Abnormality detection method and abnormality detection system for single crystal production
By using image sensors and deep learning models for automatic anomaly detection during single crystal manufacturing, the problem of operators having difficulty detecting crystal growth anomalies in existing technologies has been solved, thereby improving the quality and yield of single crystals.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-10
- Publication Date
- 2026-03-13
AI Technical Summary
In the process of single crystal cultivation, existing technologies are unable to automatically detect abnormalities in crystal growth, leading to operator oversights and affecting crystal quality and yield.
Image sensors and deep learning models are used to analyze image data in the single crystal manufacturing process in real time and automatically detect anomalies, including anomalies in the crystal pulling operation, cone growth and straight section formation processes.
It enables automated anomaly detection in the single crystal manufacturing process, reduces operator oversights, improves crystal quality and yield, and reduces the burden on operators.
Smart Images
Figure CN121660280A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an anomaly detection method and system in single crystal manufacturing. Background Technology
[0002] The Czochralski method, also known as the single crystal pulling method, involves heating and melting the raw material for single crystals in a crucible until it becomes a molten liquid. The seed crystal is then brought into contact with the molten liquid and pulled while being rotated, thereby cultivating a single crystal. This method has the advantage that the single crystal does not come into contact with the crucible, thus avoiding mechanical strain caused by the crucible, and therefore it is easy to produce high-quality, large-diameter crystals. Therefore, this method can be used to manufacture silicon (Si), germanium (Ge), gallium arsenide (GaAs), and yttrium aluminum garnet (Y3Al5O4). 12 YAG), sapphire (Al2O3), lithium niobate (LiNbO3; LN), lithium tantalate (LiTaO3; LT), bismuth germanate (Bi4Ge3O) 12 Single crystals such as BGO.
[0003] However, in the cultivation of single crystals using the Czochralski method, a seed crystal is first introduced into contact with the molten raw material, followed by crystal growth. First, a neck with the same thickness as the seed crystal is formed. Next, a tapered section with a gradually increasing diameter is formed, and then the cylindrical section is formed to the desired distance while maintaining the large diameter. Finally, a tail section with a decreasing diameter in the opposite direction to the tapered section is formed.
[0004] As an apparatus for automatically cultivating single crystals using the Czochralski method, the single crystal cultivation apparatus of Patent Document 1 is known as prior art. In the single crystal cultivation apparatus described in Patent Document 1, information related to the shape of the exposed portion of the single crystal is derived from image information output by an image sensor in an image processing device. Furthermore, based on this shape information, the single crystal cultivation apparatus of Patent Document 1 calculates the length of the portion of the single crystal in the molten liquid, or the distance between the lower end of the portion of the single crystal in the molten liquid and the bottom of the crucible, and outputs sound when the distance between the lower end of the crystal and the bottom of the crucible is below a certain value.
[0005] The device calculates the radius of the molten liquid surface or the distance from the molten liquid surface to the lower end of the seed crystal. Based on the output of a weight sensor or image processing device, it performs various calculations and controls the heater, rotation and lifting device, display device, sound output device, etc., according to the calculated distance or the calculation result.
[0006] Existing technical documents Patent documents Patent Document 1: Japanese Patent No. 6953912 Summary of the Invention
[0007] The problem that the invention aims to solve In the process of single-crystal growth using the Czochralski method, as long as the crystal growth proceeds smoothly, a qualified product can be produced without any problems. However, under certain circumstances, abnormalities may occur during crystal growth. Due to these abnormalities, sometimes crystal growth itself cannot be carried out directly, and sometimes although crystal growth can proceed, cracking or other issues may occur during cooling and extraction, resulting in a failure to obtain a qualified product.
[0008] Such anomalies in crystal growth can be perceived numerically, but cannot be detected from measured values. Therefore, in cases where it is desired to determine anomalies in crystal growth based on appearance or images, methods such as operators visually confirming the presence of anomalies or matching identification templates of devices to a pre-defined range are used to confirm the presence of anomalies.
