CN115909171A

CN115909171A - Method and system for producing steel ladle air brick

Info

Publication number: CN115909171A
Application number: CN202211630797.4A
Authority: CN
Inventors: 沈军华; 沈立宇; 焦宏宇; 林坚; 李顺良; 李马良
Original assignee: Zhejiang Jinhuihua Special Refractories Co ltd
Current assignee: Zhejiang Jinhuihua Special Refractories Co ltd
Priority date: 2022-12-19
Filing date: 2022-12-19
Publication date: 2023-04-04
Anticipated expiration: 2042-12-19
Also published as: CN115909171B

Abstract

The method comprises the steps of obtaining stirring speed values of a plurality of preset time points in a preset time period and a mixed monitoring video of the preset time period, extracting multi-scale time sequence correlation characteristics of image semantics and the stirring speed values of a monitoring frame by adopting an artificial intelligence production technology based on deep learning so as to generate a complex mapping relation between the stirring speed and the state change of raw material mixing, and adjusting the stirring speed of the current time point based on the state change of mixing. Like this, can be through mixing state change real-time adjustment stirring speed of self-adaptation to improve stirring efficiency and guarantee stirring quality, and then improve the life of fire-resistant air brick.

Description

Method and system for producing steel ladle air brick

Technical Field

The present application relates to the field of intelligent production technologies, and more particularly, to a method and a system for producing a steel ladle air brick.

Background

The ladle air brick is a key functional element for providing an argon blowing process at the bottom of a ladle installed in external refining, and has the main functions of blowing inert gas into molten steel, stirring the molten steel, quickly dispersing and melting alloy added into the molten steel, and floating harmful impurities and gas to achieve the purpose of refining. The quality and the performance of the air brick of the steel ladle directly affect the quality and the efficiency of steelmaking.

At present, the traditional ladle refractory air brick has poor thermal shock resistance in the use process, and easily causes the damage of a brick body in the high-temperature oxygen blowing process, thereby influencing the service life of the brick body.

Therefore, an optimized ladle air brick production scheme is desired.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a method and a system for producing a steel ladle air brick, which are used for acquiring stirring speed values of a plurality of preset time points in a preset time period and a mixed monitoring video of the preset time period, extracting multi-scale time sequence correlation characteristics of image semantics and the stirring speed values of a monitoring frame by adopting an artificial intelligence production technology based on deep learning so as to generate a complex mapping relation between the stirring speed and the state change of raw material mixing, and adjusting the stirring speed of the current time point based on the state change of mixing. Therefore, the stirring speed can be adaptively adjusted in real time through the change of the mixing state, so that the stirring efficiency is improved, the stirring quality is ensured, and the service life of the refractory air brick is prolonged.

According to an aspect of the present application, there is provided a method of producing a ladle gas brick, including:

acquiring stirring speed values of a plurality of preset time points in a preset time period and a mixed monitoring video of the preset time period;

extracting mixed monitoring key frames corresponding to the plurality of preset time points from the mixed monitoring video;

respectively passing the mixed monitoring key frames of the plurality of preset time points through a first convolution neural network model containing a depth feature fusion module to obtain a plurality of mixed monitoring feature matrixes;

respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors to obtain a plurality of mixed monitoring feature vectors;

splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension, and then obtaining a multi-scale state change feature vector through a first multi-scale neighborhood feature extraction module;

arranging the stirring speed values of the plurality of preset time points into a stirring speed input vector according to the time dimension, and then obtaining a multi-scale stirring speed feature vector through a second multi-scale neighborhood feature extraction module;

calculating the responsiveness estimation of the multi-scale state change characteristic vector relative to the multi-scale stirring speed characteristic vector to obtain a classification characteristic matrix;

correcting the classification feature matrix based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix; and

and passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the stirring speed value at the current time point should be increased or decreased.

In the above method for producing a steel ladle gas permeable brick, the step of obtaining a plurality of hybrid monitoring feature matrices by passing the hybrid monitoring key frames of the plurality of predetermined time points through a first convolutional neural network model including a depth feature fusion module includes: extracting a shallow feature matrix from a shallow layer of the first convolutional neural network model containing the depth feature fusion module; extracting a deep feature matrix from a deep layer of the first convolutional neural network model comprising a depth feature fusion module; and fusing the shallow feature matrix and the deep feature matrix using a deep feature fusion module of the first convolutional neural network model comprising a deep feature fusion module to obtain the plurality of hybrid monitoring feature matrices.

In the above method for producing a steel ladle gas permeable brick, the expanding the plurality of mixed monitoring feature matrices into one-dimensional feature vectors respectively to obtain a plurality of mixed monitoring feature vectors includes: and expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors along the row vectors or the column vectors respectively to obtain a plurality of mixed monitoring feature vectors.

In the method for producing the steel ladle air brick, the first multi-scale neighborhood characteristic extraction module comprises: the multi-scale fusion system comprises a first convolutional layer and a second convolutional layer which are parallel to each other, and a first multi-scale fusion layer connected with the first convolutional layer and the second convolutional layer, wherein the first convolutional layer and the second convolutional layer use one-dimensional convolution kernels with different scales.

In the above method for producing a steel ladle gas permeable brick, the splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension, and then obtaining a multi-scale state change feature vector by a first multi-scale neighborhood feature extraction module includes: performing one-dimensional convolution coding on the global mixed monitoring feature vector by using a first convolution layer of the first multi-scale neighborhood feature extraction module according to the following formula to obtain the first scale state change feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; wherein the formula is:

wherein a is the width of the first convolution kernel in the X direction, F (a) is a first convolution kernel parameter vector, G (X-a) is a local vector matrix operated with a convolution kernel function, w is the size of the first convolution kernel, and X represents the global hybrid monitoring feature vector;

performing one-dimensional convolution encoding on the global hybrid supervised feature vector by using a second convolution layer of the first multi-scale neighborhood feature extraction module with the following formula to obtain the second scale state change feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; wherein the formula is:

wherein b is the width of the second convolution kernel in the X direction, F (b) is a second convolution kernel parameter vector, G (X-b) is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the global hybrid monitoring feature vector; and cascading the first scale state change feature vector and the second scale state change feature vector to obtain the multi-scale state change feature vector.

In the above method for producing a steel ladle gas permeable brick, the second multi-scale neighborhood characteristic extraction module includes: a third convolutional layer and a fourth convolutional layer in parallel with each other, and a second multi-scale fusion layer connected to the third convolutional layer and the fourth convolutional layer, wherein the third convolutional layer and the fourth convolutional layer use one-dimensional convolution kernels having different scales.

In the above method for producing a ladle gas brick, the calculating a responsiveness estimate of the multi-scale state change eigenvector with respect to the multi-scale stirring speed eigenvector to obtain a classification feature matrix includes: calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector by the following formula to obtain a classification feature matrix; wherein the formula is:

wherein, V _s Representing a multi-scale state-change feature vector, V _l And M represents the multi-scale stirring speed feature vector, and the classification feature matrix.

