Blast furnace taphole mud bag shape prediction method and system

Abstract

The application discloses a blast furnace tap hole mud bag form prediction method and system, the method comprises the following steps: for the blast furnace tap hole area, a local sector cylindrical shell calculation domain of the inner wall of the hearth is constructed with the center of the tap hole as the reference; in the local sector cylindrical shell calculation domain, column coordinates are generated by uniform sampling of column coordinates, the normalized mud volume is combined with the column coordinates into a four-dimensional vector, the mud bag distribution neural implicit model after training is input, and the volume fraction of the taphole clay at the corresponding position is predicted; the volume fraction of the taphole clay of the unsampled points in the local sector cylindrical shell calculation domain is completed, a continuous taphole clay volume fraction field is generated, and the three-dimensional space distribution form of the mud bag is determined. The three-dimensional space form of the mud bag in the blast furnace tap hole area can be stably, quantitatively and in accordance with engineering experience predicted under the condition that there is no internal measured label data of the mud bag and no complex explicit control equation is needed.

Description

Technical Field

[0001] This application relates to the field of blast furnace ironmaking process modeling and intelligent optimization technology, and in particular to a method and system for predicting the morphology of mud bales at the blast furnace taphole. Background Technology

[0002] The working condition of the blast furnace taphole area directly affects the stability of the taphole, the life of the hearth, and the safety of tapping operations. In industrial settings, taphole mud is typically injected into the taphole channel using mud guns. This mud forms a mud-filled structure under high temperature and pressure to adjust the effective taphole depth and protect the surrounding refractory materials.

[0003] Existing analytical methods for the condition of blast furnace slag or taphole mainly include: relying on empirical formulas or operator experience to indirectly judge the condition of blast furnace slag through macroscopic indicators such as slag volume, taphole life, and tapping stability; using simplified numerical simulation or computational fluid dynamics models to analyze the flow, heat transfer, and erosion behavior of molten iron in the hearth and then indirectly inferring the working condition of the taphole area; and attempting to introduce data-driven methods for condition identification. However, due to the closed environment inside the blast furnace, limited observation methods, and difficulty in obtaining labeled data, it is difficult to form a direct prediction of the three-dimensional morphology of the blast furnace slag. Summary of the Invention

[0004] In view of this, this application provides a method and system for predicting the morphology of the mud bag at the blast furnace taphole. It can make stable, quantitative, and engineering experience-compliant predictions of the three-dimensional spatial morphology of the mud bag in the blast furnace taphole area under the condition of lacking measured and labeled data inside the mud bag and without relying on complex explicit control equations. This allows for further determination of a reasonable mud-removal volume range that meets the effective taphole depth requirements based on the prediction results.

[0005] According to one aspect of this application, a method for predicting the morphology of the blast furnace taphole mud bag is provided, the method comprising: For the blast furnace tapping area, a local sector-shaped cylindrical shell computational domain is constructed with the tapping center as the reference for the spatial structure of the hearth inner wall. Within the local sector-shaped cylindrical shell computational domain, multiple cylindrical coordinates are generated through a uniform sampling mechanism. For any cylindrical coordinate, the normalized sludge volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential, and axial. The normalized sludge volume is the net sludge volume obtained by normalizing the actual sludge volume after deducting the filling part inside the tap hole channel. The trained implicit neural model of mud distribution predicts the volume fraction of gunning mud at the spatial location corresponding to the four-dimensional vector, where the volume fraction of gunning mud is used to characterize the degree to which gunning mud occupies the spatial location corresponding to the four-dimensional vector. After completing the unsampled spatial locations within the local sector-shaped cylindrical shell computational domain with the volume fraction of the cutterhead, a continuous cutterhead volume fraction field covering the entire local sector-shaped cylindrical shell computational domain is generated. Based on the continuous cutterhead volume fraction field, the three-dimensional spatial distribution morphology of the cutterhead is determined.

[0006] According to another aspect of this application, a blast furnace taphole mud morphology prediction system is provided, the system comprising: The column shell computational domain construction module is used to construct a local fan-shaped column shell computational domain for the blast furnace tapping area, with the tapping center as the reference, targeting the spatial structure of the hearth inner wall. The cylindrical coordinate uniform sampling module is used to generate multiple cylindrical coordinates within the local fan-shaped cylindrical shell computational domain through a cylindrical coordinate uniform sampling mechanism. For any cylindrical coordinate, the normalized mud volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential, and axial. The normalized mud volume is the net mud volume obtained by normalizing the actual mud volume after deducting the filling part inside the iron outlet channel. The slurry volume fraction prediction module is used to predict the slurry volume fraction at the spatial location corresponding to the four-dimensional vector by using a trained neural implicit model of slurry distribution. The slurry volume fraction is used to characterize the degree to which the spatial location corresponding to the four-dimensional vector is occupied by slurry. The blast furnace taphole mud bag morphology prediction module is used to complete the unsampled spatial locations within the local fan-shaped cylindrical shell computational domain by adding the mud volume fraction, and then generate a continuous mud volume fraction field covering the entire local fan-shaped cylindrical shell computational domain. Based on the continuous mud volume fraction field, the three-dimensional spatial distribution morphology of the mud bag is determined.

[0007] By means of the above technical solution, the method and system for predicting the morphology of the mud bag at the blast furnace taphole provided in this application can make a stable, quantitative prediction of the three-dimensional spatial morphology of the mud bag in the blast furnace taphole area that is consistent with engineering experience, without the lack of measured and labeled data inside the mud bag and without relying on complex explicit control equations.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1A flowchart illustrating a method for predicting the morphology of mud bales at the taphole of a blast furnace, as provided in an embodiment of this application, is shown. Figure 2 This illustration shows a schematic diagram of a local sector-shaped cylindrical shell computational domain provided in an embodiment of this application; Figure 3 This illustration shows a distribution diagram of uniformly random sampling points within a local sector-shaped cylindrical shell computational domain, as provided in an embodiment of this application. Figure 4 This illustration shows a schematic diagram of a hidden neural network structure for mud-covered distribution provided in an embodiment of this application. Figure 5 This illustration shows the three-dimensional spatial distribution morphology and key features of a mud pouch provided in an embodiment of this application; Figure 6 A schematic diagram of the structure of a blast furnace taphole mud bag morphology prediction system provided in an embodiment of this application is shown. Detailed Implementation

[0010] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0011] This embodiment provides a method for predicting the morphology of the blast furnace taphole mud bag, such as... Figure 1 As shown, the method includes: Step 101: For the blast furnace tapping area, a local sector-shaped cylindrical shell computational domain is constructed based on the center of the tapping area for the spatial structure of the hearth inner wall.

