Control method for correcting collected solid-to-liquid ratio in furnace

Abstract

The invention provides a control method for correcting a collected solid-to-liquid ratio in a furnace, and the method comprises the steps: collecting information of a plurality of images in at least two rotation periods in the furnace, and forming an image set; the image set is preprocessed; linear features of a solid-liquid interface are extracted from the preprocessed image set; in combination with time axis information, a three-dimensional information set of a solid-liquid interface in the furnace is constructed based on the extracted linear features, and the three-dimensional information set comprises a material block height value and a solid-liquid area ratio; if the three-dimensional information set meets the preset condition, power reduction operation is automatically triggered or an alarm signal is sent out. According to the method, the solid-to-liquid ratio judgment condition of an existing system can be optimized, the material block height judgment condition logic item is added, and the solid-to-liquid ratio judgment condition is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and in particular to a control method for correcting a solid-liquid ratio collected in a furnace. Background Art

[0002] At present, the Jingsheng and Jingyang systems on the market use the process of adding or cooling according to whether the solid-liquid ratio in the furnace meets the judgment conditions during the feeding stage and temperature control stage during the crystal pulling preparation process. The solid-liquid ratio judgment method is: after the camera takes a photo, it is recorded into the background storage of the visual system, and the collected pictures are compared with the existing model within the set time, and the solid-liquid ratio is output. The system outputs the feeding and cooling instructions according to the solid-liquid ratio. The existing solid-liquid ratio model is established as follows: for each image, the ratio of the number of solid pixels to the number of liquid pixels in each image is used as the original solid-liquid ratio of each image, and the average of the original solid-liquid ratios of multiple images is used as the target solid-liquid ratio.

[0003] However, the prior art uses the average of the number of solid pixels and the number of liquid pixels in multiple images as the output result of the system solid-liquid ratio. In the judgment condition, there is a pixel value that cannot identify a high material block with a certain thickness below the molten silicon liquid level. After the system judges, the solid-liquid ratio is small, and the power is reduced in advance. Or, due to the large area of ​​thin sheet materials and the actual thickness is too small, the pixel value output solid-liquid ratio is large, resulting in the system not reducing the power output; the above two situations frequently occur in actual on-site production. If the production personnel do not pay enough attention, the high material block is misjudged and the temperature is reduced too early and needs to be reheated: on the one hand, the production efficiency is not improved, but the material time is wasted due to premature cooling; on the other hand, the service life of the crucible is reduced due to the thermal stress effect during the repeated heating and cooling process; the system does not cool down due to misjudgment of thin sheet material blocks: on the one hand, there are actually few solid material blocks in the furnace, and the heater does not reduce the power, resulting in baking of the straight wall of the crucible; on the other hand, due to the continuous high temperature in the furnace, the temperature does not drop for a long time during the temperature control process, which affects the efficiency of entering the temperature adjustment stage. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a control method for correcting the solid-liquid ratio collected in a furnace. The method can optimize the existing system for judging the solid-liquid ratio conditions, add a logical item for the material block height judgment condition, and make the solid-liquid ratio judgment condition more accurate.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A control method for correcting the solid-liquid ratio collected in a furnace, the method comprising the following steps:

[0007] Collecting at least one image information in at least two rotation periods in the furnace to form an image collection;

[0008] Preprocessing the image collection;

[0009] Extract linear features of the solid-liquid interface from the preprocessed image collection;

[0010] Combining the time axis information, constructing a three-dimensional information set of the solid-liquid interface in the furnace based on the linear features, wherein the three-dimensional information set includes a material block height value and a solid-liquid area ratio;

[0011] If the three-dimensional information set meets the preset conditions, a power reduction operation is automatically triggered or an alarm signal is issued;

[0012] Wherein, the preset conditions include: the material block height value is less than a height threshold and the solid-liquid area ratio is less than an area ratio threshold.