In the above method for producing a ladle gas brick, the correcting the classification feature matrix based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix includes: carrying out topological-class center fusion of class nodes on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector according to the following formula to obtain a fusion characteristic matrix; wherein the formula is:

wherein, V ₁ And V ₂ Respectively are the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector, V ₂ ^T Is the transposed vector, M, of the multi-scale stirring speed feature vector _c Is the matrix of the fused features, and,

and |, respectively, denote the Kronecker and Hadamard products of a matrix or vector, D (V) ₁ ，V ₂ ) Is a distance matrix between the multi-scale state change eigenvector and the multi-scale stirring speed eigenvector, and V ₁ And V ₂ Are column vectors, exp (-) represents an exponential operation of a matrix, which represents the calculation of a natural exponential function value raised to the power of the eigenvalue of each position in the matrix; and multiplying the fusion feature matrix and the classification feature matrix to obtain the correctionAnd (5) classifying the feature matrix after positive.

In the above method for producing a ladle gas brick, the step of passing the corrected classification characteristic matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the stirring speed value at the current time point should be increased or decreased, includes: expanding the corrected classification feature matrix into classification feature vectors according to row vectors or column vectors; performing full-join encoding on the classification feature vectors using a plurality of full-join layers of the classifier to obtain encoded classification feature vectors; and passing the encoding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

According to another aspect of the present application, there is provided a ladle gas brick production system, including:

the data acquisition module is used for acquiring the stirring speed values of a plurality of preset time points in a preset time period and the mixed monitoring video of the preset time period;

a key frame extraction module, configured to extract, from the mixed surveillance video, mixed surveillance key frames corresponding to the plurality of predetermined time points;

the convolution coding module is used for enabling the mixed monitoring key frames of the plurality of preset time points to pass through a first convolution neural network model comprising a depth feature fusion module respectively so as to obtain a plurality of mixed monitoring feature matrixes;

the vector expansion module is used for respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors so as to obtain a plurality of mixed monitoring feature vectors;

the first multi-scale feature extraction module is used for splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension and then obtaining a multi-scale state change feature vector through the first multi-scale neighborhood feature extraction module;

the second multi-scale feature extraction module is used for arranging the stirring speed values of the plurality of preset time points into a stirring speed input vector according to the time dimension and then obtaining a multi-scale stirring speed feature vector through the second multi-scale neighborhood feature extraction module;

the responsiveness estimation calculation module is used for calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector to obtain a classification feature matrix;

the correction module is used for correcting the classification characteristic matrix based on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector to obtain a corrected classification characteristic matrix; and

and the stirring speed control module is used for enabling the corrected classification characteristic matrix to pass through a classifier to obtain a classification result, and the classification result is used for indicating that the stirring speed value of the current time point should be increased or decreased.

Compared with the prior art, the method and the system for producing the steel ladle air brick have the advantages that the stirring speed values of a plurality of preset time points in the preset time period and the mixed monitoring video of the preset time period are obtained, the artificial intelligence production technology based on deep learning is adopted, the multi-scale time sequence correlation characteristic extraction is carried out on the image semantics and the stirring speed values of the monitoring frame, the complex mapping relation between the stirring speed and the state change of raw material mixing is generated, and the stirring speed of the current time point is adjusted based on the state change of mixing. Like this, can be through mixing state change real-time adjustment stirring speed of self-adaptation to improve stirring efficiency and guarantee stirring quality, and then improve the life of fire-resistant air brick.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.

Fig. 1 is a schematic view of a scene of a method for producing a ladle air brick according to an embodiment of the application.

Fig. 2 is a flowchart of a method for producing a ladle gas permeable brick according to an embodiment of the present application.

Fig. 3 is a schematic configuration diagram of a ladle gas brick production method according to an embodiment of the application.

Fig. 4 is a flowchart illustrating a sub-step of step S130 in a ladle gas brick production method according to an embodiment of the present application.

Fig. 5 is a flowchart illustrating the substeps of step S190 in the ladle gas brick production method according to an embodiment of the present application.

Fig. 6 is a block diagram of a ladle gas brick production system according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

Overview of scenes

As mentioned above, the conventional ladle refractory air brick has poor thermal shock resistance in the use process, and the brick body is easily damaged in the high-temperature oxygen blowing process, so that the service life of the brick body is influenced. Therefore, an optimized ladle air brick production scheme is desired.

Specifically, in order to solve the above problems, the chinese patent application CN 108975925B proposes a method for preparing a ladle refractory air brick, which specifically comprises the following steps: (1) Taking 70-80 parts of aluminum powder, 35-40 parts of boron powder and 65-70 parts of carbon powder in sequence by weight, heating for melting, cooling, and grinding to obtain aluminum-boron-carbon mixed powder; (2) mixing petroleum coke powder and mixed acid liquor according to the mass ratio of 1:5 to 1:8, heating and stirring for reaction, adding glycol 2-3 times of the mass of the mixed acid solution after the reaction is finished, continuously stirring and mixing, refrigerating, and washing with water to be neutral to obtain wet gel; (3) According to the weight portion, 30-40 portions of magnesia, 10-20 portions of wet gel, 8-10 portions of aluminum ground carbon mixed powder, 40-50 portions of aluminum soil, 10-20 portions of feldspar, 10-15 portions of fly ash, 4-6 portions of potassium permanganate, 8-10 portions of organic adhesive and 30-40 portions of water are taken in sequence, and after being uniformly stirred and mixed, the mixture is pressed and formed to obtain a brick blank; (4) Slowly heating the obtained green bricks to 400-500 ℃ under the protection of inert gas, preserving heat and presintering for 2-4 h, then continuously heating to 1500-1600 ℃ and preserving heat and sintering for 3-5 h, cooling and discharging to obtain sintered materials; (5) And (4) ultrasonically cleaning the sintering material by using alkali liquor, washing the sintering material to be neutral, and drying the sintering material to obtain the ladle refractory air brick.

Accordingly, it was found that the quality of the prepared ladle refractory air brick was not good and the efficiency was slow in the actual process of preparing the ladle refractory air brick, and it was found that the stirring speed could not be adaptively changed according to the mixing state in step 3. That is, the mixing state is different at each time point due to the addition of different weights of raw materials at different stages of the mixing, and the stirring speed values required at each stage are also different, so that the real-time control of the stirring speed is a key to improve the stirring efficiency and the stirring quality. In the process, the difficulty lies in how to establish the mapping relation between the stirring speed and the state change of the raw material mixing, so that the stirring speed is adaptively adjusted through the state change of the mixing so as to improve the stirring efficiency and ensure the stirring quality, and further the service life of the refractory air brick is prolonged.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.

