Mechanism data hybrid driving steel liquid temperature soft measurement method for induction melting furnace

Abstract

The present application belongs to the technical field of metallurgical automation, and particularly relates to a mechanism data hybrid driving induction melting and casting furnace liquid steel temperature soft measurement method. A mechanism model is first constructed, including a liquid steel energy balance model in the refining and heating process and a furnace heat transfer model, real-time production data in the liquid steel refining process is collected, including the initial temperature and the final temperature of the liquid steel, the heating power, the power frequency, then a neural network model is constructed with the heating power and the power frequency as the input and the electromagnetic induction heat efficiency as the output, the electromagnetic induction heat efficiency output by the neural network model is substituted into the mechanism model to obtain the predicted final temperature, and the error between the predicted final temperature and the actual final temperature is taken as the loss function to train and optimize the neural network model, based on the mechanism model and the trained neural network model, a liquid steel temperature soft measurement model is constructed to realize online real-time prediction of the liquid steel temperature.

Description

Technical Field

[0001] This invention belongs to the field of metallurgical automation technology, specifically relating to a soft measurement method for molten steel temperature in an induction casting furnace driven by a mixture of mechanism data. Background Technology

[0002] Vacuum electromagnetic casting is a core process for manufacturing clean castings and key components for high-end equipment such as aero-engines and medical devices. This process relies on a vacuum environment to avoid atmospheric pollution and combines electromagnetic induction heating and electromagnetic stirring to achieve homogenization of alloy composition and effective removal of impurities from the melt. The precision of controlling the superheated temperature of the molten pool during the refining stage directly affects the quality and service performance of the final product. Therefore, achieving accurate online measurement of the molten steel temperature is a crucial prerequisite for ensuring product quality stability and meeting the high-performance manufacturing requirements of high-end components.

[0003] However, due to the complex operating conditions of the vacuum electromagnetic induction heating refining process, the online accurate measurement of molten steel temperature faces two major technical bottlenecks, and existing conventional temperature measurement methods cannot meet the process control requirements. First, contact measurement methods have inherent limitations. During vacuum refining, the temperature of molten steel is typically in an extreme high-temperature range above 1300℃. In this combined environment of high temperature and strong chemical corrosion, contact temperature sensing elements such as thermocouples are prone to burn-out and dissolution when used for continuous temperature measurement. This not only results in a short service life but also contaminates the molten steel and damages the cleanliness of the melt. They can only obtain discrete temperature data through intermittent insertion and removal measurements, which cannot meet the actual industrial production requirements for real-time, continuous, and stable temperature measurement. Second, the accuracy of non-contact measurement methods is difficult to meet. To achieve continuous measurement of molten steel temperature, non-contact temperature measurement technology such as infrared thermometry is commonly used in industry. However, during electromagnetic casting, the intense electromagnetic stirring caused by the high-intensity induced current leads to continuous and violent churning and irregular fluctuations on the surface of the molten steel in the crucible, accompanied by slag accumulation and even splashing. This causes significant dynamic drift in the emissivity of the thermal radiation signal received by the infrared thermometer, coupled with complex background interference, resulting in low reliability of the final temperature measurement results. The actual measurement error often reaches ±50℃ or more, which far exceeds the requirements for precise temperature control in high-end casting production, making it difficult to support stable production and quality control of high-end castings.

[0004] In summary, the current lack of accurate and reliable online measurement methods for molten steel temperature in the vacuum electromagnetic casting field severely restricts the level of precision control and product quality stability of this process. Therefore, an accurate and reliable online measurement method for molten steel temperature is urgently needed. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a mechanism- and data-driven soft measurement method for the temperature of molten steel in an electromagnetic induction melting furnace, which overcomes the limitations of contact and non-contact temperature measurement methods and achieves high-precision, continuous online prediction of molten steel temperature.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] Step 1: Construct a mechanism model for the refining process of molten steel by electromagnetic induction heating, including an energy balance model of molten steel during the refining heating process and a heat transfer model within the furnace.

[0008] Step 1.1 Constructing an energy balance model for molten steel during the refining heating process:

[0009] During the refining heating process, the electromagnetic induction heating power absorbed by the molten steel is used to increase its own heat and compensate for heat loss to the surrounding environment. Based on this, a continuous form differential equation for the energy balance of the molten steel during the refining heating process is established:

[0010] (1);

[0011] in, This refers to the heating power, specifically the coil power during the refining heating process. The power supply frequency; For electromagnetic induction thermal efficiency; Specific heat of molten steel; For the quality of molten steel; The temperature of the molten steel; Heating time; This represents the power lost due to heat in molten steel.

[0012] To meet the discretization requirements for data sampling and model solving, the above continuous equations are discretized in the time domain. Let the time step be... , with T l T represents the temperature of the molten steel at the l-th sampling time. l-1 Let be the temperature of the molten steel at the previous sampling moment. Then, the discretized energy balance equation can be expressed as:

[0013] (2);

[0014] in, and These represent the heating power and power frequency at the l-th sampling time, respectively. Let represent the heat loss power of the molten steel at the corresponding time. Based on the recursive relationship of equation (2), the final temperature can be obtained by iteratively solving from the initial temperature within each discretized time interval.