[0013] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the collecting of at least one image information in at least two rotation cycles in the furnace comprises:

[0014] Use the existing furnace camera as the image acquisition device and adjust the exposure time to suit the lighting conditions in the furnace;

[0015] The acquisition frame rate is determined according to the rotation speed in the furnace and the required image resolution, and at least one image information in at least two rotation periods in the furnace is acquired according to the acquisition frame rate.

[0016] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the preprocessing of the image collection includes:

[0017] Set the image acquisition cycle. If the industrial computer sends an image every K seconds, then set the image acquisition cycle to an integer multiple of K. After the images collected in each image acquisition cycle are preliminarily processed, identify the solid area in each image, and select the image with the largest solid area as the small result image output of the acquisition cycle;

[0018] Collect M small result images of at least two rotation cycles, calculate the solid areas of the M small result images, and select the image with the largest solid area as the final judgment basis;

[0019] Among them, it is necessary to perform image processing such as grayscale linear transformation, image denoising, and feature enhancement on the collected M small result images.

[0020] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the three-dimensional information set of the solid-liquid interface in the furnace is constructed, comprising:

[0021] Use edge detection algorithms to identify boundary features of solid-liquid interfaces;

[0022] Using image segmentation technology, the image processed by edge detection is segmented into solid and liquid areas;

[0023] Copy the image after edge detection and segmentation, create multiple layers, apply different image processing techniques to different layers, and use image synthesis technology to superimpose the processed layers together to form an image with layering and depth;

[0024] Generate a depth map of the image after layer overlay processing.

[0025] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the image after layer overlay processing is processed to generate a depth map:

[0026] Choose or train a monocular depth estimation model;

[0027] Apply the model to the image after layer overlay processing to generate a depth map;

[0028] Convert each pixel value in the depth map into depth information representing the distance from the point to the camera;

[0029] Convert the depth information into linearly expressed height information so as to construct a three-dimensional solid-liquid information set;

[0030] Use 3D reconstruction technology to convert 2D images and depth information into 3D models or scenes.

[0031] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, before modifying the system power reduction judgment condition, a secondary solid-liquid ratio entry of the three-dimensional information set into a semantic segmentation model is performed, including:

[0032] Converting a three-dimensional information set including block height values ​​and solid-liquid area ratios into a format suitable for processing by a semantic segmentation model;

[0033] Annotate the converted data, i.e. mark the boundary between the solid phase and the liquid phase;

[0034] The model is trained using the labeled data. During the training process, the model is used to learn how to predict the segmentation results of the solid and liquid phases based on the input three-dimensional information or its two-dimensional representation;

[0035] Apply the trained model to the new 3D information set to perform secondary input and segmentation of the solid-liquid ratio;

[0036] The model will output the classification results for each pixel, thereby obtaining the segmentation results of the solid-liquid ratio.

[0037] According to a control method for correcting the solid-liquid ratio of a furnace collected according to the present invention, maintaining the existing power material conditions includes:

[0038] The material block capture height>threshold height & capture area>threshold area ratio, the system maintains the existing power;

[0039] When the material block capture height > threshold height and capture area < threshold area ratio, the system mainly judges based on the height threshold condition;

[0040] When the material block capture height < threshold height and capture area > threshold area ratio, the system mainly judges based on the capture area ratio threshold condition;

[0041] Among them, the capture area is the number of solid pixel values / liquid pixel values.

[0042] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the following steps are further performed:

[0043] An adaptive threshold mechanism is introduced to dynamically adjust the height threshold and area ratio threshold based on historical data and real-time furnace conditions, including:

[0044] Collect historical data related to the solid-liquid ratio measurement in the furnace;

[0045] Clean and preprocess historical data;

[0046] Extract features related to solid-liquid ratio measurement from historical data and real-time data, which will be used as input to the adaptive threshold algorithm to predict and adjust the threshold. The features include the historical change trend of the solid-liquid interface height, the range of area ratio, the stability of the temperature gradient, etc.

[0047] Use historical data to train the adaptive threshold algorithm model.

