A method for dynamically monitoring the thickness of a solidified shell of a continuously cast strand and related apparatus

Abstract

This application provides a method and related apparatus for dynamic monitoring of the solidified shell thickness of continuously cast billets, belonging to the field of continuous casting production process monitoring technology. This application constructs a finite element model covering the crystallizer, secondary cooling zone, and solidification end, generating simulation datasets of shell thickness and temperature field distribution under different combinations of process parameters; it collects multimodal measured data from continuous casting production and merges it with the simulation dataset to generate a fused dataset. A shell thickness prediction model embedded with residual constraints of the heat transfer control equation is constructed, and a dual-path feature extraction module is configured to process the temporal sequence of process parameters and spatial field distribution information respectively; it is first pre-trained using the simulation dataset and then fine-tuned using the fused dataset. Based on real-time collected continuous casting process parameters and multimodal measured data, continuous online dynamic monitoring of the shell thickness and temperature field throughout the entire process is achieved. This application has high monitoring accuracy and good spatiotemporal resolution, and can provide timely early warning of steel leakage risks, providing a reliable basis for continuous casting process optimization and safe production.

Description

Technical Field

[0001] This application relates to the field of billet quality monitoring technology in continuous casting production process, specifically to a method and related device for dynamic monitoring of the solidified shell thickness of continuously cast billets. Background Technology

[0002] The thickness of the solidified shell of a continuously cast billet is a core parameter determining the safety and quality of continuous casting production. Abnormal shell thickness can easily lead to major accidents such as steel leakage and cracking. Real-time and accurate monitoring of the shell thickness is a critical requirement for continuous casting production. Existing methods for monitoring the shell thickness of continuously cast billets have systemic defects, and there is currently no solution that can simultaneously solve multiple technical challenges related to simulation accuracy, data fusion, model consistency, operating condition adaptation, and closed-loop control.

[0003] Currently, billet shell thickness monitoring mainly falls into three technical categories: prediction methods based on finite element simulation, measurement methods based on single sensors such as ultrasonic sensors, and traditional data-driven models based on single neural networks. Existing patents mostly focus on optimizing these approaches, including improving temperature field calculations, fusing multiple types of monitoring data, and introducing machine learning algorithms. However, simulation models lack detailed characterization of the dynamic evolution of the crystallizer gap and the non-uniformity of the secondary cooling zone spray, resulting in insufficient simulation accuracy. Multi-source data fusion uses simple weighted averaging, failing to quantify data reliability and error propagation patterns. Data-driven models do not incorporate physical constraints such as heat conduction, easily leading to predictions that violate physical laws. They lack effective incremental learning mechanisms, requiring retraining when operating conditions change abruptly. Furthermore, they fail to achieve closed-loop linkage with the continuous casting machine control system, enabling only passive monitoring and preventing proactive optimization of process parameters. Summary of the Invention

[0004] This application addresses the technical problems of existing continuous casting billet shell thickness monitoring methods, such as low simulation accuracy, poor data fusion, and the tendency to produce prediction results that violate physical laws. It provides a method and related device for dynamic monitoring of the solidified billet shell thickness of continuous casting billets.

[0005] To achieve the above objectives, this application adopts the following technical solution: The first aspect of this application provides a method for dynamically monitoring the thickness of the solidified shell of a continuously cast billet, comprising the following steps: A finite element model covering the crystallizer, secondary cooling zone, and solidification end was constructed to generate a simulation dataset containing the thickness and temperature field distribution of the billet shell under different combinations of process parameters. Multimodal measured data are collected during the continuous casting production process, and the simulation dataset is fused with the multimodal measured data to generate a fused dataset. A billet shell thickness prediction model is constructed. The heat transfer control equation of the solidification process of the continuously cast billet is embedded into the hidden layer of the billet shell thickness prediction model in the form of residual constraints. A dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively. The billet shell thickness prediction model is first pre-trained using a simulation dataset, and then the pre-trained billet shell thickness prediction model is trained using a fusion dataset. Based on real-time acquired continuous casting process parameters and multimodal measured data, a billet shell thickness prediction model is used to achieve dynamic monitoring of billet shell thickness and temperature field distribution.

