the camera at one time during the shooting process. The identification assumption is: the image in the scene is processed to form a two-dimensional array, and for each row numbered odd, it is extracted and saved as a new image array, and for each row numbered even, it is extracted and saved as a new image array. After separating the odd and even row images, check the dimension of each new image array to ensure that their number of columns remains consistent. After successfully separating the odd and even row images, perform a splicing operation.

[0022] According to the above technical solution, the process of the AI ​​detection module is: image acquisition, model training, effect verification, and project delivery. During the effect verification, if the requirements are met, the project will be delivered directly. If the requirements are not met, the model training needs to be re-entered.

[0023] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention solves the problems of difficulty in detecting end face hidden crack defects and low efficiency of manual detection, improves the production efficiency of the automatic production line, improves the quality of finished silicon ingots, realizes defect classification, and can determine whether the hidden cracks can meet the standards for recycling, thereby helping to recycle some defective products. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0025] Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1 The present invention provides a technical solution: a silicon ingot appearance defect detection system, including a laser source module, a camera module, an image processing module, an AI detection module, and a control and feedback module. The laser source module uses a 200W 1300nm laser as illumination and shoots obliquely from the side. The camera module uses a 1K black-and-white line-scanning indium gallium arsenide camera with two rows of sensors, and sets two sets of exposure parameters to achieve high dynamic range imaging. The image processing module uses a specific algorithm to process the two acquired images, including alternate row extraction, splicing, and brightness uniformity comparison. The AI ​​detection module is used to analyze the selected image through the trained AI model to determine whether there is a hidden crack defect. The control and feedback module controls the coordinated work of each module, feeds back the detection results to the user interface, and provides system status and detection results.

[0028] The system works as follows:

[0029] S1. After the laser source module is started, a 200W 1300nm laser is irradiated to the N-type silicon ingot, and the image of sequential laser illumination is obtained by side oblique shooting;

[0030] S2. A black dust cover is designed around the camera module. The camera module takes pictures alternately line by line through the polling function according to the laser lighting conditions, and uses two sets of exposure parameters to cope with different light transmittances to form a light and dark image.

[0031] S3, in the image processing module, extracting and splicing the images of the odd-numbered rows and the even-numbered rows respectively to obtain two complete images;

[0032] S4, comparing the brightness and wear marks of the two images through an algorithm, and determining which exposure image to use as the final target to be detected;

[0033] S5. The selected image is sent to the AI ​​detection module, and the AI ​​formula is set. The AI ​​model is trained by distinguishing a large number of images with and without hidden cracks. The trained AI model determines whether there are hidden cracks inside, and feeds the results back to the control and feedback module to update the system status and user interface;

[0034] In S3, the captured image is rotated, and the part outside the detection area is cropped through ROI processing to brighten and magnify the defects, which provides important support for AI detection. The image and the real object are in the same direction, which is more conducive to the operator to find the defect location. During the defect screening process, the operator can modify the algorithm through the software to detect the defect area, quantity and other parameters to customize the card control tightness to meet flexible detection needs.

[0035] In S5, the model parameter settings of the AI ​​detection module include setting the target category to be detected, the network parameters of the model, the training parameters, and the augmentation parameters, selecting representative samples, and selecting pictures with various lighting, angles, and backgrounds. When a picture is marked, it is automatically added to the training sample set, and all targets in the picture are marked. The ambiguous area is covered with LMask. When you click to stop training, the tool will automatically trigger the test operation. If there is a new picture imported, and the recognition result of the previous dependent node changes, after the update operation is completed, click the test button to view the recognition effect of the model, repeat the process of marking, training, and testing until the recognition requirements are met. If a hidden crack defect is found, a signal is given to the buzzer alarm, and the picture is classified into the NG folder. If broken edges, wear marks, bright lines, and glue marks are found, they are only detected and recorded without triggering an alarm.

