A slab roll-off slag diagnosis method based on uneven state of crystallizer flow field

Abstract

The present application relates to the technical field of steel metallurgy continuous casting, and particularly relates to a slab slag entrapment diagnosis method based on a mold flow field imbalance state, the present application collects core features of the mold flow field and continuous casting process parameters and completes preprocessing, then constructs a multi-dimensional flow field feature fusion flow field imbalance index SFI, determines feature weights and divides imbalance degrees in combination with an analytic hierarchy process and an entropy weight method, then constructs an SFI time sequence to predict future SFI values, finally fuses the flow field imbalance degree, the SFI prediction value and the model diagnosis result to realize slab slag entrapment grading diagnosis, and triggers corresponding early warning and matching quantitative process adjustment scheme; that is, realizes dual judgment of current state quantization and future trend of the flow field, improves the slag entrapment diagnosis precision and reduces the missed diagnosis rate in combination with a diagnosis model of physical constraints, effectively avoids the breakout accident through the forward-looking early warning, and realizes the whole-process slag entrapment prevention and control from data collection to diagnosis early warning and process adjustment.

Description

Technical Field

[0001] This invention relates to the field of continuous casting technology in iron and steel metallurgy, and in particular to a method for diagnosing slag entrainment in slabs based on the unbalanced flow field in the crystallizer. Background Technology

[0002] The stability of the molten steel flow field within the continuous casting mold directly determines the surface quality and internal purity of the slab. However, in actual production, the coupling effect of molten steel turbulence, surface fluctuations, and process parameters easily leads to slag inclusion defects. Traditional manual judgment (such as observing surface fluctuations and sampling and analyzing slag film) suffers from strong subjectivity and delayed response. To improve the accuracy of slag inclusion identification, some patents have attempted to determine defects based on the coupling of multiple process parameters. Patent CN121199056A quantitatively determines slag inclusion defects by real-time acquisition of the molten steel surface fluctuation function F-number and the heat of slag extraction per ton of steel R-number, calculating their ratio, and comparing it with historical slag inclusion critical thresholds.

[0003] Besides the process parameter coupling method, some patents use thermocouple temperature signals to indirectly monitor the risk of slag entrapment, overcoming the limitations of a single liquid level sensor. Patent CN121339363A sets multiple thermocouples at intervals along the height direction on the narrow copper plate of the crystallizer, and forms a thermocouple group with two thermocouples in opposite positions. By calculating the temperature deviation between the two thermocouples in the group, when the deviation value is greater than or equal to a threshold and the duration exceeds a preset time, it is determined that there is a risk of slag entrapment, thus realizing the indirect perception of the uneven flow of molten steel.

[0004] Meanwhile, in the slag entrapment early warning and process control stages, conventional systems generally adopt a "post-event response" mode, that is, process parameters are only adjusted after the slag entrapment defect has been confirmed by finished product inspection or after a steel leakage accident has occurred. Therefore, developing a slab slag entrapment diagnosis method that can comprehensively quantify the flow field imbalance state, adapt to multiple working conditions with high precision, and have both graded diagnosis and real-time early warning functions has become an urgent need for improving the quality and efficiency of continuous casting production. Summary of the Invention

[0005] The purpose of this invention is to overcome the technical defects of traditional slab slag entrainment diagnosis technology, such as incomplete flow field characterization, lack of physical constraints in diagnosis, and lack of forward-looking early warning. It provides a slab slag entrainment diagnosis method based on the unbalanced state of the flow field in the crystallizer, which can achieve accurate quantification of the unbalanced state of the flow field in the crystallizer, graded forward-looking diagnosis and early warning of slag entrainment risk, and match targeted quantitative process adjustment schemes to form a closed-loop prevention and control system covering the entire process of data acquisition, feature quantification, trend prediction, diagnosis and early warning and process control. This improves the accuracy of slag entrainment diagnosis, reduces the missed diagnosis rate, effectively avoids slag entrainment and steel leakage accidents, and ensures the stability of continuous casting production and slab quality.