[0012] Currently, the common approach is to use indirect calculations based on blast furnace production data and process parameters. This involves first collecting parameters such as blast volume, blast temperature, blast pressure, number of tuyeres, tuyere area, and coal ratio to calculate the blast kinetic energy and tuyere swirl zone depth. Then, combining this with the relative positions of the tuyere swirl zone, the dead material column in the hearth, and the mud bag, the actual taphole depth is calculated. Finally, a mathematical model of the mud bag's outer contour is established to calculate its volume, thereby estimating the amount of mud removed from the blast furnace taphole. While this method can achieve a certain degree of quantitative acquisition of taphole depth and mud removal amount, it still has shortcomings, for the following reasons: Existing methods primarily rely on blast furnace production data, pre-defined geometric profiles, and analytical models to indirectly calculate the state of the mud bag. The results primarily reflect overall parameter relationships, making it difficult to directly characterize the true three-dimensional spatial distribution of the mud bag within a local area of ​​the hearth. Furthermore, existing methods are heavily dependent on model assumptions, parameter settings, and empirical boundary conditions; when blast furnace operating conditions change, the model's adaptability and predictive stability are easily affected. In addition, due to the enclosed environment and limited observation methods within the blast furnace, direct observation data on the internal morphology of the mud bag is lacking. Therefore, existing methods struggle to accurately describe the spatial variation of the mud bag morphology under different mud-removing volumes or operating conditions, thus offering limited support for the refined optimization of blast furnace mud-removing operations. In the above embodiments of this application, by representing the morphology of the clay bag at the taphole as a continuous volume fraction field, and constructing a neural implicit model of the clay bag distribution with cylindrical coordinates and normalized clay loading as input and clay volume fraction as output, the problem of clay bag geometry prediction is transformed into a continuous field learning problem. After the clay bag distribution neural implicit model outputs the clay volume fraction, the three-dimensional morphology of the clay bag can be extracted by isosurface, and the variation law of the maximum thickness of the clay bag and under different clay loading (volume) conditions can be further calculated, which can be used to form a quantitative recommendation range for reasonable clay loading.

[0013] Optionally, in step 101, a local sector-shaped cylindrical shell computational domain is constructed based on the center of the taphole, targeting the spatial structure of the hearth inner wall, including: Step 1011: Determine the boundary parameters used to define the boundary of the local sector-shaped cylindrical shell calculation domain, wherein the boundary parameters include the inner radius of the hearth, the center angle of the taphole, the circumferential half-width, the center height of the taphole, the axial half-height, and the inner diameter of the local sector-shaped cylindrical shell calculation domain. Step 1012: Using the inner diameter of the local sector-shaped cylindrical shell computational domain as the inner boundary and the inner radius of the furnace hearth as the outer boundary, construct a radial range, where the radial range is represented as... ; Step 1013: Taking the center angle of the taphole as the center, expand the circumferential half-width to both sides to construct the circumferential range, where the circumferential range is represented as... ; Step 1014: Using the center height of the taphole as the center, extend the axial half-height both above and below to construct the axial range, where the axial range is represented as... ; Based on the sector-shaped cylindrical shell structure formed by the intersection of the radial, circumferential, and axial ranges, a local sector-shaped cylindrical shell computational domain is constructed, which is represented as:

[0014] For the local sector-shaped cylindrical shell computational domain, For radial components, For the circumferential component, For the axial component, The inner radius of the hearth. For the inner diameter of the computational domain of the local sector-shaped cylindrical shell, The angle of the center of the taphole. The circumferential half-width, The height of the center of the taphole. It is the axial half-height.

[0015] In the above embodiments of this application, such as Figure 2 As shown, in Figure 2 In the diagram, the left xy plane is a top view viewed from the top of the blast furnace hearth, and the right xz plane is a side view viewed horizontally from the side of the blast furnace hearth. The specific steps for constructing a local sector-shaped cylindrical shell computational domain for the blast furnace taphole area are as follows: Using the taphole center as the reference point, a cylindrical coordinate system is employed. Describe the computational domain of the local sector-shaped cylindrical shell. Figure 2 The top view on the left side of the middle section is used to show the radial (r) and circumferential (r) directions. The dimensions are shown in the right-hand side view, which displays the radial (r) and axial (z) dimensions. Then, the radial, circumferential, and axial ranges are determined.

[0016] Furthermore, the inner radius of the furnace hearth is preset to be... The center angle of the iron outlet is Circumferential half-width is The center height of the iron outlet is axial half-height is The inner diameter of the local sector-shaped cylindrical shell computational domain is .

[0017] The inner diameter of the computational domain is calculated using a local sector-shaped cylindrical shell. inner boundary (corresponding to) Figure 2 In the top view, the radius of the inner dashed circle and the inner radius R of the furnace hearth are the outer boundaries (corresponding to...). Figure 2 The radius of the outer solid line circle in the top view (covering the annular area near the inner wall of the hearth) constitutes the radial range, i.e. .

[0018] Angle of the center of the iron taphole Expand circumferentially by half a width to both sides from the center. Forming a fan-shaped region, constructing a circumferential range, that is... Figure 2 The half-angle of the green sector in the top view has a circumferential range of [missing information]. .

[0019] At the center height of the iron outlet Centered on the axis, extend the axial half-height both above and below. The axial range of the columnar region is also known as the formation of the columnar region. ,correspond Figure 2 The upper and lower halves of the green area in the side view on the right.

