A method for controlling particle size of refractory materials

Through the convolutional neural network model, the particle size of the refractory material is automatically controlled, which solves the complex operation and error problems caused by manual sampling in traditional methods, real-time and accurate particle size control is achieved, and production efficiency and product quality are improved. .

CN118616341BActive Publication Date: 2024-12-03LIAONING GOODE NEW MATERIAL TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411082407.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-12-03
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

In the production process of traditional refractory materials, particle size control relies on manual sampling and laboratory testing, resulting in complex operations, long cycles, susceptible to human errors, and it is difficult to achieve accurate and real-time control.

Method used

The convolutional neural network model is used to perform image analysis on the sample image data, generate particle size distribution data, and determine whether it meets the set standards through the distribution judgment model, automatically mark unqualified areas and take adjustment measures.

Benefits of technology

Real-time particle size control is achieved, time saving, production efficiency is improved, product stability and reliability are ensured, unqualified products are reduced, costs are reduced, and corporate competitiveness is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118616341B_ABST
    Figure CN118616341B_ABST
Patent Text Reader

Abstract

The invention discloses a refractory material particle size control method, and relates to the field of particle size control. The refractory material particle size control method obtains sample image data in a set area during the refractory material production process and performs preprocessing; performs image analysis on the preprocessed sample image data based on a convolutional neural network model to obtain material type data; analyzes the material type data to obtain type feature data; performs statistical analysis based on the material type data and the type feature data, and generates sample material particle size distribution data. The invention automatically obtains and processes sample image data, performs image analysis using a convolutional neural network model, and realizes real-time particle size control. Compared with traditional manual sampling and laboratory testing, this automated process saves a lot of time and improves production efficiency. Real-time monitoring and automated processing during the production process shorten the production cycle and significantly improve production efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of particle size control, in particular to a method for controlling the particle size of a refractory material. Background Art

[0002] In the production process of refractory materials, the control of particle size is crucial. Refractory materials are mainly used in high-temperature environments, such as industrial furnaces, smelting furnaces, etc., so their performance directly affects the stability and service life of the equipment. The particle size has an important influence on the physical properties of refractory materials such as density, thermal conductivity, and wear resistance, so the control of particle size is extremely demanding.

[0003] In the traditional refractory production process, particle size control is usually achieved through mechanical means such as screening. However, the traditional mechanical screening method has problems such as low efficiency and low precision, and cannot meet the needs of fine particle size control. With the continuous advancement of science and technology, people's performance requirements for refractory materials are getting higher and higher, so a more efficient and accurate particle size control method is needed.

[0004] In the prior art, particle size control in the traditional refractory production process usually relies on manual sampling, laboratory testing and manual adjustment, which results in complex operation, long cycle time and susceptibility to human errors, which in turn makes it difficult to achieve accurate and real-time particle size control. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for controlling the particle size of refractory materials, which solves the problem that in the traditional refractory material production process, particle size control usually relies on manual sampling, laboratory testing and manual adjustment, resulting in complex operation, long cycle time and susceptibility to human errors, which in turn makes it difficult to achieve accurate and real-time particle size control.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for controlling the particle size of refractory materials, comprising the following steps: obtaining sample image data in a set area during the production process of refractory materials and performing preprocessing; performing image analysis on the preprocessed sample image data based on a convolutional neural network model to obtain material type data; analyzing the material type data to obtain type characteristic data; performing statistical analysis based on the material type data and the type characteristic data, and generating sample material particle size distribution data; judging whether the sample material particle size distribution data meets the set standards based on a distribution discrimination model, and marking the corresponding areas where the sample material particle size distribution data does not meet the set standards as unqualified areas, and performing screening and recovery, and taking material delivery adjustment measures.

[0007] Furthermore, the sample image data specifically refers to the voxel value of each voxel block in a sample three-dimensional image made using refractory raw materials in a set area during the refractory material production process, the material type data specifically refers to various types of refractory raw materials of different sizes obtained through image analysis of the refractory raw materials, the type characteristic data specifically refers to the average particle diameter value of each type of refractory raw material, and the sample material particle size distribution data specifically refers to the proportion of refractory raw materials in each particle diameter range.

[0008] Further, sample image data is obtained in a set area during the refractory material production process, and preprocessed as follows: the voxel value of each voxel block in the sample three-dimensional image is read and arranged in ascending order to obtain the maximum voxel value and the minimum voxel value in the sample image data; based on the maximum voxel value, the minimum voxel value and the contrast enhancement model, the voxel value of each voxel block in the sample three-dimensional image is subjected to contrast enhancement analysis to obtain the voxel processing value of each voxel block; wherein the contrast enhancement model is as follows: ;in, is the voxel processing value of the voxel block after contrast enhancement analysis, is the voxel value of the voxel block, is the maximum voxel value in the small sample image data, is the minimum voxel value in the small sample image data, is the maximum range of voxel values.

[0009] Further, the pre-processed sample image data is analyzed based on the convolutional neural network model to obtain the material type data, specifically: read the voxel processing value of each voxel block in the sample image data, and input it into the pre-trained convolutional neural network model for material identification, and obtain various types of refractory raw materials with different sizes; the pre-training steps of the convolutional neural network model are as follows: obtain a voxel value data set, and label the category and normalize each voxel value in the voxel value data set; divide the voxel value data set after the labeled category and normalization into a voxel training data set and a voxel verification training set; initialize the parameters of the convolutional neural network model, read the voxel training data set to train the convolutional neural network model, and calculate the voxel loss function, which is specifically a cross entropy loss function; analyze the gradient of the voxel loss function to each parameter based on the back propagation algorithm; iteratively train the convolutional neural network model, and evaluate the model performance based on the optimization algorithm and the voxel verification training set, and adjust the model parameters according to the verification results until the various types of refractory raw materials classified by size predicted by the model are consistent with the various types of refractory raw materials actually classified by size.