[0036] S5 also includes determining whether the hidden crack has reached a limit size, and the specific method is as follows:

[0037] S5-1. The scanned standard image with hidden cracks is enlarged along the hidden crack area until the enlarged image with hidden cracks reaches a degree that cannot be used normally, which is recorded as 100% deformation, i.e., the worst quality. The model of the deformation degree of the silicon ingot area is the limit state, while the standard image with hidden cracks is 0% deformation, i.e., the quality for recycling;

[0038] S5-2. By collecting two different sizes of hidden crack areas of silicon ingot images and the deformation degree of the image model scanned after the silicon ingot images are scanned, the deformation degrees of the two image models are between 0-100%, and the functional relationship between the size of the hidden crack area of ​​the silicon ingot image and the image model scanned after scanning is obtained, and the formula is:

[0039]

[0040] Among them, D x is the area of ​​the hidden crack region of the recyclable silicon ingot image, D y is the area of ​​the hidden crack region of the silicon ingot image during scanning, A% is the degree of deformation of the image model of the recyclable silicon ingot image, B% is the degree of deformation of the image model of the silicon ingot image during scanning, and finally k is obtained;

[0041] S5-3. After the functional relationship is obtained, the area size D0 of the hidden crack region corresponding to 100% deformation, that is, when the model of the degree of deformation of the hidden crack region is in the limit state, is obtained. This value is the maximum value allowed for the area of ​​the hidden crack region;

[0042] In S5, the specific method for comparing the wear marks of the two images is: under the premise of ensuring that the area of ​​the hidden crack region is greater than D0, the number of wear marks scanned after scanning the silicon ingot image is counted multiple times, and the quality of the silicon ingot is judged according to the number of wear marks. The fewer the number of wear marks, the better the quality of the silicon ingot. Since the number of wear marks and the area of ​​the hidden crack region are not linearly positively correlated, the silicon ingot quality corresponding to the area of ​​the hidden crack region is best when the number of wear marks decreases fastest with the change of the area of ​​the hidden crack region;

[0043] In S3, after the camera module completes shooting, a series of image data with light and dark alternating will be obtained. Each image consists of a number of rows, and the number of rows is the number of image rows captured by the camera at one time during the shooting process. The identification assumption is: the image in the scene is processed to form a two-dimensional array. For each row numbered odd, it is extracted and saved as a new image array, and for each row numbered even, it is extracted and saved as a new image array. After separating the odd-numbered and even-numbered row images, the dimension of each new image array is checked to ensure that their column numbers are still consistent. After successfully separating the odd-numbered and even-numbered row images, the splicing operation is performed;

[0044] The process of the AI ​​detection module is: image acquisition, model training, effect verification, and project delivery. During effect verification, if the requirements are met, the project will be delivered directly. If the requirements are not met, the model training needs to be re-entered.

[0045] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Silicon ingot appearance defect detection system, characterized by: It includes a laser source module, a camera module, an image processing module, an AI detection module, and a control and feedback module. The laser source module uses a 200W 1300nm laser as lighting and shoots obliquely from the side. The camera module uses a 1K black-and-white line-scan InGaAs camera with two rows of sensors and sets two sets of exposure parameters to achieve high dynamic range imaging. The image processing module uses a specific algorithm to process the two acquired images, including alternate row extraction, splicing, and brightness uniformity comparison. The AI ​​detection module is used to analyze the selected image through a trained AI model to determine whether there is a hidden crack defect. The control and feedback module controls the collaborative work of each module, feeds back the detection results to the user interface, and provides system status and detection results.

2. The silicon ingot appearance defect detection system according to claim 1, characterized in that: The system works as follows: S1. After the laser source module is started, a 200W 1300nm laser is irradiated to the N-type silicon ingot, and the image of sequential laser illumination is obtained by side oblique shooting; S2. A black dust cover is designed around the camera module. The camera module takes pictures alternately line by line through the polling function according to the laser lighting conditions, and uses two sets of exposure parameters to cope with different light transmittances to form a light and dark image. S3, in the image processing module, extracting and splicing the images of the odd-numbered rows and the even-numbered rows respectively to obtain two complete images; S4, comparing the brightness and wear marks of the two images through an algorithm, and determining which exposure image to use as the final target to be detected; S5. The selected image is handed over to the AI ​​detection module, and the AI ​​formula is set. The AI ​​model is trained by using a large number of pictures with and without hidden cracks to determine whether there are hidden cracks inside. The trained AI model then feeds the results back to the control and feedback module to update the system status and user interface.