[0020] Finally, based on the intersection of the radial, circumferential, and axial ranges mentioned above, a fan-shaped cylindrical shell structure is formed, that is... Figure 2 In the middle, the three-dimensional combination of the green areas forms a local fan-shaped cylindrical shell computational domain. Meanwhile, the volume of the local sector-shaped cylindrical shell computational domain. The formula for calculation using cylindrical coordinate integration is as follows: , Therefore, the radial, circumferential, and axial boundaries are defined using cylindrical coordinates, combined with... Figure 2 A three-dimensional sector-shaped cylindrical shell computational domain is constructed from the two-dimensional view as a local sector-shaped cylindrical shell computational domain to achieve efficient modeling of key areas. The formed local sector-shaped cylindrical shell computational domain can focus on the key areas of the mud-covered inner wall of the hearth around the taphole, avoiding modeling the entire hearth, which can significantly reduce the amount of computation and improve the efficiency of numerical solution. At the same time, it can also limit the solution focus to the key areas where the mud-covered actually forms, thereby reducing the dependence on modeling the entire hearth and improving the solution efficiency.

[0021] Step 102: Within the local sector-shaped cylindrical shell computational domain, multiple cylindrical coordinates are generated through a uniform sampling mechanism. For any cylindrical coordinate, the normalized mud volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential, and axial. The normalized mud volume is the net mud volume obtained by normalizing the actual mud volume after deducting the filling part inside the tap hole channel.

[0022] Next, a four-dimensional vector of "cylindrical coordinates + normalized sludge volume" is generated within the local sector-shaped cylindrical shell computational domain. This four-dimensional vector directly binds each spatial point (r, θ, z) to the corresponding sludge volume, characterizing the spatial distribution of the initial sludge volume in key areas of the hearth inner wall. This provides fine-grained input for the numerical simulation of the sludge bale formation process, accurately reflecting the sludge filling situation at different locations and improving the accuracy of sludge bale morphology prediction. The net sludge volume is the effective sludge volume used for constructing the sludge bale on the hearth inner wall after deducting the filling portion inside the taphole channel, avoiding interference from invalid data. Through normalization, the differences in absolute sludge volume between different blast furnaces and different operating batches can be eliminated, making the data comparable and facilitating cross-scenario model training or parameter optimization.

[0023] Vectors generated by uniform sampling in cylindrical coordinates can cover key areas within the local computational domain, ensuring that the mud-making situation at each spatial location is considered, avoiding omission of important areas for mud bag formation, and improving the comprehensiveness and reliability of simulation results.

[0024] The sampling range within the local computational domain focuses on the key areas of the mud bag, and the number of generated four-dimensional vectors is controllable. This ensures the relevance of the data and avoids redundant calculations in global sampling, which is consistent with the core objective of local modeling (improving solution efficiency).

[0025] Optionally, in step 102, multiple cylindrical coordinates are generated through a uniform cylindrical coordinate sampling mechanism, including: Step 1021, perform the following steps multiple times to obtain multiple cylindrical coordinates: Generate a set of random variables that follow a uniform distribution between 0 and 1, wherein the set of random variables includes a first random variable used to control the radial range. A second random variable used to control the circumferential range and a third random variable used to control the axial range , ; Step 1022: Using the radial component calculation formula, and by offsetting the change in radial area with the increase of the radial component using the square root form, the radial component is obtained. The radial component calculation formula is as follows: , For radial components, The inner radius of the hearth. The inner diameter of the computational domain for a local sector-shaped cylindrical shell; Step 1023: Using the circumferential component calculation formula, obtain the circumferential component, where the circumferential component calculation formula is: , For the circumferential component, The angle of the center of the taphole. It is half the width in the circumference direction; Step 1024: Using the axial component calculation formula, obtain the axial component, where the axial component calculation formula is: , For the axial component, The height of the center of the taphole. Half-height of the axis; Step 1025: Construct a cylindrical coordinate system using the radial, circumferential, and axial components generated from a set of random variables.

[0026] In the above embodiments of this application, it is assumed that the parameters of the local sector-shaped cylindrical shell computational domain are: The inner radius of the furnace hearth is R = 5 m; the inner diameter of the local sector-shaped cylindrical shell calculation domain is... =3 m; Center angle of the taphole =30°, circumferential half-width =10°, center height of the taphole =2 m, axial half height =0.5 m.

[0027] Next, generate three independent, uniformly distributed random variables within the range [0,1], for example: The first random variable u = 0.25 is used to control the radial range, the second random variable v = 0.6 is used to control the circumferential range, and the third random variable w = 0.3 is used to control the axial range.

[0028] Next, the radial component (r) is mapped using the formula... To eliminate the change in radial area as r increases, ensuring uniform area sampling: , ; Mapping the circumferential components using the formula Map v to : , Map the axial component (z) using the formula Map w to : , Next, generate multiple sets of coordinates repeatedly. For example, generate another set of random variables u=0.75, v=0.3, w=0.8, and you will get: , , ; The final two cylindrical coordinates are (3.606, 32°, 1.8) and (4.583, 26°, 2.3).

[0029] Furthermore, within the local sector-shaped cylindrical shell computational domain, after generating multiple cylindrical coordinates through a uniform cylindrical coordinate sampling mechanism, the local sector-shaped cylindrical shell computational domain and the distribution of uniformly random sampling points are shown in the figure below. Figure 3 As shown.

[0030] To this end, in radial mapping, a square root transformation is used to achieve uniform area sampling, avoiding points that are too dense or too sparse near the center of the furnace. In circumferential / axial mapping, a linear transformation maps random variables to a preset range, ensuring coverage of the sector and height intervals of the computational domain. Repeating the above steps yields a uniform set of cylindrical coordinates covering the entire local computational domain, which can provide fine-grained spatial input for subsequent mud-bag simulation.

[0031] This mechanism ensures both the uniformity of sampling and precise matching of the geometric features of the blast furnace taphole area, making it a key link between the mud-scraping operation and numerical simulation.

[0032] In particular, in discrete implementation, for from Obtained by uniform sampling within sampling points Arbitrary function The volume average can be written as: .

[0033] Optionally, in step 102, the normalized sludge volume is combined with the cylindrical coordinates to form a four-dimensional vector for the cylindrical coordinates, including: Step 1026: Convert the cylindrical coordinates to Cartesian coordinates using trigonometric functions, and then rearrange the Cartesian coordinates into a three-dimensional vector to obtain the Cartesian coordinate vector, where the Cartesian coordinate vector is: ; x is a Cartesian coordinate vector, T is the transpose, and (x, y, z) are Cartesian coordinates. , , () represents cylindrical coordinates; Step 1027: Combine the Cartesian coordinate vector with the normalized mud volume to form a four-dimensional vector, where the four-dimensional vector is: ; s is a four-dimensional vector. To normalize the volume of the clay.