[0010] Furthermore, the input into the pre-trained convolutional neural network model for material identification is specifically as follows: the voxel processing value of each voxel block in the small sample image data is read and processed based on the three-dimensional convolution layer in the convolutional neural network model, specifically: each convolution layer convolves the input voxel processing with the convolution kernel to generate multiple feature maps; after each convolution operation, the ReLU activation function is applied to introduce nonlinear factors; for the multiple feature maps generated, feature dimensionality reduction is performed in the pooling layer; the feature map processed by multiple layers of convolution and pooling is flattened in the fully connected layer of the convolutional neural network model; the output of the last fully connected layer is processed based on the Softmax function in the Softmax layer to predict the probability of each refractory raw material category.

[0011] Further, the material type data is analyzed to obtain the type characteristic data as follows: the various types of refractory raw materials obtained by the convolutional neural network model analysis are read for quantitative analysis to obtain the geometric center of the particle voxel and the number of voxel blocks of each type of refractory raw material; based on the geometric center of the particle voxel and the number of voxel blocks, the diameter of each type of refractory raw material is analyzed to obtain the average particle diameter value of each type of refractory raw material, wherein the specific calculation formula of the average particle diameter value is as follows: ;in, is the average particle diameter of the refractory raw material, is the geometric center of the particle voxel of the refractory raw material, It is the first refractory raw material The voxel value of each voxel block, , is the total number of voxel blocks in the refractory raw material.

[0012] Furthermore, statistical analysis is performed based on material type data and type characteristic data, and sample material particle size distribution data is generated, specifically: various types of refractory raw materials are read and the total number of particles of the refractory raw materials is counted; the average particle diameter value of each type of refractory raw material is divided according to the set particle diameter range, and the number of refractory raw material particles in each particle diameter range is obtained; based on the number of refractory raw material particles in each particle diameter range and the total number of refractory raw material particles, a proportion analysis is performed to obtain the proportion of refractory raw materials in each particle diameter range.

[0013] Furthermore, judging whether the particle size distribution data of the sample material meets the set standard based on the distribution discrimination model is as follows: performing deviation analysis on the distribution of the proportion of refractory raw materials within the particle diameter interval and the preset standard refractory raw material proportion distribution based on the Euclidean distance analysis, and performing comparative analysis based on the distribution discrimination model to judge whether the distribution of the proportion of refractory raw materials within the particle diameter interval meets the set standard; the deviation analysis formula is as follows: ;in, It is the deviation between the proportion distribution of refractory raw materials within the particle diameter range and the preset standard proportion distribution of refractory raw materials. For the preset The proportion of standard refractory raw materials within a particle diameter range.

[0014] Furthermore, the distribution discrimination model is specifically as follows: ;in, The deviation threshold is set. The specific discrimination logic of the distribution discrimination model is that if the deviation value between the distribution of the proportion of refractory raw materials within the particle diameter range and the preset standard distribution of the proportion of refractory raw materials is Less than or equal to the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range meets the set standard. If the deviation value between the distribution of the proportion of refractory raw materials within the particle diameter range and the preset standard distribution of refractory raw materials Greater than the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range does not meet the set standards.

[0015] Furthermore, the specific steps of taking material delivery adjustment measures are as follows: based on the material delivery adjustment model, the proportion of refractory raw materials in each current particle diameter interval is adjusted and analyzed to obtain the adjusted delivery ratio of refractory raw material particles in each particle diameter interval; wherein the material delivery adjustment model is specifically: ;in, For the Adjust the proportion of refractory raw material particles within the particle diameter range. is the adjustment factor, is the particle adjustment factor, To adjust sensitive parameters.

[0016] The present invention has the following beneficial effects:

[0017] (1) This refractory material particle size control method realizes real-time particle size control by automatically acquiring and processing small sample image data and performing image analysis using a convolutional neural network model. Compared with traditional manual sampling and laboratory testing, this automated process saves a lot of time and improves production efficiency. Real-time monitoring and automated processing during the production process shorten the production cycle and significantly improve production efficiency.

[0018] (2) This refractory material particle size control method can timely discover problems in the production process and take adjustment measures through precise particle size control and automatic identification of unqualified areas. This refined quality control ensures the stability and reliability of the product, reduces the production and circulation of unqualified products, and improves the product quality level.

[0019] (3) This refractory material particle size control method reduces manpower and time costs through an automated particle size control process, while avoiding waste and rework caused by unqualified products. This cost-saving effect has a significant positive impact on the profitability and competitive advantage of the enterprise. By effectively controlling various costs in the production process, the enterprise can improve production efficiency, reduce production costs, and thus enhance competitiveness and profitability.

[0020] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flow chart of a method for controlling the particle size of refractory materials.

[0022] Figure 2 The present invention is a specific flow chart of the pre-training steps of a convolutional neural network model in a method for controlling refractory material particle size. DETAILED DESCRIPTION

[0023] The embodiment of the present application provides a method for controlling the particle size of refractory materials, thereby solving the problem that particle size control in traditional refractory material production processes usually relies on manual sampling, laboratory testing and manual adjustment, resulting in complex operations, long cycle times, and susceptibility to human errors, which in turn makes it difficult to achieve accurate and real-time particle size control.