[0006] The objective of this invention can be achieved through the following technical solution: a method for diagnosing slag entrapment in slabs based on the unbalanced flow field state in a crystallizer, comprising the following steps: Step 1: Collect the core characteristic parameters of the crystallizer flow field and the continuous casting process parameters. The collected continuous casting process parameters are used as the physical constraint input of the diagnostic model. At the same time, the collected raw data is preprocessed to obtain standardized raw data. Step 2: Construct a flow field imbalance index (SFI) based on standardized raw data, integrating multi-dimensional flow field features, and determine the feature weights using the analytic hierarchy process (AHP) and entropy weight method to classify the degree of imbalance. Step 3: Construct the time series of the flow field imbalance index SFI and output the predicted value of the flow field imbalance index SFI in the future t2. Step 4: Setting physical constraints for the continuous casting process in the diagnostic model, designing the diagnostic model structure, and training and optimizing the diagnostic model; Step 5: Based on the degree of imbalance classification, the predicted value of the flow field imbalance index SFI in the future t2, and the final diagnostic results of the diagnostic model, output the diagnostic results of the slag entrainment classification and trigger the early warning operation.

[0007] Preferably, the process for constructing the flow field imbalance index SFI is as follows: S1: Determine the core characteristics of flow field imbalance: steel flow velocity asymmetry F1, liquid surface fluctuation amplitude difference F2, steel turbulent kinetic energy distribution difference F3, nozzle impact depth difference F4, steel temperature gradient difference F5; S2: The feature weights are determined by combining the analytic hierarchy process (AHP) and the entropy weighting method. S3: Calculation of the flow field imbalance index SFI, based on the combined weights and normalized eigenvalues.

[0008] Preferably, the asymmetry of molten steel flow velocity F1 represents the ratio of the velocity difference at the same position on both sides of the wide face to the mean, F1 = |v1-v2| / (v1+v2) / 2, where v1 and v2 are the velocities at corresponding positions on both sides; the difference in liquid surface fluctuation amplitude F2 represents the difference in liquid surface fluctuation amplitude on both sides of the nozzle, F2 = |A1-A2|, where A1 and A2 are the fluctuation amplitudes on both sides of the liquid surface; the difference in turbulent kinetic energy distribution F3 represents the difference in the mean turbulent kinetic energy on both sides of the wide face of the crystallizer, calculated based on high-speed camera images, F3 = |E1-E2|, where E1 and E2 are the turbulent kinetic energies on both sides; the difference in impact depth F4 represents the difference in the impact depth of molten steel on both sides of the nozzle, F4 = |D1-D2|, where D1 and D2 are the impact depths on both sides; and the difference in temperature gradient F5 represents the difference in the mean temperature gradient on both sides of the narrow face of the crystallizer, F5 = |G1-G2|, where G1 and G2 are the temperature gradients on both sides.

[0009] Preferably, the degree of imbalance is divided into three levels according to the flow field imbalance index SFI. The preset thresholds [SFImin, SFImax] are retrieved. 0 ≤ flow field imbalance index SFI < SFImin indicates a balanced state, SFImin ≤ flow field imbalance index SFI < SFImax indicates an unbalanced state, and flow field imbalance index SFI ≥ SFImax indicates a severely unbalanced state.

[0010] Preferably, the construction of the flow field imbalance index SFI time series and the output of the predicted value of the flow field imbalance index SFI within the future t2 specifically includes: Based on the preprocessed flow field characteristic data and the calculated real-time flow field imbalance index (SFI) value, a flow field imbalance index (SFI) time series with a time step of t1 is constructed. At the same time, the change rate of continuous casting process parameters at the corresponding time step is correlated to form a basic dataset for trend prediction. The pre-constructed flow field imbalance trend prediction model is retrieved, and the output layer is the predicted value of the flow field imbalance index (SFI) in the future t2.

[0011] Preferably, the specific implementation of step four is as follows: T1: Physical constraint embedding: Two types of physical constraints for continuous casting process are embedded in the input and training process of the diagnostic model, including process parameter boundary constraints and flow field dynamic constraints; T2: Diagnostic model structure design: The diagnostic model structure consists of 4 layers, including the input layer, CNN spatial feature extraction layer and BiLSTM temporal feature extraction layer, attention fusion layer and output layer; The output layer outputs the probability distribution of three types of slag risk levels (including mild, moderate, and severe), and marks the risk level with the highest probability of slag as the final diagnostic result. T3: Diagnostic model training and optimization.