[0006] Furthermore, a dynamic calibration algorithm combining Kalman filtering and Bayesian inference is used to fuse the simulation dataset and multimodal measured data to generate a fused dataset.

[0007] Furthermore, the heat transfer control equation for the solidification process of the continuously cast billet is specifically as follows:

[0008] In the formula: Density of steel billet; The specific heat capacity of the steel billet under constant pressure; The temperature of the continuously cast billet predicted by the neural network; The thermal conductivity of the steel billet; For gradient operators; The latent heat of solidification of steel; The solid fraction of the continuously cast billet predicted by the neural network; This refers to the solidification time; The physical residual loss of the structure when the heat transfer control equation of the solidification process of the continuously cast billet is applied is:

[0009] In the formula, This refers to the physical constraint residual loss.

[0010] Furthermore, the dual-path feature extraction module includes a temporal feature extraction branch and a spatial feature extraction branch; the temporal feature extraction branch is used to extract the time-dependent features of the time sequence of process parameters, and the spatial feature extraction branch is used to extract the spatial correlation features between the temperature field and the solidification field.

[0011] Furthermore, when training the pre-trained billet shell thickness prediction model using a fused dataset, a composite loss function with a dynamic weight adjustment mechanism is used to correct the billet shell thickness prediction model during the training process; the composite loss function is as follows:

[0012] In the formula, Dynamic weights for data fitting terms; The fitting loss is the result of fusing measured and simulated data. For the dynamic weights of physical constraint terms; This refers to the physical constraint residual loss. The fixed weights are for the gradient smoothing regularization term; This is a gradient smoothing regularization term; To counteract the loss of weight; For regression loss.

[0013] Furthermore, the online dynamic monitoring also includes: obtaining the billet growth curve, early warning information of solidification danger areas, and process parameter optimization suggestions based on the predicted results of billet thickness and temperature field distribution.

[0014] Furthermore, the online dynamic monitoring also includes: Based on the predicted results of billet shell thickness and temperature field distribution, the uniformity of billet shell thickness and the location of solidification endpoint are quantitatively evaluated and the results are determined. Combined with the continuous casting production qualification standards, the deviation between the uniformity evaluation results and the location of solidification endpoint is obtained. The deviation between the uniformity evaluation results and the location of solidification endpoint is used as the basis for adjustment to achieve adaptive adjustment of cooling water volume and dynamic optimization of billet drawing speed.

[0015] A second aspect of this application provides a dynamic monitoring system for the solidified shell thickness of a continuously cast billet, comprising: The simulation modeling module is used to construct finite element models covering the crystallizer, secondary cooling zone and solidification end, and generate simulation datasets containing the thickness and temperature field distribution of the billet shell under different combinations of process parameters. The data fusion module is used to collect multimodal measured data during the continuous casting production process and fuse the simulation dataset with the multimodal measured data to generate a fused dataset. The model building and training module is used to build a billet shell thickness prediction model. The heat transfer control equation of the solidification process of the continuously cast billet is embedded into the hidden layer of the model in the form of residual constraints. A dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively. It is also used to first pre-train the billet shell thickness prediction model with a simulation dataset, and then train the pre-trained model with a fusion dataset. The online monitoring module is used to dynamically monitor the billet shell thickness and temperature field distribution based on real-time acquired continuous casting process parameters and multimodal measured data, using a billet shell thickness prediction model.

[0016] A third aspect of this application provides a computer device, including: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the above-described method for dynamic monitoring of the solidified shell thickness of continuously cast billets.

[0017] A fourth aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed as described above for the dynamic monitoring method of solidified shell thickness of continuously cast billets.