[0024] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0025] First, by acquiring and processing the sample image data, the convolutional neural network model is used to perform image analysis, and then the material type and characteristic data are analyzed to generate the particle size distribution data. Then, the distribution discrimination model is used to determine whether the particle size data meets the standard. If not, the unqualified area is marked and adjustment measures are taken. The specific steps are: first, the sample image data is acquired and processed, and then the convolutional neural network model is used to perform image analysis. Then, the material type and characteristic data are analyzed, and then the particle size distribution data is generated by statistical analysis. Finally, the distribution discrimination model is used to determine and take adjustment measures.

[0026] See also Figure 1The embodiment of the present invention provides a technical solution: a method for controlling the particle size of refractory materials, comprising the following steps: obtaining sample image data in a set area during the production process of refractory materials and performing preprocessing; performing image analysis on the preprocessed sample image data based on a convolutional neural network model to obtain material type data; analyzing the material type data to obtain type characteristic data; performing statistical analysis based on the material type data and the type characteristic data, and generating sample material particle size distribution data; judging whether the sample material particle size distribution data meets the set standards based on a distribution discrimination model, marking the corresponding area where the sample material particle size distribution data does not meet the set standards as an unqualified area, and performing screening and recovery, and taking material delivery adjustment measures at the same time.

[0027] The sample image data specifically refers to the voxel value of each voxel block in the sample three-dimensional image made of refractory raw materials in a set area during the refractory material production process; the material type data specifically refers to the various types of refractory raw materials of different sizes obtained through image analysis of the refractory raw materials; the type characteristic data specifically refers to the average particle diameter value of each type of refractory raw material; and the sample material particle size distribution data specifically refers to the proportion of refractory raw materials in each particle diameter range.

[0028] Specifically, sample image data is obtained in a set area during the refractory material production process, and preprocessed as follows: the voxel value of each voxel block in the sample three-dimensional image is read and arranged in ascending order to obtain the maximum voxel value and the minimum voxel value in the sample image data; based on the maximum voxel value, the minimum voxel value and the contrast enhancement model, the voxel value of each voxel block in the sample three-dimensional image is subjected to contrast enhancement analysis to obtain the voxel processing value of each voxel block; wherein the contrast enhancement model is as follows: ;in, is the voxel processing value of the voxel block after contrast enhancement analysis, is the voxel value of the voxel block, is the maximum voxel value in the small sample image data, is the minimum voxel value in the small sample image data, For example, if each voxel value is 8 bits, this means that each voxel can take values ​​from 0 to 255, so , if each voxel value is 16 bits, the voxel value range is from 0 to 65535, so .

[0029] Among them, the voxel value contrast enhancement analysis calculation data example is as follows:

[0030] Table 1 Example of voxel value contrast enhancement analysis calculation data

[0031]

[0032] In Table 1, is the voxel processing value of the voxel block after contrast enhancement analysis, is the voxel value of the voxel block, is the maximum voxel value in the small sample image data, is the minimum voxel value in the small sample image data, The maximum range for voxel values ​​is 256.

[0033] The first set of data: the voxel value of the voxel block with coordinates (1, 1, 1): 100; the maximum voxel value in the small sample image data: 300; the minimum voxel value in the small sample image data: 100; the maximum range of voxel values: 256; the voxel processing value of the voxel block with coordinates (1, 1, 1) after contrast enhancement analysis: 0.

[0034] The second set of data: the voxel value of the voxel block with coordinates (1, 1, 2): 150; the maximum voxel value in the small sample image data: 300; the minimum voxel value in the small sample image data: 100; the maximum range of voxel values: 256; the voxel processing value of the voxel block with coordinates (1, 1, 2) after contrast enhancement analysis: 85.

[0035] The third set of data: the voxel value of the voxel block with coordinates (1, 1, 3): 200; the maximum voxel value in the small sample image data: 300; the minimum voxel value in the small sample image data: 100; the maximum range of voxel values: 256; the voxel processing value of the voxel block with coordinates (1, 1, 3) after contrast enhancement analysis: 170.

[0036] The fourth set of data: the voxel value of the voxel block with coordinates (1, 1, 4): 250; the maximum voxel value in the small sample image data: 300; the minimum voxel value in the small sample image data: 100; the maximum range of voxel values: 256; the voxel processing value of the voxel block after contrast enhancement analysis of the voxel block with coordinates (1, 1, 4): 255.

[0037] The fifth set of data: the voxel value of the voxel block with coordinates (1, 1, 5): 300; the maximum voxel value in the small sample image data: 300; the minimum voxel value in the small sample image data: 100; the maximum range of voxel values: 256; the voxel processing value of the voxel block after contrast enhancement analysis of the voxel block with coordinates (1, 1, 5): 255.

[0038] In this embodiment, during the refractory material production process, sample image data is acquired in a set area and preprocessed, which includes reading the voxel value of each voxel block and arranging them in ascending order, thereby obtaining the maximum voxel value and the minimum voxel value in the sample image data. Through preprocessing, key information in the image data can be effectively extracted to provide an accurate data basis for subsequent contrast enhancement analysis. Based on the maximum voxel value, the minimum voxel value and the contrast enhancement model, the voxel value of each voxel block in the sample image data is subjected to contrast enhancement analysis to obtain the voxel processing value of each voxel block. Contrast enhancement can make subtle features in the image more clearly visible. , which helps to accurately analyze and control the particle size of refractory materials. This analysis can improve the accuracy and stability of particle size in the production process, thereby improving the performance and quality of refractory materials. The specific calculation of the model includes standardizing the voxel values ​​of the voxel block to a predetermined range, and performing contrast enhancement according to the dynamic range of the image data, so that the visual effect of the image is more prominent. The contrast enhancement model can dynamically adjust the degree of enhancement according to the characteristics of the image data, so that the contrast enhancement effect is more in line with actual needs. The application of this model makes the particle size control method more intelligent and adaptable, and can adapt to changes in different production environments and raw material characteristics.