[0012] Preferably, the specific implementation of step five is as follows: Step 1: Combine the flow field imbalance index SFI, the predicted value of the flow field imbalance index SFI within the next t2, and the final diagnostic results of the diagnostic model; Step 2: Slag Grading and Diagnosis Process Mild slag entrainment: The flow field imbalance index SFI is in an unbalanced state, SFImin≤SFI<SFIz (SFIz∈[(SFImin+SFImax) / 2,SFImax)), and the final diagnosis result is mild slag entrainment risk level, or the current flow field imbalance index SFI<SFImin, but the predicted value of the flow field imbalance index SFI in the future t2 enters [SFImin,SFIz). Moderate slag entrainment: The flow field imbalance index SFI is in an imbalanced state, SFIz≤SFI<SFImax, and the final diagnosis result is a moderate slag entrainment risk level, or the current flow field imbalance index SFI∈[SFImin, SFIz), but the predicted flow field imbalance index SFI in the future t2 enters [SFIz, SFImax). Severe slag entrainment: The flow field imbalance index SFI is in a severely imbalanced state, SFI≥SFImax, and the final diagnosis result is a severe slag entrainment risk level, or the current flow field imbalance index SFI∈[SFIz, SFImax), but the predicted value of the flow field imbalance index SFI in the future t2 is ≥SFImax. Step 3: Real-time early warning mechanism: When the diagnosis result of the slag grading diagnosis is one of mild slag entanglement, moderate slag entanglement, or severe slag entanglement, an early warning operation is triggered. Step 4: Based on the diagnostic results of the slag classification diagnosis, match the corresponding preset process adjustment plan.

[0013] The beneficial effects of this invention are as follows: (1) This invention constructs a multi-dimensional crystallizer flow field imbalance quantification system for slag entrapment and combines it with a metallurgical physical constraint deep learning model to achieve hierarchical diagnosis. It can accurately identify hidden slag entrapment causes such as local slag tumbling, vortex, and flow velocity deviation in the crystallizer from the source of molten steel flow, thus solving the defects of traditional single-parameter detection and judgment that are one-sided and unable to identify potential slag entrapment risks.

[0014] (2) Industrial online testing shows that the overall diagnostic accuracy of the present invention for slag shavings can reach 94.1%, which can reduce the defect rate of slag shavings in slabs by more than 30%.

[0015] (3) Compared with the traditional single-parameter diagnostic method for liquid surface fluctuation (accuracy 78.3%) and the ordinary unconstrained neural network diagnostic method (accuracy 82.5%), the diagnostic accuracy of the present invention is improved by 29.8% and 15.6% respectively, and the false negative rate and false negative rate are both less than 1%, which is significantly better than the existing conventional technical solutions.