[0018] Compared with the prior art, this application has the following beneficial effects: This application generates a simulation dataset by constructing a full-process finite element model covering the crystallizer, secondary cooling zone, and solidification end, thus fully reproducing the continuous evolution of the billet shell throughout the solidification process. By fusing the simulation dataset with multimodal measured data, it not only compensates for the insufficient sample size of purely measured data but also enhances the model's adaptability to complex on-site conditions. The continuous casting solidification heat transfer control equation is embedded into the hidden layer of the prediction model in the form of residual constraints, ensuring the rationality of the prediction results and avoiding abnormal outputs that violate heat transfer laws in purely data-driven models. At the same time, a dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively, realizing the collaborative modeling of global process influences and local spatial features. Furthermore, a two-stage training strategy of pre-training with simulation data and fine-tuning with fused data is adopted, which significantly improves the prediction accuracy and generalization ability of the model under small sample conditions in industrial settings. It can output the billet shell thickness and temperature field distribution in real time and accurately throughout the entire process, providing reliable data support for steel leakage warning, quality control, and process optimization in continuous casting production. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the dynamic monitoring method for solidified shell thickness of continuously cast billets provided in this application; Figure 2 A schematic diagram of the multimodal data fusion process in the method provided in this application; Figure 3 A schematic diagram of the physical constraint embedding structure in the method provided in this application; Figure 4 This is a schematic diagram of the dynamic monitoring system for the solidified shell thickness of continuously cast billets provided in this application; Figure 5 This is a schematic diagram of the electronic device structure according to a preferred embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The thickness of the solidified shell of a continuously cast billet is a core parameter affecting the quality and efficiency of continuous casting production, directly determining the internal quality of the billet, the incidence of surface defects, and the risk of leakage. However, existing methods for monitoring the shell thickness of continuously cast billets face several technical bottlenecks: traditional offline sampling and detection methods suffer from severe lag, failing to provide real-time guidance for production process adjustments, and can only obtain information from local cross-sections, making it difficult to reflect the solidification uniformity along the entire length of the billet; single-sensor detection methods (such as ultrasonic detection and infrared thermometry) are highly susceptible to environmental interference, exhibiting the problem of superposition of systematic and random errors, making it difficult to guarantee detection accuracy in the high-temperature, high-humidity, and strong-vibration continuous casting environment; and numerical simulation methods based on purely physical mechanisms require numerous simplification assumptions regarding the complex solidification process, making accurate simulation impossible. Actual operating conditions such as the dynamic evolution of the air gap in the simulated crystallizer and the non-uniformity of the secondary cooling zone spray lead to significant deviations between the predicted results and the actual results. Purely data-driven machine learning methods lack physical mechanism constraints, are prone to producing prediction results that violate the basic laws of heat transfer, and have poor generalization ability, with performance deteriorating sharply when process parameters fluctuate or operating conditions change abruptly. Existing methods have not achieved effective fusion of multi-source heterogeneous data, and cannot simultaneously utilize the global physical consistency of finite element simulation and the local high-precision information of multimodal measured data, making it difficult to meet the monitoring requirements of modern continuous casting production for high precision, high real-time performance, and high robustness.

[0023] Based on this, this application provides a method for dynamic monitoring of the solidified shell thickness of continuously cast billets. Through a technical approach that combines physical information-driven methods with multimodal data fusion, it achieves high-precision real-time monitoring of the entire process and all cross-sections of the continuously cast billet. For example... Figure 1 As shown, the specific steps include: S1. Construct a finite element model covering the crystallizer, secondary cooling zone, and solidification end, and generate a simulation dataset containing the shell thickness and temperature field distribution under different combinations of process parameters.

[0024] Specifically, a full-process finite element model is constructed based on the physical mechanism of multi-field coupling of heat, fluid, and phase change in the continuous casting process. This finite element model integrates a dynamic evolution sub-model of the crystallizer air gap and a non-uniformity sub-model of spray cooling in the secondary cooling zone to achieve a refined simulation of the dynamic behavior of the solidification front. Among them, the dynamic evolution sub-model of the crystallizer air gap is used to simulate the formation and development process of the air gap between the billet and the crystallizer wall due to thermal contraction, accurately reflecting the influence of the air gap on the heat transfer efficiency; the non-uniformity sub-model of spray cooling in the secondary cooling zone is used to simulate the spatial distribution differences of the spray intensity of each nozzle in the secondary cooling zone, achieving accurate modeling of heat transfer in the secondary cooling zone.