[0039] Specifically, Figure 2 As shown, the pre-processed small sample image data is analyzed based on the convolutional neural network model to obtain the material type data, specifically: read the voxel processing value of each voxel block in the small sample image data, and input it into the pre-trained convolutional neural network model for material identification, and obtain various types of refractory raw materials with different sizes; the pre-training steps of the convolutional neural network model are as follows: obtain a voxel value data set from the three-dimensional image of refractory material production, and annotate and normalize each voxel value in the voxel value data set, that is, refractory materials of different sizes; divide the annotated category and normalized voxel value data set into a voxel training data set and voxel validation training set; initialize the parameters of the convolutional neural network model, usually randomly initialize the weights and biases, read the voxel training data set to train the convolutional neural network model, and calculate the voxel loss function, which is specifically the cross entropy loss function; analyze the gradient of the voxel loss function for each parameter based on the back propagation algorithm; iteratively train the convolutional neural network model, and evaluate the model performance based on optimization algorithms such as SGD, Adam, etc. and the voxel validation training set, and adjust the model parameters according to the verification results until the various types of refractory raw materials classified by size predicted by the model are consistent with the various types of refractory raw materials actually classified by size.

[0040] The input and input data of the convolutional neural network model are voxel value data and the probability of predicting each refractory raw material category, respectively. The convolutional neural network model includes: an input layer for receiving preprocessed three-dimensional voxel data as input, a three-dimensional convolution layer for extracting local features in the voxel data, by sliding the three-dimensional convolution kernel on the volume data, calculating the dot product of the convolution kernel and the voxel, these layers can capture the shape, size and other spatial features of the particles, an activation layer for introducing nonlinearity so that the model can learn more complex feature representations, specifically applying the nonlinear function ReLU to the output of the convolution layer to enhance the expression ability of the model, a pooling layer for reducing the spatial size of the feature map and improving the displacement and deformation invariance of the feature, specifically performing maximum pooling or average pooling on the three-dimensional feature map, simplifying network calculations, and reducing overfitting, a fully connected layer for mapping the learned high-dimensional features to the final classification result, specifically after the feature map is flattened, it is processed through one or more fully connected layers to output the score of each category, and an output layer for converting the output of the fully connected layer into a probability distribution, specifically using the Softmax function to process the result of the output layer to ensure that the output is a valid probability distribution.

[0041] In this embodiment, a voxel value dataset is obtained from a three-dimensional image of refractory material production, and different types of refractory materials are labeled and the dataset is normalized. The dataset is then divided into a training set and a validation set for model training and evaluation. The pre-training step makes full use of the existing dataset so that the model can converge faster in subsequent training and can effectively improve the recognition accuracy of the model for different types of refractory materials. The convolutional neural network model consists of an input layer, a three-dimensional convolutional layer, an activation layer, a pooling layer, a fully connected layer and an output layer. The input layer receives the pre-processed three-dimensional voxel data, the convolutional layer is used to extract local features, the activation layer introduces nonlinearity, the pooling layer reduces the spatial size of the feature map, the fully connected layer processes the features and outputs the score of each category, and the output layer uses The Softmax function converts the score into a probability distribution. The CNN model gradually extracts features from the image by stacking layers, and can automatically learn the abstract features in the data, thus having a strong classification ability. In addition, convolution operations and pooling operations can effectively reduce the number of parameters, reduce the complexity of the model, and improve the generalization ability of the model. The model is trained through iterative training and optimization algorithms such as SGD and Adam, and the voxel loss function is calculated and the model parameters are updated through the back propagation algorithm until the model's prediction results are consistent with the actual types of refractory raw materials divided by size. The training process continuously adjusts the model parameters so that the model can gradually improve the recognition ability of different types of refractory materials, thereby improving the accuracy and reliability of refractory material type recognition.

[0042] Specifically, the input into the pre-trained convolutional neural network model for material identification is as follows: the voxel processing value of each voxel block in the sample image data is read and processed based on the three-dimensional convolution layer in the convolutional neural network model. Specifically, each convolution layer convolves the input voxel processing with the convolution kernel to generate multiple feature maps, each of which represents a type of feature in the image. These layers automatically learn to capture particle features by sliding a multidimensional filter, i.e., a convolution kernel, in three-dimensional space. The specific formula is: ;in, is the new voxel value after the convolution operation, is the input voxel value, is the weight of the convolution kernel; after each convolution operation, the ReLU activation function is applied to introduce nonlinear factors, so that the network can learn and express more complex patterns, avoid the network output being just a linear combination of the input, and increase the expressive power of the model. The specific formula is as follows: ;in, is the original value output by the convolutional layer, , is the output of the ReLU activation function, which sets all negative values ​​to 0, thereby introducing nonlinearity and allowing the network to learn complex functions. For the generated multiple feature maps, feature dimensionality reduction is performed in the pooling layer, specifically sliding a window on the feature map and taking the maximum value from each window to reduce the feature dimension and enhance the displacement invariance of the feature. The specific formula is as follows: ;in, is the position in the voxel block arrive The voxel set of the interval, is the size of the pooling window, usually a positive integer such as 2 or 3, which means considering adjacent Individual elements, After the pooling operation, the position The result is the maximum voxel value in the selected area; the feature map that has undergone multi-layer convolution and pooling processing is flattened in the fully connected layer of the convolutional neural network model, specifically: it is processed through one or more fully connected layers to map the extracted high-dimensional feature vector to the final classification output; the output of the last fully connected layer is processed based on the Softmax function in the Softmax layer to predict the probability of each refractory raw material category. The Softmax layer ensures that the output is a valid probability distribution, and the sum of the predicted probabilities of each category is 1.