[0048] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, real-time data is input into a trained adaptive threshold algorithm model under real-time furnace conditions to calculate the current height threshold and area ratio threshold;

[0049] The power reduction decision is made based on the calculated height threshold and area ratio threshold combined with the real-time furnace conditions.

[0050] According to a control method for correcting the solid-liquid ratio collected in a furnace provided by the present invention, the grayscale linear transformation includes:

[0051] Read M images using the image processing library.

[0052] Grayscale conversion: convert color images to grayscale images;

[0053] Linear transformation: Apply a linear transformation formula to each grayscale image;

[0054] Overflow processing: overflow processing is performed on the transformed pixel value to ensure that it is within the valid range;

[0055] The image noise reduction comprises:

[0056] Filter denoising: Use filters to smooth the image to remove noise;

[0057] The feature enhancements include:

[0058] Histogram equalization: By adjusting the histogram distribution of the image, the contrast of the image is made more uniform, thereby enhancing the visual effect of the image;

[0059] Sharpening filter: Use the sharpening filter to sharpen the image to highlight the edges and details of the image;

[0060] Morphological operations: Morphological operations are used to extract and enhance image features.

[0061] It can be seen that compared with the prior art, the method provided by the present invention has the following beneficial effects:

[0062] 1. Improved precision and accuracy: The present invention collects multiple image information within at least two rotation cycles in the furnace. This method can obtain more comprehensive furnace state data, thereby more accurately extracting the linear features of the solid-liquid interface and constructing a three-dimensional information set. This fusion of multi-cycle and multi-image data significantly improves the precision and accuracy of solid-liquid ratio measurement.

[0063] 2. Multi-dimensional judgment criteria: The present invention introduces the height threshold condition and the area ratio threshold condition as dual judgment criteria, so that the system can more comprehensively evaluate the furnace conditions. This multi-dimensional judgment method is more flexible and reliable than a single condition, can better adapt to different furnace working conditions, and reduce the occurrence of misjudgment and missed judgment.

[0064] 3. Automation and intelligence: The present invention realizes automation and intelligence of furnace management by automatically triggering power reduction operation or issuing alarm signals, which not only reduces the burden of manual monitoring, but also improves response speed and decision-making efficiency, helps to timely discover and deal with abnormal situations in the furnace, and ensures the stability and safety of the production process.

[0065] 4. Flexibility and adaptability: The present invention can flexibly adjust the judgment logic and power reduction strategy according to the different situations of the height and area ratio of the solid-liquid interface in the collected image. For example, when the height of the solid-liquid interface is greater than the threshold and the area ratio is much greater than K times the threshold, the system can judge that the furnace is in good condition and maintain the existing power; when the height of the solid-liquid interface is less than the threshold, a more detailed judgment is made based on the ratio of solid-liquid pixel values. This flexibility and adaptability enable the method to better adapt to different production needs and furnace conditions.

[0066] 5. Improve production efficiency: The present invention helps optimize the melting and reaction process of materials in the furnace and improve production efficiency through accurate solid-liquid ratio calculation and timely power reduction operation. At the same time, it reduces downtime and production losses caused by abnormalities in the furnace, which also helps to reduce production costs and improve economic benefits.

[0067] In summary, the control method for correcting the solid-liquid ratio collected in the furnace of the present invention has significant advantages in improving accuracy, degree of automation, flexibility and adaptability, and is of great significance for improving the efficiency and safety of furnace management.

[0068] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 The present invention is a flow chart of a control method embodiment for correcting the solid-liquid ratio collected in a furnace.