[0034] In the above embodiments of this application, the cylindrical coordinates are (r, θ, z), which are converted to Cartesian coordinates (x, y, z) using trigonometric functions: , Then rearrange it into a three-dimensional Cartesian vector: .

[0035] Next, the three-dimensional Cartesian vector is combined with the normalized mud volume. By concatenating the elements, a four-dimensional vector is formed: It integrates spatial location (derived from cylindrical coordinates) and operational parameters (normalized sludge volume).

[0036] Step 103: Using the trained implicit neural model of mud distribution, predict the volume fraction of gunning mud at the spatial location corresponding to the four-dimensional vector, wherein the volume fraction of gunning mud is used to characterize the degree to which the spatial location corresponding to the four-dimensional vector is occupied by gunning mud.

[0037] Next, the trained neural implicit model of mud distribution is used to predict the volume fraction of gun mud corresponding to the four-dimensional vector. By transforming discrete observation data into a continuous, high-precision spatial distribution field, it has the following advantages compared to traditional simulation: Traditional mesh simulations, limited by computational resources, struggle to capture the subtle boundaries of mud pockets near the taphole. Neural implicit models, expressed through continuous functions, can calculate volume fractions in real-time at any coordinate point (not just mesh points), achieving a leap from "discrete pixels" to "high-dimensional vectors" and significantly improving the detail in depicting mud pocket shapes. The trained model can then provide results through lightweight neural network forward computation, achieving faster computation speeds than physical fluid simulations.

[0038] Optionally, in step 103, the volume fraction of the gun clay at the spatial location corresponding to the four-dimensional vector is predicted using the trained implicit neural model of clay distribution, including: Step 1031: The four-dimensional vector is received through the trained mud-bun distribution neural implicit model, so that the trained mud-bun distribution neural implicit model, with a multilayer perceptron as the main body, performs forward propagation calculation on the four-dimensional vector through the trained network parameters, and obtains the intermediate result input into the Sigmoid function. Step 1032: Map the output value of the Sigmoid function to the interval between 0 and 1 to obtain the volume fraction of the mud. The expression for the implicit neural model of mud bag distribution is: , c represents the volume fraction of the clay, which ranges from 0 to 1. c=1 indicates that the clay is completely occupied by the clay, and c=0 indicates that the clay is not occupied by the clay. x is a Cartesian coordinate vector. To normalize the volume of the clay, Here, T is the sigmoid function, and T is the transpose. For learnable network parameters, It is a multilayer perceptron.

[0039] In the above embodiments of this application, the architecture of the mud-bag distribution neural implicit model is as follows: Figure 4 As shown, Figure 4 In this context, "inputlayer" refers to the input layer, which is the layer in the neural network that receives input data. "Spatialposition" refers to the spatial position, specifically the coordinates in space. "x," "y," and "z" represent the three-dimensional spatial coordinates (x, y, z), respectively. "V" and "mudvolume" together represent the normalized mud volume. Hidden layers are located between the input and output layers. They perform complex processing and feature extraction on the input data. There are six hidden layers, each with 128 neurons. The output layer is the last layer of the neural network, responsible for outputting the final result. Volume fraction refers to the volume fraction c of the clay, which is the final output of the neural network and is used to characterize the degree to which the clay occupies a corresponding spatial location.

[0040] In the neural implicit model of mud bag distribution, the activation function of the hidden layer is SiLU, and the output layer is Sigmoid, so that the output mud volume fraction is between 0 and 1.

[0041] Specifically, the input for prediction is a four-dimensional vector, consisting of two parts: three-dimensional spatial coordinates: x, y, z corresponding to Cartesian coordinates. Figure 4 The x, y, and z nodes of the input layer represent spatial locations; the normalized sludge volume... ,correspond Figure 4 The V node of the input layer is the normalized value of the net sludge removal amount, reflecting the operating parameters.

[0042] The trained model, based on a multilayer perceptron (MLP), performs the following operations: The input layer receives a four-dimensional vector and passes it to the hidden layer; The hidden layer consists of 6 layers, each with 128 neurons. The SILU activation function is used for feature extraction and nonlinear transformation to perform complex fusion of the input spatial location and mud volume information. After layer-by-layer computation in the hidden layer, the intermediate result is obtained and input into the Sigmoid function of the output layer.

[0043] The output layer maps the intermediate results to the 0~1 range using the Sigmoid function to obtain the final gun clay volume fraction c: c=1 indicates that the spatial location is completely occupied by gun clay; c=0 indicates that it is not occupied by gun clay. Values ​​between 0 and 1 reflect the degree of occupancy of the stemming material (e.g., c=0.6 means 60% is covered by stemming material).

[0044] Therefore, the trained neural implicit model for mud bale distribution receives a four-dimensional input of spatial location and mud volume, performs feature fusion using a trained MLP, and finally outputs a volume fraction of 0 to 1 through the Sigmoid function to achieve accurate prediction of mud bale distribution.

[0045] Step 104: After completing the unsampled spatial locations within the local sector-shaped cylindrical shell computational domain with the volume fraction of the cutterhead, a continuous cutterhead volume fraction field covering the entire local sector-shaped cylindrical shell computational domain is generated. Based on the continuous cutterhead volume fraction field, the three-dimensional spatial distribution morphology of the cutterhead is determined.

[0046] Next, by completing the volume fraction of the unsampled clay and generating a continuous field, the three-dimensional morphology of the clay bag is finally determined. Essentially, it transforms discrete observation data into a high-precision, full-space digital model of the clay bag, providing key support for the maintenance of the blast furnace taphole from "experience-based judgment" to "quantitative decision-making". Specifically, there are a large number of unsampled "blank areas" in the local fan-shaped cylindrical shell computational domain, such as the inside of the clay bag and the boundary transition zone. Traditional discrete sampling can only obtain fragmented data and cannot reflect the complete outline of the clay bag.

[0047] The completed continuous volume fractional field can cover every spatial point in the computational domain, accurately capturing the fine structure of the mud pouch, such as internal pores and boundary gradient layers, avoiding morphological misjudgment due to insufficient sampling.