[0009] For example, after the seed crystal contacting the molten raw material, the temperature of the molten raw material sometimes deviates from the appropriate temperature (the appropriate temperature being slightly higher than the melting point of the molten raw material). If the temperature of the molten raw material is too high compared to the appropriate temperature, the temperature of the seed crystal becomes above its melting point, the seed crystal melts, and the contact between the molten raw material and the seed crystal breaks. Conversely, if the temperature of the molten raw material is too low compared to the appropriate temperature, overcooling occurs around the area where the seed crystal is in contact, causing the solidified portion to expand.
[0010] The former cannot continue crystal growth because the contact between the molten raw material and the seed crystal breaks. Even if the latter can continue crystal growth, the solidified part produced by overcooling has poor crystallinity, so the crystallinity of the subsequently formed crystals also becomes poor, and the probability of cracking after cooling increases.
[0011] As another example, during the formation of the cone after the neck, growth ridges (here referring to the protrusions formed along the direction of the small crystal faces) appear in a prescribed direction of the crystal structure. These growth ridges can extend regularly in one direction, but if the growth ridges branch due to foreign matter or secondary nucleation, the subsequent crystal will become polycrystalline and crack upon cooling.
[0012] Previously, operators would visually identify such anomalies and, if found, repeat the crystal growth process. However, visually identifying anomalies is not easy, and even skilled operators can sometimes miss them.
[0013] Furthermore, in cited reference 1, when using a single crystal growth apparatus to grow single crystals, no method is described for determining whether there are any abnormalities in the crystal growth of the single crystal. Therefore, regarding the single crystal growth apparatus described in patent reference 1, even if an abnormality occurs during the crystal growth of the single crystal, the cultivation of the single crystal continues directly.
[0014] To address the aforementioned problems, the present invention aims to provide an anomaly detection method and system for single-crystal manufacturing, which eliminates defects caused by operator oversights and performs automatic judgment, thus eliminating the need for operator confirmation and saving labor.
[0015] Solution for solving the problem The inventors conducted in-depth research and discovered that in the single-crystal manufacturing process, after the crystal pulling operation, the neck, cone, cylindrical portion, and crystal growth are completed, followed by separation, an image sensor monitors the internal structure, and a computing device processes the obtained image data. Based on the results, any abnormalities are automatically detected, thereby solving the aforementioned problems and completing this invention. The invention is as follows. [1] An anomaly detection method for single crystal manufacturing includes a heating and melting process in which a single crystal raw material is heated and melted in a crucible by a heating source to produce a raw material molten liquid, and a pulling process in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured by a rotation pulling method. The method is characterized in that, in the pulling process, a computing device equipped with a pre-learned model is used to perform computational processing on image data of the part of the seed crystal in contact with the raw material molten liquid or image data of the single crystal obtained by an image sensor, and anomalies in the manufacturing of the single crystal are detected based on the result of the computational processing. [2] The abnormal detection method for single crystal manufacturing as described in [1] above is characterized in that the pulling process includes a first process of bringing the seed crystal into contact with the raw material melt and pulling the neck through the rotation pulling method. [3] The abnormal detection method for single crystal manufacturing as described in [2] above is characterized in that the pulling process includes a second process of pulling a cone portion with an enlarged diameter. [4] The abnormal detection method for single crystal manufacturing as described in [3] above is characterized in that the pulling process includes a third process of pulling the straight part into the maximum diameter. [5] The abnormal detection method for single crystal manufacturing as described in [4] above is characterized in that the pulling process includes a fourth process of separating the pulled single crystal and the raw material melt. [6] The anomaly detection method for single crystal manufacturing as described in any one of [1] to [5] above, characterized in that the pre-learned model is formed through deep learning. [7] Anomaly detection method for single crystal manufacturing as described in any one of [1] to [6] above, wherein the single crystal is an oxide single crystal. [8] The anomaly detection method for single crystal manufacturing as described in [5] above is characterized in that all processes, including the first to fourth processes, are automatically executed. [9] An anomaly detection system for single crystal manufacturing includes a heating and melting process in which a single crystal raw material is heated and melted in a crucible by a heating source to produce a raw material molten liquid, and a pulling process in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured by a rotation pulling method. The system includes a computing device equipped with a pre-learned model. In the pulling process, the computing device performs computational processing on image data of the part of the seed crystal in contact with the raw material molten liquid or image data of the single crystal obtained by an image sensor, and detects anomalies in the manufacturing of the single crystal based on the result of the computational processing.