[0043] In this embodiment, a convolution operation is performed by processing the convolution kernel with the input voxel to generate multiple feature maps, which represent different features in the image. The convolution operation can effectively extract local features in the image, so that the model can better capture the shape, size and other spatial features of the refractory particles, thereby improving the accuracy and robustness of recognition. The ReLU activation function is applied after each convolution operation to introduce nonlinear factors, increase the expressive power of the model, and avoid the output being merely a linear combination of the input. The use of the ReLU activation function can enable the model to have better nonlinear fitting capabilities, so that it can learn more complex and abstract features, further improving the performance of the model, and performing a pooling operation on the generated feature map. The operation is to take the maximum value in each window through the sliding window, reduce the feature dimension, and enhance the displacement invariance of the feature. The pooling operation can effectively reduce the feature dimension and reduce the computational complexity. At the same time, it enhances the robustness of the model to image displacement and deformation, making the model more stable and reliable. The feature map processed by convolution and pooling is flattened, and the feature vector is mapped to the final classification output through the fully connected layer. Finally, the probability of each refractory raw material category is predicted through the Softmax function. The fully connected layer can convert the high-dimensional feature vector into a classification output, so that the model can directly predict the type of refractory material. The use of the Softmax function can ensure that the output is a valid probability distribution, making the model output more interpretable and credible.

[0044] Specifically, the material type data is analyzed to obtain the type characteristic data as follows: the various types of refractory raw materials obtained by the convolutional neural network model analysis are read for quantitative analysis to obtain the geometric center of the particle voxel and the number of voxel blocks of each type of refractory raw material; based on the geometric center of the particle voxel and the number of voxel blocks, the diameter of each type of refractory raw material is analyzed to obtain the average particle diameter value of each type of refractory raw material, wherein the specific calculation formula of the average particle diameter value is as follows: ;in, is the average particle diameter of the refractory raw material, is the geometric center of the particle voxel of the refractory raw material, It is the first refractory raw material The voxel value of each voxel block, , is the total number of voxel blocks in the refractory raw material.

[0045] In this implementation scheme, various types of refractory raw materials obtained by reading the convolutional neural network model analysis are subjected to quantitative analysis to obtain the geometric center of the particle voxel and the number of voxel blocks of each type of refractory raw material. The quantitative analysis can help understand the distribution of each type of refractory raw material in the sample, including the distribution density and distribution range of the particle voxel, and provide a data basis for subsequent diameter analysis. Based on the geometric center of the particle voxel and the number of voxel blocks, diameter analysis is performed on various types of refractory raw materials to obtain the average particle diameter value of each type of refractory raw material. The diameter analysis can quantify the size characteristics of each type of refractory raw material particles, and provide an important reference for particle size control in the production process. By analyzing the average particle diameter of different types of refractory materials, the physical properties of the materials can be better understood, the production process can be optimized, and the performance and quality of the materials can be improved.

[0046] Specifically, statistical analysis is performed based on material type data and type characteristic data, and sample material particle size distribution data is generated as follows: various types of refractory raw materials are read and the total number of particles of the refractory raw materials is counted; the average particle diameter value of each type of refractory raw material is divided according to the set particle diameter interval to obtain the number of refractory raw material particles in each particle diameter interval; based on the number of refractory raw material particles in each particle diameter interval and the total number of refractory raw material particles, a proportion analysis is performed to obtain the proportion of refractory raw materials in each particle diameter interval, and the specific formula is: ;in, For the The proportion of refractory raw materials in the particle diameter range, For the The number of refractory raw material particles within a particle diameter range, is the total number of particles of refractory raw materials, , is the total number of set particle diameter intervals.

[0047] In this implementation scheme, various types of refractory raw materials are read and the total number of particles of each type of refractory raw materials is counted. The total number of particles can help understand the quantity distribution of each type of refractory raw materials in the sample, and provide basic data for subsequent particle size distribution analysis. The average particle diameter value of each type of refractory raw material is divided according to the set particle diameter interval to obtain the number of refractory raw material particles in each particle diameter interval. The particle average diameter interval division can classify the particles according to the size range, making the subsequent analysis clearer and more specific. Based on the proportion analysis of the number of refractory raw material particles in each particle diameter interval and the total number of refractory raw material particles, the value of each particle diameter interval is obtained. The proportion of refractory raw materials in the production process can be analyzed by the proportion analysis, which can intuitively reflect the distribution of refractory raw materials in different particle diameter ranges, and help to evaluate the uniformity and range of material particle size distribution. Through this statistical analysis method, the particle size distribution data of small sample materials can be generated, so as to fully understand the particle size distribution of various refractory raw materials. This data analysis can provide an important basis for particle size control and optimization in the production process, help to adjust the production process, improve product quality, and increase production efficiency. Therefore, this analysis method has an important contribution to solving technical problems in the production process of refractory materials, and can provide scientific support for the refined management and optimization of the production process.