[0048] Continuous volume fraction fields can directly calculate key parameters of the mud bag, for example: Geometric features: thickness distribution of the mud mound, radial extension length, and symmetry of the radial profile; Density distribution: the degree of clay occupation in different areas, such as whether the volume fraction in the central area of ​​the taphole meets the standard; These quantitative indicators can replace the traditional "visual observation + experience judgment" to accurately assess whether the mud bag is thick enough and whether there are any weak areas, providing a scientific basis for mud-making operations.

[0049] The three-dimensional morphology predicted by the continuous field can also simulate the influence of different mud-making parameters, such as mud-making amount and mud-making pressure, on the growth of mud blobs, and predict the evolution trend of mud blobs in advance. When it is found that the volume fraction of the mud bag is too low in some areas, such as insufficient occupancy in the side wall area, the mud application position and dosage can be adjusted accordingly to avoid mud bag damage or taphole blockage caused by blind operation, extend taphole life, and reduce the risk of unplanned blast furnace shutdown.

[0050] In particular, the three-dimensional spatial distribution morphology and key characteristics of the clay packing inside the hearth are as follows: Figure 5 As shown, Figure 5 In this context, Z-AXIS stands for "Z-axis," representing the height direction in three-dimensional space, used to locate the vertical position of the mud bag. Figure 5 On the left, the gray solid simulates a partial fan-shaped columnar shell structure of the blast furnace hearth. The hearth is the area at the bottom of the blast furnace where molten iron is stored. The clump-like structure attached to its inner wall simulates the morphology of "mud clumps attached to the inner wall of the hearth near the taphole." The key features correspond to the Z-axis coordinate diagram on the right, showing the volume fraction distribution of the mud clumps in the height direction. The shades of color represent the degree of mud occupation. The overall features show the core characteristic of extending outward from the taphole and varying thickness with position. Figure 5The radial extension length (X-axis value) and height coverage range (Z-axis 0.5-1.5 range) of the mud pack can be read intuitively. These parameters are the key basis for evaluating whether the mud pack can effectively protect the hearth lining.

[0051] Optionally, in step 104, determining the three-dimensional spatial distribution morphology of the mud bag based on the continuous mud volume fraction field includes: Step 1041: In the continuous shot clay volume fraction field, the isosurface extraction algorithm is used to connect all spatial points whose shot clay volume fraction is equal to the preset clay bag entity determination threshold to form a closed surface, and the closed surface is used as the entity boundary of the clay bag. Step 1042: Obtain the three-dimensional spatial distribution morphology of the mud bag based on the entity boundary.

[0052] In the embodiments described above, the process of extracting entity boundaries in a continuous shotcrete volume fraction field is essentially a transformation from a probability distribution to a geometric entity. The volume fraction predicted by the model is a continuously varying value from 0 to 1, representing the probability that a point in space is shotcrete. You need to set a preset threshold for determining the entity in the shotcrete as a logical boundary between "entity" and "non-entity". This threshold is like the height of sea level, determining which parts will emerge as land.

[0053] Next, spatial interpolation and point connection are performed. The algorithm traverses every tiny grid within the local sector-shaped cylindrical shell computational domain. If a vertex of a grid has a volume fraction higher than a threshold, while an adjacent vertex has a volume fraction lower than the threshold, it means that there must exist a point between these two points with a volume fraction exactly equal to the threshold. The algorithm uses linear interpolation to accurately calculate the location of this equilibrium point. In this way, all points satisfying the threshold condition are found and located throughout the entire computational domain.

[0054] Next, the algorithm constructs a closed surface and its shape. Based on topological consistency, the algorithm connects these scattered equilibrium points in space in an orderly manner to form countless tiny triangular patches. These patches are connected end to end, eventually weaving a continuous "skin" covering the entire high-concentration area, that is, a closed surface. The surface completely encloses the high-concentration clay area, and its undulations and extensions in the cylindrical coordinate system intuitively constitute the three-dimensional spatial distribution of the clay around the taphole.

[0055] Optionally, in step 1031, before receiving the four-dimensional vector through the trained mud-bag distribution neural implicit model, the method further includes: Step 105: Construct a multi-objective loss function and train the mud-bag distribution neural implicit model using the multi-objective loss function. The multi-objective loss function is as follows: ; For multi-objective loss functions, , , and These are volume constraints, iron taphole aggregation constraints, interface smoothing and phase separation constraints, and reference point enrichment constraints, respectively. , , and These are the weighting coefficients for volume constraints, iron tapping constraints, interface smoothing and phase separation constraints, and reference point enrichment constraints, respectively. Step 106: When training the mud-bag distribution neural implicit model using a multi-objective loss function, the gradient of the multi-objective loss function with respect to the network parameters is obtained. The Adam optimizer is then used to iteratively update the learnable network parameters based on the gradient until a preset number of iterations are reached to complete the training of the mud-bag distribution neural implicit model. The mud-bag distribution neural implicit model includes network parameters, and the iterative update expression for the network parameters is: ; Characterizing the multi-objective loss function with respect to network parameters The gradient of Adam is the adaptive momentum estimate.

[0056] In the embodiments described above, a multi-objective loss function is constructed and used to train a neural implicit model of mud bale distribution. Volume constraints, taphole aggregation constraints, interface smoothing and phase separation constraints, and reference point enrichment constraints are all incorporated into the optimization process. A cylindrical geometry-related offset potential function is used to directly guide the mud bale to aggregate in the vicinity of the taphole and spread along the inner wall of the hearth. Since the training process does not rely on manually labeled mud bale morphology data, it can adapt to application scenarios where direct observation inside the blast furnace is difficult.

[0057] Specifically, the mud-bag distribution neural implicit model constrains the learning direction through a multi-objective loss function, the expression of which is: , In each multi-objective loss function, the roles and weights of the constraint terms are as follows: Volume constraints Used to ensure that the predicted "mud bag volume" matches the actual mud volume, avoiding volume deviation; Iron mouth cluster constraint : Used to guide the mud bags to gather near the taphole, which conforms to the physical laws of mud bags adhering to the taphole area; Interface smoothing and phase separation constraints Used to smooth the boundaries of the clay bag, while enhancing the separation between the clay and non-clay areas; Reference point enrichment constraint Used to guide the model to learn the distribution of mud bales in key areas by utilizing known reference points, such as the location of the taphole.