The deep learning and the development of the neural network provide a new solution for excavating the complex mapping relation between the mixing speed and the state change of the raw material mixing. Those skilled in the art will appreciate that the deep neural network model based on deep learning can be adapted by an appropriate training strategy, for example by a back propagation algorithm with gradient descent, to parameters of the deep neural network model to enable it to simulate complex non-linear correlations between things, which is obviously suitable for simulating and establishing complex mapping relationships between stirring speed and state changes of raw material mixing.

Specifically, in the technical scheme of the application, firstly, the stirring speed values of a plurality of predetermined time points in a predetermined time period and the mixed monitoring video of the predetermined time period are obtained. Then, it is considered that in the mixed monitoring video, the state change characteristics of the raw material mixture can be represented by the difference between the adjacent monitoring frames in the mixed monitoring video, that is, the state change condition of the mixture is represented by the image representation of the adjacent image frames. However, considering that the difference between adjacent frames in the mixed monitoring video is small and a large amount of data redundancy exists, in order to reduce the calculation amount and avoid adverse effects on detection caused by the data redundancy, the key frame sampling is performed on the mixed monitoring video at a predetermined sampling frequency so as to extract the mixed monitoring key frames corresponding to the plurality of predetermined time points from the mixed monitoring video. Here, it is worth mentioning that the sampling frequency may be adjusted based on the application requirements of the actual scene, rather than a default value.

Then, a convolutional neural network model with excellent performance in terms of implicit feature extraction of the image is used for feature mining of the mixed monitoring key frames of the plurality of predetermined time points, and particularly, in order to be able to more accurately extract implicit feature information about a mixed state to improve accuracy of stirring speed control when mining hidden features of the mixed monitoring key frames of the plurality of predetermined time points, attention should be paid to shallow features such as color, texture and the like of a mixture in the mixed monitoring key frames during feature extraction, and the shallow features have significance for feature mining of the mixed state to control stirring speed, and when coding, the convolutional neural network can become fuzzy and even be submerged by noise as the depth of the convolutional neural network deepens. Therefore, in the technical scheme of this application, use the convolutional neural network model that contains the depth feature fusion module to come to carry out the difference to the mixed monitoring key frame of a plurality of predetermined time points is handled in order to obtain a plurality of mixed monitoring feature matrixes, compare in standard convolutional neural network model, according to this application convolutional neural network model can keep the shallow feature and the deep feature of mixed monitoring key frame to not only make characteristic information richer, and the feature of the different degree of depth can be kept, in order to improve stirring speed real-time control's precision.

Further, it is considered that the mixing state characteristics of the mixture in the plurality of monitoring key frames have time sequence relevance in the time dimension, that is, the mixing state of the mixture has dynamic change characteristics in the time dimension and has fluctuation in time sequence due to the mixing state. Therefore, in the technical scheme of the application, in order to accurately extract the state change feature of the mixture, the plurality of mixed monitoring feature matrices are respectively expanded into one-dimensional feature vectors along row vectors or column vectors to obtain a plurality of mixed monitoring feature vectors, and the plurality of mixed monitoring feature vectors are spliced into a global mixed monitoring feature vector along a sample dimension and then subjected to feature mining in a first multi-scale neighborhood feature extraction module to extract dynamic multi-scale neighborhood associated feature distribution information of the mixed state feature of the mixture under different time spans, so that the multi-scale state change feature vectors are obtained. In particular, here, the first multi-scale neighborhood feature extraction module includes a first convolutional layer and a second convolutional layer that are parallel to each other, and a first multi-scale fusion layer connected to the first convolutional layer and the second convolutional layer, where the first convolutional layer and the second convolutional layer use one-dimensional convolution kernels having different scales.

Then, for the extraction of the dynamic change feature of the stirring speed, similarly, the stirring speed values have volatility and have different mode state features under different time cycle spans in a time sequence, so that the stirring speed values at the plurality of predetermined time points are arranged into a stirring speed input vector according to a time dimension, and then are encoded in a second multi-scale neighborhood feature extraction module to extract a dynamic multi-scale neighborhood associated feature of the stirring speed under different time spans, so as to obtain a multi-scale stirring speed feature vector. Here, the second multi-scale neighborhood feature extraction module includes a third convolutional layer and a fourth convolutional layer in parallel with each other, and a second multi-scale fusion layer connected to the third convolutional layer and the fourth convolutional layer, wherein the third convolutional layer and the fourth convolutional layer use one-dimensional convolution kernels having different scales.

Then, further calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector to represent the relevance feature distribution information between the dynamic change feature of the stirring speed and the dynamic change feature of the mixing state of the mixture, and using the relevance feature distribution information as a classification feature matrix to perform classification processing in a classifier so as to obtain a classification result used for representing that the stirring speed value at the current time point should be increased or decreased.

That is, in the technical solution of the present application, the label of the classifier includes that the stirring speed value at the current time point should be increased or decreased, wherein the classifier determines to which classification label the classification feature matrix belongs through a soft maximum function. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control strategy label of the stirring speed value at the current time point, and therefore, after the classification result is obtained, the stirring speed at the current time point can be adaptively adjusted based on the classification result, so as to improve the stirring efficiency and ensure the stirring quality.

Particularly, in the technical solution of the present application, when the classification feature matrix is obtained by calculating the responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector, since the multi-scale state change feature vector and the multi-scale stirring speed feature vector respectively express the image semantics of a monitoring frame and the multi-scale time sequence association of the stirring speed value, the feature distribution of the multi-scale state change feature vector may deviate from the classification probability representation of the feature distribution of the multi-scale stirring speed feature vector on the classification probability representation, that is, there is class center deviation between the multi-scale state change feature vector and the multi-scale stirring speed feature vector, thereby affecting the accuracy of calculating the responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector.

Therefore, the feature vector V is preferably changed for the multi-scale state ₁ And the multi-scale stirring speed eigenvector V ₂ Performing topology-class center fusion of class nodes, expressed as:

and [ ] respectively represent the Kronecker and Hadamard products of the matrix (vector), D (V) ₁ ，V ₂ ) As a feature vector V ₁ And V ₂ A matrix of distances between, i.e. d _i，j ＝d(v _1i ，v _2j ) And V is ₁ And V ₂ Are column vectors.

The applicant of the present application considers that in the two-classification problem of the classifier, if the multi-scale state change feature vector V is used ₁ And the multi-scale stirring speed eigenvector V ₂ If the fused class node is represented as a tree form, the multi-scale state change feature vector V ₁ And the multi-scale stirring speed eigenvector V ₂ The respective class nodes are distributed as subtrees based on the root nodes, so that the node distribution of the fused class nodes can be represented as a subgraph structure centering on the respective nodes based on the graph topology by utilizing the associated graph topology among the nodes, thereby expressing the multi-scale state change characteristic vector V ₁ And the multi-scale stirring speed eigenvector V ₂ Subtree structure with respective class node as root to implement said multiscale state change eigenvector V ₁ And the multi-scale stirring speed eigenvector V ₂ So as to eliminate the multi-scale state change characteristic vector V ₁ And the multi-scale stirring speed eigenvector V ₂ Class center offset between.