[10] The single crystal manufacturing anomaly detection system as described in [9] above is characterized in that the pulling process includes a first process of bringing the seed crystal into contact with the raw material melt and pulling the neck through the rotation pulling method.
[11] The anomaly detection system for single crystal manufacturing as described in
[10] above is characterized in that the pulling process includes a second process of pulling a cone portion with an enlarged diameter.
[12] The anomaly detection system for single crystal manufacturing as described in
[11] above is characterized in that the pulling process includes a third process of pulling the straight section into the maximum diameter.
[13] The anomaly detection system for single crystal manufacturing as described in
[12] above is characterized in that the pulling process includes a fourth process of separating the pulled single crystal and the raw material melt.
[14] The anomaly detection system for single crystal manufacturing as described in any one of [9] to
[13] above, characterized in that the pre-learned model is formed through deep learning.
[15] An anomaly detection system made of single crystal as described in any one of [9] to
[14] above, wherein the single crystal is an oxide single crystal.
[16] The anomaly detection system for single crystal manufacturing as described in
[13] above is characterized in that it automatically executes all processes, including the first to fourth processes.
[0016] According to the present invention, an anomaly detection method and an anomaly detection system for single crystal manufacturing can be provided, which can eliminate non-conformities caused by operator oversights and perform automatic judgment, so that the operator does not need to confirm, thus saving labor. Attached Figure Description
[0017] Figure 1 This is a schematic diagram showing an example of a single crystal manufacturing apparatus. Figure 2 It is assembled in Figure 1 A schematic diagram of an example of an anomaly detection system for single crystal manufacturing in a single crystal manufacturing apparatus. Figure 3This is a schematic diagram of a single crystal produced by the spin-pulling method. Figure 4 (a) is a schematic diagram of the contact between the seed crystal and the molten raw material. Figure 4 (b) is a schematic diagram showing the solidification that occurs when the seed crystal comes into contact with the molten raw material. Figure 4 (c) is a schematic diagram of the seed crystal leaving the raw material molten liquid. Figure 5 (a) is a schematic diagram of the cone-shaped part of a single crystal viewed from the side. Figure 5 (b) is a schematic diagram of the cone-shaped part of a single crystal as viewed from above. Figure 5 (c) is a schematic diagram of the cone of a single crystal after the growth of the ridge branch, viewed from above. Detailed Implementation
[0018] The fabrication of a single crystal includes a heating and melting process, in which the raw material for the single crystal is placed in a crucible and melted by a heat source to create a raw material molten liquid, and a pulling process, in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is fabricated using a spin-pulling method. (See reference...) Figure 1 Detailed explanation of single crystal manufacturing. Figure 1 This is a schematic diagram illustrating an example of a single crystal manufacturing apparatus. In the single crystal manufacturing apparatus 1, a single crystal raw material is added to a crucible 4 and heated using a heating source 5. The heating source 5 can be a resistance heater or similar source, provided the crucible 4 is conductive; it can also be a high-frequency heating source. By heating, the temperature of the single crystal raw material is increased, causing it to melt and forming a raw material melt 3. The temperature of the raw material melt 3 can be measured indirectly, either by a thermocouple 6 placed at the bottom of the crucible, or directly from above using a radiation thermometer (though not shown in the diagram). Furthermore, this temperature data can be input to a control unit 13 for PID control. The control unit 13 can be a PLC (Programmable Logic Controller) or a personal computer, etc.