[0070] Figure 2 It is a schematic diagram of using an image edge algorithm to identify the boundary characteristics of an object in an embodiment of a control method for correcting the solid-liquid ratio collected in a furnace according to the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0072] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0073] See also Figure 1 The present invention provides a control method for correcting the solid-liquid ratio of furnace collection, the method comprising the following steps:

[0074] Step S1, collecting at least one image information in at least two rotation periods in the furnace to form an image collection;

[0075] Step S2, preprocessing the image collection;

[0076] Step S3, extracting linear features of the solid-liquid interface from the preprocessed image collection;

[0077] Step S4, combining the time axis information and constructing a three-dimensional information set of the solid-liquid interface in the furnace based on the extracted linear features, the three-dimensional information set including the material block height value and the solid-liquid area ratio;

[0078] Step S5: If the three-dimensional information set meets the preset conditions, a power reduction operation is automatically triggered or an alarm signal is issued.

[0079] The preset conditions in step S5 include: the material block height value is less than the height threshold and the solid-liquid area ratio is less than the area ratio threshold.

[0080] If the height of the solid-liquid interface in the collected image is greater than the threshold height, and the area ratio is greater than the area ratio threshold K times (such as 10 times, which can be set), it can be determined that the furnace condition is good, there is no need to reduce the power, and the existing power is maintained; if the height of the solid-liquid interface in the collected image is greater than the threshold height, and the area ratio is close to 1, the height threshold condition is used for judgment; if the height of the solid-liquid interface in the collected image is less than the threshold height, it is judged by the ratio of solid-liquid pixel values.

[0081] In the above step S1, when performing image acquisition, the existing furnace table camera is used as the image acquisition device, and the exposure time is adjusted to adapt to the lighting conditions in the furnace.

[0082] According to the rotation speed in the furnace and the required image resolution, the acquisition frame rate is determined, and at least one image information in at least two rotation cycles in the furnace is acquired according to the acquisition frame rate. For example, the acquisition frequency of 2 seconds / image and the crucible rotation speed based on 1 revolution / minute are set to ensure that sufficient image data (such as 30 images) are evenly acquired in each rotation cycle. In this way, the dynamic changes of the solid-liquid interface in the furnace can be more comprehensively reflected.

[0083] In the above step S2, the image collection is preprocessed, including:

[0084] Set the image acquisition cycle. If the industrial computer sends an image every K seconds, set the image acquisition cycle to an integer multiple of K. After preliminary processing of the images collected in each image acquisition cycle, identify the solid area in each image and select the image with the largest solid area as the small result image output of the acquisition cycle.

[0085] Collect M small result images of at least two rotation cycles, calculate the solid areas of the M small result images, and select the image with the largest solid area as the final judgment basis.

[0086] Among them, it is necessary to perform image processing such as grayscale linear transformation, image denoising, and feature enhancement on the collected M small result images.

[0087] Specifically, in the preprocessing mechanism, since the industrial computer sends a picture every 2 seconds, the cycle must be set to an integer multiple of 2. The following takes setting the acquisition cycle to 10 seconds as an example: the industrial computer sends a picture every 2 seconds, and the acquisition cycle is 10 seconds, that is, 5 pictures are collected within the set cycle, and the 5 pictures with the largest solid area are identified as 1 small result; since the crucible rotates at 1 revolution / min, in order to make the solid-liquid ratio judgment more accurate, it is necessary to collect pictures of at least 2 rotation cycles in the furnace for solid-liquid ratio judgment. The two rotation cycles are 120 seconds, which just outputs 12 small result pictures, and the maximum solid area in the 12 small result pictures is output as the final judgment value, wherein the collected 12 small result pictures need to be subjected to image denoising. In the digital signal processing box: grayscale image linear transformation is used to improve image quality, increase image contrast, selectively highlight image solid-liquid surface information, suppress background area information, and enhance solid-liquid characteristics.

[0088] In the above step S3, a three-dimensional information set of the solid-liquid interface in the furnace is constructed, including:

[0089] Use edge detection algorithms to identify boundary features of solid-liquid interfaces. For example, using a Canny edge detector, which combines Gaussian filtering and gradient detection, can effectively identify edges in images. Figure 2 As shown: Solids usually have clear, regular linear edges, while liquids have relatively fuzzy and less regular edges due to their fluidity and irregularity.