[0016] (4) This invention differentiates and retains the characteristics of high-frequency transient flow field to accurately capture the instantaneous vortex rolling behavior of slag, while embedding the flow physics constraints of the crystallizer to correct the irrational prediction deviation of the data-driven model. Combined with the spatiotemporal multi-feature fusion mechanism, it fundamentally improves the identification ability and diagnostic stability of slag rolling hazards under complex working conditions from the perspective of fluid mechanics and metallurgical mechanism. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings; Figure 1 This is a reference diagram of the method of the present invention; Figure 2 This is a reference diagram for the analysis of step two of the present invention; Figure 3 This is a flowchart of step five of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments; Example 1: Please refer to Figures 1 to 3 As shown, this invention is a method for diagnosing slag entrainment in slabs based on the unbalanced flow field state in the crystallizer, comprising the following steps: Step 1: Collect the core characteristic parameters of the crystallizer flow field and the continuous casting process parameters. The collected continuous casting process parameters are used as the physical constraint input of the diagnostic model. At the same time, the collected raw data is preprocessed to obtain standardized raw data. Step 2: Construct a flow field imbalance index (SFI) based on standardized raw data, integrating multi-dimensional flow field features, and determine the feature weights using the analytic hierarchy process (AHP) and entropy weight method to classify the degree of imbalance. Step 3: Construct the time series of the flow field imbalance index SFI and output the predicted value of the flow field imbalance index SFI in the future t2. Step 4: Setting physical constraints for the continuous casting process in the diagnostic model, designing the diagnostic model structure, and training and optimizing the diagnostic model; Step 5: Based on the degree of imbalance classification, the predicted value of the flow field imbalance index SFI in the future t2, and the final diagnostic results of the diagnostic model, output the diagnostic results of the slag entrainment classification diagnosis and trigger the early warning operation. The various processes work together to achieve full coverage from data collection to diagnosis and early warning, and process adjustment; Step 1: Collect core characteristic parameters of the flow field in the crystallizer and continuous casting process parameters (such as nozzle depth and argon flow rate) to solve the problems of invisible flow field and single data dimension. Use online sensor acquisition and synchronous acquisition of process parameters to ensure data comprehensiveness and real-time performance. The collected continuous casting process parameters serve as the physical constraint inputs for the diagnostic model, including: billet casting speed (0.8-1.2 m / min), submerged entry nozzle depth (120-180 mm), argon flow rate (3-17 L / min), molten steel superheat (20-50 °C), and crystallizer vibration parameters (amplitude ±3~±5 mm, frequency 100-280 times / min). Flow field characteristic parameters are acquired at a frequency of 10 Hz, and process parameters are acquired at a frequency of 1 Hz. Data is stored synchronously and timestamped to ensure spatiotemporal alignment. The collected raw data undergoes preprocessing to obtain standardized raw data. This is because the raw data contains issues such as noise interference, missing values, and inconsistent dimensions, necessitating preprocessing to ensure the accuracy of subsequent quantification and diagnosis. The specific steps are as follows: Noise Removal: For random noise from electromagnetic flowmeters and thermocouples, a combination of wavelet transform and moving average is used for noise removal. The wavelet basis is selected as db4, the number of decomposition layers is 5, and the sliding window size is set to 10 to remove high-frequency interference and burst noise. Missing value imputation: For missing values ​​caused by temporary sensor failures, short-term missing values ​​(≤5 sampling points) are imputed by linear interpolation, and long-term missing values ​​(>5 sampling points) are imputed by the average of historical data under the same operating conditions to ensure data continuity; Data normalization: Map all feature parameters to the [0,1] interval to eliminate dimensional differences, using the min-max normalization formula: Where x is the original data, , These represent the minimum and maximum values ​​of this parameter under the same working conditions. Spatiotemporal alignment: Based on the timestamp, the flow field characteristic parameters (10Hz) and process parameters (1Hz) are interpolated and aligned to a unified 10Hz sampling frequency to ensure data integrity within the same time step; Outlier removal: The 3σ criterion is used to remove outlier data. When the data exceeds the mean ± 3 times the standard deviation, it is judged as an outlier and replaced with the mean under the same working conditions to avoid interfering with subsequent calculations. Step 2: Construct a multi-dimensional flow field imbalance index (SFI) based on standardized raw data, and determine the feature weights using the analytic hierarchy process (AHP) and entropy weight method to classify the degree of imbalance. This involves constructing the SFI to accurately quantify the flow field imbalance state. The specific steps are as follows: S1: Determine the core characteristics of flow field imbalance: Select 5 flow field characteristics that have the most significant impact on slag entrainment, namely: steel flow velocity asymmetry F1, liquid surface fluctuation amplitude difference F2, steel turbulent kinetic energy distribution difference F3, nozzle impact depth difference F4, and steel temperature gradient difference F5. Among them, the asymmetry of molten steel flow velocity F1 represents the ratio of the difference in flow velocity at the same position on both sides of the wide face to the mean value, F1=|v1-v2| / (v1+v2) / 2, where v1 and v2 are the flow velocities at corresponding positions on both sides; The difference in liquid level fluctuation amplitude F2 represents the difference in the fluctuation amplitude of the liquid level on both sides of the inlet. F2 = |A1-A2|, where A1 and A2 are the fluctuation amplitudes of the liquid level on both sides. The turbulent kinetic energy distribution difference F3 represents the difference in the mean turbulent kinetic energy on both sides of the wide face of the crystallizer. The turbulent kinetic energy is calculated based on high-speed camera images. F3 = |E1-E2|, where E1 and E2 are the turbulent kinetic energies on both sides. The impact depth difference F4 represents the difference in impact depth of molten steel on both sides of the nozzle. F4 = |D1-D2|, where D1 and D2 are the impact depths on both sides. The temperature gradient difference F5 represents the difference in the mean temperature gradient between the two sides of the narrow face of the crystallizer. F5 = |G1-G2|, where G1 and G2 are the temperature gradients on both sides. S2: The feature weights are determined by combining the Analytic Hierarchy Process (AHP) with the entropy weight method. Specifically, this includes: Analytic Hierarchy Process (AHP): Construct a judgment matrix and determine the subjective weights of the five features, such as ω1[0.32, 0.25, 0.20, 0.13, 0.10]; Entropy weight method: Based on historical production data, calculate the information entropy of each feature and determine the objective weight, such as ω2 [0.28, 0.27, 0.27, 0.12, 0.11]; Combined weights: ω = αω1 + (1-α)ω2, where α is 0.5, to determine the final weights, such as ω[0.30, 0.26, 0.21, 0.12, 0.11]; S3: Calculation of the flow field imbalance index SFI, based on the combined weights and normalized eigenvalues. The formula is as follows: , where fi is the i-th eigenvalue after normalization; Unevenness Classification: Based on the flow field unevenness index SFI, the degree of flow field unevenness is divided into 3 levels. The preset thresholds [SFImin, SFImax] are retrieved. 0 ≤ flow field unevenness index SFI < SFImin indicates an even state, SFImin ≤ flow field unevenness index SFI < SFImax indicates an uneven state, and flow field unevenness index SFI ≥ SFImax indicates a severe uneven state. This provides a basis for subsequent slag entrainment classification and diagnosis.