[0015] Step 1.2 Constructing a furnace heat transfer model for the refining heating process: Under vacuum conditions, calculate the heat loss of the molten steel based on radiative heat dissipation from the surface of the molten steel and conductive heat dissipation from the molten steel along the crucible wall.

[0016] (3);

[0017] in, This refers to the outward radiative heat flow from the surface of molten steel. This refers to the heat flow conducted outwards along the crucible wall from the molten steel.

[0018] Step 1.3: Calculate the outward radiative heat flow from the surface of the molten steel. .

[0019] Step 1.4: Calculate the outward heat conduction of molten steel along the crucible wall. .

[0020] Step 2: Collect real-time production data during the electromagnetic induction heating refining process of molten steel to construct a production dataset, including the initial and final temperatures of the molten steel, heating power, and power supply frequency. Data is automatically recorded at fixed sampling intervals to ensure temporal correspondence, providing a complete data foundation for subsequent modeling and training.

[0021] Step 3: Preprocess the real-time production data collected in Step 2, including removing abnormal data that deviates significantly from reality, normalizing the heating power and power frequency, and dividing the preprocessed production dataset into training set, validation set and test set.

[0022] Step 4: Construct a neural network model with heating power and power frequency as inputs and electromagnetic induction thermal efficiency as output. Determine the number of hidden layer nodes and activation functions in the neural network model, and initialize the neural network model parameters based on the intelligent optimization algorithm.

[0023] Step 5: Substitute the electromagnetic induction thermal efficiency output by the neural network model in Step 4 into the mechanism model in Step 1 to solve for the termination temperature of the molten steel and obtain the predicted termination temperature; and use the mean square error between the predicted termination temperature and the actual termination temperature as the loss function to train and optimize the neural network model through an intelligent optimization algorithm.

[0024] Step 6: Based on the mechanism model in Step 1 and the neural network model trained in Step 5, construct a soft measurement model for molten steel temperature to achieve online real-time prediction of molten steel temperature.

[0025] The specific operation includes the following: real-time production data of multiple independent heating zones during the refining process of molten steel heated by electromagnetic induction are collected. Each zone covers different combinations of starting and ending temperatures, and the zone duration and temperature rise range vary. Before the start of the heating zone, the heating power is reduced to stabilize the surface fluctuations of the molten steel. The initial temperature value is measured and recorded using an infrared thermometer as the starting point for the iterative solution of the neural network model. Subsequently, the power is restored for heating. The neural network model collects heating power and power frequency data in real time and continuously outputs the predicted temperature value of the molten steel. When the heating zone ends, the power is reduced again and the actual ending temperature value is measured. This value is compared with the predicted ending temperature output by the neural network model to optimize the neural network model.

[0026] The present invention, employing the above-mentioned technical solution, has the following beneficial effects and significant advantages: It provides a mechanism- and data-driven soft measurement method for molten steel temperature in electromagnetic induction casting furnaces, achieving high-precision, continuous online prediction of molten steel temperature, effectively overcoming the shortcomings of discontinuous contact temperature measurement and severe distortion in non-contact infrared temperature measurement. The model built by the mechanism- and data-driven modeling method possesses both physical interpretability and complex nonlinear fitting capabilities, and its prediction accuracy is superior to that of pure mechanism modeling or pure data-driven modeling.

[0027] Following this line of thought, this invention uses energy balance and heat transfer mechanisms as the main framework and introduces a neural network to fit the electromagnetic induction thermal efficiency. Unlike conventional neural network training, this scenario does not provide training data pairs with heating power and power frequency as inputs and electromagnetic induction thermal efficiency as output. Therefore, the electromagnetic induction thermal efficiency output by the neural network model is substituted into the mechanism model to solve for the termination temperature. Then, the temperature prediction error is used to inversely optimize the network parameters. This effectively solves the modeling problem caused by the inability to directly measure electromagnetic induction thermal efficiency, constituting the core innovation of this invention.

[0028] This invention achieves an organic combination of "white-box" reliability and "black-box" adaptability by constructing a hybrid model architecture that integrates mechanism and data-driven approaches. This effectively enhances the model's generalization ability and reliability, and improves prediction accuracy. Secondly, by introducing intelligent optimization algorithms to train and fine-tune the neural network, the system can efficiently and accurately identify and approximate the optimal model parameters under complex operating conditions. Furthermore, this solution only requires the collection of easily measurable process variables such as heating power and power frequency, without requiring modifications to existing equipment, exhibiting good engineering applicability and low-cost deployment. The soft measurement technology provided by this invention, by constructing a model and utilizing easily measurable variables to infer key process parameters, offers a highly promising technical path for overcoming engineering bottlenecks in existing technologies and achieving accurate online measurement of molten steel temperature. Attached Figure Description

[0029] Figure 1The flowchart is for the mechanism data-driven soft measurement method of molten steel temperature in an induction casting furnace provided in Embodiment 1 of the present invention.