[0090] Use image segmentation technology to segment the image after edge detection and divide the image into solid and liquid areas. In the single crystal furnace, the solid part can be segmented according to its brightness, grayscale and texture characteristics.

[0091] Copy the image that has been processed by edge detection and segmentation, create multiple layers, and apply different image processing techniques to different layers, such as blur (Gaussian blur), displacement (affine transformation), deformation (perspective transformation or elastic transformation), etc. Use image synthesis technology (such as Alpha blending) to superimpose the processed layers together to form an image with a sense of hierarchy and depth. Among them, due to the pre-processing mechanism, the collected pictures have obvious solid and liquid characteristics, and the processed pictures are layered. Copy the pictures that have been processed by edge detection and segmentation into multiple layers, blur, displace, deform, etc. on different layers, and then superimpose them together to obtain a set of pictures with a sense of hierarchy and depth.

[0092] The image after layer overlay processing is used to generate a depth map. Each pixel value represents the distance from the point to the camera. By performing a linear transformation on the depth map, the depth information can be converted into linear height information, thereby generating three-dimensional information of the captured image.

[0093] Specifically, the image after the layer overlay processing is used to generate a depth map, including:

[0094] Choose or train a monocular depth estimation model (such as Monodepth, MiDaS).

[0095] The model is applied to the image after layer overlay processing to generate a depth map.

[0096] Convert each pixel value in the depth map into depth information representing the distance from the point to the camera.

[0097] The depth information is converted into linearly represented height information in order to construct a three-dimensional solid-liquid information set.

[0098] Use 3D reconstruction techniques (such as point cloud generation, meshing, etc.) to convert 2D images and depth information into 3D models or scenes.

[0099] In the above step S4, the above processed image information set is entered into the three-dimensional information set: mainly including the material block height value and the solid-liquid area ratio. The above processing time for the image information starts after the last barrel is filled with material, and is mainly used for the system to identify the solid-liquid information and automatically reduce the power judgment.

[0100] In this embodiment, before modifying the system power reduction judgment condition, a second solid-liquid ratio entry of the three-dimensional information set into the semantic segmentation model is performed, including:

[0101] Convert the three-dimensional information set including the block height value and the solid-liquid area ratio into a format suitable for processing by the semantic segmentation model, such as projecting the three-dimensional data onto a two-dimensional plane, or encoding the three-dimensional features into a two-dimensional image in some way.

[0102] The converted data is labeled, that is, the boundaries between the solid and liquid phases are marked to provide the model with the labels required for training.

[0103] The model is trained using labeled data, during which it is used to learn how to predict the segmentation results of the solid and liquid phases based on the input three-dimensional information or its two-dimensional representation.

[0104] The trained model is applied to the new three-dimensional information set for secondary entry and segmentation of the solid-liquid ratio.

[0105] The model will output the classification result (solid phase or liquid phase) of each pixel, thereby obtaining the segmentation result of the solid-liquid ratio.

[0106] In this embodiment, the height threshold condition is: a threshold for the height of the material block is set. When the height of the material block exceeds the threshold, the system should automatically trigger a power reduction operation or issue an alarm. This threshold can be set based on factors such as the stacking speed of the material in the furnace and the furnace capacity.

[0107] In this embodiment, the area ratio threshold condition is: a threshold value of the solid-liquid area ratio is set. When the solid-liquid area ratio is lower than the threshold value, a power reduction operation or an alarm is also triggered. This threshold value reflects the solid-liquid distribution in the furnace, which helps to promptly discover and deal with the problem of solid-liquid imbalance.

[0108] Furthermore, the height threshold condition and the area ratio threshold condition are integrated into the power reduction judgment logic of the system. The power reduction operation can be set to be triggered when any one of the conditions is met, or it can be set to be triggered only when both conditions are met.