[0048] Specifically, judging whether the particle size distribution data of the sample material meets the set standard based on the distribution discriminant model is as follows: based on the Euclidean distance analysis, the distribution of the proportion of refractory raw materials in the particle diameter range is compared with the preset standard distribution of refractory raw materials, and a comparative analysis is performed based on the distribution discriminant model to judge whether the distribution of the proportion of refractory raw materials in the particle diameter range meets the set standard; the deviation analysis formula is as follows: ;in, It is the deviation between the proportion distribution of refractory raw materials within the particle diameter range and the preset standard proportion distribution of refractory raw materials. For the preset The proportion of standard refractory raw materials within a particle diameter range.

[0049] In this embodiment, the deviation analysis is performed by calculating the Euclidean distance between the distribution of the proportion of refractory raw materials in the particle diameter range and the preset standard refractory raw material proportion distribution. As a commonly used distance measurement method, the Euclidean distance can intuitively quantify the degree of difference between the distribution of the proportion of refractory raw materials in the particle diameter range and the standard. Through the deviation analysis, the deviation between the actual distribution and the expected standard can be clearly understood. Based on the results of the deviation analysis, a distribution discrimination model is used for comparative analysis to determine whether the distribution of the proportion of refractory raw materials in the particle diameter range meets the set standard. The distribution discrimination model can perform a deeper analysis and evaluation of the data based on the results of the Euclidean distance analysis to determine whether the material particle size distribution meets the set standard. Such comparative analysis can provide guidance for adjustments and optimizations in the production process to ensure that product quality meets the requirements. Through this analysis method based on the distribution discrimination model, the particle size distribution data of the sample material can be comprehensively evaluated to determine whether it meets the set standard requirements. This method can timely discover the particle size distribution deviation, provide effective guidance and basis for the adjustment and control of the production process, thereby ensuring stable and reliable product quality.

[0050] Specifically, the distribution discriminant model is as follows: ;in, The deviation threshold is set. The specific judgment logic of the distribution judgment model is that if the deviation value between the proportion distribution of refractory raw materials within the particle diameter range and the preset standard refractory raw materials proportion distribution is Less than or equal to the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range meets the set standard. If the deviation value between the distribution of the proportion of refractory raw materials within the particle diameter range and the preset standard distribution of refractory raw materials Greater than the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range does not meet the set standards.

[0051] In this implementation scheme, the model first needs to set a deviation threshold value, which can be formulated according to actual needs and standards, and is used to determine whether the deviation between the proportion distribution of refractory raw materials and the preset standard is within an acceptable range. According to the deviation threshold value, the model determines whether the proportion distribution of refractory raw materials meets the set standard. The specific logic is that if the deviation value between the proportion distribution of refractory raw materials in the particle diameter range and the preset standard is less than or equal to the set deviation threshold value, the output is "satisfied"; otherwise, the output is "not satisfied". This judgment logic is simple and clear, easy to understand and operate. By setting a reasonable deviation threshold value, the proportion distribution of refractory raw materials can be quickly judged, thereby timely discovering the deviation of particle size distribution, which is beneficial to the production process. The contribution of this distribution discrimination model is that it provides a simple and effective method to evaluate whether the particle size distribution of refractory materials meets the standard requirements. By quickly and accurately judging whether the material proportion distribution meets the standard, it can help production personnel to adjust production parameters in time, optimize production processes, ensure that product quality meets the requirements, and improve production efficiency and economic benefits. This distribution discrimination model is of great significance for the quality control and management of the refractory production process. It can effectively help the production department to conduct real-time monitoring and discrimination to ensure that the particle size distribution of the product meets the expected standards. It provides enterprises with a reliable quality management tool, which helps to enhance their competitiveness and market reputation.

[0052] Specifically, the specific steps of taking material delivery adjustment measures are as follows: based on the material delivery adjustment model, the proportion of refractory raw materials in each current particle diameter interval is adjusted and analyzed to obtain the adjusted delivery ratio of refractory raw material particles in each particle diameter interval; wherein the material delivery adjustment model is specifically: ;in, For the Adjust the proportion of refractory raw material particles within the particle diameter range. is the adjustment coefficient, which is used to control the adjustment range. is the particle adjustment factor, To adjust sensitive parameters, and When the difference increases, the adjustment range is reduced to prevent over-adjustment.

[0053] Among them, the calculation data example of the adjusted input ratio of refractory raw material particles in the first particle diameter range is as follows:

[0054] Table 2 Example of calculation data for the adjusted feed ratio of refractory raw material particles within the first particle diameter range

[0055]

[0056] In Table 2, is the proportion of standard refractory raw materials within the preset first particle diameter range, is the proportion of refractory raw materials in the first particle diameter interval, is the adjusted proportion of refractory raw material particles within the first particle diameter interval, is the adjustment coefficient and the value is assigned to 0.5, To adjust the sensitive parameter and assign a value of 0.1, is the particle adjustment factor in the first particle diameter interval.

[0057] The first set of data: the proportion of standard refractory raw materials in the preset first particle diameter range: 0.375; the proportion of refractory raw materials in the first particle diameter range: 0.156; adjustment coefficient: 0.5; adjustment sensitive parameter: 0.1; particle adjustment factor: 0.112; the adjusted proportion of refractory raw material particles in the first particle diameter range: 0.0123, which means increasing the proportion of refractory raw material particles in this diameter range, and the value is 0.0123.