[0019] Next, a Czochralski process is performed to bring the seed crystal into contact with the molten raw material 3 and to produce a single crystal using a spin-pulling method. This Czochralski process may include a first step of bringing the seed crystal into contact with the molten raw material 3 and pulling the neck portion using a spin-pulling method. In addition to the first step, this Czochralski process may also include a second step of pulling a tapered portion with an enlarged diameter. Furthermore, in addition to the first and second steps, this Czochralski process may also include a third step of pulling a straight cylindrical portion to its maximum diameter. Furthermore, in addition to the first, second, and third steps, this Czochralski process may also include a fourth step of separating the pulled single crystal from the molten raw material. Through these steps, the single crystal is pulled. An example of a single crystal produced by the spin-pulling method... Figure 3 As shown. The control of single-crystal pulling can also be performed by the control unit 13. Control can be achieved using a PLC or a personal computer, etc. As input data, the crystal weight measured by the weighing sensor 11 can be used, or the diameter of the single crystal calculated from an image obtained from the image sensor 12 can be used. The output is adjusted by the heating source 5, and PID control or similar methods can be used on the input data.
[0020] [Anomaly Detection Methods in Single Crystal Manufacturing] The following describes an anomaly detection method for single-crystal manufacturing according to one embodiment of the present invention. According to an embodiment of the present invention, in any of the first to fourth steps, the image data obtained by the image sensor 12 is used by the computing device in the control unit 13 to determine whether there is an abnormality in the single crystal manufacturing process based on the image data, and can be automatically detected if there is an abnormality in the manufacturing process of the single crystal. Therefore, even inexperienced operators can easily detect anomalies in single-crystal manufacturing, thus preventing operator oversights. Furthermore, automated detection reduces the operator's workload.
[0021] In the first process, the image sensor 12 monitors the contact area between the raw material melt 3 and the seed crystal and its surroundings. The computing device in the control unit 13 uses a pre-learned model to determine the presence or absence of abnormalities in the single crystal manufacturing process based on the image data. If an abnormality exists in the single crystal manufacturing process, it can be automatically detected. If the temperature of the raw material molten liquid 3 is not appropriate after the seed crystal is brought into contact with the raw material molten liquid 3, an abnormality will occur in the single crystal manufacturing process. Images of raw material melt 3 prepared in advance at appropriate temperatures (refer to) Figure 4 (a) An image of a solidified portion formed due to the low temperature of the raw material melt (refer to...) Figure 4 (b) and an image showing the breakage of the contact between the molten raw material and the seed crystal due to the high temperature of the molten raw material (see reference). Figure 4(c) Deep learning-based learning is implemented. In this case, the goal of learning-based image classification is to extract regularities, i.e., image features, for classifying "normal" and "abnormal" from images of the raw material molten liquid 3 at an appropriate temperature, images of solidified portions formed due to a lower temperature of the raw material molten liquid, and images of the raw material molten liquid and seed crystals breaking contact due to a higher temperature of the raw material molten liquid. This allows for the correct classification of all images where the seed crystal contacts the raw material molten liquid 3 as "normal" or "abnormal." Furthermore, the observation data in the learning process consists of pairs of input data and their corresponding labels. For example, if the input data is an image of the raw material molten liquid 3 at an appropriate temperature, the label becomes "normal." If the input data is an image of solidified portions formed due to a lower temperature of the raw material molten liquid, the label becomes "abnormal." If the input data is an image of the raw material molten liquid and seed crystals breaking contact due to a higher temperature, the label becomes "abnormal." Then, the model is optimized as follows: when an image of the raw material molten liquid 3 at an appropriate temperature is input into the model (e.g., a multilayer neural network), the output is "normal"; when an image of a solidified portion due to a low temperature of the raw material molten liquid is input into the model, the output is "abnormal"; and when an image of a disconnection between the raw material molten liquid and the seed crystal due to a high temperature of the raw material molten liquid is input into the model, the output is "abnormal". This optimization of the model is called "learning". Based on the learning results, the presence or absence of anomalies in single crystal manufacturing can be determined with high probability. Additionally, an alarm sound, such as an alarm tone, can be activated upon detecting an anomaly in single crystal manufacturing.
