Automatic filling detection control method and device for reaction kettle

Abstract

The invention discloses an automatic filling detection control method and device for a reaction kettle. The method comprises the following steps: acquiring image data of material mixing states at different positions in a reaction kettle at intervals of preset time, preprocessing a color image to enhance the material particle feature identification degree, extracting features by using a machine learning algorithm, identifying feature units meeting a mixing uniformity standard through a classification model, and counting the number, and adopting a corresponding control method for the specific image features based on the statistical result and a preset rule. The device comprises an image acquisition module, an image preprocessing module, an image feature processing module and a control module which are respectively used for realizing image acquisition, preprocessing, feature processing and rule-based automatic control. Through the image detection and intelligent control technology, the problems that an existing reaction kettle is lagged in mixing effect monitoring and insufficient in automatic control are solved, the production efficiency and the mixing uniformity are improved, and the reaction kettle is suitable for the fields of new energy materials, chemical engineering and the like.

Description

Technical Field

[0001] This invention relates to the field of reaction vessels, and more specifically, to an automatic filling detection and control method and apparatus for reaction vessels. Background Technology

[0002] In industries such as new energy materials, chemicals, and pharmaceuticals, reaction vessels are key equipment for achieving material mixing, reaction, and processing. Their main functions include uniformly dispersing materials through a mechanical stirring system, quantitatively adding solid and liquid raw materials with a feeding device, and promoting chemical reactions or physical mixing processes under set process conditions (such as temperature and pressure). The working principle of a reaction vessel is typically based on mechanical transmission and fluid dynamics design. The rotation of the stirring blades generates a flow field, driving the material circulation to achieve uniform mixing. Simultaneously, sensors monitor basic parameters such as liquid level and weight, providing data support for process control.

[0003] In existing technologies, reactors have significant shortcomings in automated filling and detection control: First, there is a lack of real-time quantitative detection methods for the mixing state of materials during the filling process. The filling progress can only be indirectly judged by a single physical quantity (such as liquid level and pressure), and it is impossible to accurately identify the microscopic characteristics of material particles such as uniformity and agglomeration. Second, the linkage between the control system and the detection unit is insufficient. When the material mixing effect is not up to standard or abnormalities occur during filling, it is difficult to automatically adjust parameters such as filling speed and stirring speed based on real-time detection data, resulting in frequent manual intervention in the production process. Third, the ability to integrate and analyze multiple sets of detection data is insufficient. It is impossible to achieve dynamic optimization of the filling and mixing process through multi-dimensional information (such as image features and quality data). Especially in new energy fields such as ternary lithium cathode materials, where the requirements for mixing uniformity are extremely high, the detection accuracy and control efficiency of traditional reactors are difficult to meet the needs of industrial production. Summary of the Invention

[0004] The purpose of this application is to propose an automatic filling detection and control method and device for a reactor, which aims to solve the problems of lagging monitoring of mixing effect and insufficient automatic control in existing reactors by using image detection and intelligent control technology, thereby improving production efficiency and mixing uniformity.

[0005] The technical solution of the present invention is as follows: An automatic filling and detection control method for a reaction vessel includes the following steps: Image data of the material mixing state at different locations inside the reactor are collected at preset time intervals; The acquired color images are preprocessed to enhance the feature recognition of material particles in the images; The preprocessed image is used to extract basic features using machine learning algorithms, the features are classified using a classification model, feature units that meet the mixing uniformity standard are identified, and the number of feature units is counted using a clustering algorithm. Based on the statistical results of the number of feature units, corresponding control methods are adopted for specific image features detected according to preset rules.

[0006] In one possible implementation, the image preprocessing includes: Grayscale conversion converts a color image into a grayscale image. It simplifies image processing calculations by fusing the values ​​of the red, green, and blue channels, and enhances the grayscale difference between material particles and the background. Noise reduction is achieved by using a Gaussian filtering algorithm to remove noise points from the image. By adjusting the filtering parameters to adapt to the image characteristics of different materials, image clarity is improved. Edge detection uses gradient operators to extract boundary features of material particles in an image and identify areas with significant changes in brightness or color, providing a basis for subsequent material analysis. Contrast adjustment enhances the detail of material particles in the image through histogram equalization algorithm, stretches the range of pixel grayscale distribution, and makes particle distribution and contours easier to identify.