[0058] The second set of data: the proportion of standard refractory raw materials in the preset first particle diameter range: 0.951; the proportion of refractory raw materials in the first particle diameter range: 0.058; adjustment coefficient: 0.5; adjustment sensitive parameter: 0.1; particle adjustment factor: 0.000133; the adjusted delivery ratio of refractory raw material particles in the first particle diameter range: 0.000059, which means increasing the delivery ratio of refractory raw material particles in this diameter range, and the value is 0.000059.

[0059] The third set of data: the proportion of standard refractory raw materials in the preset first particle diameter range: 0.732; the proportion of refractory raw materials in the first particle diameter range: 0.866; adjustment coefficient: 0.5; adjustment sensitive parameter: 0.1; particle adjustment factor: 0.261; the adjusted proportion of refractory raw material particles in the first particle diameter range: -0.0175, which means reducing the proportion of refractory raw material particles in this diameter range, and the value is -0.0175.

[0060] The fourth set of data: the proportion of standard refractory raw materials in the preset first particle diameter range: 0.599; the proportion of refractory raw materials in the first particle diameter range: 0.601; adjustment coefficient: 0.5; adjustment sensitive parameter: 0.1; particle adjustment factor: 0.976; the adjusted delivery ratio of refractory raw material particles in the first particle diameter range: -0.0012, which means reducing the delivery ratio of refractory raw material particles in this diameter range, and the value is -0.0012.

[0061] The fifth set of data: the proportion of standard refractory raw materials in the preset first particle diameter range: 0.156; the proportion of refractory raw materials in the first particle diameter range: 0.708; adjustment coefficient: 0.5; adjustment sensitive parameter: 0.1; particle adjustment factor: 0.004; the adjusted delivery ratio of refractory raw material particles in the first particle diameter range: -0.0011, which means reducing the delivery ratio of refractory raw material particles in this diameter range, and the value is -0.0011.

[0062] In this implementation scheme, based on the material delivery adjustment model, the current proportion of refractory raw materials in each particle diameter range is adjusted and analyzed. Through the adjustment analysis, the proportion of refractory raw materials in the particle diameter range can be optimized according to actual conditions to make it more in line with production requirements and quality standards, thereby improving the overall performance and quality of the product. The adjusted delivery ratio of refractory raw material particles in each particle diameter range is obtained according to the material delivery adjustment model. The obtained adjusted delivery ratio can guide specific operations in production practice and help production personnel accurately adjust the delivery amount of refractory raw materials in each particle diameter range to achieve precise control of particle size distribution. The model specifically includes two parameters, namely the adjustment coefficient and the particle adjustment factor, which are used to control the adjustment amplitude and adjustment sensitivity. The material delivery adjustment model can dynamically adjust the delivery ratio of refractory raw materials in each particle diameter range according to actual conditions. It has certain intelligence and adaptability, can better cope with changes and fluctuations in the production process, and ensure production stability and efficiency.

[0063] In summary, this application has at least the following effects:

[0064] By automatically acquiring and processing small sample image data and using convolutional neural network models for image analysis, real-time particle size control is achieved. Compared with traditional manual sampling and laboratory testing, this automated process saves a lot of time and improves production efficiency. Real-time monitoring and automated processing during the production process shorten the production cycle and significantly improve production efficiency.

[0065] Through precise granularity control and automatic identification of unqualified areas, problems in the production process can be discovered in a timely manner and adjustment measures can be taken. This refined quality control ensures the stability and reliability of the product, reduces the production and circulation of unqualified products, and improves the product quality level.

[0066] The automated granularity control process reduces labor and time costs, while avoiding waste and rework caused by unqualified products. This cost-saving effect has a significant positive impact on the profitability and competitive advantage of the company. By effectively controlling various costs in the production process, the company can improve production efficiency, reduce production costs, and thus enhance competitiveness and profitability.

[0067] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0068] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for controlling the particle size of refractory materials, It is characterized in that The following steps are involved: Acquire sample image data in a set area during the refractory production process and perform preprocessing; Perform image analysis on the preprocessed sample image data based on the convolutional neural network model to obtain material type data; Analyze the material type data to obtain type characteristic data; Conduct statistical analysis based on material type data and type characteristic data, and generate sample material particle size distribution data; Based on the distribution discrimination model, determine whether the particle size distribution data of the sample material meets the set standard, and mark the corresponding area where the particle size distribution data of the sample material does not meet the set standard as an unqualified area, and perform screening and recycling, and take material delivery adjustment measures; The sample image data is specifically the voxel value of each voxel block in the sample three-dimensional image made of refractory raw materials in the set area in the refractory material production process, the material type data is specifically various types of refractory raw materials of different sizes obtained by image analysis of the refractory raw materials, the type characteristic data is specifically the average particle diameter value of each type of refractory raw material, and the sample material particle size distribution data is specifically the proportion of refractory raw materials in each particle diameter range; Acquire sample image data in a set area during the refractory production process and perform preprocessing as follows: Read the voxel value of each voxel block in the small sample three-dimensional image and arrange them in ascending order to obtain the maximum voxel value and the minimum voxel value in the small sample image data; Based on the maximum voxel value, the minimum voxel value and the contrast enhancement model, the voxel value of each voxel block in the small sample three-dimensional image is respectively analyzed for contrast enhancement to obtain the voxel processing value of each voxel block; Among them, the contrast enhancement model is as follows: ; in, is the voxel processing value of the voxel block after contrast enhancement analysis, is the voxel value of the voxel block, is the maximum voxel value in the small sample image data, is the minimum voxel value in the small sample image data, is the maximum range of voxel values.