[0020] Example 2: Step 3: Construct the time series of the flow field imbalance index (SFI) and output the predicted SFI value for the next t2, specifically including: Based on the preprocessed flow field characteristic data and the calculated real-time flow field imbalance index SFI value, a flow field imbalance index SFI time series with a time step of t1 (t1>0, such as t1=1s) is constructed. At the same time, the change rate of continuous casting process parameters (such as casting speed change rate, nozzle depth adjustment rate, and argon flow rate change rate) at the corresponding time step is associated to form a basic dataset for trend prediction. The pre-built flow field imbalance trend prediction model is retrieved. The input layer consists of the SFI time series data and process parameter change rate of the past 30s, and the output layer consists of the predicted value of the flow field imbalance index SFI in the future t2 (t2>0, such as t1=15s). Physically constrained CNN-BiLSTM slag contamination diagnosis is the core of slag contamination diagnosis. It overcomes the shortcomings of traditional machine learning, such as lack of physical constraints and weak generalization. By embedding physical constraints of the continuous casting process and combining the advantages of existing CNN and BiLSTM, high-precision slag contamination diagnosis is achieved. This is step four: setting the physical constraints of the continuous casting process for the diagnostic model, designing the diagnostic model structure, and training and optimizing the diagnostic model. The specific scheme is as follows: T1: Physical Constraint Embedding: Two types of continuous casting process physical constraints are embedded during the input and training of the diagnostic model, including process parameter boundary constraints and flow field dynamic constraints, to improve the generalization ability and diagnostic rationality of the diagnostic model. Process parameter boundary constraints: The preset ranges of casting speed, nozzle depth, and argon flow rate are used as hard constraints. When the input parameters exceed the range, the diagnostic model directly outputs a parameter anomaly warning and does not participate in slag entrainment diagnosis. At the same time, the process parameters are used as prior features to input into the diagnostic model to guide the diagnostic model to learn the coupling relationship between process parameters and flow field and slag entrainment. Flow field dynamics constraints: Based on the principle of continuous casting flow field dynamics, the following constraints are added: molten steel flow velocity ≤ preset a1 (e.g., 2.5 m / s), liquid surface fluctuation amplitude ≤ preset a2 (e.g., 5 mm), nozzle impact depth ≥ preset a3 (e.g., 120 mm). When the diagnostic model prediction results violate these constraints, the output weights of the diagnostic model are adjusted to correct the diagnostic results and avoid misjudgments that do not conform to physical laws. T2: Diagnostic Model Structure Design: Integrating the spatial feature extraction capability of CNN with the temporal feature capture capability of BiLSTM, and combining it with an attention mechanism to improve feature utilization, the diagnostic model structure consists of 4 layers: Input layer: The input dimension is (time step T, feature dimension D), T is 30 (corresponding to 3s data), and D is 10 (5 flow field features + 5 process features). CNN spatial feature extraction layer: contains 3 convolutional layers + 2 pooling layers, convolutional kernel size 3×3, stride 1, padding method same, activation function ReLU, pooling kernel size 2×2, stride 2, extracts the spatial distribution pattern of flow field features (spatial differences in flow velocity and temperature). BiLSTM Temporal Feature Extraction Layer: Contains 2 BiLSTM layers, each with 128 hidden neurons and a dropout rate of 0.2. It captures the temporal variation patterns of flow field features (temporal trends of liquid surface fluctuations), overcoming the limitation of a single LSTM layer that can only extract unidirectional temporal features. Attention Fusion Layer: Employs a spatiotemporal attention mechanism to weightedly fuse spatial features extracted by CNN and temporal features extracted by BiLSTM, highlighting key features and time steps that affect convolutional defects and improving diagnostic accuracy; Output layer: The softmax activation function is used to output the probability distribution of three types of slag risk levels (mild, moderate and severe), and the risk level with the highest probability of slag is marked as the final diagnosis result; T3: Diagnostic model training and optimization: Combining high-precision CFD numerical simulation (e.g., 1000 sets) with industrial field data (e.g., 500 sets), labeling the slag entrainment level (light, moderate, heavy, no slag entrainment), and dividing it into training set (e.g., 70%), validation set (e.g., 20%), and test set (e.g., 10%). Training parameters: Adam optimizer, learning rate 0.001, batch size 32, number of iterations 200, cross-entropy loss function, and L2 regularization (weight decay coefficient 0.001) added to prevent overfitting; Diagnostic model optimization: Bayesian optimization was used to adjust hyperparameters such as the number of convolutional kernels and the number of neurons in the BiLSTM hidden layer to determine the optimal structure and ensure the diagnostic accuracy of the diagnostic model on the test set. The