[0030] Figure 2 This is a schematic diagram of a three-surface enclosed cavity radiation heat transfer network model provided in Embodiment 1 of the present invention. Detailed Implementation

[0031] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. It should be noted that the embodiments described in this invention are only for further explanation and illustration, and are not intended to limit its application scope. Based on this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the protection scope of this invention.

[0032] To more clearly and completely illustrate the technical solution of the present invention, further description is provided below in conjunction with the accompanying drawings and embodiments. It should be noted that the specific embodiments below are only used to explain the present invention and are not intended to limit the scope of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] This invention employs a hybrid modeling approach driven by both mechanism and data. By constructing a model of the energy balance of molten steel and the heat transfer mechanism within the furnace, and combining this with a neural network trained on actual process data to identify key parameters, a soft measurement model of molten steel temperature that can adapt to complex working conditions is finally established. This enables high-precision, continuous online prediction of molten steel temperature during the refining process, providing a reliable basis for the refined control of vacuum electromagnetic induction casting process.

[0034] Example 1

[0035] This embodiment uses the temperature control of the refining process in a medium-frequency vacuum induction furnace at an aerospace parts manufacturing plant as an application scenario to provide a detailed description of the technical solution of the present invention. The method flow is as follows: Figure 1 As shown.

[0036] Step 1: Construct a mechanism model for the refining process of molten steel by electromagnetic induction heating, including an energy balance model of molten steel during the refining heating process and a heat transfer model within the furnace.

[0037] Step 1.1: Construct an energy balance model for molten steel during the refining heating process.

[0038] During the refining heating process, the electromagnetic induction heating power absorbed by the molten steel is used to increase its own heat and compensate for heat loss to the surrounding environment. Based on this, a continuous form differential equation for the energy balance of the molten steel during the refining heating process is established:

[0039] (1);

[0040] in, This refers to the heating power, specifically the coil power during the refining heating process. The power supply frequency; For electromagnetic induction thermal efficiency; Specific heat of molten steel; For the quality of molten steel; The temperature of the molten steel; Heating time; This represents the power lost due to heat in molten steel.

[0041] To meet the discretization requirements for data sampling and model solving, the continuous form differential equation of the energy balance of molten steel in the refining heating process, as shown in equation (1), is discretized in the time domain. Let the time step be... , with T l T represents the temperature of the molten steel at the l-th sampling time. l-1 Let be the temperature of the molten steel at the previous sampling moment. Then, the discretized energy balance equation can be expressed as:

[0042] (2);

[0043] in, and These represent the heating power and power frequency at the l-th sampling time, respectively. This represents the heat loss power of the molten steel at the corresponding time point. Based on the above recursive relationship, the final temperature can be obtained by iteratively solving from the initial temperature within each discretized time interval.

[0044] Step 1.2: Construct a furnace heat transfer model for the refining heating process.

[0045] In a vacuum environment, the heat loss of the molten steel is calculated based on the outward radiation heat dissipation from the surface of the molten steel and the outward conduction heat dissipation from the molten steel along the crucible wall:

[0046] (3);

[0047] in, This refers to the outward radiative heat flow from the surface of molten steel. This refers to the heat flow conducted outwards along the crucible wall from the molten steel.

[0048] Step 1.3: Calculate the outward radiative heat flow from the surface of the molten steel. .

[0049] This embodiment uses a three-surface closed-cavity radiation network model to perform thermal radiation analysis on the upper surface of the molten steel, the crucible wall not in contact with the molten steel, and the crucible opening, and then calculates the outward radiative heat flow from the surface of the molten steel. Figure 2As shown, let the upper surface of the molten steel be surface 1, the crucible wall not in contact with the molten steel be surface 2, and the opening of the crucible be surface 3. Surface 3 can be considered a blackbody. The effective radiation of surface 3 is then... Equal to its own radiation force ,Right now .set up Let be the angle coefficient of surface a with respect to surface b. , .

[0050] First, calculate the angle coefficients between the surfaces: based on the geometric relationship between surface 1 and surface 3, obtain the angle coefficient of surface 1 with respect to surface 3. Then, based on the integrity of the angle coefficient, the angle coefficient of surface 1 relative to surface 2 is obtained. The angle coefficient of surface 2 with respect to surface 3 is obtained based on the relativity and symmetry of the angle coefficient. .

[0051] Construct a three-surface closed cavity radiation network model and define the surface thermal resistance and spatial thermal resistance of each surface:

[0052] (4);

[0053] (5);

[0054] in, Let a be the surface thermal resistance of surface a. The spatial thermal resistance of surface a to surface b. Let a be the emissivity of surface a. Let be the area of ​​surface a.

