Method for online rapid detection of unevenness of flow field of continuous casting crystallizer
By deploying a multi-source sensor array and a deep learning model on the continuous casting crystallizer, flow field inhomogeneity can be detected and diagnosed in real time, solving the problems of detection lag and false alarms and missed alarms in the existing technology, and realizing efficient and accurate flow field monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BENGANG STEEL PLATES CO LTD
- Filing Date
- 2025-11-10
- Publication Date
- 2026-06-23
AI Technical Summary
Existing technologies lack methods for real-time, rapid, and accurate detection and diagnosis of non-uniform flow fields in continuous casting molds, leading to frequent quality defects and steel leakage accidents.
By deploying a multi-source sensor array on the crystallizer, flow field state signals are collected, feature values are extracted using signal processing technology, and real-time evaluation and diagnosis are performed using a deep learning neural network model, outputting flow field uniformity indicators and abnormal patterns.
It achieves second-level detection time, complete synchronization with the production process, improves the accuracy and reliability of detection, reduces false alarms and false negatives, and has self-learning capabilities to adapt to different process conditions.
Smart Images
Figure CN121535146B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of steelmaking technology, and in particular to a method for online rapid detection of flow field inhomogeneity in continuous casting molds. Background Technology
[0002] The continuous casting mold is a crucial part of continuous casting production, and the flow field of the molten steel inside it directly determines the quality of the cast billet. Uneven flow field can lead to a series of quality defects, such as slag entrapment, bubble entrainment, uneven melting of the protective slag, and surface and subsurface cracks in the cast billet. In severe cases, it can even cause steel leakage accidents.
[0003] Currently, the main methods for detecting and evaluating the flow field in a crystallizer include: physical simulation (water model experiment), which involves establishing a scaled-down transparent crystallizer model in the laboratory, using water to simulate molten steel, and observing the flow field using techniques such as PIV (particle image velocimetry) by injecting dyes or tracer particles. This method cannot be applied online and has limitations due to the "similarity criterion," resulting in deviations from actual conditions. Numerical simulation (computer simulation), which uses CFD (computational fluid dynamics) software to establish mathematical models to simulate the flow, heat transfer, and solidification processes of molten steel. This method consumes significant computational resources and is time-consuming, typically requiring several hours or even days for a single simulation, failing to meet the real-time monitoring and feedback requirements of production lines. Direct temperature measurement, which involves arranging thermocouples on the copper plates of the crystallizer and indirectly inferring the flow field state by monitoring changes in the temperature field. However, temperature field changes lag behind flow field changes and are susceptible to interference from various factors, resulting in a slow response and inability to accurately capture rapid dynamic changes in the flow field. Existing online methods include some patents proposing online monitoring methods based on crystallizer friction or thermocouple temperature. However, these methods are usually based on simple threshold judgments or traditional statistical process control (SPC) diagrams, which have limited ability to recognize complex and nonlinear flow field inhomogeneities, have high false alarm and false negative rates, and cannot accurately diagnose the type and degree of inhomogeneity.
[0004] Therefore, existing technologies lack a method that can truly achieve real-time, rapid, and accurate detection and diagnosis of crystallizer flow field inhomogeneity during the production process. This invention aims to fill this technological gap. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a method for rapid online detection of flow field inhomogeneity in continuous casting molds. The invention utilizes a multi-source sensor array deployed on the mold to acquire high-frequency raw signals that directly or indirectly reflect the flow field state; it extracts feature values strongly correlated with flow field characteristics using signal processing techniques; these feature values are input into a pre-trained deep learning neural network model; this model outputs a comprehensive evaluation index of flow field uniformity in real time and automatically identifies inhomogeneity patterns and levels.
[0006] The technical means employed in this invention are as follows:
[0007] A method for rapid online detection of flow field inhomogeneity in a continuous casting mold includes: acquiring signals from the hydraulic vibration device of the mold and converting them into digital signals as time-series synchronization data and preprocessing them; slicing the time-series synchronization data into fixed time windows, extracting multi-dimensional features from the data within each time window, and constructing a feature vector; the multi-dimensional features include temperature field features, vibration field features, and friction force features; using a hybrid model combining supervised and unsupervised learning for flow field evaluation; and using a multi-task learning model to output alarm level prediction and diagnostic modes.
[0008] Furthermore, the signals in the crystallizer hydraulic vibration device include signals that reflect the frictional force between the crystallizer and the billet shell in real time, as well as the pulling speed and superheat; the preprocessing includes filtering, noise reduction, and signal amplification.