[0025] Furthermore, a simulation dataset was generated through multi-condition parametric scanning and adaptive mesh refinement optimization. Key process parameters included billet pulling speed, casting superheat, crystallizer cooling water flow rate, crystallizer electromagnetic stirring parameters, cooling water flow rate and temperature in each section of the secondary cooling zone, etc. The simulation dataset was augmented and subjected to controllable noise injection to improve the diversity and generalization ability of the data samples. Data augmentation methods included process parameter interpolation and time series shifting; controllable noise injection was used to simulate random disturbances in actual production, with the noise amplitude controlled within the normal fluctuation range of the process parameters.

[0026] This step solves the problem of a lack of high-quality labeled data in actual production by constructing a high-fidelity finite element model and generating a large number of simulation datasets, providing sufficient basic data support for the training of the subsequent billet shell thickness prediction model.

[0027] S2 collects multimodal measured data during the continuous casting production process and merges the simulation dataset with the multimodal measured data to generate a fused dataset.

[0028] Specifically, the collected multimodal measured data includes ultrasonic shell thickness detection data, infrared array temperature measurement data, crystallizer vibration state signals, and cooling water flow / temperature fluctuation information. A dynamic calibration algorithm combining Kalman filtering and Bayesian inference is used to fuse the finite element simulation results with the multi-source measured data, simultaneously compensating for the propagation of systematic and random errors, and constructing a high-confidence multimodal fusion dataset. The multimodal data fusion process is as follows: Figure 2 As shown, the details are as follows: It can simultaneously access finite element simulation data and multimodal measured data such as ultrasonic, infrared, vibration, and cooling water. Based on the system state vector, control input vector, and finite element simulation state vector from the previous moment, the prior prediction of the system state at the current moment is completed through the state equation of Kalman filtering. The observation equations using Kalman filtering map the a priori predicted system state to multimodal observation theoretical values. Bayesian inference is used to quantify the confidence weight of the measured data for each modality, and the error compensation amount and confidence weight are output. Substituting the error compensation amount and confidence weight into the Kalman filter gain calculation and state update formula, the optimal estimation of the system state is completed. The optimal estimates of the system state at all times are integrated to generate a fused dataset.

[0029] As one example, the state equation of the Kalman filter is:

[0030] In the formula, Let k be the system state vector at time k; Let k be the adaptive state transition matrix at time k, whose elements are determined by the thermal conductivity of the billet shell. Dynamic adjustment; Let k be the adaptive control input matrix at time k; The control input vector at time k-1; The fusion matrix of finite element simulation results at time k; Let k-1 be the finite element simulation state vector, and... The dimensions are consistent, including simulated blank thickness, simulated temperature, and simulated thermal conductivity; Let k-1 be the process noise vector, which follows a Gaussian distribution and has units of 1 / 2. The inconsistency mainly stems from random disturbances such as crystallizer vibration and fluctuations in molten steel composition.

[0031] The observation equation is:

[0032] In the formula, Let k be the multimodal observation vector at time k; Let k be the adaptive observation matrix at time k; The observation error compensation matrix at time k is used to compensate for systematic errors in ultrasonic detection and infrared thermometry (such as instrument calibration errors). Its elements are dynamically adjusted by Bayesian inference results. Let be the systematic error vector of the multimodal measured data at time k, corresponding to the ultrasonic detection error and the infrared thermometry error, with units of and . Consistent; Let K be the noise vector observed at time k, which follows a Gaussian distribution and has units of 1 / k. The inconsistency mainly stems from environmental interference and random instrument noise during the actual measurement process.

[0033] This step, through the organic combination of Kalman filtering and Bayesian inference, achieves complementary advantages between finite element simulation data and multimodal measured data, effectively eliminating the influence of systematic and random errors, and significantly improving the confidence and accuracy of the fused dataset.

[0034] S3. Construct a billet shell thickness prediction model. Embed the heat transfer control equation of the continuous casting billet solidification process into the hidden layer of the billet shell thickness prediction model in the form of residual constraints. Configure a dual-path feature extraction module to process the time sequence of process parameters and the spatial field distribution information respectively. First, pre-train the billet shell thickness prediction model with a simulation dataset, and then train the pre-trained billet shell thickness prediction model with a fusion dataset.