[0055] right Figure 2 Effective radiation intermediate nodes of surface 1 and surface 2 in the radiation network , Applying Kirchhoff's laws, a system of nodal equations is established, and the effective radiation is obtained by solving the system simultaneously. , Then, the radiative heat flow of surface 1 and surface 2 is calculated. and :

[0056] (6);

[0057] (7);

[0058] Among them, the self-radiation force of surface 1 Surface 2 self-radiation force , Let be the blackbody radiation constant, and its value is . Let T1 and T2 be the temperatures of surface 1 and surface 2, respectively. Assuming the heat transfer system inside the furnace is in steady state, the radiant heat absorbed by surface 2 is in equilibrium with the heat it conducts outwards and the radiant heat emitted from the outer wall of the sleeve surrounding the crucible. Establish the heat balance equation between conduction and radiation:

[0059] (8);

[0060] (9);

[0061] (10);

[0062] in, The heat flow is conducted outward from surface 2 of the sleeve. This refers to the radiative heat flow dissipated from the outer wall of the sleeve. The total thermal resistance from the inner wall of the crucible to the outer wall of the sleeve. Surface 2 corresponds to the outer wall area of ​​the sleeve. The emissivity of the outer wall of the sleeve. This refers to the temperature of the furnace inner wall.

[0063] Solving the heat balance equations for heat conduction and radiation allows us to determine the temperature of the crucible's inner wall in the portion not in contact with the molten steel. and the temperature of the outer wall of the sleeve in the part not in contact with the molten steel. This leads to the conclusion of the outward radiative heat flow from the surface of the molten steel. .

[0064] Step 1.4: Calculate the outward heat conduction of molten steel along the crucible wall. .

[0065] In this embodiment, the crucible structure has cylindrical sidewalls and a hemispherical bottom, with a cylindrical sleeve on the outside. Based on this structural feature, the heat loss of molten steel transferred outward through the crucible wall can be divided into two parts: the heat loss of the sidewalls (cylindrical part) and the heat loss of the bottom (hemispherical part).

[0066] (1) The heat loss of molten steel through the cylindrical part of the side wall is calculated using the following method:

[0067] (8');

[0068] (9');

[0069] in, This refers to the heat flow that is conducted out of the sleeve by the molten steel through the cylindrical sidewall. This refers to the radiative heat flow emitted by the molten steel through the outer wall of the sleeve corresponding to the side cylinder. The total thermal resistance from the inner wall of the crucible to the outer wall of the sleeve. This represents the area of ​​the outer wall of the sleeve corresponding to the cylindrical sidewall of the molten steel. The emissivity of the outer wall of the sleeve. This refers to the temperature of the furnace inner wall.

[0070] Solving equations (8') and (9') yields the temperature of the crucible's inner wall as the molten steel passes through the cylindrical section of the sidewall. And the temperature of the outer wall of the sleeve corresponding to the side-wall cylinder as the molten steel passes through. This leads to the conclusion that the heat flow of molten steel is conducted out of the sleeve through the side-wall cylinder. This refers to the heat loss of molten steel through the cylindrical section of the side wall.

[0071] The heat conduction loss of the sidewall portion of the molten steel is calculated using the same method as the heat conduction from surface 2 in step 1.3, both based on a multi-layer cylindrical wall heat conduction model. The main difference is that the inner wall temperature of the crucible portion in contact with the molten steel is known and equal to the molten steel temperature, so the heat loss can be calculated simply by combining formulas (8') and (9'); however, the inner wall temperature of the crucible portion not in contact with the molten steel in step 1.3 is unknown and requires further calculation. Figure 2 The radiative heat transfer network and the conduction process in the equation are solved simultaneously. For simplicity, the specific formulas will not be repeated here.

[0072] (2) The heat loss of molten steel through the bottom hemispherical conduction is calculated as follows:

[0073] To simplify the calculation, the hemispherical portion of the crucible is discretized vertically along the longitudinal section, and it is assumed that heat dissipation from the discretized bottom hemispherical portion occurs only in the radial and vertically downward directions. Starting from the center of the hemispherical circle, a fixed step size s is taken downwards, with each fixed step size yielding a discrete point. After a fixed step size s, discrete points are obtained. , The vertical height from the center of the circle is .

[0074] Then discrete points Horizontal distance to the inner wall, outer wall and outer wall of the crucible , , They are respectively:

[0075] (11);

[0076] (12);

[0077] (13);

[0078] in, , , These represent the inner radius, outer radius, and sleeve radius, respectively. Discrete points Height from the bottom of the crucible From this, the bottom of the crucible can be obtained. Thermal resistance of molten steel in the radial direction at a certain height for:

[0079] (14);

[0080] in, , Let be the thermal conductivity of the crucible and the sleeve, respectively, and x represent the radial direction. Let the temperature of the molten steel and the corresponding temperature of the outer wall of the sleeve be respectively... , , The radial heat loss of the molten steel at the bottom of the crucible is then... for:

[0081] (15);

[0082] For the k-th ring segmented by step size s, its corresponding sphere center angle for:

[0083] (16);

[0084] Based on the ratio of the area of ​​the spherical cap, the surface area of ​​the k-th annulus The following recursive formula can be used to calculate:

[0085] (17);

[0086] in, This represents the surface area of ​​the p-th annulus preceding the k-th annulus.