[0022] In the second process, image sensor 12 is used to monitor the growth ridges formed on the single crystal, particularly the cone portion (see reference). Figure 5 (a) and (b) utilize a computing device equipped with a pre-learned model to process image data obtained from image sensor 12, and automatically detect abnormalities in growth edges based on the processing results, thereby detecting abnormalities in single-crystal manufacturing. If foreign matter or secondary nucleation occurs during cone growth, abnormalities in growth edge branching occur (see reference). Figure 5 (c)). Prepare images of the growth ridges under normal conditions in advance (refer to...). Figure 5 (b) and images of abnormal growth ridges (refer to) Figure 5(c) Implement learning based on deep learning. Specifically, the model is optimized in such a way that when an image with normal growth edges is input into the model, the output is "normal," and when an image with normal growth edges is input into the model, the output is "abnormal." Based on this learning result, it is possible to determine with high probability whether there is an anomaly in growth edges, and as a result, it is possible to detect anomalies in single-crystal manufacturing with high probability.
[0023] Preferably, a pre-learned model is formed through deep learning, which can determine with high probability whether there are abnormalities in the growth edges. In particular, the shapes of these growth edges are not identical; all single crystals have different shapes. Therefore, branching anomalies also have completely different shapes. If a deterministic method such as template matching is used for judgment, it is difficult to obtain a high accuracy rate. However, if a method like deep learning is used for judgment, a high accuracy rate can be obtained.
[0024] As a model for deep learning, Convolutional Neural Networks (CNNs) are considered to be the preferred choice due to their advantages in image processing, but are not limited to this.
[0025] The single crystal can be an oxide single crystal. Oxide single crystals tend to have a higher melting point compared to other substances. Therefore, the light from the furnace during crystal growth is too strong, and while light reduction is achieved using a light shield or the like, it is not easily confirmed by the operator's visual inspection. Therefore, by using an image obtained by the image sensor 12, edges and the like can be shown in more detail than by visual inspection.
[0026] The feature is that, based on an image of the contact area between the molten raw material 3 and the seed crystal, and its surrounding area, a computing device calculates the width of the solidified portion generated around the contact area, and automatically detects anomalies based on the width of the solidified portion. Since the temperature of the solidified portion is lower than that of the molten liquid, it appears slightly darker in the image. The length used as a reference can be determined within the image data, or it can be determined by the number of pixels, etc., and the width of the solidified portion is calculated based on the result. For example, programming software such as Python can also be used to automatically calculate the length based on the image data. The width of the solidified portion is preferably 4 mm or less, more preferably 2 mm or less.
[0027] Its key feature is that it automatically executes all processes, including the first through fourth steps. Since all processes are automated, labor savings are expected. Furthermore, operator-induced deviations are reduced, leading to stable utilization and yield rates.
[0028] The anomaly detection method for single-crystal manufacturing according to one embodiment of the present invention is an example of the anomaly detection method for single-crystal manufacturing of the present invention, and does not limit the anomaly detection method for single-crystal manufacturing of the present invention.