Further, the fused feature matrix M _c And the classification feature matrix M _f Matrix multiplying to matrix M the classification feature _f Mapping into a fused feature space with class center offset removed to promote the classification feature matrix M _f And representing the accuracy of the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector, thereby obtaining a corrected classification feature matrix. And carrying out classification processing on the corrected classification characteristic matrix through a classifier to obtain a real-time control result of the stirring speed value. Like this, can be through mixing state change real-time adjustment stirring speed of self-adaptation with improvement stirring efficiency and guarantee the stirring quality, and then improve the life of fire-resistant air brick.

Based on this, the application provides a method for producing a steel ladle air brick, which comprises the following steps: acquiring stirring speed values of a plurality of preset time points in a preset time period and a mixed monitoring video of the preset time period; extracting mixed monitoring key frames corresponding to the plurality of preset time points from the mixed monitoring video; respectively passing the mixed monitoring key frames of the plurality of preset time points through a first convolution neural network model containing a depth feature fusion module to obtain a plurality of mixed monitoring feature matrixes; respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors to obtain a plurality of mixed monitoring feature vectors; splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension, and then obtaining a multi-scale state change feature vector through a first multi-scale neighborhood feature extraction module; arranging the stirring speed values of the plurality of preset time points into a stirring speed input vector according to the time dimension, and then obtaining a multi-scale stirring speed feature vector through a second multi-scale neighborhood feature extraction module; calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector to obtain a classification feature matrix; correcting the classification feature matrix based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix; and passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the stirring speed value at the current time point should be increased or decreased.

Fig. 1 is a schematic view of a scene of a method for producing a ladle air brick according to an embodiment of the application. As shown in fig. 1, in this application scenario, first, the stirring speed values at a plurality of predetermined time points within a predetermined time period (e.g., C1 as illustrated in fig. 1) and the mixed monitoring video (e.g., C2 as illustrated in fig. 1) of the predetermined time period are acquired; then, the acquired stirring speed value and the hybrid surveillance video are input into a server (e.g., S as illustrated in fig. 1) deployed with a ladle air brick production algorithm, wherein the server is capable of processing the stirring speed value and the hybrid surveillance video for the predetermined period of time based on the ladle air brick production algorithm to generate a classification result indicating that the stirring speed value at the current point in time should be increased or decreased.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary method

Fig. 2 is a flowchart of a method for producing a ladle gas permeable brick according to an embodiment of the present application. As shown in fig. 2, the method for producing the ladle gas permeable brick according to the embodiment of the application comprises the following steps: s110, acquiring stirring speed values of a plurality of preset time points in a preset time period and a mixed monitoring video of the preset time period; s120, extracting mixed monitoring key frames corresponding to the plurality of preset time points from the mixed monitoring video; s130, enabling the mixed monitoring key frames of the plurality of preset time points to pass through a first convolutional neural network model comprising a depth feature fusion module respectively to obtain a plurality of mixed monitoring feature matrixes; s140, respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors to obtain a plurality of mixed monitoring feature vectors; s150, splicing the plurality of mixed monitoring feature vectors into a global mixed monitoring feature vector along the sample dimension, and then obtaining a multi-scale state change feature vector through a first multi-scale neighborhood feature extraction module; s160, arranging the stirring speed values of the preset time points into a stirring speed input vector according to a time dimension, and then obtaining a multi-scale stirring speed feature vector through a second multi-scale neighborhood feature extraction module; s170, calculating the responsiveness estimation of the multi-scale state change characteristic vector relative to the multi-scale stirring speed characteristic vector to obtain a classification characteristic matrix; s180, correcting the classification characteristic matrix based on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector to obtain a corrected classification characteristic matrix; and S190, passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the stirring speed value at the current time point should be increased or decreased.

Fig. 3 is a schematic configuration diagram of a ladle gas brick production method according to an embodiment of the application. As shown in fig. 3, in the network architecture, first, the stirring speed values of a plurality of predetermined time points in a predetermined time period and the mixed monitoring video of the predetermined time period are obtained; then, extracting the mixed monitoring key frames corresponding to the plurality of preset time points from the mixed monitoring video; then, the mixed monitoring key frames of the plurality of preset time points respectively pass through a first convolution neural network model containing a depth feature fusion module to obtain a plurality of mixed monitoring feature matrixes; then, respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors to obtain a plurality of mixed monitoring feature vectors; then, splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along the sample dimension, and then obtaining a multi-scale state change feature vector through a first multi-scale neighborhood feature extraction module; then, arranging the stirring speed values of the plurality of preset time points into a stirring speed input vector according to the time dimension, and then obtaining a multi-scale stirring speed feature vector through a second multi-scale neighborhood feature extraction module; then, calculating the responsiveness estimation of the multi-scale state change characteristic vector relative to the multi-scale stirring speed characteristic vector to obtain a classification characteristic matrix; then, based on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector, correcting the classification characteristic matrix to obtain a corrected classification characteristic matrix; and finally, passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the stirring speed value at the current time point should be increased or decreased.

Specifically, in step S110, the stirring speed values at a plurality of predetermined time points within a predetermined time period and the mixed monitoring video of the predetermined time period are acquired. As mentioned above, the conventional ladle refractory air brick has poor thermal shock resistance in the use process, and the brick body is easily damaged in the high-temperature oxygen blowing process, so that the service life of the brick body is influenced. Therefore, an optimized ladle air brick production scheme is desired.

Specifically, in order to solve the above problems, chinese patent application CN 108975925B proposes a method for preparing a ladle refractory air brick, which comprises the following specific steps: (1) Taking 70-80 parts of aluminum powder, 35-40 parts of boron powder and 65-70 parts of carbon powder in sequence by weight, heating for melting, cooling, and grinding to obtain aluminum-boron-carbon mixed powder; (2) mixing petroleum coke powder and mixed acid liquor according to the mass ratio of 1:5 to 1:8, heating and stirring for reaction, adding glycol 2-3 times of the mass of the mixed acid solution after the reaction is finished, continuously stirring and mixing, refrigerating, and washing with water to be neutral to obtain wet gel; (3) According to the weight portion, 30-40 portions of magnesia, 10-20 portions of wet gel, 8-10 portions of aluminum ground carbon mixed powder, 40-50 portions of aluminum soil, 10-20 portions of feldspar, 10-15 portions of fly ash, 4-6 portions of potassium permanganate, 8-10 portions of organic adhesive and 30-40 portions of water are taken in sequence, and after being uniformly stirred and mixed, the mixture is pressed and formed to obtain a brick blank; (4) Slowly heating the obtained green bricks to 400-500 ℃ under the protection of inert gas, preserving heat and presintering for 2-4 h, then continuously heating to 1500-1600 ℃ and preserving heat and sintering for 3-5 h, cooling and discharging to obtain sintered materials; (5) And ultrasonically cleaning the sintering material by alkali liquor, washing the sintering material to be neutral, and drying to obtain the ladle refractory air brick.