[0035] Specifically, the billet shell thickness prediction model is a hybrid neural network that integrates prior physical knowledge and data-driven capabilities, employing a heterogeneous fusion structure of graph neural networks and convolutional neural networks. The graph neural network is used to model the spatial topological relationships of the continuously cast billet, capturing the heat transfer coupling effect between different locations on the billet; the convolutional neural network is used to extract local features of the temperature and solidification fields, achieving fine-scale field distribution characterization. The physical constraint embedding structure is as follows: Figure 3 As shown.

[0036] As one embodiment, the heat transfer control equations for the solidification process of continuously cast billets are embedded in the network in the form of residual constraints, specifically:

[0037] In the formula: The density of the steel billet varies slightly with temperature; the average value is taken as kg / m³. Specific heat capacity of steel billet under constant pressure, J / (kg·℃); The temperature of the continuously cast billet predicted by the neural network is in °C. The value is the thermal conductivity of the steel billet, in W / (m·℃). This is a gradient operator used to calculate the spatial rate of change of the temperature field; is the latent heat of solidification of steel, J / kg; The solid fraction of the continuously cast billet, predicted by the neural network, characterizes the degree of solidification. The solidification time is expressed in seconds (s).

[0038] The physical residual loss constructed using it is:

[0039] In the formula: This is the physical constraint residual loss, used to constrain the neural network output to meet the physical laws of solidification heat transfer and reduce physical deviations.

[0040] Furthermore, the dual-path feature extraction module includes a temporal feature extraction branch and a spatial feature extraction branch. The temporal feature extraction branch is used to extract the time-dependent features of the time sequence of process parameters, and uses a Long Short-Term Memory (LSTM) network to process process parameters such as billet pulling speed and cooling water volume that change over time. The spatial feature extraction branch is used to extract the spatial correlation features between the temperature field and the solidification field, and uses multi-scale convolutional kernels of a convolutional neural network to extract field distribution information at different scales.

[0041] This step embeds the physical control equations into the neural network in the form of residual constraints, ensuring that the model must follow the basic laws of solidification heat transfer while fitting data, effectively solving the problem of poor physical consistency in purely data-driven models; the dual-path feature extraction module achieves the decoupling and fusion of temporal and spatial information, significantly improving the model's ability to characterize complex solidification processes.

[0042] Specifically, when training the pre-trained billet thickness prediction model using a fused dataset, a composite loss function with a dynamic weight adjustment mechanism is employed. The composite loss function integrates a data fitting term, a physical constraint residual term, a gradient smoothing regularization term, and an adversarial loss term, and is configured with a dynamic weight adjustment mechanism to adaptively adjust the weight ratio of physical constraints to data fitting based on the rate of change of the prediction error on the validation set. The composite loss function is as follows:

[0043] In the formula: The dynamic weights of the data fitting terms are adaptively adjusted according to the rate of change of the prediction error on the validation set. The fitting loss between measured and simulated data is used to measure the model's prediction deviation of the actual billet thickness. For the dynamic weights of physical constraint terms; This refers to the physical constraint residual loss. The fixed weights (hyperparameters) for the gradient smoothing regularization term are used to prevent model overfitting and training oscillations. This is a gradient smoothing regularization term that suppresses abrupt gradient changes in the model and improves generalization ability. To counteract the loss of weight; For regression loss.

[0044] In some embodiments of this application, the training process employs a two-stage collaborative training strategy of pre-training and fine-tuning. The first stage involves pre-training the billet thickness prediction model based on a simulation dataset. The second stage introduces a fused dataset for parameter fine-tuning, adapting the billet thickness prediction model to actual production conditions. In the later stages of training, adversarial example enhancement techniques are used to improve the anti-interference capability of the billet thickness prediction model. Adversarial example enhancement is achieved by adding a disturbance signal to the input data that conforms to the noise distribution of the industrial scenario, with the amplitude of the disturbance signal controlled within the allowable range of sensor measurement accuracy.