[0029] [Anomaly Detection System in Single Crystal Manufacturing] The following describes an anomaly detection system for single-crystal manufacturing according to one embodiment of the present invention. The anomaly detection system for single crystal manufacturing of the present invention includes a heating and melting process in which a single crystal raw material is heated and melted in a crucible using a heating source to create a raw material molten liquid, and a pulling process in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured by a spin-pulling method. Furthermore, one embodiment of the single crystal manufacturing anomaly detection system of the present invention includes a computing device equipped with a pre-learned model. During the pulling process, the computing device performs computational processing on image data of the portion of the seed crystal in contact with the raw material molten liquid, or on image data of the single crystal itself, obtained from an image sensor. Moreover, the single crystal manufacturing anomaly detection system of one embodiment of the present invention detects anomalies in single crystal manufacturing based on the results of its computational processing. Therefore, even inexperienced operators can easily detect anomalies in single-crystal manufacturing, thus preventing operator oversights. Furthermore, automated detection reduces the operator's workload.
[0030] Furthermore, regarding the manufacturing of single crystals, an anomaly detection method for single crystal manufacturing according to one embodiment of the present invention has been described, so the description of single crystal manufacturing is omitted. Also, regarding the computing device in the anomaly detection system for single crystal manufacturing according to one embodiment of the present invention, an anomaly detection method for single crystal manufacturing according to one embodiment of the present invention has been described, so the description of the computing device is omitted.
[0031] The anomaly detection system for single-crystal manufacturing according to one embodiment of the present invention is an example of the anomaly detection system for single-crystal manufacturing of the present invention, and does not limit the anomaly detection system for single-crystal manufacturing of the present invention.
[0032] [Example] The following examples and comparative examples are provided to further illustrate the present invention, but the present invention is not limited to these examples.
[0033] [Example 1] The raw materials for lithium tantalate single crystal were put into a crucible, heated and melted using a high-frequency induction heating furnace, and a 6-inch diameter lithium tantalate single crystal was manufactured by the spin-pulling method. Figure 1 A schematic diagram of the manufacturing apparatus used in the fabrication of lithium tantalate single crystals is shown. In the lithium tantalate manufacturing process, this device is used to automatically perform the following first to fourth processes. Specifically, the process includes a first step of bringing the lithium tantalate seed crystal into contact with the molten lithium tantalate raw material and pulling out the neck; a second step of pulling out the tapered portion to expand the diameter; a third step of pulling out the straight cylindrical portion to its maximum diameter; and a fourth step of separating the pulled single crystal from the molten raw material. Finally, the lithium tantalate single crystal is cooled and extracted. This series of processes (the first to fourth processes) is considered as one lifting operation, and a total of 491 lifting operations were performed.
[0034] In a series of processes, from the first to the fourth process, an image sensor 12, located on the upper part of the device, monitors the interior of the furnace. For the images acquired by the image sensor 12, in... Figure 2 The anomaly detection system described herein performs computational processing. Within the anomaly detection system, a computer is assembled as the control unit 13, and the system is configured to: process video from the image sensor 12 in real time; detect anomalies when they are found; and sound an alarm via the sound generator 14. Furthermore, in anomaly detection, YOLOv5 (a model capable of object detection based on deep learning results) is installed in the Python programming software. 500 images of crystal-derived anomalies and 200 images of growth ridge anomalies were prepared for pre-learning.
[0035] In 491 pulling operations, 82 crystal pulling anomalies occurred in the first process. Among these anomalies, 70 instances involved the formation of solidified portions around the seed crystal, and 12 instances involved the breakage of the contact between the molten raw material and the seed crystal. Regarding the formation of solidified portions, the neck was observed after the lithium tantalate single crystal was extracted to determine if there were any portions with a diameter larger than the set diameter. These anomalies were confirmed during the pulling process using an anomaly detection system, which detected 75 out of 82 anomalies, achieving a detection rate of 91.5%.
[0036] [Example 2] Following the first step in Example 1, an abnormality in the growth ridge of the cone was also confirmed in the second step.
[0037] In 491 Czochralski pulls, abnormalities in the growth edges of the cone occurred 41 times in the second process. These abnormalities were also observed and determined after the lithium tantalate single crystal was derived. These abnormalities were also confirmed during the Czochralski pull using an anomaly detection system, which detected 39 out of 41 abnormalities, achieving a detection rate of 95.1%.