[0007] In one possible implementation, the machine learning algorithm is a convolutional neural network, and the specific steps are as follows: The preprocessed image is extracted layer by layer through multiple convolutional layers. The first convolutional layer identifies basic features such as the edge and texture of the material particles, and the deep convolutional layers combine the basic features to form composite features such as particle shape and aggregation degree. Pooling layers are used to downsample the feature maps output by convolutional layers, reducing feature dimensions while retaining key information to form abstract feature vectors, which provide input for subsequent classification models. The network parameters were trained based on the labeled material mixing image dataset. By comparing the model's predicted results with the actual mixing state labels, the convolution kernel weights and biases were adjusted to improve the accuracy of feature extraction. The final extracted abstract feature vector is then input into the classification model.

[0008] In one possible implementation, the classification model employs a support vector machine algorithm, with the following specific steps: The basic features extracted from the preprocessed image are converted into feature vectors, which are then used as input data for the classification model. Based on labeled material mixing state samples, the model classification boundary is optimized by adjusting the regularization parameter, thereby improving the recognition accuracy of feature units that meet the mixing uniformity standard. The classification results output by the model are used to identify whether the feature units in the image meet the preset mixing uniformity standard, providing a basis for subsequent feature unit quantity statistics.

[0009] In one possible implementation, the clustering algorithm is the K-means clustering algorithm, and the specific steps are as follows: The feature units classified by the support vector machine are grouped according to attributes such as spatial location and size, and K cluster centers are initialized. Calculate the distance between each feature unit and each cluster center, assign it to the nearest cluster, and recalculate the cluster center position. Repeat this process until the cluster center position no longer changes significantly. The number of feature units that meet the mixing uniformity standard and the total number of feature units in each cluster are counted and output to the subsequent judgment process.

[0010] In one possible implementation, the method of taking corresponding control measures for the detected specific image features according to preset rules includes: When the number of consecutive n feature units in the images captured by each lens reaches a preset threshold and the images at different locations all reach the preset threshold, the automatic unloading process is triggered. When bubble feature units are detected in the image, the mixing speed is automatically reduced.

[0011] In one possible implementation, the corresponding control method further includes: When the unloading conditions are not met for an extended period of time, liquid level data is collected, and the loading speed is automatically reduced and the mixing speed is increased based on the liquid level data to optimize the mixing effect. When the liquid level in the reactor reaches the preset full-fill threshold, the feeding process should be stopped immediately.

[0012] An automatic filling detection and control device for a reaction vessel includes: The image acquisition module acquires image data of the material mixing state at different locations within the reactor at preset time intervals; The image preprocessing module performs image preprocessing on the acquired color images to enhance the feature recognition of material particles in the images; The image feature processing module uses machine learning algorithms to extract basic features from the preprocessed image, classifies the features using a classification model, identifies feature units that meet the mixing uniformity standard, and uses a clustering algorithm to count the number of feature units. The control module, based on the statistical results of the number of feature units, adopts corresponding control methods for the specific image features detected according to preset rules.

[0013] In one possible implementation, the image acquisition module further includes several detection lenses for acquiring images of material mixing inside the reactor.

[0014] In one possible implementation, the control module further includes a stirring motor, stirring blades, a liquid level sensor, a discharge solenoid ball valve, and several loading speed control valves. The stirring motor is used to drive the stirring blades, the stirring blades are used to mix the materials in the reactor, the liquid level sensor is used to detect the liquid level of the materials in the reactor, the discharge solenoid ball valve is used to control the discharge of the mixed materials, and the loading speed control valves are used to control the loading of materials.

[0015] This application has the following beneficial effects: This application achieves multi-dimensional, real-time material state monitoring by collecting image data of material mixing status at different locations within the reactor at preset time intervals, thus solving the problem of lagging mixing effect monitoring in existing technologies. It enhances the identification of material particle features through color image preprocessing, optimizing image quality and providing a foundation for accurate analysis, thus overcoming the limitations of existing single-physical-quantity detection. Furthermore, it achieves quantitative analysis of material mixing uniformity through feature extraction using machine learning algorithms, screening of qualified feature units using classification models, and statistical counting using clustering algorithms, thus addressing the problem of large errors in subjective human judgment. Finally, it achieves closed-loop automation of detection and execution through a linkage control method based on the statistical results of feature unit counts and preset rules, solving the problems of insufficient multi-parameter collaborative control capabilities and frequent human intervention in existing technologies. Attached Figure Description