[0109] In this embodiment, maintaining the existing power material conditions includes:

[0110] The material block capture height>threshold height & capture area>threshold area ratio, the system maintains the existing power;

[0111] When the material block capture height > threshold height and capture area < threshold area ratio, the system mainly judges based on the height threshold condition;

[0112] When the material block capture height < threshold height and capture area > threshold area ratio, the system mainly judges based on the capture area ratio threshold condition;

[0113] Among them, the capture area is the number of solid pixel values / liquid pixel values.

[0114] In this embodiment, the following is also performed:

[0115] An adaptive threshold mechanism is introduced to dynamically adjust the height threshold and area ratio threshold based on historical data and real-time furnace conditions, including:

[0116] Collect historical data related to the solid-liquid ratio measurement in the furnace.

[0117] Clean and preprocess historical data.

[0118] Features related to solid-liquid ratio measurement are extracted from historical data and real-time data and used as input to the adaptive threshold algorithm for predicting and adjusting the threshold. The features include the historical change trend of the solid-liquid interface height, the range of change of the area ratio, the stability of the temperature gradient, etc.

[0119] Use historical data to train the adaptive threshold algorithm model. During the training process, methods such as cross-validation can be used to evaluate the performance of the model.

[0120] Under real-time furnace conditions, real-time data is input into the trained adaptive threshold algorithm model to calculate the current height threshold and area ratio threshold, which will be dynamically adjusted as furnace conditions change to reflect the current actual situation.

[0121] According to the calculated height threshold and area ratio threshold, combined with the real-time furnace conditions, the power reduction judgment is made. If the height of the solid-liquid interface is greater than the threshold and the area ratio meets certain conditions (such as greater than K times the area ratio threshold or close to 1), it is judged that the furnace is in good condition and does not need to reduce power; otherwise, the power reduction operation is performed or an alarm signal is issued according to the specific situation.

[0122] The real-time judgment results and actual operation conditions are fed back to the adaptive threshold algorithm model to further optimize the parameters and performance of the model.

[0123] The model is updated and maintained regularly to ensure it can adapt to changes in furnace conditions and maintain a high level of accuracy.

[0124] In the above step S2, the grayscale linear transformation includes:

[0125] Read images: Use the image processing library to read M images.

[0126] Grayscale conversion: Convert a color image to a grayscale image.

[0127] Linear transformation: Apply the linear transformation formula to each grayscale image. Specifically, the linear transformation formula is usually s=ar+b, where s is the transformed pixel value, r is the original pixel value, and a and b are the coefficients of the linear transformation. By adjusting the values ​​of a and b, different transformation effects can be achieved, such as contrast enhancement, brightness adjustment, etc. Contrast enhancement: Increasing the value of a can enhance the contrast of the image; Brightness adjustment: Adjusting the value of b can change the brightness of the image; Inverse transformation: When a=-1, b=255, the inverse transformation of the image is achieved.

[0128] Overflow handling: Since pixel values ​​must be within a certain range (such as 0-255), overflow handling is required for the transformed pixel values ​​to ensure that they are within the valid range.

[0129] In this embodiment, image noise reduction includes:

[0130] Filter denoising: Use filters (such as mean filter, Gaussian filter, median filter, etc.) to smooth the image to remove noise;

[0131] In this embodiment, feature enhancement includes:

[0132] Histogram equalization: By adjusting the histogram distribution of the image, the contrast of the image is made more uniform, thereby enhancing the visual effect of the image.

[0133] Sharpening filter: Use sharpening filters (such as Laplacian filter, high-pass filter, etc.) to sharpen the image to highlight the edges and details of the image.

[0134] Morphological operations: Use morphological operations (such as dilation, erosion, opening, closing, etc.) to extract and enhance image features.