Accordingly, it was found that the quality of the prepared ladle refractory air brick was not good and the efficiency was slow in the actual process of preparing the ladle refractory air brick, and it was found that the stirring speed could not be adaptively changed according to the mixing state in step 3. That is, in different stages of the mixing, the states of mixing caused by adding different weights of raw materials are different at each time point, and the stirring speed values required in each stage are also different, so that the real-time control of the stirring speed is a key to improve the stirring efficiency and the stirring quality. In the process, the difficulty lies in how to establish the mapping relation between the stirring speed and the state change of the raw material mixing, so as to adaptively adjust the stirring speed through the state change of the mixing to improve the stirring efficiency and ensure the stirring quality, and further improve the service life of the refractory air brick.

The deep learning and the development of the neural network provide a new solution for mining the complex mapping relation between the mixing speed and the state change of raw material mixing. Those skilled in the art will appreciate that the deep neural network model based on deep learning can be adapted by an appropriate training strategy, for example by a back propagation algorithm with gradient descent, to parameters of the deep neural network model to enable it to simulate complex non-linear correlations between things, which is obviously suitable for simulating and establishing complex mapping relationships between stirring speed and state changes of raw material mixing.

Specifically, in the technical scheme of the application, firstly, the stirring speed values of a plurality of predetermined time points in a predetermined time period and the mixed monitoring video of the predetermined time period are acquired to establish a complex mapping relation between the stirring speed and the state change of raw material mixing.

Specifically, in step S120, the mixed monitoring key frames corresponding to the plurality of predetermined time points are extracted from the mixed monitoring video. Then, it is considered that in the mixed monitoring video, the state change characteristics of the raw material mixture can be represented by the difference between the adjacent monitoring frames in the mixed monitoring video, that is, the state change condition of the mixture is represented by the image representation of the adjacent image frames. However, considering that the difference between adjacent frames in the mixed monitoring video is small and a large amount of data redundancy exists, in order to reduce the amount of calculation and avoid adverse effects on detection caused by the data redundancy, key frame sampling is performed on the mixed monitoring video at a predetermined sampling frequency, so as to extract the mixed monitoring key frames corresponding to the plurality of predetermined time points from the mixed monitoring video. Here, it is worth mentioning that the sampling frequency may be adjusted based on the application requirements of the actual scene, rather than a default value.

Specifically, in step S130, the hybrid monitoring keyframes at the plurality of predetermined time points are respectively passed through a first convolution neural network model including a depth feature fusion module to obtain a plurality of hybrid monitoring feature matrices. Then, feature mining is performed on the mixed monitoring key frames of the plurality of predetermined time points by using a convolutional neural network model which has excellent performance in terms of implicit feature extraction of the image, and particularly, in order to extract implicit feature information about a mixed state more accurately to improve accuracy of stirring speed control when mining hidden features of the mixed monitoring key frames of the plurality of predetermined time points, light features such as color and texture of a mixture in the mixed monitoring key frames should be focused on in feature extraction, and the light features have significance for controlling the stirring speed by performing feature mining on the mixed state to submerge, and when coding the convolutional neural network, the light features become fuzzy and even are affected by noise as the depth of the convolutional neural network increases.

Therefore, in the technical scheme of this application, use the convolutional neural network model that contains the depth feature fusion module to come to carry out the difference to the mixed monitoring key frame of a plurality of predetermined time points is handled in order to obtain a plurality of mixed monitoring feature matrixes, compare in standard convolutional neural network model, according to this application convolutional neural network model can keep the shallow feature and the deep feature of mixed monitoring key frame to not only make characteristic information richer, and the feature of the different degree of depth can be kept, in order to improve stirring speed real-time control's precision.

In this embodiment of the present application, fig. 4 is a flowchart of a sub-step of step S130 in a ladle gas brick production method according to an embodiment of the present application, and as shown in fig. 4, the passing the mixed monitoring keyframes at a plurality of predetermined time points through a first convolutional neural network model including a depth feature fusion module to obtain a plurality of mixed monitoring feature matrices includes: s210, extracting a shallow feature matrix from the shallow layer of the first convolution neural network model containing the depth feature fusion module; s220, extracting a deep feature matrix from the deep layer of the first convolution neural network model containing the depth feature fusion module; and S230, fusing the shallow feature matrix and the deep feature matrix by using the depth feature fusion module of the first convolution neural network model containing the depth feature fusion module to obtain the plurality of mixed monitoring feature matrices.

Specifically, in step S140, the plurality of hybrid monitoring feature matrices are respectively expanded into one-dimensional feature vectors to obtain a plurality of hybrid monitoring feature vectors. Further, it is considered that the mixing state characteristics of the mixture in the plurality of monitoring key frames have time sequence relevance in the time dimension, that is, the mixing state of the mixture has dynamic change characteristics in the time dimension and has fluctuation in time sequence due to the mixing state. Therefore, in the technical solution of the present application, in order to accurately extract the state change feature of the mixture, the plurality of hybrid monitoring feature matrices are respectively expanded into one-dimensional feature vectors along the row vector or the column vector to obtain a plurality of hybrid monitoring feature vectors.

Wherein, expand a plurality of mixed control feature matrix respectively as one-dimensional eigenvector in order to obtain a plurality of mixed control feature vector, include: and expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors along the row vectors or the column vectors respectively to obtain a plurality of mixed monitoring feature vectors.

Specifically, in step S150, the multiple mixed monitoring feature vectors are spliced into a global mixed monitoring feature vector along a sample dimension, and then the global mixed monitoring feature vector passes through a first multi-scale neighborhood feature extraction module to obtain a multi-scale state change feature vector. And splicing the plurality of mixed monitoring feature vectors into a global mixed monitoring feature vector along the sample dimension, and then performing feature mining in a first multi-scale neighborhood feature extraction module to extract dynamic multi-scale neighborhood associated feature distribution information of the mixed state features of the mixture under different time spans, thereby obtaining a multi-scale state change feature vector. In particular, here, the first multi-scale neighborhood feature extraction module includes a first convolutional layer and a second convolutional layer parallel to each other, and a first multi-scale fusion layer connected to the first convolutional layer and the second convolutional layer, where the first convolutional layer and the second convolutional layer use one-dimensional convolution kernels having different scales.