[0058] The weighting coefficients are used to balance the importance of each constraint and can be adjusted according to actual needs.

[0059] Next, the network parameters are iteratively optimized through the following steps: Calculate the gradient: for multi-objective loss functions Taking the derivative with respect to the network parameter w, we obtain the gradient. ; Adam optimizer update: Parameters are updated based on gradients using the Adam adaptive momentum estimator. , The termination condition is to repeat the iteration until a preset number of training rounds is reached, such as 50 rounds. The above embodiments of this application do not rely on manually labeled mud bag morphology data, such as the mud bag shape obtained by dissecting a blast furnace. The model is guided to learn only by physical constraints and reference points. Therefore, it can adapt to application scenarios where it is difficult to directly observe the inside of a blast furnace and reduce data acquisition costs.

[0060] Optionally, step 106 trains the mud bag distribution neural implicit model using a multi-objective loss function, including: Step 1061: The volume constraint in the multi-objective loss function is used to make the mud volume fraction output by the mud bag distribution neural implicit model consistent with the target mud volume input for training. Step 1062: The iron tapping aggregation constraint in the multi-objective loss function guides the mud to aggregate in the vicinity of the tapping outlet and distribute along the inner wall of the hearth. Step 1063: Maintain the continuous smoothness of the mud bag interface through the interface smoothing and phase separation constraints in the multi-objective loss function, and promote the polarization of the volume fraction of the gun mud towards the binary state. Step 1064: Increase the probability of the mud occupying the neighborhood of the taphole reference point by enriching constraints in the multi-objective loss function, so as to avoid the prediction result deviating from the reference point in the center region of the taphole.

[0061] In the above embodiments of this application, the volume constraint in the multi-objective loss function ensures that the mud volume fraction output by the mud bag distribution neural implicit model is consistent with the target mud volume input for training. The volume constraint is as follows: ; Let X be the volume fraction of the gun clay at position X. The actual (target) sludge volume is input; the taphole aggregation constraint is characterized by a normalized geometric potential function, and the taphole aggregation constraint is: ; ; For reference position of the taphole, This indicates the radial inward contraction relative to the inner wall of the furnace hearth. This represents the circumferential arc length offset relative to the center angle of the taphole. Indicates axial offset. This represents the maximum value of the geometric potential function within the local sectoral cylindrical shell computational domain. To prevent tiny positive numbers with a denominator of zero.

[0062] Interface smoothing and phase separation constraints are obtained by coupling interface smoothing constraints and binarization constraints. The interface smoothing constraint reduces jagged edges, oscillations, and non-physical fine wrinkles by penalizing the spatial abrupt changes in the volume fraction field. , ; Binarization constraints suppress intermediate states and promote the separation of the mud-covered and non-mud-covered phases, as follows: ; This item is in or Take zero at time, Taking a larger value nearby can suppress intermediate states and promote the separation of the two phases in the mud-covered region and the non-mud-covered region; The interface smoothing and phase separation constraints obtained by coupling interface smoothing constraints and binarization constraints are as follows: ; To adjust the parameters for the relative weights of the two items; The reference point enrichment constraint is obtained by defining a distance decay weight function, which is: ; The attenuation coefficient; The reference point enrichment loss corresponding to the distance decay weighting function is: .

[0063] Optionally, in step 104, after determining the three-dimensional spatial distribution morphology of the mud bag based on the continuous mud volume fraction field, the method further includes: Step 107: In the cylindrical coordinate system, with the center of the taphole as the reference, measure the maximum distance from the outer boundary of the three-dimensional spatial distribution of the mud bag to the inner wall of the hearth along the radial direction as the maximum thickness of the mud bag. Step 108, and / or along the circumferential and axial directions, statistically analyze the three-dimensional spatial distribution of the mud pack on the inner wall of the furnace hearth, as the protection range of the mud pack on the inner wall of the furnace hearth. Step 109, and / or, record the maximum mud pack thickness corresponding to different target mud volumes, and determine the relationship between mud volume and maximum mud pack thickness based on the recorded results.

[0064] In the above embodiments of this application, for the maximum thickness of the clay bag, the outer boundary points of the three-dimensional shape of the clay bag are extracted in the generated continuous clay volume fraction field, with the center of the taphole as the origin of the cylindrical coordinate system. The radial component of each boundary point is used to calculate the distance R from the boundary point to the inner wall of the hearth. The maximum thickness of the mudbag is the maximum value of r among all distances.

[0065] To determine the extent of protection provided by the mud pack to the inner wall of the furnace hearth, the points in the mud pack's shape that cover the inner wall of the furnace hearth are counted. The superposition of these two points forms a fan-shaped protective area, which represents the extent of protection provided by the mud pack to the inner wall of the furnace hearth.

[0066] To investigate the relationship between the volume of mud and its maximum thickness, the maximum thickness of the mud bag corresponding to different net mud volumes can be recorded, and the function mapping or trend curve between the two can be obtained by fitting multiple sets of data.

[0067] When the maximum thickness is below the threshold, it is necessary to replenish the mud in time to avoid the molten iron from washing over the inner wall of the hearth and causing burn-through. The smaller the protection range, the larger the area of ​​the hearth refractory material directly washed by the molten iron. It is necessary to optimize the mud-repairing strategy to extend the blast furnace life. By utilizing the relationship between mud-repairing volume and thickness, the required amount of mud can be quantitatively calculated to avoid waste of taphole mud or poor mud pack formation, thus achieving refined operation and maintenance.

[0068] By applying the technical solution of this embodiment, the morphology of the mud bag is represented as a continuous volume fraction field, and a neural implicit model of the mud bag distribution is constructed with spatial coordinates and normalized mud loading as inputs and the volume fraction of the taphole mud as output, thereby enabling the prediction of the three-dimensional spatial distribution of the mud bag in the taphole area of ​​the blast furnace.