[0135] In summary, the present invention aims to improve the solid-liquid ratio correction in the later stage. When collecting a collection of images within at least one rotation cycle, the collection frequency is 2 seconds per image. The crucible rotation in the current material sintering process is 1 revolution / min, that is, at least 30 images are collected, and the image collection within the collected cycle is subjected to image denoising, and the complex function is used to improve the image quality and enhance the image features. The processed images are subjected to linear extraction of the solid-liquid distribution contours, and the comprehensive output is performed, and finally the three-dimensional characteristic relationship of the solid-liquid relationship is output. The above processing is to add three-dimensional model data on the basis of the existing solid-liquid ratio judgment conditions of the number of solid pixel values ​​and the number of liquid pixel values, so as to correct the solid-liquid ratio in the existing solid-liquid ratio relationship and improve the accuracy of the actual material block feature judgment in the furnace.

[0136] Furthermore, by collecting multiple image information within at least two rotation cycles in the furnace, the present invention can obtain more comprehensive furnace state data, thereby more accurately extracting the linear features of the solid-liquid interface and constructing a three-dimensional information set. This fusion of multi-cycle and multi-image data significantly improves the precision and accuracy of solid-liquid ratio measurement.

[0137] Furthermore, the present invention introduces the height threshold condition and the area ratio threshold condition as dual judgment criteria, so that the system can more comprehensively evaluate the furnace conditions. This multi-dimensional judgment method is more flexible and reliable than a single condition, and can better adapt to different furnace working conditions and reduce the occurrence of misjudgment and missed judgment.

[0138] Furthermore, the present invention realizes automation and intelligence of furnace management by automatically triggering power reduction operation or issuing alarm signals, which not only reduces the burden of manual monitoring, but also improves response speed and decision-making efficiency, helps to timely discover and deal with abnormal situations in the furnace, and ensures the stability and safety of the production process.

[0139] Furthermore, the present invention can flexibly adjust the judgment logic and power reduction strategy according to the different situations of the height and area ratio of the solid-liquid interface in the collected image. For example, when the height of the solid-liquid interface is greater than the threshold and the area ratio is much greater than K times the threshold, the system can judge that the furnace is in good condition and maintain the existing power; when the height of the solid-liquid interface is less than the threshold, a more detailed judgment is made based on the ratio of solid-liquid pixel values. This flexibility and adaptability enable the method to better adapt to different production needs and furnace conditions.

[0140] Furthermore, the present invention helps optimize the melting and reaction process of materials in the furnace and improve production efficiency through accurate solid-liquid ratio measurement and timely power reduction operation. At the same time, it reduces downtime and production losses caused by abnormalities in the furnace, which also helps to reduce production costs and improve economic benefits.

[0141] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0142] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A control method for correcting the solid-liquid ratio of furnace collection, characterized in that: The method comprises the following steps: Collecting at least one image information in at least two rotation periods in the furnace to form an image collection; Preprocessing the image collection; Extract linear features of the solid-liquid interface from the preprocessed image collection; Combining the time axis information, constructing a three-dimensional information set of the solid-liquid interface in the furnace based on the linear features, wherein the three-dimensional information set includes a material block height value and a solid-liquid area ratio; If the three-dimensional information set meets the preset conditions, a power reduction operation is automatically triggered or an alarm signal is issued; Wherein, the preset conditions include: the material block height value is less than a height threshold and the solid-liquid area ratio is less than an area ratio threshold.

2. The method according to claim 1, characterized in that The collecting of at least one image information in at least two rotation cycles in the furnace comprises: Use the existing furnace camera as the image acquisition device and adjust the exposure time to suit the lighting conditions in the furnace; The acquisition frame rate is determined according to the rotation speed in the furnace and the required image resolution, and at least one image information in at least two rotation periods in the furnace is acquired according to the acquisition frame rate.

3. The method according to claim 1, characterized in that The preprocessing of the image collection includes: Set the image acquisition cycle. If the industrial computer sends an image every K seconds, then set the image acquisition cycle to an integer multiple of K. After the images collected in each image acquisition cycle are preliminarily processed, identify the solid area in each image, and select the image with the largest solid area as the small result image output of the acquisition cycle; Collect M small result images of at least two rotation cycles, calculate the solid areas of the M small result images, and select the image with the largest solid area as the final judgment basis; Among them, it is necessary to perform image processing such as grayscale linear transformation, image denoising, and feature enhancement on the collected M small result images.