Further, in this embodiment of the present application, the obtaining a multi-scale state change feature vector by a first multi-scale neighborhood feature extraction module after splicing the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension includes: performing one-dimensional convolution coding on the global mixed monitoring feature vector by using a first convolution layer of the first multi-scale neighborhood feature extraction module according to the following formula to obtain the first scale state change feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; wherein the formula is:

It is worth mentioning that compared to the traditional feature engineering, the multi-scale neighborhood feature extraction module is essentially a deep neural network model based on deep learning, which can fit any function through a predetermined training strategy and has higher feature extraction generalization capability.

Specifically, in step S160, the stirring speed values of the plurality of predetermined time points are arranged as a stirring speed input vector according to a time dimension, and then the stirring speed input vector is passed through a second multi-scale neighborhood feature extraction module to obtain a multi-scale stirring speed feature vector. Then, for the extraction of the dynamic change feature of the stirring speed, similarly, the stirring speed values have volatility and have different mode state features under different time cycle spans in a time sequence, so that the stirring speed values at the plurality of predetermined time points are arranged into a stirring speed input vector according to a time dimension, and then are encoded in a second multi-scale neighborhood feature extraction module to extract a dynamic multi-scale neighborhood associated feature of the stirring speed under different time spans, so as to obtain a multi-scale stirring speed feature vector.

Here, the second multi-scale neighborhood feature extraction module includes a third convolutional layer and a fourth convolutional layer in parallel with each other, and a second multi-scale fusion layer connected to the third convolutional layer and the fourth convolutional layer, wherein the third convolutional layer and the fourth convolutional layer use one-dimensional convolution kernels having different scales.

Further, in this embodiment of the present application, after the stirring speed values at the plurality of predetermined time points are arranged as a stirring speed input vector according to the time dimension, the multi-scale stirring speed feature vector is obtained by the second multi-scale neighborhood feature extraction module, including: performing one-dimensional convolution encoding on the stirring speed input vector by using a third convolution layer of the second multi-scale neighborhood feature extraction module according to the following formula to obtain the first scale stirring speed feature vector, wherein the third convolution layer has a third one-dimensional convolution kernel with a third length; wherein the formula is:

wherein c is the width of the third convolution kernel in the x direction, F (c) is a parameter vector of the third convolution kernel, G (x-c) is a local vector matrix operated with the convolution kernel function, n is the size of the third convolution kernel, and Y represents the input vector of the stirring speed;

performing one-dimensional convolution encoding on the stirring speed input vector by using a fourth convolution layer of the second multi-scale neighborhood feature extraction module according to the following formula to obtain the second-scale stirring speed feature vector, wherein the fourth convolution layer has a fourth one-dimensional convolution kernel with a fourth length, and the third length is different from the fourth length; wherein the formula is:

wherein d is the width of the fourth convolution kernel in the x direction, F (d) is a fourth convolution kernel parameter vector, G (x-d) is a local vector matrix operated with the convolution kernel function, p is the size of the fourth convolution kernel, and Y represents the stirring speed input vector; and cascading the first scale stirring speed feature vector and the second scale stirring speed feature vector to obtain the multi-scale stirring speed feature vector.

Specifically, in step S170, a responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector is calculated to obtain a classification feature matrix. Then, further calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector to represent the relevance feature distribution information between the dynamic change feature of the stirring speed and the dynamic change feature of the mixing state of the mixture, and using the relevance feature distribution information as a classification feature matrix to perform classification processing in a classifier so as to obtain a classification result used for representing that the stirring speed value at the current time point should be increased or decreased.

Further, in this embodiment of the present application, the calculating a responsiveness estimate of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector to obtain a classification feature matrix includes: calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector by the following formula to obtain a classification feature matrix; wherein the formula is:

Specifically, in step S180, the classification feature matrix is corrected based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix. Particularly, in the technical solution of the present application, when the classification feature matrix is obtained by calculating the responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector, since the multi-scale state change feature vector and the multi-scale stirring speed feature vector respectively express the image semantics of a monitoring frame and the multi-scale time sequence association of the stirring speed value, the feature distribution of the multi-scale state change feature vector may deviate from the classification probability representation of the feature distribution of the multi-scale stirring speed feature vector on the classification probability representation, that is, there is class center deviation between the multi-scale state change feature vector and the multi-scale stirring speed feature vector, thereby affecting the accuracy of calculating the responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector.

Therefore, the feature vector V is preferably changed for the multi-scale state ₁ And the multi-scale stirring speed eigenvector V ₂ Performing topology-class center fusion of class nodes, namely performing the topology-class center fusion of the class nodes on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector according to the following formula to obtain a fusion characteristic matrix; wherein the formula is:

wherein, V ₁ And V ₂ Respectively, the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector, V ₂ ^T Is the transposed vector, M, of the multi-scale stirring speed feature vector _c Is the matrix of the fused features, and,

and [ ] represent the Kronecker product and Hadamard product, D (V), of a matrix or vector, respectively ₁ ，V ₂ ) Is a distance matrix between the multi-scale state change eigenvector and the multi-scale stirring speed eigenvector, and V ₁ And V ₂ Are column vectors, exp (-) represents an exponential operation of a matrix, which represents the calculation of a natural exponential function value raised to the power of the eigenvalue of each position in the matrix; and, mixing the aboveAnd performing matrix multiplication on the fusion characteristic matrix and the classification characteristic matrix to obtain the corrected classification characteristic matrix.

The applicant of the present application considers that in the two-classification problem of the classifier, if the multi-scale state change feature vector V is divided into ₁ And the multi-scale stirring speed eigenvector V ₂ The merged class node is expressed as a tree form, and then the multi-scale state change characteristic vector V ₁ And the multi-scale stirring speed eigenvector V ₂ The respective class nodes are distributed as subtrees based on the root nodes, so that the node distribution of the fused class nodes can be represented as a sub-graph structure centering on the respective nodes based on the graph topology by utilizing the graph topology associated among the nodes, and the multi-scale state change characteristic vector V is expressed ₁ And the multi-scale stirring speed eigenvector V ₂ Subtree structure with respective class node as root to implement said multiscale state change eigenvector V ₁ And the multi-scale stirring speed eigenvector V ₂ So as to eliminate the multi-scale state change characteristic vector V ₁ And the multi-scale stirring speed eigenvector V ₂ Class center offset between.

Further, fusing the feature matrix M _c And the classification feature matrix M _f Matrix multiplying to matrix M the classification feature _f Mapping into a fused feature space with class center offset removed to promote the classification feature matrix M _f And representing the accuracy of the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector, thereby obtaining a corrected classification feature matrix. And carrying out classification processing on the corrected classification characteristic matrix through a classifier to obtain a real-time control result of the stirring speed value. Like this, can be through mixing state change real-time adjustment stirring speed of self-adaptation with improvement stirring efficiency and guarantee stirring quality, and then improve the life of fire-resistant air brick.