slag entrainment classification diagnosis and early warning system enables slag entrainment classification diagnosis and real-time early warning, coupled with a process adjustment plan, forming a closed-loop control system. Specifically, step five involves: based on the degree of imbalance classification, the predicted value of the flow field imbalance index (SFI) within the next t2, and the final diagnostic results of the diagnostic model, outputting the diagnosis result of the slag entrainment classification diagnosis and triggering an early warning operation. The specific scheme is as follows: Step 1: Combine the flow field imbalance index SFI, the predicted value of the flow field imbalance index SFI within the next t2, and the final diagnostic results of the diagnostic model; Step 2: Slag Grading and Diagnosis Process Mild slag entrainment: The flow field imbalance index SFI is in an unbalanced state, SFImin≤SFI<SFIz (SFIz∈[(SFImin+SFImax) / 2,SFImax)), and the final diagnosis result is mild slag entrainment risk level, or the current flow field imbalance index SFI<SFImin, but the predicted value of the flow field imbalance index SFI in the future t2 enters [SFImin,SFIz). Moderate slag entrainment: The flow field imbalance index SFI is in an imbalanced state, SFIz≤SFI<SFImax, and the final diagnosis result is a moderate slag entrainment risk level, or the current flow field imbalance index SFI∈[SFImin, SFIz), but the predicted flow field imbalance index SFI in the future t2 enters [SFIz, SFImax). Severe slag entrainment: The flow field imbalance index SFI is in a severely imbalanced state, SFI≥SFImax, and the final diagnosis result is a severe slag entrainment risk level, or the current flow field imbalance index SFI∈[SFIz, SFImax), but the predicted value of the flow field imbalance index SFI in the future t2 is ≥SFImax. For example: SFImin = 0.3, SFIz = 0.5, SFImax = 0.7; Mild slag entrainment: The flow field imbalance index (SFI) is in an unbalanced state, 0.3 ≤ SFI < 0.5, and the final diagnosis result is mild slag entrainment risk level, or the current flow field imbalance index (SFI) < 0.3, but the predicted value of the flow field imbalance index (SFI) within the next t2 is in the range [0.3, 0.5). Moderate slag entrainment: The flow field imbalance index SFI is in an imbalanced state, 0.5≤SFI<0.7, and the final diagnosis result is a moderate slag entrainment risk level, or the current flow field imbalance index SFI∈[0.3, 0.5), but the predicted flow field imbalance index SFI in the future t2 enters [0.5, 0.7). Severe slag entrainment: The flow field imbalance index SFI is in a severely imbalanced state, SFI≥0.7, and the final diagnosis result is a severe slag entrainment risk level, or the current flow field imbalance index SFI∈[0.5,0.7), but the predicted value of the flow field imbalance index SFI in the future t2 is ≥0.7; Step 3: Real-time early warning mechanism: When the slag grading diagnosis result is one of mild slag entrapment, moderate slag entrapment, or severe slag entrapment, an early warning operation is triggered. That is, the slag entrapment grade, SFI value, and other information are displayed in real time on the display screen in the continuous casting main control room, and the operator is alerted through an audible and visual alarm. Step 4: Based on the diagnostic results of the slag classification diagnosis, match the corresponding preset process adjustment plan; For example: slight slag entrainment: based on the database, the optimal reduction amount of argon flow rate under the current operating conditions (3-5 L / min) is matched, and the improvement of SFI after adjustment (expected to drop to below 0.3) and the effective time (within 5 seconds) are predicted. For moderate slag entrainment: a combined optimization scheme of reducing casting speed and increasing nozzle depth was used to quantify the respective contributions of reducing casting speed by 0.1-0.2 m / min and increasing nozzle depth by 10-20 mm to improving SFI (e.g., reducing casting speed contributes 60% and increasing nozzle depth contributes 40%), and the total effective time (within 10 seconds) for the adjusted SFI to drop below 0.3 was predicted. For severe slag entrapment: an emergency optimization plan is proposed, which involves reducing the casting speed to 0.6 m / min, increasing the nozzle depth to over 160 mm, and stopping the argon gas supply. The effective time of the adjustment is determined based on the SFI prediction trend (within 15 seconds). Emergency preparations are made in advance to avoid steel leakage accidents. In summary, this invention constructs a core dataset for trend prediction based on real-time SFI values ​​and flow field time-series data. Simultaneously, the trend prediction results are fed back into the SFI state assessment, achieving a dual flow field state judgment based on current quantification and future trends. Furthermore, in collaboration with physically constrained CNN-BiLSTM diagnostics, the model not only captures the current spatial and temporal characteristics of the flow field but also identifies abrupt trend changes, further improving diagnostic accuracy and reducing the missed diagnosis rate. Additionally, by integrating SFI trend predictions, it enables proactive early warning of slag entrainment risks, triggering warnings in advance. Finally, based on the database, it matches quantified process adjustment schemes to graded warnings, achieving precise connection between warnings and adjustments.