[0038] The results of the above embodiments are shown in the table below. [Table 1]
[0039] Explanation of reference numerals in the attached figures 1: Single crystal manufacturing equipment 2: Single crystal 2S: Seed section 2N: Neck 2C: Conical part 2B: Straight section 2T: Rear 2R: Growth ridge 3: Raw material molten liquid 4: Crucible 5: Heating source 6: Thermocouple 11: Weighing sensor 12: Image Sensor 13: Control Department 14: Sound generating device SS: Curing Section AR: Abnormal growth ridge.
Claims
1. A method for detecting anomalies in single crystal manufacturing, comprising a heating and melting process in which a single crystal raw material is heated and melted in a crucible using a heating source to create a raw material molten liquid, and a pulling process in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured using a spin-pulling method, characterized in that, In the pulling process, a computing device equipped with a pre-learned model is used to process image data of the part of the seed crystal in contact with the raw material melt or the image data of the single crystal obtained by the image sensor. Based on the result of the processing, abnormalities in the manufacturing of the single crystal are detected.
2. The anomaly detection method for single crystal manufacturing according to claim 1, characterized in that, The pulling process includes a first step of bringing the seed crystal into contact with the raw material molten liquid and pulling the neck through the rotary pulling method.
3. The anomaly detection method for single crystal manufacturing according to claim 2, characterized in that, The lifting process includes a second step of lifting the cone portion to enlarge its diameter.
4. The anomaly detection method for single crystal manufacturing according to claim 3, characterized in that, The lifting process includes a third step of lifting the section into a straight cylinder with the maximum diameter.
5. The anomaly detection method for single crystal manufacturing according to claim 4, characterized in that, The pulling process includes a fourth step of separating the pulled single crystal from the raw material melt.
6. The anomaly detection method for single crystal manufacturing according to claim 1, characterized in that, The pre-learned model is formed through deep learning.
7. The anomaly detection method for single crystal manufacturing according to claim 1, wherein, The single crystal is an oxide single crystal.
8. The anomaly detection method for single crystal manufacturing according to claim 5, characterized in that, All processes, including the first to fourth processes, are executed automatically.
9. An anomaly detection system for single crystal manufacturing, comprising a heating and melting process in which a single crystal raw material is heated and melted in a crucible using a heating source to create a raw material molten liquid, and a pulling process in which a seed crystal is brought into contact with the raw material molten liquid and a single crystal is manufactured using a spin-pulling method, wherein, It possesses a computing device equipped with a pre-learned model. In the pulling process, the computing device performs calculations on image data of the part of the seed crystal in contact with the raw material melt or the image data of the single crystal obtained by the image sensor, and detects abnormalities in the manufacturing of the single crystal based on the results of the calculations.
10. The anomaly detection system for single-crystal manufacturing according to claim 9, characterized in that, The pulling process includes a first step of bringing the seed crystal into contact with the raw material molten liquid and pulling the neck through the rotary pulling method.
11. The anomaly detection system for single crystal manufacturing according to claim 10, characterized in that, The lifting process includes a second step of lifting the cone portion to enlarge its diameter.
12. The anomaly detection system for single-crystal manufacturing according to claim 11, characterized in that, The lifting process includes a third step of lifting the section into a straight cylinder with the maximum diameter.
13. The anomaly detection system for single-crystal manufacturing according to claim 12, characterized in that, The pulling process includes a fourth step of separating the pulled single crystal from the raw material melt.
14. The anomaly detection system for single crystal manufacturing according to claim 9, characterized in that, The pre-learned model is formed through deep learning.
15. The anomaly detection system for single-crystal manufacturing according to claim 9, wherein, The single crystal is an oxide single crystal.
16. The anomaly detection system for single crystal manufacturing according to claim 13, characterized in that, All processes, including the first to fourth processes, are executed automatically.