2. A method for controlling refractory material particle size according to claim 1, Features: Based on the convolutional neural network model, the preprocessed sample image data is analyzed to obtain the material type data as follows: Read the voxel processing value of each voxel block in the sample image data and input it into the pre-trained convolutional neural network model for material recognition to obtain various types of refractory raw materials of different sizes; The pre-training steps of the convolutional neural network model are as follows: Acquire a voxel value dataset, and label a category for each voxel value in the voxel value dataset and perform normalization processing; Divide the labeled categories and the normalized voxel value dataset into a voxel training dataset and a voxel validation training set; Initialize the parameters of the convolutional neural network model, read the voxel training data set to train the convolutional neural network model, and calculate the voxel loss function, wherein the voxel loss function is specifically a cross entropy loss function; Analyze the gradient of the voxel loss function with respect to each parameter based on the back-propagation algorithm; The convolutional neural network model is iteratively trained, and the model performance is evaluated based on the optimization algorithm and voxel validation training set. The model parameters are adjusted according to the validation results until the model's predictions for the various types of refractory raw materials classified by size are consistent with the actual types of refractory raw materials classified by size.

3. A method for controlling refractory material particle size according to claim 2, Features: Input into the pre-trained convolutional neural network model for material recognition is as follows: The voxel processing value of each voxel block in the sample image data is read and processed based on the three-dimensional convolution layer in the convolutional neural network model. Specifically, each convolution layer performs a convolution operation on the input voxel processing and the convolution kernel to generate multiple feature maps; Applying ReLU activation function after each convolution operation introduces nonlinear factors; For the multiple feature maps generated, feature dimension reduction is performed in the pooling layer; Flatten the feature maps processed by multiple layers of convolution and pooling in the fully connected layer of the convolutional neural network model; The output of the last fully connected layer is processed based on the Softmax function in the Softmax layer to predict the probability of each refractory raw material category.

4. A method for controlling refractory material particle size according to claim 2, Features: By analyzing the material type data, the type characteristic data obtained are as follows: Read various types of refractory raw materials obtained by the convolutional neural network model analysis to perform quantitative analysis, and obtain the geometric center of the particle voxel and the number of voxel blocks of each type of refractory raw material; Based on the geometric center of the particle voxel and the number of voxel blocks, the diameter of each type of refractory raw material is analyzed to obtain the average particle diameter value of each type of refractory raw material. The specific calculation formula of the average particle diameter value is as follows: ; in, is the average particle diameter of the refractory raw material, is the geometric center of the particle voxel of the refractory raw material, It is the first refractory raw material The voxel value of each voxel block, , is the total number of voxel blocks in the refractory raw material.

5. A method for controlling refractory material particle size according to claim 4, Features: Based on the material type data and type characteristic data, statistical analysis is performed and the particle size distribution data of the sample material is generated as follows: Read various types of refractory raw materials and count the total number of particles of refractory raw materials; The average particle diameter value of each type of refractory raw material is divided according to the set particle diameter interval to obtain the number of refractory raw material particles in each particle diameter interval; Based on the number of refractory raw material particles in each particle diameter range and the total number of refractory raw material particles, a proportion analysis is performed to obtain the proportion of refractory raw materials in each particle diameter range.

6. A method for controlling refractory material particle size according to claim 5, Features: The distribution discrimination model is used to determine whether the particle size distribution data of the sample material meets the set standards: Based on the Euclidean distance analysis, the distribution of the proportion of refractory raw materials within the particle diameter range is analyzed for deviation from the preset standard distribution of the proportion of refractory raw materials, and a comparative analysis is performed based on the distribution discrimination model to determine whether the distribution of the proportion of refractory raw materials within the particle diameter range meets the set standard; The deviation analysis formula is as follows: ; in, It is the deviation between the proportion distribution of refractory raw materials within the particle diameter range and the preset standard proportion distribution of refractory raw materials. For the preset The proportion of standard refractory raw materials within a particle diameter range.

7. A method for controlling refractory material particle size according to claim 6, Features: The distribution discrimination model is specifically as follows: ; in, The deviation threshold is set. The specific discrimination logic of the distribution discrimination model is that if the deviation value between the distribution of the proportion of refractory raw materials within the particle diameter range and the preset standard distribution of the proportion of refractory raw materials is Less than or equal to the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range meets the set standard. If the deviation value between the distribution of the proportion of refractory raw materials within the particle diameter range and the preset standard distribution of refractory raw materials Greater than the set deviation threshold , then the output , that is, the distribution of the proportion of refractory raw materials within the particle diameter range does not meet the set standards.

8. A method for controlling refractory material particle size according to claim 6, Features: The specific steps to take material delivery adjustment measures are as follows: Based on the material delivery adjustment model, the proportion of refractory raw materials in each current particle diameter range is adjusted and analyzed to obtain the adjusted delivery ratio of refractory raw material particles in each particle diameter range; The material delivery adjustment model is as follows: ; in, For the Adjust the proportion of refractory raw material particles within the particle diameter range. is the adjustment factor, is the particle adjustment factor, To adjust sensitive parameters.

Citation Information

Patent Citations

  • Three-dimensional particle category detection method and system based on convolutional neural network

    CN112001218A

  • Adaptive sorting for particle analyzers

    CN113316713A