[0021] The threshold is set for comparative analysis of results to determine whether they are good or bad. The value of the threshold is determined by a combination of large-scale model analysis of sample data and human experience. It can also be adjusted appropriately based on seasonal or common-sense influencing factors. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing slag entrainment in slabs based on the unbalanced flow field in the crystallizer, characterized in that, Includes the following steps: Step 1: Collect the core characteristic parameters of the crystallizer flow field and the continuous casting process parameters. The collected continuous casting process parameters are used as the physical constraint input of the diagnostic model. At the same time, the collected raw data is preprocessed to obtain standardized raw data. Step 2: Construct a flow field imbalance index (SFI) based on standardized raw data, integrating multi-dimensional flow field features, and determine the feature weights using the analytic hierarchy process (AHP) and entropy weight method to classify the degree of imbalance. Step 3: Construct the time series of the flow field imbalance index SFI and output the predicted value of the flow field imbalance index SFI in the future t2. Step 4: Setting physical constraints for the continuous casting process in the diagnostic model, designing the diagnostic model structure, and training and optimizing the diagnostic model; Step 5: Based on the degree of imbalance classification, the predicted value of the flow field imbalance index SFI in the future t2, and the final diagnostic results of the diagnostic model, output the diagnostic results of the slag entrainment classification and trigger the early warning operation.

2. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 1, characterized in that, The process of constructing the flow field imbalance index SFI is as follows: S1: Determine the core characteristics of flow field imbalance: steel flow velocity asymmetry F1, liquid surface fluctuation amplitude difference F2, steel turbulent kinetic energy distribution difference F3, nozzle impact depth difference F4, steel temperature gradient difference F5; S2: The feature weights are determined by combining the analytic hierarchy process (AHP) and the entropy weighting method. S3: Calculation of the flow field imbalance index SFI, based on the combined weights and normalized eigenvalues.

3. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 2, characterized in that, The asymmetry of molten steel flow velocity F1 represents the ratio of the velocity difference at the same position on both sides of the wide face to the mean, F1 = |v1-v2| / (v1+v2) / 2, where v1 and v2 are the velocities at the corresponding positions on both sides; the difference in liquid surface fluctuation amplitude F2 represents the difference in liquid surface fluctuation amplitude on both sides of the nozzle, F2 = |A1-A2|, where A1 and A2 are the fluctuation amplitudes on both sides; the difference in turbulent kinetic energy distribution F3 represents the difference in the mean turbulent kinetic energy on both sides of the wide face of the crystallizer, calculated based on high-speed camera images, F3 = |E1-E2|, where E1 and E2 are the turbulent kinetic energies on both sides; the difference in impact depth F4 represents the difference in the impact depth of molten steel on both sides of the nozzle, F4 = |D1-D2|, where D1 and D2 are the impact depths on both sides; the difference in temperature gradient F5 represents the difference in the mean temperature gradient on both sides of the narrow face of the crystallizer, F5 = |G1-G2|, where G1 and G2 are the temperature gradients on both sides.

4. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 2, characterized in that, Unevenness Classification: Based on the flow field unevenness index SFI, the degree of flow field unevenness is divided into 3 levels. The preset thresholds [SFImin, SFImax] are retrieved. 0 ≤ flow field unevenness index SFI < SFImin indicates an even state, SFImin ≤ flow field unevenness index SFI < SFImax indicates an uneven state, and flow field unevenness index SFI ≥ SFImax indicates a severe uneven state.

5. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 1, characterized in that, The construction of the flow field imbalance index SFI time series and the output of the predicted value of the flow field imbalance index SFI in the future t2 specifically include: Based on the preprocessed flow field characteristic data and the calculated real-time flow field imbalance index (SFI) value, a flow field imbalance index (SFI) time series with a time step of t1 is constructed. At the same time, the change rate of continuous casting process parameters at the corresponding time step is correlated to form a basic dataset for trend prediction. The pre-constructed flow field imbalance trend prediction model is retrieved, and the output layer is the predicted value of the flow field imbalance index (SFI) in the future t2.

6. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 1, characterized in that, The specific plan for step four is as follows: T1: Physical constraint embedding: Two types of physical constraints for continuous casting process are embedded in the input and training process of the diagnostic model, including process parameter boundary constraints and flow field dynamic constraints; T2: Diagnostic model structure design: The diagnostic model structure consists of 4 layers, including the input layer, CNN spatial feature extraction layer and BiLSTM temporal feature extraction layer, attention fusion layer and output layer; The output layer outputs the probability distribution of the three types of slag risk levels and marks the risk level with the highest probability of slag as the final diagnosis result. T3: Diagnostic model training and optimization.

7. The slab slag entrainment diagnosis method based on the unbalanced flow field state in the crystallizer according to claim 1, characterized in that, The specific plan for step five is as follows: Step 1: Combine the flow field imbalance index SFI, the predicted value of the flow field imbalance index SFI within the next t2, and the final diagnostic results of the diagnostic model; Step 2: Slag Grading and Diagnosis Process Mild slag entrainment: The flow field imbalance index SFI is in an unbalanced state, SFImin≤SFI<SFIz, SFIz∈[(SFImin+SFImax) / 2,SFImax), and the final diagnosis result is mild slag entrainment risk level, or the current flow field imbalance index SFI<SFImin, but the predicted value of the flow field imbalance index SFI in the future t2 enters [SFImin,SFIz). Moderate slag entrainment: The flow field imbalance index SFI is in an imbalanced state, SFIz≤SFI<SFImax, and the final diagnosis result is a moderate slag entrainment risk level, or the current flow field imbalance index SFI∈[SFImin, SFIz), but the predicted flow field imbalance index SFI in the future t2 enters [SFIz, SFImax). Severe slag entrainment: The flow field imbalance index SFI is in a severely imbalanced state, SFI≥SFImax, and the final diagnosis result is a severe slag entrainment risk level, or the current flow field imbalance index SFI∈[SFIz, SFImax), but the predicted value of the flow field imbalance index SFI in the future t2 is ≥SFImax. Step 3: Real-time early warning mechanism: When the diagnosis result of the slag grading diagnosis is one of mild slag entanglement, moderate slag entanglement, or severe slag entanglement, an early warning operation is triggered. Step 4: Based on the diagnostic results of the slag classification diagnosis, match the corresponding preset process adjustment plan.

Images