[0009] Furthermore, the temperature field characteristics include transverse temperature gradient, longitudinal temperature gradient, temperature distribution standard deviation, and dominant frequency of temperature fluctuation;
[0010] The transverse temperature gradient Used to quantify the symmetry of heat distribution on the left and right sides of the same height layer, for each height layer of the wide face. The thermocouples are divided into left and right regions. Calculate the average temperature of all thermocouples in the left region within this time window. The average temperature of all thermocouples in the right region Then the height layer The transverse temperature gradient is:
[0011] ;
[0012] The longitudinal temperature gradient This is used to quantify the heat transfer intensity between different height layers, reflecting the intensity of the downward impact of molten steel and the growth of the solidified billet shell. Two height layers are combined into a single layer. Calculate combinations In the middle and upper layers, the average temperature of all thermocouples within the current time window. The average temperature of all thermocouples in the lower layer ;
[0013]
[0014] The standard deviation of the temperature distribution This is used to quantify the uniformity of temperature distribution within the same height layer; for each height layer of the wide face... Take the temperature values of all thermocouples in the row within the time window and calculate the standard deviation of the temperature values.
[0015] The temperature fluctuation main frequency This is used to characterize the periodic fluctuations of the temperature signal and the periodic oscillations of the flow stream; for each height layer of the wide surface... Take the temperature time series data of the thermocouple in the middle column of the row over the entire time window, perform a fast Fourier transform on the signal to obtain a spectrum, and extract the high-energy frequency components in the spectrum as the main frequency of the height layer.
[0016] Furthermore, the vibration field characteristics include vibration energy, vibration asymmetry index, and dominant vibration frequency;
[0017] The vibration energy , used to characterize the vibration energy directly converted from the energy of the steel flow impacting the crystallizer wall; for the signal of each vibration sensor channel k, a fast Fourier transform is performed to obtain the spectrum, and the integral value of the spectrum energy is calculated as the vibration energy of that channel;
[0018] The vibration asymmetry index The lateral asymmetry of the flow field is detected by directly comparing the vibration energy on the left and right sides; the current vibration energy is calculated from the sensors on the left and right sides of the wide face, and the ratio between the two is calculated:
[0019]
[0020] The dominant vibration frequency For each vibration sensor channel k, a fast Fourier transform is performed to extract the frequency component with the highest energy in the entire spectrum as the dominant vibration frequency.
[0021] Furthermore, the frictional characteristics include the average frictional force and the frictional force trend;
[0022] The average friction force This is used to reflect the average frictional force level within the time window; the arithmetic average of the frictional force signals within the time window is calculated.
[0023] The frictional trend This is used to reflect the changing trend of friction force within a time window; the friction force within this time window is linearly fitted, and the slope value is taken.
[0024] Furthermore, the use of a hybrid model combining supervised and unsupervised learning for flow field evaluation specifically includes:
[0025] Collect historical production data, including data from normal periods and known periods of flow field anomalies. Based on process records and billet quality inspection results, assign dual labels to the historical data time periods, including alarm level labels and diagnostic mode labels.
[0026] Furthermore, the multi-task learning model consists of a one-dimensional convolutional neural network and a long short-term memory network, with two branches as the output; the alarm levels include normal, level one warning, level two alarm, and emergency alarm; the level one alarm indicates that the flow field shows a slight abnormal trend, which needs attention but does not require immediate action; the level two alarm indicates that the flow field has shown a clear abnormality, and the operating parameters need to be adjusted as soon as possible; the emergency alarm indicates that the flow field is severely abnormal and needs to be dealt with immediately.
[0027] The diagnostic modes include normal, left-side flow, right-side flow, excessive fluctuation / vortex, slag entrainment risk, and nozzle blockage / abnormality.
[0028] Compared with the prior art, the present invention has the following advantages:
[0029] The method for online rapid detection of flow field inhomogeneity in continuous casting crystallizers provided by this invention realizes online real-time monitoring, shortens the detection time to the second level, achieves complete synchronization with the production process, and can promptly capture transient flow field anomalies.
[0030] The method for online rapid detection of flow field inhomogeneity in continuous casting molds provided by this invention has higher accuracy and reliability. By fusing multi-source information (temperature, vibration, friction, and process parameters), it avoids the limitations of a single signal source. Utilizing the powerful nonlinear mapping and pattern recognition capabilities of deep learning models, it can detect complex and subtle early-stage anomalies that are undetectable by the human eye and traditional algorithms, significantly reducing false alarms and missed alarms.
[0031] The method for online rapid detection of flow field inhomogeneity in continuous casting molds provided by this invention has universality and self-learning ability. The algorithm model can continuously adapt to the flow field characteristics under different steel grades, cross sections and process parameters by continuously collecting new production data and performing incremental learning. It has extremely strong generalization ability and increasingly higher accuracy. Attached Figure Description
[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0033] Figure 1 This is a flowchart of the method for online rapid detection of flow field inhomogeneity in continuous casting crystallizers according to the present invention.
[0034] Figure 2 This is a schematic diagram of the crystallizer sensor layout structure in an embodiment of the present invention.
[0035] Figure 3This is a flowchart of the multi-task learning model in this invention. Detailed Implementation
[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. 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.
[0038] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0039] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0040] like Figure 1As shown, the present invention provides a method for online rapid detection of flow field non-uniformity in continuous casting molds, comprising: acquiring signals from the hydraulic vibration device of the mold and converting them into digital signals, using them as time-series synchronization data and performing preprocessing; in a preferred embodiment of the present invention, the signals from the hydraulic vibration device of the mold include signals reflecting the frictional force between the mold and the billet shell in real time, as well as casting speed and superheat; the preprocessing includes filtering, noise reduction, signal amplification and A / D conversion.