4. The method according to claim 1, characterized in that The three-dimensional information set of constructing the solid-liquid interface in the furnace includes: Use edge detection algorithms to identify boundary features of solid-liquid interfaces; Using image segmentation technology, the image processed by edge detection is segmented into solid and liquid areas; Copy the image after edge detection and segmentation, create multiple layers, apply different image processing techniques to different layers, and use image synthesis technology to superimpose the processed layers together to form an image with layering and depth; Generate a depth map of the image after layer overlay processing.

5. The method according to claim 4, characterized in that The step of generating a depth map from the image after the layer overlay processing comprises: Choose or train a monocular depth estimation model; Apply the model to the image after layer overlay processing to generate a depth map; Convert each pixel value in the depth map into depth information representing the distance from the point to the camera; Convert the depth information into linearly expressed height information so as to construct a three-dimensional solid-liquid information set; Use 3D reconstruction technology to convert 2D images and depth information into 3D models or scenes.

6. The method according to claim 1, characterized in that: Before modifying the system power reduction judgment condition, a secondary solid-liquid ratio entry of the three-dimensional information set into the semantic segmentation model is performed, including: Converting a three-dimensional information set including block height values ​​and solid-liquid area ratios into a format suitable for processing by a semantic segmentation model; Annotate the converted data, i.e. mark the boundary between the solid phase and the liquid phase; The model is trained using the labeled data. During the training process, the model is used to learn how to predict the segmentation results of the solid and liquid phases based on the input three-dimensional information or its two-dimensional representation; Apply the trained model to the new 3D information set to perform secondary input and segmentation of the solid-liquid ratio; The model will output the classification results for each pixel, thereby obtaining the segmentation results of the solid-liquid ratio.

7. The method according to claim 1, characterized in that: Maintaining the existing power material conditions include: The material block capture height>threshold height & capture area>threshold area ratio, the system maintains the existing power; When the material block capture height > threshold height and capture area < threshold area ratio, the system mainly judges based on the height threshold condition; When the material block capture height < threshold height and capture area > threshold area ratio, the system mainly judges based on the capture area ratio threshold condition; Among them, the capture area is the number of solid pixel values / liquid pixel values.

8. The method according to claim 1, characterized in that Also execute: An adaptive threshold mechanism is introduced to dynamically adjust the height threshold and area ratio threshold based on historical data and real-time furnace conditions, including: Collect historical data related to the solid-liquid ratio measurement in the furnace; Clean and preprocess historical data; Extract features related to solid-liquid ratio measurement from historical data and real-time data, which will be used as input to the adaptive threshold algorithm to predict and adjust the threshold. The features include the historical change trend of the solid-liquid interface height, the range of area ratio, the stability of the temperature gradient, etc. Use historical data to train the adaptive threshold algorithm model.

9. The method according to claim 8, characterized in that: Under real-time furnace conditions, the real-time data is input into the trained adaptive threshold algorithm model to calculate the current height threshold and area ratio threshold; The power reduction decision is made based on the calculated height threshold and area ratio threshold combined with the real-time furnace conditions.

10. The method according to claim 1, characterized in that: The grayscale linear transformation includes: Image reading: Use the image processing library to read M images; Grayscale conversion: convert color images to grayscale images; Linear transformation: Apply a linear transformation formula to each grayscale image; Overflow processing: overflow processing is performed on the transformed pixel value to ensure that it is within the valid range; The image noise reduction comprises: Filter denoising: Use filters to smooth the image to remove noise; The feature enhancements include: Histogram equalization: By adjusting the histogram distribution of the image, the contrast of the image is made more uniform, thereby enhancing the visual effect of the image; Sharpening filter: Use the sharpening filter to sharpen the image to highlight the edges and details of the image; Morphological operations: Morphological operations are used to extract and enhance image features.

Images