[0045] This step achieves a balance between data fitting accuracy and physical consistency by constructing a multi-dimensional composite loss function and a dynamic weight adjustment mechanism. The two-stage training strategy and adversarial example enhancement technology further improve the model's generalization ability and robustness, ensuring that the model runs stably in complex and ever-changing industrial scenarios.

[0046] S5, based on real-time acquired continuous casting process parameters and multimodal measured data, uses a billet shell thickness prediction model to predict the billet shell thickness and temperature field distribution.

[0047] Specifically, the trained billet shell thickness prediction model is deployed on edge computing devices or cloud servers, and combined with real-time collected continuous casting process parameters and multimodal measured data to achieve millisecond-level prediction of billet shell thickness and temperature field distribution. Online dynamic monitoring also includes obtaining billet shell growth curves, early warning information on solidification danger zones, and process parameter optimization suggestions based on the billet shell thickness and temperature field distribution prediction results, which are then visualized through a human-computer interaction interface.

[0048] Furthermore, online dynamic monitoring also includes: based on the predicted results of billet shell thickness and temperature field distribution, completing the quantitative evaluation of billet shell thickness uniformity and solidification endpoint positioning; combined with the continuous casting production qualification standards, obtaining the deviation between the uniformity evaluation results and the solidification endpoint positioning; using the deviation between the uniformity evaluation results and the solidification endpoint positioning as the adjustment basis, realizing the adaptive adjustment of cooling water volume and dynamic optimization of billet drawing speed, and finally forming a closed-loop control loop with the continuous casting machine control system.

[0049] Among them, the solidification endpoint location is obtained based on the temperature field distribution prediction results and through the temperature field gradient abrupt change point identification algorithm.

[0050] The quantitative evaluation of billet shell thickness uniformity is based on the predicted billet shell thickness. The standard deviation of the billet shell thickness is calculated and obtained through the billet shell thickness uniformity evaluation index. This index uses a combination of standard deviation and coefficient of variation, specifically:

[0051] In the formula, It serves as an evaluation index for the uniformity of blank thickness; Let be the standard deviation of the blank thickness, in mm; Here is the coefficient of variation (relative standard deviation) of the billet shell thickness, in mm; The weighting factor is the standard deviation. The weighting coefficients of the coefficient of variation satisfy the following conditions: .

[0052] This step enables real-time online monitoring and closed-loop control of the solidification process of continuously cast billets, allowing for timely detection of solidification anomalies and process adjustments, effectively reducing the incidence of billet defects and improving the stability and product quality of continuous casting production.

[0053] This method also includes: when the continuous casting process conditions change abruptly or the prediction deviation of the billet shell thickness prediction model is detected to continuously exceed a preset threshold, an online incremental learning mechanism is activated. The preset threshold is set according to the quality requirements of the continuously cast billet, and its value ranges from ±3% to ±5% of the relative error.

[0054] Specifically, by combining finite element rapid simulation data and real-time measured data under new working conditions, the model's local parameters are updated using elastic weight solidification and replay buffer techniques. Elastic weight solidification is used to protect the core weights learned by the billet thickness prediction model under the original working conditions, avoiding catastrophic forgetting; the replay buffer is used to store historical typical working condition data, and replay training is performed during incremental learning to maintain the model's adaptability to historical working conditions.

[0055] This step enables the model to adaptively update under varying operating conditions, ensuring that the model can maintain high-precision prediction capabilities even when the continuous casting process is adjusted or the equipment status changes, thus extending the model's lifespan and reducing model maintenance costs.