[0087] Introducing projection angle Then the projected area of ​​the k-th annulus on the bottom surface for:

[0088] (18);

[0089] By combining geometric relationships, the vertical downward thermal resistance of the k-th ring can be calculated. for:

[0090] (19);

[0091] Where H is the distance from the plane containing the center of the sphere at the bottom of the crucible to the bottom of the sleeve. Let the temperature of the circular region corresponding to the bottom of the sleeve and the bottom of the crucible be... Therefore, the heat loss of the molten steel at the bottom of the crucible along the vertical downward direction can be obtained. for:

[0092] (20);

[0093] Then, based on the thermal balance relationship between the heat conduction of the molten steel and the radiation of the sleeve, the heat dissipation of the molten steel at the bottom of the crucible can be solved simultaneously, that is, the heat loss of the molten steel through the bottom hemispherical part.

[0094] Step 2: Collect real-time production data in segments during the electromagnetic refining process to construct a production dataset, including the initial and final temperatures of the molten steel, heating power, and power supply frequency. Data is automatically recorded at fixed sampling intervals to ensure temporal correspondence, providing a complete data foundation for subsequent modeling and training.

[0095] In this embodiment, K188 high-temperature alloy was used as the casting object. The data collected consisted of multiple independent heating zones, each covering different combinations of starting and ending temperatures. The duration and temperature rise range of each zone varied, and the sampling interval was fixed at 1 second. The actual measured values ​​of the initial and ending temperatures of each zone were obtained as follows: before and after the start of the heating zone, the heating power was reduced to stabilize the surface fluctuations of the molten steel. Then, an infrared thermometer was used to measure and record the values, which were used as the reference temperature for model training and verification.

[0096] Step 3: Preprocess the real-time production data collected in Step 2, including removing outliers that significantly deviate from reality, normalizing the heating power and power frequency, and dividing the preprocessed production dataset into training, validation, and test sets, specifically:

[0097] Step 3.1: Preprocess the real-time production data collected in Step 2, removing temperature noise data that significantly deviates from the normal process range. The heating power and power frequency are normalized using the maximum-minimum normalization method, as shown in the following formula:

[0098] (twenty one);

[0099] in, The raw data collected, The minimum value in the original data. The maximum value in the original data. This is the data after normalization.

[0100] Step 3.2: Divide the real-time production dataset processed in Step 3.1 into training set, validation set and test set in a ratio of 8:1:1.

[0101] Step 4: Construct a neural network model with heating power and power frequency as inputs and electromagnetic induction thermal efficiency as output. Determine the number of hidden layer nodes and activation functions in the neural network model, and initialize the neural network model parameters based on an intelligent optimization algorithm. As a preferred implementation, this embodiment uses the differential evolution algorithm to implement the initialization process. Specifically:

[0102] Step 4.1: In this embodiment, a single-hidden-layer network is used as the neural network model. The number of nodes in the hidden layer is calculated using the following formula:

[0103] (twenty two);

[0104] Where q is the initial number of nodes in the hidden layer, n and m are the number of nodes in the input and output layers, respectively, and c is an adjustment constant ranging from 1 to 10. The sigmoid function is used as the activation function.

[0105] Step 4.2: Encode all weights and thresholds in the neural network model into a D-dimensional vector. , Let W be the population size and G be the generation number in the iteration process. The initial population is randomly generated within the search space using the following method:

[0106] (twenty three);

[0107] Where rand is a random number between [0,1]. , .

[0108] Step 5: Substitute the electromagnetic induction thermal efficiency output by the neural network model in Step 4 into the mechanism model in Step 1 to solve for the termination temperature of the molten steel, obtaining the predicted termination temperature; and use the mean square error between the predicted termination temperature and the actual termination temperature as the loss function, train and optimize the neural network model using the differential evolution algorithm, specifically:

[0109] Differential evolution algorithm generates intermediate individuals through the following mutation operation. :

[0110] (twenty four);

[0111] in, and F is a variation factor between 0 and 1.

[0112] Utilize target individuals and intermediate individuals obtained through mutation Perform crossover operations to obtain undetermined experimental individuals. The specific operation for the components is as follows:

[0113] (25);

[0114] Where CR is the cross factor between 0 and 1. A dimension is randomly chosen from all dimensions. The numerical values ​​at each position in a D-dimensional vector are called components. Corresponding to , , .

[0115] In this embodiment, the mean squared error (MSE) is selected as the loss function, and it is also used as the fitness value function in the differential evolution algorithm. Its expression is as follows:

[0116] (26);

[0117] Where M is the number of samples in the training set. For training set samples The actual measured temperature value, For training set samples The predicted termination temperature.

[0118] By selecting the test individual Perform fitness evaluation:

[0119] (27);

[0120] and The fitness function is used. At this point, the population completes its evolution and continues to iterate in this manner until the differential evolution algorithm reaches its maximum iteration limit, or the optimal solution of the population meets the predetermined error accuracy requirements. Only then does the entire optimization process terminate, and the neural network model training end.