Specifically, in step S190, the corrected classification feature matrix is passed through a classifier to obtain a classification result, where the classification result is used to indicate that the stirring speed value at the current time point should be increased or decreased. In the technical solution of the present application, the label of the classifier includes that the stirring speed value at the current time point should be increased or decreased, wherein the classifier determines which classification label the classification feature matrix belongs to through a soft maximum function. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control strategy label of the stirring speed value at the current time point, and therefore, after the classification result is obtained, the stirring speed at the current time point can be adaptively adjusted based on the classification result, so as to improve the stirring efficiency and ensure the stirring quality.

Fig. 5 is a flowchart illustrating the substeps of step S190 in the method for producing a ladle ventilating brick according to an embodiment of the present application, and as shown in fig. 5, the step of passing the corrected classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the stirring speed value at the current time point should be increased or decreased, includes: s310, unfolding the corrected classification feature matrix into classification feature vectors according to row vectors or column vectors; s320, carrying out full-connection coding on the classification feature vector by using a plurality of full-connection layers of the classifier to obtain a coding classification feature vector; and S330, passing the coded classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

In a specific example of the present application, the classifier is used to process the corrected classification feature matrix according to the following formula to obtain the classification result; wherein the formula is:

O＝softmax{(W _n ，B _n )：...：(W ₁ ，B ₁ ) L Project (F) }, where W ₁ To W _n As a weight matrix, B ₁ To B _n Project (F) is to Project the corrected classification feature matrix as a vector for the bias vector.

In summary, according to the method for producing the air brick for the steel ladle in the embodiment of the application, the stirring speed values of a plurality of predetermined time points in a predetermined time period and the mixed monitoring video of the predetermined time period are obtained, an artificial intelligence production technology based on deep learning is adopted, so that the image semantics and the stirring speed values of the monitoring frame are subjected to multi-scale time sequence associated feature extraction, a complex mapping relation between the stirring speed and the state change of raw material mixing is generated, and the stirring speed of the current time point is adjusted based on the state change of mixing. Like this, can be through mixing state change real-time adjustment stirring speed of self-adaptation to improve stirring efficiency and guarantee stirring quality, and then improve the life of fire-resistant air brick.

Exemplary System

Fig. 6 is a block diagram of a ladle gas brick production system according to an embodiment of the present application. As shown in fig. 6, the ladle gas brick production system 100 according to the embodiment of the present application includes: the data acquisition module 110 is configured to acquire stirring speed values at a plurality of predetermined time points within a predetermined time period and a mixed monitoring video of the predetermined time period; a key frame extracting module 120, configured to extract, from the mixed surveillance video, mixed surveillance key frames corresponding to the plurality of predetermined time points; a convolution coding module 130, configured to pass the mixed monitoring key frames at the multiple predetermined time points through a first convolution neural network model including a depth feature fusion module, respectively, to obtain multiple mixed monitoring feature matrices; a vector expansion module 140, configured to expand the multiple hybrid monitoring feature matrices into one-dimensional feature vectors respectively to obtain multiple hybrid monitoring feature vectors; the first multi-scale feature extraction module 150 is configured to splice the multiple mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension, and then obtain a multi-scale state change feature vector through the first multi-scale neighborhood feature extraction module; the second multi-scale feature extraction module 160 is configured to arrange the stirring speed values of the multiple predetermined time points into a stirring speed input vector according to a time dimension, and then obtain a multi-scale stirring speed feature vector through the second multi-scale neighborhood feature extraction module; a responsiveness estimation calculation module 170, configured to calculate a responsiveness estimation of the multi-scale state change feature vector with respect to the multi-scale stirring speed feature vector to obtain a classification feature matrix; a correcting module 180, configured to correct the classification feature matrix based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix; and the stirring speed control module 190 is configured to pass the corrected classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the stirring speed value at the current time point should be increased or decreased.

In one example, in the above ladle gas brick production system 100, the convolutional encoding module includes: the shallow feature extraction unit is used for extracting a shallow feature matrix from the shallow layer of the first convolution neural network model containing the depth feature fusion module; a deep feature extraction unit, configured to extract a deep feature matrix from a deep layer of the first convolutional neural network model including the depth feature fusion module; and a fusion unit, configured to fuse the shallow feature matrix and the deep feature matrix using a deep-shallow feature fusion module of the first convolutional neural network model including the deep-shallow feature fusion module to obtain the plurality of hybrid monitoring feature matrices.

In one example, in the above ladle gas brick production system 100, the vector deployment module is configured to: and expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors along the row vectors or the column vectors respectively to obtain a plurality of mixed monitoring feature vectors.

In one example, in the above ladle gas brick production system 100, the first multi-scale neighborhood feature extraction module is configured to: the multi-scale fusion system comprises a first convolutional layer and a second convolutional layer which are parallel to each other, and a first multi-scale fusion layer connected with the first convolutional layer and the second convolutional layer, wherein the first convolutional layer and the second convolutional layer use one-dimensional convolution kernels with different scales.

In one example, in the above ladle gas brick production system 100, the first multi-scale feature extraction module includes: a first scale state change feature extraction unit, configured to perform one-dimensional convolutional coding on the global hybrid monitoring feature vector by using a first convolutional layer of the first multi-scale neighborhood feature extraction module according to a formula to obtain the first scale state change feature vector, where the first convolutional layer has a first one-dimensional convolutional kernel with a first length; wherein the formula is:

a second scale state change feature extraction unit, configured to perform one-dimensional convolutional encoding on the global hybrid monitoring feature vector using a second convolutional layer of the first multi-scale neighborhood feature extraction module according to a formula to obtain the second scale state change feature vector, where the second convolutional layer has a second one-dimensional convolutional kernel with a second length, and the first length is different from the second length; wherein the formula is:

wherein b is the width of the second convolution kernel in the X direction, F (b) is a parameter vector of the second convolution kernel, G (X-b) is a local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the global mixed monitoring feature vector; and the cascade unit is used for cascading the first scale state change feature vector and the second scale state change feature vector to obtain the multi-scale state change feature vector.

In one example, in the above ladle gas brick production system 100, the second multi-scale neighborhood feature extraction module is configured to: a third convolutional layer and a fourth convolutional layer in parallel with each other, and a second multi-scale fusion layer connected to the third convolutional layer and the fourth convolutional layer, wherein the third convolutional layer and the fourth convolutional layer use one-dimensional convolution kernels having different scales.

In one example, in the above ladle gas brick production system 100, the responsiveness estimation calculation module is further configured to: calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector by the following formula to obtain a classification feature matrix; wherein the formula is:

wherein, V _s Representing a multi-scale state change feature vector, V _l And M represents the multi-scale stirring speed feature vector, and the classification feature matrix.