3. The silicon ingot appearance defect detection system according to claim 2, characterized in that: In S3, the captured image is rotated, and the portion outside the detection area is cropped through ROI processing, the defects are brightened and magnified, and the image is made to have the same direction as the real object.

4. The silicon ingot appearance defect detection system according to claim 3, characterized in that: In S5, the model parameter setting of the AI ​​detection module includes setting the target category to be detected, the network parameters of the model, the training parameters, and the augmentation parameters, selecting representative samples, and selecting pictures with various lighting, angles, and representative backgrounds. When a picture is marked, it is automatically added to the training sample set, all the targets in the picture are marked, and the ambiguous areas are covered with LMask. When you click to stop training, the tool will automatically trigger the test operation. If a new picture is imported and the recognition result of the previous dependent node changes, after the update operation is completed, click the test button to view the recognition effect of the model, repeat the marking, training, and testing process until the recognition requirements are met, and if a hidden crack defect is found, a signal is given to the buzzer alarm, and the picture is classified into the NG folder. If broken edges, wear marks, bright lines, and glue marks are found, they are only detected and recorded without triggering an alarm.

5. The silicon ingot appearance defect detection system according to claim 4, characterized in that: In S5, it is also included to determine whether the hidden crack has reached the limit size, and the specific method is as follows: S5-1. The scanned standard image with hidden cracks is enlarged along the hidden crack area until the enlarged image with hidden cracks reaches a degree that cannot be used normally, which is recorded as 100% deformation, i.e., the worst quality. The model of the deformation degree of the silicon ingot area is the limit state, while the standard image with hidden cracks is 0% deformation, i.e., the quality for recycling; S5-2. By collecting two different sizes of hidden crack areas of silicon ingot images and the deformation degree of the image model scanned after the silicon ingot images are scanned, the deformation degrees of the two image models are between 0-100%, and the functional relationship between the size of the hidden crack area of ​​the silicon ingot image and the image model scanned after scanning is obtained, and the formula is: Among them, D x is the area of ​​the hidden crack region of the recyclable silicon ingot image, D y is the area of ​​the hidden crack region of the silicon ingot image during scanning, A% is the degree of deformation of the image model of the recyclable silicon ingot image, B% is the degree of deformation of the image model of the silicon ingot image during scanning, and finally k is obtained; S5-3. After the functional relationship is obtained, the size D0 of the hidden crack area corresponding to 100% deformation, that is, when the model of the degree of deformation of the hidden crack area is in the limit state, is obtained. This value is the maximum value allowed for the area of ​​the hidden crack area.

6. The silicon ingot appearance defect detection system according to claim 5, characterized in that: In S5, the specific method for comparing the wear marks of the two images is: under the premise of ensuring that the area of ​​the hidden crack region is greater than D0, the number of wear marks scanned after scanning the silicon ingot image is counted multiple times, and the quality of the silicon ingot is judged according to the number of wear marks. The fewer the number of wear marks, the better the quality of the silicon ingot. Since the number of wear marks and the area of ​​the hidden crack region are not in a linear positive correlation, the quality of the silicon ingot corresponding to the area of ​​the hidden crack region is best when the number of wear marks decreases fastest with the change in the area of ​​the hidden crack region.

7. The silicon ingot appearance defect detection system according to claim 6, characterized in that: In S3, after the camera module completes shooting, a series of image data with alternating light and dark will be obtained, each image is composed of a number of rows, and the number of rows is the number of image rows captured by the camera at one time during the shooting process. The identification assumption is: the image in the scene is processed to form a two-dimensional array, for each row numbered odd, it is extracted and saved as a new image array, for each row numbered even, it is extracted and saved as a new image array, after separating the odd and even row images, check the dimension of each new image array to ensure that their number of columns remains consistent, and after successfully separating the odd and even row images, perform a splicing operation.

8. The silicon ingot appearance defect detection system according to claim 7, characterized in that: The process of the AI ​​detection module is: image acquisition, model training, effect verification, and project delivery. During the effect verification, if the requirements are met, the project will be delivered directly. If the requirements are not met, the model training needs to be re-entered.

Images