[0016] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method in Example 1; Figure 2 This is a flowchart of the image preprocessing process in Example 1; Figure 3 This is a diagram of the convolutional neural network feature extraction structure in Example 1; Figure 4 This is a flowchart of the support vector machine classification process in Example 1; Figure 5 This is a flowchart of the K-means clustering process in Example 1; Figure 6 This is a block diagram of the device structure in Example 2; Figure 7 This is a schematic diagram of the reactor structure in Example 2; Figure 8 This is a schematic diagram of the lower structure of the support in Example 2; Figure 9 This is a schematic diagram of the internal structure of the reactor body in Example 2.

[0018] The following are the symbols and their meanings: 1. Liquid tank; 2. Solid tank; 3. Loading speed control valve; 4. Heating layer; 5. Support frame; 6. Solid feed motor; 7. Vibrator; 8. Screwdriver; 9. Stirring motor; 10. Reactor body; 11. Solenoid ball valve; 12. Discharge motor; 13. Detection lens; 14. Stirring blades. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] Example 1

[0021] like Figures 1-5 As shown in the figure, this embodiment proposes an automatic filling detection and control method for a reaction vessel, including the following steps: S100: Collect image data of the material mixing state at different locations inside the reactor at preset intervals; Step S100 acquires information on the mixing state of materials in different regions of the reactor in real time through multi-dimensional and periodic image acquisition, providing a rich data foundation for subsequent analysis, avoiding judgment bias caused by a single perspective or non-real-time detection, helping to detect abnormalities such as uneven mixing in a timely manner, and improving the comprehensiveness and timeliness of detection.

[0022] S200. Perform image preprocessing on the acquired color images to enhance the feature recognition of material particles in the images; Step S200 optimizes image quality through preprocessing, enabling subsequent machine learning algorithms to extract effective features more accurately, improving the reliability of feature analysis, and laying the foundation for accurate judgment of mixing uniformity.

[0023] S300. The preprocessed image is processed using machine learning algorithms to extract basic features. The features are then classified using a classification model to identify feature units that meet the mixing uniformity standard. Finally, a clustering algorithm is used to count the number of feature units. Step S300 utilizes the powerful feature extraction capabilities of machine learning algorithms to automatically identify key features related to mixing uniformity in the image (such as particle shape, distribution density, etc.), filters out feature units that meet the standards through a classification model, and then uses a clustering algorithm to count their quantity and distribution, thereby achieving a quantitative analysis of the mixing uniformity of materials. Compared with subjective human judgment, this is more objective and accurate, reduces human error, and improves the scientificity and accuracy of the detection.

[0024] S400: Based on the statistical results of the number of feature units, and in accordance with preset rules, take corresponding control methods for the specific image features detected; Step S400 links the feature analysis results with the preset control logic. Based on specific image features such as whether the number of feature units meets the standard and whether there are bubbles, it automatically triggers control actions such as unloading, adjusting the loading speed or stirring speed, so as to realize closed-loop automated control from detection to execution, avoid the lag of manual intervention, improve production efficiency, and at the same time ensure that the material mixing state always meets the process requirements, thereby improving product quality stability and production safety.

[0025] In this embodiment, image preprocessing includes: Grayscale conversion converts a color image into a grayscale image. It simplifies image processing calculations by fusing the values ​​of the red, green, and blue channels, and enhances the grayscale difference between material particles and the background. Grayscale conversion makes the overall outline and distribution of material particles in an image easier to distinguish, reduces color information interference and computational complexity, and provides a simpler and more effective data foundation for subsequent feature extraction and analysis.

[0026] Specifically, the calculation formula for grayscale processing is as follows: , in, For pixels The converted grayscale value; a , b , c This represents the sensitivity coefficient for the corresponding color; , , These are the channel values ​​corresponding to the colors.

[0027] Noise reduction is achieved by using a Gaussian filtering algorithm to remove noise points from the image. By adjusting the filtering parameters to adapt to the image characteristics of different materials, image clarity is improved. The noise reduction uses a Gaussian filtering algorithm to remove noise points in the image. By adjusting the filtering parameters to adapt to the image characteristics of different materials, it can effectively suppress random noise points caused by equipment noise or environmental interference during image acquisition, improve image clarity, avoid misjudgment or loss of material particle features caused by noise, ensure that subsequent analysis is based on clean and reliable image data, and improve the accuracy of detection results.