[0056] Example 1 (1) Hardware configuration: The computing device uses an Intel Core i7-12700K processor, 32GB DDR5 memory, and NVIDIA RTX4090 graphics card. The interface module supports Modbus TCP / IP protocol and can realize real-time data interaction with on-site ultrasonic sensors, infrared array thermometers, flow sensors, etc. (2) Finite element model parameters: The full process model was constructed using ANSYS Fluent. The crystallizer air gap dynamic evolution sub-model was based on the thermoelastic mechanical equation. The non-uniformity sub-model of the secondary cooling zone spray was simulated by the discrete phase model (DPM) to simulate the water droplet impact characteristics. The mesh size was 5mm×5mm×5mm, and the adaptive mesh refinement threshold was set to temperature gradient ≥50℃ / mm. (3) Process parameter range: billet speed 0.8-1.5m / min, casting superheat 20-30℃, crystallizer cooling water flow 80-120m³ / h, cooling water flow rate of each section of the secondary cooling zone 5-20m³ / h, cooling water temperature 25-35℃; (4) Structure of the billet shell thickness prediction model: The graph neural network adopts GAT (Graph Attention Network), and the input nodes are 100 cross sections of the continuous casting billet along the casting direction. Each cross section contains 5 features such as temperature and stress. The convolutional neural network adopts a 3-layer convolution + 2-layer pooling structure, and the convolution kernel sizes are 3×3, 5×5, and 7×7, respectively. The temporal branch of the dual-path feature extraction module adopts a 2-layer LSTM with 256 hidden layer neurons. (5) Training parameters: 1000 pre-training iterations, 500 fine-tuning iterations, initial learning rate of 1e-4, Adam optimizer used; error change rate threshold of dynamic weight adjustment mechanism is set to 0.05, initial weight of physical constraint residual term is 0.3, initial weight of data fitting term is 0.7; (6) Incremental learning parameters: The replay buffer capacity is set to 10,000 samples, the weight retention ratio of the elastic weight is 80%, and the model update cycle is 5 min / time.

[0057] The performance of the proposed method was verified using continuous casting production data of Q235 steel from a steel plant, and compared with existing technologies (traditional finite element simulation, single ultrasonic testing, and traditional neural networks). Table 1. Experimental results using the method of this application and conventional methods.

[0058] The verification results are shown in Table 1. The method of this application is significantly better than the existing technology in terms of prediction accuracy, robustness, and process optimization effect, and fully meets the needs of industrial production.

[0059] In one embodiment of this application, such as Figure 4 As shown, a dynamic monitoring system for the solidified shell thickness of a continuously cast billet is provided, comprising: The simulation modeling module is used to construct finite element models covering the crystallizer, secondary cooling zone and solidification end, and generate simulation datasets containing the thickness and temperature field distribution of the billet shell under different combinations of process parameters. The data fusion module is used to collect multimodal measured data during the continuous casting production process and fuse the simulation dataset with the multimodal measured data to generate a fused dataset. The model building and training module is used to build a billet shell thickness prediction model. The heat transfer control equation of the solidification process of the continuously cast billet is embedded into the hidden layer of the model in the form of residual constraints. A dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively. It is also used to first pre-train the billet shell thickness prediction model with a simulation dataset, and then train the pre-trained model with a fusion dataset. The online monitoring module is used to dynamically monitor the billet shell thickness and temperature field distribution based on real-time acquired continuous casting process parameters and multimodal measured data, using a billet shell thickness prediction model.

[0060] Specific limitations regarding the dynamic monitoring system for the solidified shell thickness of continuously cast billets can be found in the limitations of the dynamic monitoring method for the solidified shell thickness of continuously cast billets described above. The corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned dynamic monitoring system for the solidified shell thickness of continuously cast billets can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0061] Figure 5 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 5 As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for dynamically monitoring the thickness of the solidified shell of continuously cast billets. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0062] As will be understood by those skilled in the art, computer equipment Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0063] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0064] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0065] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0066] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method of dynamic monitoring of the solidified shell thickness of a continuously cast strand, characterized in that Includes the following steps: A finite element model covering the crystallizer, secondary cooling zone, and solidification end was constructed to generate a simulation dataset containing the thickness and temperature field distribution of the billet shell under different combinations of process parameters. Multimodal measured data are collected during the continuous casting production process, and the simulation dataset is fused with the multimodal measured data to generate a fused dataset. A billet shell thickness prediction model is constructed. The heat transfer control equation of the solidification process of the continuously cast billet is embedded into the hidden layer of the billet shell thickness prediction model in the form of residual constraints. A dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively. The billet shell thickness prediction model is first pre-trained using a simulation dataset, and then the pre-trained billet shell thickness prediction model is trained using a fusion dataset. Based on real-time acquired continuous casting process parameters and multimodal measured data, a billet shell thickness prediction model is used to achieve dynamic monitoring of billet shell thickness and temperature field distribution.