[0121] Step 6: Based on the mechanism model in Step 1 and the neural network model trained in Step 5, construct a soft-sensing model for molten steel temperature to achieve online real-time prediction of molten steel temperature. Specifically:

[0122] The system uses a constructed soft sensor model to predict the temperature of molten steel online. The system collects heating power and power frequency data in real time, inputs them into a neural network model, and outputs the electromagnetic induction thermal efficiency value under the current operating conditions. Combined with the molten steel energy balance model and furnace heat transfer model established in step 1, the real-time predicted value of molten steel temperature is obtained through iterative solution.

[0123] After deploying the trained model to the on-site industrial control system, for the refining process of K188 steel, before the start of each heating interval, the heating power was reduced to stabilize the surface fluctuations of the molten steel. An infrared thermometer was used to measure and record the initial true temperature value, which served as the starting point for the model's iterative solution. Subsequently, the power was restored for heating, and the model continuously collected power and frequency data in real time, outputting predicted molten steel temperatures. At the end of each heating interval, the power was reduced again, and the actual termination temperature was measured and compared with the predicted termination temperature output by the model to optimize the model. Some comparison results are shown in Table 1.

[0124] Table 1 Comparison of predicted and actual termination temperatures from the soft sensor model

[0125]

[0126] As shown in Table 1, under the five different heating ranges, the absolute error between the model-predicted termination temperature and the actual termination temperature is within 10℃. The maximum absolute error is 7.9℃, the minimum is 0.6℃, and the average absolute error is 3.06℃. The overall prediction accuracy meets the requirements of engineering practice for predicting the temperature of molten steel.

[0127] This invention proposes a soft measurement method for molten steel temperature applicable to the electromagnetic casting refining process by integrating mechanistic modeling and neural networks. Based on easily measurable variables such as heating power and power supply frequency, and combined with energy balance and heat transfer mechanisms, this method achieves high-precision, continuous online prediction of molten steel temperature. It effectively overcomes the shortcomings of discontinuous traditional contact temperature measurement and large errors in non-contact infrared temperature measurement, and has good engineering applicability and promotional value.

[0128] It should be noted that the above description is merely a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, several improvements, substitutions, or modifications can be made without departing from the technical principles of the present invention, and these improvements, substitutions, or modifications should all be considered within the scope of protection of the present invention.

Claims

1. A soft measurement method for molten steel temperature in an induction casting furnace driven by hybrid mechanism data, characterized in that, Includes the following: Step 1: Construct a mechanism model for the refining process of molten steel by electromagnetic induction heating, including an energy balance model of molten steel during the refining heating process and a heat transfer model within the furnace; Step 1.1: Constructing an energy balance model for molten steel during the refining heating process: During the refining heating process, the electromagnetic induction heating power absorbed by the molten steel is used to increase its own heat and compensate for heat loss to the surrounding environment. Based on this, a continuous form differential equation for the energy balance of the molten steel during the refining heating process is established: （1）； in, For heating power, For power supply frequency, For electromagnetic induction thermal efficiency, For the specific heat of molten steel, For the quality of molten steel, The temperature of the molten steel. Heating time, This refers to the power lost due to heat in molten steel. Step 1.2: Constructing a furnace heat transfer model for the refining heating process: In a vacuum environment, the heat loss of the molten steel is calculated based on the outward radiation heat dissipation from the surface of the molten steel and the outward conduction heat dissipation from the molten steel along the crucible wall: （3）； in, This refers to the radiative heat flow from the surface of the molten steel outwards. This refers to the heat flow conducted outwards from the crucible wall by the molten steel. Step 2: Collect real-time production data during the refining process of molten steel by electromagnetic induction heating to construct a production dataset, including the initial and final temperatures of the molten steel, heating power, and power frequency. The data is automatically recorded at fixed sampling intervals to ensure time sequence correspondence. Step 3: Preprocess the real-time production data collected in Step 2, including removing temperature noise data that deviates from the normal process range, normalizing the heating power and power frequency using the maximum-minimum normalization method, and dividing the preprocessed real-time production dataset into training set, validation set and test set in a ratio of 8:1:

1. Step 4: Construct a neural network model with heating power and power frequency as inputs and electromagnetic induction thermal efficiency as output; determine the number of hidden layer nodes and activation functions in the neural network model; and initialize the neural network model parameters based on the intelligent optimization algorithm. Step 5: Substitute the electromagnetic induction thermal efficiency output by the neural network model in Step 4 into the mechanism model in Step 1 to solve for the termination temperature of the molten steel and obtain the predicted termination temperature; and use the error between the predicted termination temperature and the actual termination temperature as the loss function to train and optimize the neural network model through an intelligent optimization algorithm. Step 6: Based on the mechanism model in Step 1 and the neural network model trained in Step 5, construct a soft measurement model for molten steel temperature to achieve online real-time prediction of molten steel temperature; In step 1.2, the heat conduction of the molten steel along the crucible wall outwards... The calculation method is as follows: The crucible structure is set as having cylindrical sidewalls and a hemispherical bottom, with a cylindrical sleeve on the outside. The heat loss of the molten steel transferred outward through the crucible wall is divided into two parts: the heat loss of the cylindrical sidewall and the heat loss of the hemispherical bottom. (1) The calculation method for the heat conduction loss of molten steel through the cylindrical part of the side wall is as follows: （8’）； （9’）； in, This refers to the heat flow that is conducted out of the sleeve by the molten steel through the cylindrical sidewall. This refers to the radiative heat flow emitted by the molten steel through the outer wall of the sleeve corresponding to the side cylinder. The total thermal resistance from the inner wall of the crucible to the outer wall of the sleeve. This represents the area of ​​the outer wall of the sleeve corresponding to the cylindrical sidewall of the molten steel. The emissivity of the outer wall of the sleeve. This refers to the temperature of the furnace inner wall. Solve equations (8') and (9') to obtain the temperature of the crucible's inner wall as the molten steel passes through the cylindrical section of the sidewall. And the temperature of the outer wall of the sleeve corresponding to the side-wall cylinder as the molten steel passes through. The heat flow of molten steel conducted out of the sleeve through the side-wall cylinder was obtained. This refers to the heat loss of molten steel through the cylindrical part of the side wall; (2) The calculation method for the heat loss of molten steel through the bottom hemisphere is as follows: The crucible hemispherical part is discretized along the vertical direction of the longitudinal section, and the heat dissipation of the discretized bottom hemispherical part is set to occur only in the radial and vertical downward heat conduction on the side. Starting from the center of the hemisphere, take a fixed step downwards. s Each fixed step size yields a discrete point, and the process continues downwards. a fixed step size s Then, discrete points are obtained. , The vertical height from the center of the circle is ; Then discrete points Horizontal distance to the inner wall, outer wall and outer wall of the crucible , , They are respectively: （11）； （12）； （13）； in, , , These are the inner radius, outer radius, and sleeve radius of the crucible, respectively. discrete points Height from the bottom of the crucible Thus, the bottom of the crucible is obtained. Thermal resistance of molten steel in the radial direction at a certain height for: （14）； in, , , respectively, are the thermal conductivity of the crucible and the sleeve, where x represents the radial direction; Let the temperature of the molten steel and the corresponding temperature of the outer wall of the sleeve be respectively... , , The radial heat loss of the molten steel at the bottom of the crucible is then... for: （15）； For the k Each step size s The segmented rings, their corresponding central angles for: （16）； Based on the proportional relationship of the area of ​​the spherical cap, the first k Surface area of ​​each ring Calculate using the following recursive formula: （17）； in, Indicates the first k The first one before p The surface area of ​​each ring band; Introducing projection angle Then the first k The projected area of ​​each ring on the bottom surface for: （18）； Find the first k Each ring conducts thermal resistance vertically downwards. for: （19）； in, H This is the distance from the plane containing the center of the sphere at the bottom of the crucible to the bottom of the sleeve; Let the temperature of the circular region corresponding to the bottom of the sleeve and the bottom of the crucible be... Therefore, the heat loss of the molten steel at the bottom of the crucible along the vertical downward direction is obtained. for: （20）； Based on the thermal balance relationship between the heat conduction of molten steel and the radiation of the sleeve, the heat dissipation of molten steel at the bottom of the crucible can be solved simultaneously, that is, the heat loss of molten steel through the bottom hemispherical part.

2. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data as described in claim 1, characterized in that, Step 1.1 includes the following: In order to meet the discretization requirements of data sampling and model solution, the continuous form differential equation of the energy balance of molten steel in the refining heating process of equation (1) is discretized in the time domain; Let the time step be ,by T l Indicates the first l The temperature of the molten steel at each sampling time. T l-1 Let be the temperature of the molten steel at the previous sampling moment. Then, the discretized energy balance equation can be expressed as: （2）； in, and The first l Heating power and power supply frequency at each sampling time, This represents the heat loss power of the molten steel at the corresponding moment; Based on the recursive relationship of equation (2), the final temperature is obtained by iteratively solving from the initial temperature in each discretized time interval.

3. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data as described in claim 1, characterized in that, The outward radiative heat flow from the surface of the molten steel in step 1.2 The calculation method is as follows: A three-surface closed cavity radiation network model was used to perform thermal radiation analysis on the upper surface of the molten steel, the crucible wall that was not in contact with the molten steel, and the crucible opening, and then the radiative heat flow from the surface of the molten steel to the outside was calculated. Let the upper surface of the molten steel be surface 1, the crucible wall not in contact with the molten steel be surface 2, and the opening of the crucible be surface 3. Treating surface 3 as a blackbody, the effective radiation of surface 3 is... Equal to its own radiation force ,Right now ; set up For surface a For surface b angular coefficient, , ; First, calculate the angle coefficients between the surfaces: based on the geometric relationship between surface 1 and surface 3, obtain the angle coefficient of surface 1 with respect to surface 3. Then, based on the integrity of the angle coefficient, the angle coefficient of surface 1 relative to surface 2 is obtained. The angle coefficient of surface 2 with respect to surface 3 is obtained based on the relativity and symmetry of the angle coefficient. ; Construct a three-surface closed cavity radiation network model and define the surface thermal resistance and spatial thermal resistance of each surface: （4）； （5）； in, For surface a Surface thermal resistance, For surface a For surface b Spatial thermal resistance, For surface a blackness, For surface a The area; Effective radiation intermediate nodes of surfaces 1 and 2 in the radiation network , Applying Kirchhoff's laws, a system of nodal equations is established, and the effective radiation is obtained by solving the system simultaneously. , Then, the radiative heat flow of surface 1 and surface 2 is calculated. and : （6）； （7）； Among them, the self-radiation force of surface 1 Surface 2 self-radiation force , Let be the blackbody radiation constant. T 1. T 2 represents the temperatures of surface 1 and surface 2; Assuming the heat transfer system inside the furnace is in steady state, the radiant heat absorbed by surface 2 is in equilibrium with the heat conducted outward and the radiant heat emitted from the outer wall of the sleeve surrounding the crucible. The heat balance equation between conduction and radiation is then established: （8）； （9）； （10）； in, The heat flow is conducted outward from surface 2 of the sleeve. This refers to the radiative heat flow dissipated from the outer wall of the sleeve. The total thermal resistance from the inner wall of the crucible to the outer wall of the sleeve. Surface 2 corresponds to the outer wall area of ​​the sleeve. The emissivity of the outer wall of the sleeve. This refers to the temperature of the furnace inner wall. Solving the heat balance equations for heat conduction and radiation yields the temperature of the crucible's inner wall in the portion not in contact with the molten steel. and the temperature of the outer wall of the sleeve in the part not in contact with the molten steel. This leads to the conclusion of the outward radiative heat flow from the surface of the molten steel. .

4. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data as described in claim 1, characterized in that, Step 4 includes the following: Step 4.1: Set the neural network model to use a single hidden layer network. The number of nodes in the hidden layer is calculated using the following formula: （22）； in, q This represents the initial number of nodes in the hidden layer. n and m These represent the number of nodes in the input layer and the output layer, respectively. c The Sigmoid function is chosen as the activation function, with a value range of 1 to 10 as the adjustment constant. Step 4.2: Encode all weights and thresholds in the neural network model into a single... D dimensional vector , , W For population size, G To represent the algebraic progression during the iteration process, the initial population is randomly generated within the search space using the following method: （23）； in, rand A random number between [0,1] , .

5. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data as described in claim 1, characterized in that, Step 5 includes the following: using the mean square error between the predicted and actual termination temperatures as the loss function, the neural network model is trained and optimized using the differential evolution algorithm, specifically: Differential evolution algorithm generates intermediate individuals through the following mutation operation. : （24）； in, and , F A variable factor between 0 and 1; Utilize target individuals and intermediate individuals obtained through mutation Perform crossover operations to obtain undetermined experimental individuals. The specific operation for the components is as follows: （25）； in, CR The cross factor is between 0 and 1. A dimension is randomly selected from all dimensions; The mean squared error (MSE) is chosen as the loss function, and this is also used as the fitness value function in the differential evolution algorithm. Its expression is as follows: （26）； in, M The number of samples in the training set. For training set samples The actual measured temperature value, For training set samples The predicted termination temperature; By selecting the test individual Perform fitness evaluation: （27）； At this point, the population completes its evolution and continues to iterate in this manner until the differential evolution algorithm reaches its maximum number of iterations, or the optimal solution of the population meets the predetermined error accuracy requirements. Only then does the entire optimization process end, and the training of the neural network model conclude.

6. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data as described in claim 1, characterized in that, Step 6 includes the following: Real-time heating power and power frequency data are collected and input into a neural network model. The electromagnetic induction thermal efficiency value under the current operating conditions is output. Combined with the molten steel energy balance model and furnace heat transfer model established in step 1, the real-time predicted value of molten steel temperature is obtained through iterative solution.

7. The method for soft measurement of molten steel temperature in an induction casting furnace driven by hybrid mechanism data according to claim 1, characterized in that, Includes the following: Real-time production data from multiple independent heating zones during the refining process of molten steel heated by electromagnetic induction were collected. Each zone encompassed different combinations of initial and final temperatures, with varying durations and temperature rise ranges. Before the start of each heating zone, the heating power was reduced to stabilize the surface fluctuations of the molten steel. An infrared thermometer was used to measure and record the initial temperature, which served as the starting point for the iterative solution of the neural network model. Subsequently, the power was restored for heating, and the neural network model continuously collected heating power and power frequency data in real time, continuously outputting predicted molten steel temperatures. At the end of each heating zone, the power was reduced again, and the final temperature was measured and compared with the predicted final temperature output by the neural network model to optimize the model.

Images