[0028] Specifically, the calculation formula for noise reduction is as follows: , in, For pixels Pixel values ​​after Gaussian filtering and noise reduction; For Gaussian kernel; k For the size of the kernel;i and j This is the local coordinate offset within the Gaussian filter kernel (window), used to traverse the current pixel. Surrounding neighboring pixels; I The pixel values ​​of the input image.

[0029] Edge detection uses gradient operators to extract boundary features of material particles in an image and identify areas with significant changes in brightness or color, providing a basis for subsequent material analysis. Edge detection uses gradient operators to extract the boundary features of material particles in an image, identify areas with significant changes in brightness or color, accurately locate the edge contours of material particles, and distinguish the boundaries between particles and between particles and the background. This provides key boundary information for subsequent analysis of the shape, size, and aggregation state of material particles, enabling the evaluation of mixing uniformity to be based on more intuitive particle morphology features, thus enhancing the scientific rigor and reliability of the analysis.

[0030] Specifically, the calculation formula for edge detection processing is as follows: , , in, This represents the rate of grayscale change in the image along the horizontal direction, used for detecting vertical edges. This represents the rate of grayscale change in the vertical direction of the image, used for detecting horizontal edges; and The Sobel operator matrix; i and j This refers to the local coordinate offset within the convolution kernel, used to traverse the current pixel. Surrounding neighboring pixels; I This represents the pixel values ​​of the input image.

[0031] gradient value ;direction .

[0032] Contrast adjustment enhances the detail of material particles in the image through histogram equalization algorithm, stretches the range of pixel grayscale distribution, and makes particle distribution and contours easier to identify. Contrast adjustment enhances the detail of material particles in the image through histogram equalization algorithm, stretches the range of pixel grayscale distribution, makes the originally blurry particle edges and dense areas clearer, and makes particle distribution and contours easier to identify. It effectively solves the problem of overlapping or indistinguishable particle features caused by insufficient image contrast, improves the visualization of material state in the image, and creates favorable conditions for accurately counting the number of feature units and judging the uniformity of mixing.

[0033] Specifically, the histogram equalization algorithm includes calculating the cumulative distribution function (CDF) and constructing a mapping function. The formula for calculating the cumulative distribution function (CDF) is as follows: , in, grayscale value i Histogram; N The total number of pixels in the image; v It is a grayscale level.

[0034] The formula for constructing the mapping function is as follows: , in, After histogram equalization, the coordinates in the image are... The new grayscale value of the pixel; Coordinates in the original image The grayscale value of the pixel at that location; L This represents the total number of gray levels.

[0035] In this embodiment, the machine learning algorithm is a convolutional neural network, and the specific steps are as follows: S311. The preprocessed image is extracted layer by layer through multiple convolutional layers. The first convolutional layer identifies the basic features such as the edge and texture of the material particles, and the deep convolutional layers combine the basic features to form composite features such as particle shape and aggregation degree. Step 311 extracts features from the preprocessed image layer by layer through multiple convolutional layers. The first convolutional layer uses the local connectivity and weight sharing mechanism between the convolutional kernel and the image pixels to identify basic features such as the edges and textures of the material particles. This method of capturing low-level visual cues from pixel-level information can provide basic units for subsequent feature abstraction. The deep convolutional layers, through stacked convolutional operations, nonlinearly combine the basic features extracted by the first layer to form composite features such as particle shape and aggregation degree. This achieves hierarchical construction of features from local details to overall structure, enabling the network to capture image content with more semantic information and enhancing its ability to represent the complex morphology of material particles.

[0036] S312. Use pooling layers to downsample the feature maps output by the convolutional layers, reduce the feature dimensions and retain key information to form abstract feature vectors, which provide input for subsequent classification models. Step S312 uses pooling layers to downsample the feature maps output by the convolutional layers. Through operations such as max pooling or average pooling, the spatial dimension of the feature maps is reduced while the main feature responses are retained. This reduces the computational complexity of subsequent steps and avoids overfitting. The resulting abstract feature vectors can highlight key information and suppress redundant details, making the feature representation more robust and providing more compact input features for subsequent classification models.