In one example, in the above ladle gas brick production system 100, the calibration module includes: the topology-class center fusion unit of the class nodes is used for performing the topology-class center fusion of the class nodes on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector according to the following formula to obtain a fusion characteristic matrix; wherein the formula is:

wherein, V ₁ And V ₂ Respectively are the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector, V ₂ ^T Is the transposed vector, M, of the multiscale stirring speed eigenvector _c Is the matrix of the fused features, and,

and [ ] represent the Kronecker product and Hadamard product, D (V), of a matrix or vector, respectively ₁ ，V ₂ ) Is a distance matrix between the multi-scale state change eigenvector and the multi-scale stirring speed eigenvector, and V ₁ And V ₂ Are column vectors, exp (-) represents the exponential operation of the matrix, saidThe exponential operation of the matrix represents the calculation of a natural exponential function value with the characteristic value of each position in the matrix as power; and the multiplying unit is used for carrying out matrix multiplication on the fusion characteristic matrix and the classification characteristic matrix to obtain the corrected classification characteristic matrix.

In one example, in the above ladle gas brick production system 100, the stirring speed control module includes: the characteristic matrix unfolding unit is used for unfolding the corrected classification characteristic matrix into classification characteristic vectors according to row vectors or column vectors; a full-concatenation encoding unit, configured to perform full-concatenation encoding on the classification feature vector using a plurality of full-concatenation layers of the classifier to obtain an encoded classification feature vector; and the classification result generating unit is used for enabling the encoding classification feature vector to pass through a Softmax classification function of the classifier to obtain the classification result.

Here, it will be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described ladle gas brick production system 100 have been described in detail in the above description of the ladle gas brick production method with reference to fig. 1 to 5, and thus, a repetitive description thereof will be omitted.

The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations should be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. The production method of the steel ladle air brick is characterized by comprising the following steps:

calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector to obtain a classification feature matrix;

2. The method for producing the steel ladle gas permeable brick as claimed in claim 1, wherein the step of passing the mixed monitoring key frames of the plurality of predetermined time points through a first convolutional neural network model including a depth feature fusion module to obtain a plurality of mixed monitoring feature matrices comprises:

extracting a shallow feature matrix from a shallow layer of the first convolution neural network model containing the depth feature fusion module;

extracting a deep feature matrix from a deep layer of the first convolutional neural network model comprising the depth feature fusion module; and

fusing the shallow feature matrix and the deep feature matrix using a deep feature fusion module of the first convolutional neural network model comprising a deep feature fusion module to obtain the plurality of hybrid monitoring feature matrices.

3. The method for producing the ladle gas permeable brick according to claim 2, wherein the expanding the plurality of hybrid monitoring feature matrices into one-dimensional feature vectors to obtain a plurality of hybrid monitoring feature vectors respectively comprises: and respectively expanding the plurality of mixed monitoring feature matrixes into one-dimensional feature vectors along the row vectors or the column vectors to obtain a plurality of mixed monitoring feature vectors.

4. The method for producing the steel ladle gas permeable brick as claimed in claim 3, wherein the first multi-scale neighborhood characteristic extraction module comprises: the multi-scale fusion system comprises a first convolutional layer and a second convolutional layer which are parallel to each other, and a first multi-scale fusion layer connected with the first convolutional layer and the second convolutional layer, wherein the first convolutional layer and the second convolutional layer use one-dimensional convolution kernels with different scales.

5. The method for producing the ladle gas permeable brick according to claim 4, wherein the step of splicing the plurality of mixed monitoring feature vectors into a global mixed monitoring feature vector along a sample dimension and then obtaining a multi-scale state change feature vector through a first multi-scale neighborhood feature extraction module comprises the steps of:

performing one-dimensional convolution coding on the global mixed monitoring feature vector by using a first convolution layer of the first multi-scale neighborhood feature extraction module according to the following formula to obtain the first scale state change feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length;

wherein the formula is:

performing one-dimensional convolution encoding on the global hybrid surveillance feature vector by using a second convolution layer of the first multi-scale neighborhood feature extraction module according to a formula to obtain the second scale state change feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length;

wherein the formula is:

b is the width of the second convolution kernel in the X direction, F (b) is a parameter vector of the second convolution kernel, G (X-b) is a local vector matrix operated with a convolution kernel function, m is the size of the second convolution kernel, and X represents the global mixed monitoring characteristic vector; and

and cascading the first scale state change feature vector and the second scale state change feature vector to obtain the multi-scale state change feature vector.

6. The method for producing the steel ladle gas permeable brick as claimed in claim 5, wherein the second multi-scale neighborhood characteristic extraction module comprises: a third convolutional layer and a fourth convolutional layer in parallel with each other, and a second multi-scale fusion layer connected to the third convolutional layer and the fourth convolutional layer, wherein the third convolutional layer and the fourth convolutional layer use one-dimensional convolution kernels having different scales.

7. The method for producing the ladle gas permeable brick as claimed in claim 6, wherein the calculating the responsiveness estimation of the multi-scale state change eigenvector relative to the multi-scale stirring speed eigenvector to obtain a classification feature matrix comprises: calculating the responsiveness estimation of the multi-scale state change feature vector relative to the multi-scale stirring speed feature vector by the following formula to obtain a classification feature matrix;

wherein the formula is:

8. The method for producing the ladle gas permeable brick according to claim 7, wherein the correcting the classification feature matrix based on the multi-scale state change feature vector and the multi-scale stirring speed feature vector to obtain a corrected classification feature matrix comprises:

carrying out topological-class center fusion of class nodes on the multi-scale state change characteristic vector and the multi-scale stirring speed characteristic vector according to the following formula to obtain a fusion characteristic matrix;

wherein the formula is:

and |, respectively, denote the Kronecker and Hadamard products of a matrix or vector, D (V) ₁ ,V ₂ ) Is a distance matrix between the multi-scale state change eigenvector and the multi-scale stirring speed eigenvector, and V ₁ And V ₂ Are column vectors, exp (-) represents the exponential operation of the matrix representing the calculation for each position in the matrixThe characteristic value is a natural exponent function value of the power; and

and multiplying the fusion feature matrix and the classification feature matrix to obtain the corrected classification feature matrix.

9. The method for producing the ladle gas permeable brick according to claim 8, wherein the step of passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the stirring speed value at the current time point should be increased or decreased, comprises the steps of:

expanding the corrected classification feature matrix into classification feature vectors according to row vectors or column vectors;

performing full-join coding on the classification feature vectors using a plurality of full-join layers of the classifier to obtain coded classification feature vectors; and

and passing the encoding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

10. A ladle air brick production system is characterized by comprising:

the responsiveness estimation calculation module is used for calculating the responsiveness estimation of the multi-scale state change characteristic vector relative to the multi-scale stirring speed characteristic vector to obtain a classification characteristic matrix;