[0069] Furthermore, as Figure 1 In terms of specific implementation, this application provides a blast furnace taphole mud morphology prediction system, such as... Figure 6 As shown, the system includes: The column shell calculation domain construction module 201 is used to construct a local fan-shaped column shell calculation domain for the blast furnace tapping area, with the tapping center as the reference, targeting the spatial structure of the hearth inner wall. The cylindrical coordinate uniform sampling module 202 is used to generate multiple cylindrical coordinates in the local fan-shaped cylindrical shell calculation domain through the cylindrical coordinate uniform sampling mechanism. For any cylindrical coordinate, the normalized mud volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential and axial. The normalized mud volume is the net mud volume after deducting the filling part inside the iron outlet channel from the actual mud volume and then normalizing it. The 203 is used to predict the volume fraction of the gunning clay at the spatial location corresponding to the four-dimensional vector by using a trained neural implicit model of the clay bag distribution. The volume fraction of the gunning clay is used to characterize the degree to which the spatial location corresponding to the four-dimensional vector is occupied by gunning clay. The blast furnace taphole mud bag morphology prediction module 204 is used to complete the unsampled spatial positions in the local fan-shaped cylindrical shell calculation domain by adding the mud volume fraction, and then generate a continuous mud volume fraction field covering the entire local fan-shaped cylindrical shell calculation domain. Based on the continuous mud volume fraction field, the three-dimensional spatial distribution morphology of the mud bag is determined.

[0070] It should be noted that other corresponding descriptions of the functional units involved in the blast furnace taphole mud morphology prediction system provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in the method will not be repeated here.

[0071] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented using software and necessary general-purpose hardware platforms, or it can be implemented using hardware to construct a local fan-shaped cylindrical shell computational domain of the hearth inner wall with the taphole center as the reference for the blast furnace taphole area; within the domain, spatial position points are generated by uniformly sampling using cylindrical coordinates; the normalized mud volume is combined with the cylindrical coordinates to form a four-dimensional vector, which is input into the trained mud bag distribution neural implicit model to predict the mud volume fraction at the corresponding position; the volume fraction of unsampled points within the domain is supplemented to generate a continuous mud volume fraction field, thereby determining the three-dimensional spatial distribution morphology of the mud bag. This method enables stable, quantitative, and engineering-experience-compliant prediction of the three-dimensional spatial morphology of the mud bag in the blast furnace taphole area, even in the absence of measured and labeled data of the mud bag interior and without relying on complex explicit control equations.

[0072] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the system of the embodiment scenario can be distributed throughout the system of the embodiment scenario as described, or they can be modified to reside in one or more systems different from this embodiment scenario. The modules of the above-described embodiment scenario can be combined into one module, or further divided into multiple sub-modules.

[0073] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any modifications that can be made by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for predicting the morphology of the mud bale at the blast furnace taphole, characterized in that, The method includes: For the blast furnace tapping area, a local sector-shaped cylindrical shell computational domain is constructed with the tapping center as the reference for the spatial structure of the hearth inner wall. Within the local sector-shaped cylindrical shell computational domain, multiple cylindrical coordinates are generated through a uniform sampling mechanism. For any cylindrical coordinate, the normalized mud volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential, and axial. The normalized mud volume is the net mud volume obtained by normalizing the actual mud volume after deducting the filling part inside the tap hole channel. The trained implicit neural model of mud distribution predicts the volume fraction of gunning mud at the spatial location corresponding to the four-dimensional vector, where the volume fraction of gunning mud is used to characterize the degree to which gunning mud occupies the spatial location corresponding to the four-dimensional vector. After completing the unsampled spatial locations within the local sector-shaped cylindrical shell computational domain with the volume fraction of the cutterhead, a continuous cutterhead volume fraction field covering the entire local sector-shaped cylindrical shell computational domain is generated. Based on the continuous cutterhead volume fraction field, the three-dimensional spatial distribution morphology of the cutterhead is determined.

2. The method according to claim 1, characterized in that, Multiple cylindrical coordinates are generated through a uniform cylindrical coordinate sampling mechanism, including: To obtain multiple cylindrical coordinates, perform the following steps multiple times: Generate a set of random variables that follow a uniform distribution between 0 and 1, wherein the set of random variables includes a first random variable used to control the radial range. A second random variable used to control the circumferential range and a third random variable used to control the axial range , ; The radial component is obtained by using the formula for calculating the radial component and offsetting the change in radial area with increasing radial component using the square root form. The formula for calculating the radial component is as follows: ， For radial components, The inner radius of the hearth. The inner diameter of the computational domain for a local sector-shaped cylindrical shell; The circumferential components are obtained using the formula for calculating the circumferential components. The formula for calculating the circumferential components is as follows: ， For the circumferential component, The angle of the center of the taphole. It is half the width in the circumference direction; The axial components are obtained using the axial component calculation formula, which is as follows: ， For the axial component, The height of the center of the taphole. Half-height of the axis; A cylindrical coordinate system is constructed using radial, circumferential, and axial components generated from a set of random variables.

3. The method according to claim 1, characterized in that, The step of combining the normalized sludge volume with the cylindrical coordinates to form a four-dimensional vector for the cylindrical coordinates includes: The cylindrical coordinates are converted to Cartesian coordinates using trigonometric functions. The Cartesian coordinates are then rearranged into a three-dimensional vector to obtain the Cartesian coordinate vector, which is: ； x is a Cartesian coordinate vector, T is the transpose, and (x, y, z) are Cartesian coordinates. , , () represents cylindrical coordinates; By combining the Cartesian coordinate vector with the normalized mud volume, a four-dimensional vector is formed, which is: ； s is a four-dimensional vector. To normalize the volume of the clay.

4. The method according to claim 1, characterized in that, Using a trained neural implicit model of mud distribution, the volume fraction of gun mud at the spatial location corresponding to the four-dimensional vector is predicted, including: The trained mud-bun distribution neural implicit model receives a four-dimensional vector, and the trained mud-bun distribution neural implicit model, with a multilayer perceptron as the main body, performs forward propagation calculation on the four-dimensional vector through the trained network parameters, and obtains the intermediate result input into the Sigmoid function. Mapping the output of the Sigmoid function to the interval between 0 and 1 yields the volume fraction of the mud, where the expression for the implicit neural model of mud bag distribution is: ， c represents the volume fraction of the clay, which ranges from 0 to 1. c=1 indicates that the clay is completely occupied by the clay, and c=0 indicates that the clay is not occupied by the clay. x is a Cartesian coordinate vector. To normalize the volume of the clay, Here, T is the sigmoid function, and T is the transpose. For learnable network parameters, It is a multilayer perceptron.