[0037] S313. Train network parameters based on labeled material mixing image datasets, and adjust convolution kernel weights and biases by comparing model prediction results with actual mixing state labels to improve the accuracy of feature extraction. Step S313 trains network parameters based on labeled material mixing image datasets, calculates the error between the model's prediction results and the actual mixing state labels using the backpropagation algorithm, and adjusts the convolutional kernel weights and biases accordingly. This data-driven parameter optimization process enables the network to automatically learn feature representation patterns related to the material mixing state, continuously improves the accuracy of feature extraction, and allows the convolutional neural network to better adapt to the distribution characteristics of specific material mixing images, enhancing the model's generalization ability and classification accuracy in practical applications.

[0038] S314. Input the final extracted abstract feature vector into the classification model; In this embodiment, the classification model uses the support vector machine algorithm, and the specific steps are as follows: S321. Convert the basic features extracted from the preprocessed image into feature vectors, which are then used as input data for the classification model. Step S321 uses a structured data representation method to transform the material mixing state information (such as basic features like edges and textures) contained in the image into a numerical vector that can be processed by the support vector machine, realizing the mapping from the image pixel space to the feature space. This enables the model to analyze the material mixing state based on a quantitative feature dimension, providing a standardized input format for the subsequent classification process and ensuring that the classification model can effectively capture feature patterns related to mixing uniformity.

[0039] S322. Based on labeled material mixing state samples, the model classification boundary is optimized by adjusting the regularization parameter to improve the recognition accuracy of feature units that meet the mixing uniformity standard. Step S322, adjusting the regularization parameter, can balance the classification error and model complexity of the model on the training set. It searches for the optimal hyperplane in the feature space that maximizes the inter-class margin, thereby improving the recognition accuracy of feature units that meet the mixing uniformity standard. This data-driven parameter optimization process enables the support vector machine to adapt to the distribution characteristics of material mixing state samples, reduce the risk of overfitting, enhance the model's generalization ability to different feature combinations, and ensure that the classification boundary can accurately distinguish feature units that meet and do not meet the mixing uniformity standard.

[0040] S323. Based on the classification results output by the model, identify whether the feature units in the image meet the preset mixing uniformity standard, and provide a basis for subsequent feature unit quantity statistics; Step S323 uses nonlinear mapping or linear classification to discriminate the input feature vector and outputs a clear category label (such as "compliant" or "non-compliant"). This binary classification decision mechanism can transform complex image feature analysis into quantifiable classification results, enabling subsequent statistical processes to count and statistically analyze feature units based on clear category labels. This provides objective and traceable data support for evaluating the uniformity of material mixing and improves the efficiency and accuracy of mixing state analysis.

[0041] In this embodiment, the clustering algorithm is the K-means clustering algorithm, and the specific steps are as follows: S331. Group the feature units classified by the support vector machine according to attributes such as spatial location and size, and initialize K cluster centers; Step S331 performs initial grouping based on the spatial distribution and geometric attributes of feature units, providing a reasonable starting point for K-means clustering. This allows the initial cluster centers to roughly cover the main distribution area of ​​feature units, reducing the randomness of clustering iterations and improving the convergence efficiency of the clustering process. At the same time, grouping based on attributes with clear physical meaning (such as spatial location and size) can make the subsequent cluster division more closely match the actual spatial distribution characteristics of the material mixing state, laying the foundation for accurate clustering.

[0042] S332. Calculate the distance between each feature unit and each cluster center, assign it to the nearest cluster, and recalculate the cluster center position. Repeat this process until the cluster center position no longer changes significantly. Step S332 uses an iteratively optimized distance metric mechanism to continuously aggregate feature units toward the core region of their respective clusters, forming compact and separated cluster structures. This dynamic adjustment process based on Euclidean distance can automatically divide clusters according to the similarity of feature unit attributes, effectively identifying groups of feature units with similar spatial locations and sizes in material mixing images. At the same time, the convergence condition ensures the stability of the clustering results, enabling the cluster division to accurately reflect the inherent distribution law of feature units and providing a structured grouping basis for subsequent statistical analysis.