2. The method of dynamic monitoring of the solidified shell thickness of a continuously cast billet according to claim 1, characterized in that, A dynamic calibration algorithm combining Kalman filtering and Bayesian inference is used to fuse the simulation dataset and multimodal measured data to generate a fused dataset.

3. The method of dynamic monitoring of the solidified shell thickness of a continuously cast billet according to claim 1, characterized in that, The heat transfer control equation for the solidification process of the continuously cast billet is as follows: wherein: is the density of the billet; is the specific heat capacity at constant pressure of the billet; is the temperature of the continuously cast billet predicted by the neural network; is the thermal conductivity of the billet; is the gradient operator; is the latent heat of solidification of the steel; is the solid fraction of the continuously cast billet predicted by the neural network; is the solidification time; The physical residual loss of the structure when the heat transfer control equation of the solidification process of the continuously cast billet is applied is: In the formula, This refers to the physical constraint residual loss.

4. The method for dynamic monitoring of the solidified shell thickness of continuously cast billets according to claim 1, characterized in that, The dual-path feature extraction module includes a temporal feature extraction branch and a spatial feature extraction branch; the temporal feature extraction branch is used to extract the time dependence features of the time sequence of process parameters, and the spatial feature extraction branch is used to extract the spatial correlation features between the temperature field and the solidification field.

5. The method for dynamic monitoring of the solidified shell thickness of continuously cast billets according to claim 1, characterized in that, When training the pre-trained billet shell thickness prediction model using a fused dataset, a composite loss function with a dynamic weight adjustment mechanism is used to correct the billet shell thickness prediction model during the training process; the composite loss function is as follows: In the formula, Dynamic weights for data fitting terms; The fitting loss is the result of fusing measured and simulated data. For the dynamic weights of physical constraint terms; This refers to the physical constraint residual loss. The fixed weights are for the gradient smoothing regularization term; This is a gradient smoothing regularization term; To counteract the loss of weight; For regression loss.

6. The method for dynamic monitoring of the solidified shell thickness of continuously cast billets according to claim 1, characterized in that, The online dynamic monitoring also includes: obtaining the billet growth curve, early warning information of solidification danger areas, and process parameter optimization suggestions based on the prediction results of billet thickness and temperature field distribution.

7. The method for dynamic monitoring of the solidified shell thickness of continuously cast billets according to claim 1, characterized in that, The online dynamic monitoring also includes: Based on the predicted results of billet shell thickness and temperature field distribution, the uniformity of billet shell thickness and the location of solidification endpoint are quantitatively evaluated and the results are determined. Combined with the continuous casting production qualification standards, the deviation between the uniformity evaluation results and the location of solidification endpoint is obtained. The deviation between the uniformity evaluation results and the location of solidification endpoint is used as the basis for adjustment to achieve adaptive adjustment of cooling water volume and dynamic optimization of billet drawing speed.

8. A dynamic monitoring system for the thickness of the solidified shell of a continuously cast billet, characterized in that, include: The simulation modeling module is used to construct finite element models covering the crystallizer, secondary cooling zone and solidification end, and generate simulation datasets containing the thickness and temperature field distribution of the billet shell under different combinations of process parameters. The data fusion module is used to collect multimodal measured data during the continuous casting production process and fuse the simulation dataset with the multimodal measured data to generate a fused dataset. The model building and training module is used to build a billet shell thickness prediction model. The heat transfer control equation of the solidification process of the continuously cast billet is embedded into the hidden layer of the model in the form of residual constraints. A dual-path feature extraction module is configured to process the time sequence of process parameters and the spatial field distribution information respectively. It is also used to first pre-train the billet shell thickness prediction model with a simulation dataset, and then train the pre-trained model with a fusion dataset. The online monitoring module is used to dynamically monitor the billet shell thickness and temperature field distribution based on real-time acquired continuous casting process parameters and multimodal measured data, using a billet shell thickness prediction model.

9. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the method for dynamic monitoring of the solidified shell thickness of a continuously cast billet as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-7, the method for dynamic monitoring of solidified billet shell thickness in continuous casting.

Images