5. The method according to claim 4, characterized in that, Before receiving the four-dimensional vector through the trained mud-bun distribution neural implicit model, the method further includes: A multi-objective loss function is constructed, and the mud-bag distribution neural implicit model is trained using the multi-objective loss function. The multi-objective loss function is as follows: ； For multi-objective loss functions, , , and These are volume constraints, iron taphole aggregation constraints, interface smoothing and phase separation constraints, and reference point enrichment constraints, respectively. , , and These are the weighting coefficients for volume constraints, iron tapping constraints, interface smoothing and phase separation constraints, and reference point enrichment constraints, respectively. When training the mud-bag distribution neural implicit model using a multi-objective loss function, the gradient of the multi-objective loss function with respect to the network parameters is obtained. The Adam optimizer is then used to iteratively update the learnable network parameters based on this gradient until a preset number of iterations are reached to complete the training of the mud-bag distribution neural implicit model. The mud-bag distribution neural implicit model includes network parameters, and the iterative update expression for the network parameters is as follows: ； Characterizing the multi-objective loss function with respect to network parameters The gradient of Adam is the adaptive momentum estimate.

6. The method according to claim 5, characterized in that, The method of training the mud-bag distribution neural implicit model using a multi-objective loss function includes: The volume constraint in the multi-objective loss function ensures that the mud volume fraction output by the mud bag distribution neural implicit model is consistent with the target mud volume input for training. The iron tapping aggregation constraint in the multi-objective loss function guides the mud bales to aggregate in the vicinity of the iron tapping tap and distribute along the inner wall of the hearth. The continuous smoothness of the mud bag interface is maintained by the interface smoothing and phase separation constraints in the multi-objective loss function, and the volume fraction of the gun mud is promoted to polarize towards the binary state. The probability of occupancy of the taphole reference point neighborhood is increased by enrichment constraints in the multi-objective loss function.

7. The method according to claim 1, characterized in that, The three-dimensional spatial distribution morphology of the mud bag is determined based on the continuous mud volume fraction field, including: In the continuous shot clay volume fraction field, the isosurface extraction algorithm is used to connect all spatial points whose shot clay volume fraction is equal to the preset clay bag entity determination threshold to form a closed surface, and the closed surface is used as the entity boundary of the clay bag. The three-dimensional spatial distribution of the mud bag is obtained based on the entity boundary.

8. The method according to any one of claims 1 to 7, characterized in that, After determining the three-dimensional spatial distribution morphology of the mud bag based on the continuous mud volume fraction field, the method further includes: In a cylindrical coordinate system, with the center of the taphole as the reference, the maximum distance from the outer boundary of the three-dimensional spatial distribution of the mud bag to the inner wall of the hearth is measured along the radial direction as the maximum thickness of the mud bag. And / or along the circumferential and axial directions, the three-dimensional spatial distribution of the mud packs is statistically analyzed to determine the coverage area of ​​the mud packs on the inner wall of the furnace hearth, which is used as the protection range of the mud packs on the inner wall of the furnace hearth. And / or, record the maximum mud pack thickness corresponding to different target mud volumes, and determine the relationship between mud volume and maximum mud pack thickness based on the recorded results.

9. The method according to claim 8, characterized in that, Using the taphole center as a reference, a local sector-shaped cylindrical shell computational domain is constructed for the internal structure of the hearth, including: Determine the boundary parameters used to define the boundary of the local sector-shaped cylindrical shell computational domain, wherein the boundary parameters include the inner radius of the hearth, the center angle of the taphole, the circumferential half-width, the center height of the taphole, the axial half-height, and the inner diameter of the local sector-shaped cylindrical shell computational domain; Using the inner diameter of the local sector-shaped cylindrical shell computational domain as the inner boundary and the inner radius of the furnace hearth as the outer boundary, a radial range is constructed, where the radial range is represented as... ; Centered on the center angle of the iron outlet, extend half a width circumferentially to both sides to construct a circumferential range, where the circumferential range is represented as... ; With the center height of the taphole as the center, extend the axial half-height both above and below to construct the axial range, where the axial range is represented as... ; Based on the sector-shaped cylindrical shell structure formed by the intersection of the radial, circumferential, and axial ranges, a local sector-shaped cylindrical shell computational domain is constructed, which is represented as: For the local sector-shaped cylindrical shell computational domain, For radial components, For the circumferential component, For the axial component, The inner radius of the hearth. For the inner diameter of the computational domain of the local sector-shaped cylindrical shell, The angle of the center of the taphole. The circumferential half-width, The height of the center of the taphole. It is the axial half-height.

10. A blast furnace taphole mud bale morphology prediction system, characterized in that, The system includes: The column shell computational domain construction module is used to construct a local fan-shaped column shell computational domain for the blast furnace tapping area, with the tapping center as the reference, targeting the spatial structure of the hearth inner wall. The cylindrical coordinate uniform sampling module is used to generate multiple cylindrical coordinates within the local fan-shaped cylindrical shell computational domain through a cylindrical coordinate uniform sampling mechanism. For any cylindrical coordinate, the normalized mud volume is combined with the cylindrical coordinate to form a four-dimensional vector for the cylindrical coordinate. The cylindrical coordinate corresponds to a spatial position characterized by three components: radial, circumferential, and axial. The normalized mud volume is the net mud volume obtained by normalizing the actual mud volume after deducting the filling part inside the iron outlet channel. The slurry volume fraction prediction module is used to predict the slurry volume fraction at the spatial location corresponding to the four-dimensional vector by using a trained neural implicit model of slurry distribution. The slurry volume fraction is used to characterize the degree to which the spatial location corresponding to the four-dimensional vector is occupied by slurry. The blast furnace taphole mud bag morphology prediction module is used to complete the unsampled spatial locations within the local fan-shaped cylindrical shell computational domain by adding the mud volume fraction, and then generate a continuous mud volume fraction field covering the entire local fan-shaped cylindrical shell computational domain. Based on the continuous mud volume fraction field, the three-dimensional spatial distribution morphology of the mud bag is determined.

Images