[0043] S333: Count the number of feature units that meet the mixing uniformity standard and the total number of feature units in each cluster, and output them to the subsequent judgment process; Step S333 transforms the abstract cluster structure into measurable numerical indicators (such as the proportion of characteristic units that meet the standards) by quantifying and statistically analyzing the clustering results. This allows the evaluation of material mixing uniformity to be based on a comprehensive judgment of the local statistical results of each cluster. This clustered statistical method not only preserves the spatial distribution information of characteristic units but also provides a quantified dimension for uniformity evaluation, providing detailed and comprehensive data support for subsequent judgment processes. It facilitates the analysis of differences in mixing states in different regions and improves the accuracy and interpretability of material mixing uniformity evaluation.

[0044] In this embodiment, the control method for specific detected image features based on preset rules includes: When the number of consecutive n feature units captured by each lens reaches a preset threshold and images from different locations all reach the preset threshold, the automatic unloading process is triggered.

[0045] This rule uses the number of feature units from multiple lenses and frames to make a joint judgment, avoiding misjudgments caused by a single lens or accidental fluctuations. It ensures that the material reaches the uniform mixing standard in different areas of the reactor before triggering unloading, effectively improving the accuracy and reliability of unloading timing judgment, preventing insufficiently mixed material from entering subsequent processes, and ensuring the consistency and stability of product quality.

[0046] When bubble feature units are detected in the image, the mixing speed is automatically reduced.

[0047] This rule reduces gas entrapment caused by excessive stirring speed by identifying abnormal features (bubbles) in the image in real time and adjusting the speed accordingly. This lowers the probability of bubble formation and allows existing bubbles more time to escape, improving the physical state of the mixed materials. It also prevents bubbles from negatively impacting the uniformity of mixing and subsequent reaction processes, thus enhancing the safety and stability of the mixing process.

[0048] When the unloading conditions are not met for an extended period, liquid level data is collected, and the loading speed is automatically reduced based on the liquid level data to increase the mixing speed and optimize the mixing effect.

[0049] This rule addresses the abnormal situation of low mixing efficiency by judging the matching between the loading amount and the mixing effect through liquid level data. Under the premise of keeping the total amount of material in the reactor stable, the loading amount is reduced and the stirring intensity is increased, so that the material can be more fully mixed under the existing stock. This avoids the continuous substandard mixing uniformity caused by overloading or insufficient stirring, and achieves dynamic synergistic optimization of loading and stirring parameters, thereby improving production efficiency and mixing quality.

[0050] When the liquid level in the reactor reaches the preset full-fill threshold, the feeding process should be stopped immediately.

[0051] This rule establishes a safety protection mechanism for the loading process by imposing a strict limit on the liquid level threshold. It prevents safety hazards such as overpressure in the reactor and material overflow caused by overloading, ensuring that the equipment operates within its designed capacity range. At the same time, it avoids interference with the stirring effect and image detection accuracy caused by a full liquid state, thus ensuring the safety of the production process and the reliability of the detection data.

[0052] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0053] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0054] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0055] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0056] Example 2

[0057] like Figures 6-9 As shown, this embodiment proposes an automatic filling detection and control device for a reaction vessel, comprising: The image acquisition module acquires image data of the material mixing state at different locations within the reactor at preset time intervals; The image acquisition module collects image data of the material mixing state at different locations within the reactor at preset time intervals. This enables multi-dimensional, real-time dynamic monitoring of the material mixing process, acquiring image information covering different areas within the reactor. This provides comprehensive and timely data support for subsequent image analysis, avoiding misjudgments of the mixing state caused by single perspectives or non-real-time detection, and effectively improving the accuracy and reliability of the detection results.

[0058] The image preprocessing module performs image preprocessing on the acquired color images to enhance the feature recognition of material particles in the images; The image preprocessing module performs image preprocessing on the acquired color images to enhance the feature recognition of material particles. Through operations such as grayscale conversion, noise reduction, and edge detection, it optimizes image quality, removes noise interference, and highlights key features such as the edges and textures of material particles, making the material particles in the image more significantly different from the background. This provides high-quality input data for the image feature processing module, ensuring that subsequent machine learning algorithms can accurately extract effective features and improve the accuracy and efficiency of feature analysis.

[0059] The image feature processing module uses machine learning algorithms to extract basic features from the preprocessed image, classifies the features using a classification model, identifies feature units that meet the mixing uniformity standard, and uses a clustering algorithm to count the number of feature units. The image feature processing module uses machine learning algorithms to extract basic features from the preprocessed image, identifies feature units that meet the mixing uniformity standard through a classification model, and counts their number using a clustering algorithm. It can automatically learn multi-level features of material particles (such as edges, shapes, and agglomeration) with the help of algorithms such as convolutional neural networks, and achieve accurate classification of feature units by combining classification models such as support vector machines. Then, it uses the K-means clustering algorithm to spatially group and count the number of feature units, thereby realizing the quantitative analysis of material mixing uniformity, replacing manual subjective judgment, and improving the objectivity, scientificity, and automation level of detection.

[0060] The control module, based on the statistical results of the number of feature units, adopts corresponding control methods for the detected specific image features according to preset rules; The control module adopts corresponding control methods for specific image features based on the statistical results of the number of feature units and preset rules. It can convert the image analysis results into specific control commands in real time, such as triggering unloading based on the number of feature units meeting the standard, adjusting the mixing speed based on bubble features, and optimizing the loading and stirring parameters based on liquid level data. This achieves closed-loop automation from detection to control, responds promptly to abnormal states in the material mixing process, avoids the lag of manual intervention, improves production efficiency while ensuring mixing effect and equipment operation safety, and ensures that the entire filling and mixing process is efficient, stable and reliable.

[0061] In this embodiment, the reactor structure includes a reactor body 10, a liquid tank 1, a solid tank 2, a heating layer 4, a support 5, a solid feeding motor 6, a vibrator 7, an auger 8, and a discharge motor 12. The liquid tank 1 and solid tank 2 are both located on the upper part of the support 5, and the reactor body 10 is located on the lower part of the support 5. The heating layer 4 is fitted over the outside of the liquid tank 1 to maintain the temperature of the liquid inside and ensure normal liquid flow. Two augers 8 are provided, located at the bottom of the solid tank 2 and the bottom of the reactor body 10, respectively. The solid feeding motor 6 is connected to the auger 8 at the bottom of the solid tank 2, and the discharge motor 12 is connected to the auger 8 at the bottom of the reactor body 10. The vibrator 7 is fixed to the outside of the solid tank 2 to assist in the discharge of solid materials.

[0062] In this embodiment, the image acquisition module further includes several detection lenses 13, which are used to acquire images of material mixing inside the reactor. The detection lenses 13 are located at the bottom of the reactor body 10, with their probes facing the interior of the reactor body 10.

[0063] In this embodiment, the control module also includes a stirring motor 9, stirring blades 14, a liquid level sensor, a discharge solenoid ball valve 11, and several loading speed control valves 3. The stirring motor 9 drives the stirring blades 14, which mix the materials in the reactor. The liquid level sensor detects the liquid level in the reactor. The discharge solenoid ball valve 11 controls the discharge of the mixed materials. The loading speed control valves 3 control the loading of materials. The stirring motor 9 is located at the top of the reactor body 10. The stirring blades 14 are connected to the drive shaft of the stirring motor 9. Both the stirring blades 14 and the liquid level sensor are located inside the reactor body 10. The discharge solenoid ball valve 11 is located between the reactor body 10 and the auger 8 at the bottom of the reactor body 10. The loading speed control valves 3 are located at the discharge end of the liquid tank 1.

[0064] The automatic filling and detection control device for a reactor provided in Embodiment 2 of the present invention can realize all the processes of the automatic filling and detection control method for a reactor in Embodiment 1 above. The functions and technical effects of each module in the device are the same as those of the automatic filling and detection control method for a reactor in Embodiment 1 above, and will not be repeated here.

[0065] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An automatic filling detection and control method for a reaction vessel, characterized in that, Includes the following steps: Image data of the material mixing state at different locations inside the reactor are collected at preset time intervals; The acquired color images are preprocessed to enhance the feature recognition of material particles in the images; The preprocessed image is used to extract basic features using machine learning algorithms, the features are classified using a classification model, feature units that meet the mixing uniformity standard are identified, and the number of feature units is counted using a clustering algorithm. Based on the statistical results of the number of feature units, corresponding control methods are adopted for specific image features detected according to preset rules.

2. The automatic filling and detection control method for a reaction vessel according to claim 1, characterized in that, The image preprocessing includes: Grayscale conversion converts a color image into a grayscale image. It simplifies image processing calculations by fusing the values ​​of the red, green, and blue channels, and enhances the grayscale difference between material particles and the background. Noise reduction is achieved by using a Gaussian filtering algorithm to remove noise points from the image. By adjusting the filtering parameters to adapt to the image characteristics of different materials, image clarity is improved. Edge detection uses gradient operators to extract boundary features of material particles in an image and identify areas with significant changes in brightness or color, providing a basis for subsequent material analysis. Contrast adjustment enhances the detail of material particles in the image through histogram equalization algorithm, stretches the range of pixel grayscale distribution, and makes particle distribution and contours easier to identify.

3. The automatic filling and detection control method for a reaction vessel according to claim 1, characterized in that, The machine learning algorithm is a convolutional neural network, and the specific steps are as follows: The preprocessed image is extracted layer by layer through multiple convolutional layers. The first convolutional layer identifies basic features such as the edge and texture of the material particles, and the deep convolutional layers combine the basic features to form composite features such as particle shape and aggregation degree. Pooling layers are used to downsample the feature maps output by convolutional layers, reducing feature dimensions while retaining key information to form abstract feature vectors, which provide input for subsequent classification models. The network parameters were trained based on the labeled material mixing image dataset. By comparing the model's predicted results with the actual mixing state labels, the convolution kernel weights and biases were adjusted to improve the accuracy of feature extraction. The final extracted abstract feature vector is then input into the classification model.

4. The automatic filling and detection control method for a reaction vessel according to claim 3, characterized in that, The classification model uses the support vector machine algorithm, and the specific steps are as follows: The basic features extracted from the preprocessed image are converted into feature vectors, which are then used as input data for the classification model. Based on labeled material mixing state samples, the model classification boundary is optimized by adjusting the regularization parameter, thereby improving the recognition accuracy of feature units that meet the mixing uniformity standard. The classification results output by the model are used to identify whether the feature units in the image meet the preset mixing uniformity standard, providing a basis for subsequent feature unit quantity statistics.

5. The automatic filling and detection control method for a reaction vessel according to claim 4, characterized in that, The clustering algorithm used is the K-means clustering algorithm, and the specific steps are as follows: The feature units classified by the support vector machine are grouped according to attributes such as spatial location and size, and K cluster centers are initialized. Calculate the distance between each feature unit and each cluster center, assign it to the nearest cluster, and recalculate the cluster center position. Repeat this process until the cluster center position no longer changes significantly. The number of feature units that meet the mixing uniformity standard and the total number of feature units in each cluster are counted and output to the subsequent judgment process.

6. The automatic filling and detection control method for a reaction vessel according to claim 1, characterized in that, The control method for specific detected image features based on preset rules includes: When the number of consecutive n feature units in the images captured by each lens reaches a preset threshold and the images at different locations all reach the preset threshold, the automatic unloading process is triggered. When bubble feature units are detected in the image, the mixing speed is automatically reduced.

7. The automatic filling and detection control method for a reaction vessel according to claim 6, characterized in that, The corresponding control methods also include: When the unloading conditions are not met for an extended period of time, liquid level data is collected, and the loading speed is automatically reduced and the mixing speed is increased based on the liquid level data to optimize the mixing effect. When the liquid level in the reactor reaches the preset full-fill threshold, the feeding process should be stopped immediately.

8. An automatic filling detection and control device for a reaction vessel, characterized in that, include: The image acquisition module acquires image data of the material mixing state at different locations within the reactor at preset time intervals; The image preprocessing module performs image preprocessing on the acquired color images to enhance the feature recognition of material particles in the images; The image feature processing module uses machine learning algorithms to extract basic features from the preprocessed image, classifies the features using a classification model, identifies feature units that meet the mixing uniformity standard, and uses a clustering algorithm to count the number of feature units. The control module, based on the statistical results of the number of feature units, adopts corresponding control methods for the specific image features detected according to preset rules.

9. The automatic filling and detection control device for a reaction vessel according to claim 8, characterized in that, The image acquisition module also includes several detection lenses, which are used to acquire images of material mixing inside the reactor.

10. An automatic filling and detection control device for a reaction vessel according to claim 9, characterized in that, The control module also includes a stirring motor, stirring blades, a liquid level sensor, a discharge solenoid ball valve, and several loading speed control valves. The stirring motor is used to drive the stirring blades, which are used to mix materials in the reactor. The liquid level sensor is used to detect the liquid level of the materials in the reactor. The discharge solenoid ball valve is used to control the discharge of the mixed materials. The loading speed control valves are used to control the loading of materials.

Images