[0041] like Figure 2 As shown, the thermocouple array is deployed on the copper plate of the crystallizer, with 6 rows and 11 columns on the wide side and 6 rows and 2 columns on the narrow side (the original copper plate of the crystallizer has been deployed, but the damaged thermocouples have been replaced), and the sampling frequency is ≥10Hz, which is used to collect temperature field signals (T).
[0042] Vibration sensor array: Four high-frequency acceleration sensors (with a frequency ≥ 1 kHz) are installed on the outer wall of the crystallizer (wide and narrow sides) to collect the crystallizer vibration signal (V) caused by the impact and flow of molten steel.
[0043] The time-series synchronization data is sliced into fixed time windows. Multidimensional features are extracted from the data within each time window, and feature vectors are constructed. These multidimensional features include temperature field features, vibration field features, and friction force features. In implementation, the fixed time window is 30 seconds. Figure 2 As shown, the 11 columns of thermocouples are divided into two regions, left and right. The left region contains thermocouples from columns 1 to 5; the right region contains thermocouples from columns 7 to 11.
[0044] In a specific implementation, as a preferred embodiment of the present invention, the temperature field characteristics include the transverse temperature gradient, the longitudinal temperature gradient, the standard deviation of temperature distribution, and the dominant frequency of temperature fluctuation.
[0045] Lateral temperature gradient Used to quantify the symmetry of heat distribution on the left and right sides of the same height layer, for each height layer of the wide face. The thermocouples are divided into left and right regions. Calculate the average temperature of all thermocouples in the left region within this time window. The average temperature of all thermocouples in the right region Then the height layer The transverse temperature gradient is:
[0046] .
[0047] Output 6 feature values ( , , , , , This represents the symmetry of the flow field at different depths below the meniscus.
[0048] Longitudinal temperature gradient This is used to quantify the heat transfer intensity between different height layers, reflecting the intensity of the downward impact of molten steel and the growth of the solidified billet shell. Two height layers are combined into a single layer. Calculate combinations In the middle and upper layers, the average temperature of all thermocouples within the current time window. The average temperature of all thermocouples in the lower layer ;
[0049]
[0050] For combination j=1, the first row (top layer) and the third row (bottom layer); for combination j=2, the third row (top layer) and the sixth row (bottom layer). Output two feature values ( , ).
[0051] Temperature distribution standard deviation This is used to quantify the uniformity of temperature distribution within the same height layer; for each height layer of the wide face... Take the temperature values of all thermocouples in that row within the given time window and calculate the standard deviation of the temperature values. Output 6 characteristic values ( The larger the standard deviation, the more uneven the temperature distribution, and the more likely eddies are to exist in the flow field.
[0052] Temperature fluctuation frequency This is used to characterize the periodic fluctuations of the temperature signal and the periodic oscillations of the flow stream; for each height layer of the wide surface... Take the temperature time-series data of the thermocouple in the middle column of the row over the entire time window, perform a Fast Fourier Transform on the signal to obtain a spectrum, and extract the high-energy frequency components in the spectrum as the dominant frequency of that height layer. Output 6 eigenvalues ( ).
[0053] In a specific implementation, as a preferred embodiment of the present invention, the vibration field characteristics include vibration energy, vibration asymmetry index, and dominant vibration frequency.
[0054] Vibrational energy This is used to characterize the vibrational energy directly converted from the energy of the steel flow impacting the crystallizer wall, and can directly reflect the magnitude of the flow impact force; for the signal of each vibration sensor channel k, a fast Fourier transform is performed to obtain the spectrum, and the integral value of the spectrum energy is calculated as the vibrational energy of that channel. Four characteristic values are output ( ).
[0055] Vibration asymmetry index The lateral asymmetry of the flow field is detected by directly comparing the vibration energy on the left and right sides; the current vibration energy is calculated from the sensors on the left and right sides of the wide face, and the ratio between the two is calculated:
[0056]
[0057] Output 1 feature value ( ).
[0058] The dominant vibration frequency is determined by performing a Fast Fourier Transform on the signal from each vibration sensor channel k and extracting the frequency component with the highest energy in the entire spectrum. Output 4 feature values ( ).
[0059] In a specific implementation, as a preferred embodiment of the present invention, the friction characteristics include the average friction force and the friction force trend.
[0060] The mean friction force reflects the average friction force level within a given time window; the arithmetic mean of the friction force signals within this time window is calculated. An abnormal increase or decrease in overall friction force indicates a process abnormality.
[0061] Frictional tendency This is used to reflect the changing trend of frictional force within a time window; a linear fit is performed on the frictional force within this time window, and the slope value is taken. A continuous upward trend is one of the precursors to steel leakage.
[0062] Constructing the feature vector V:
[0063]
[0064] The eigenvector V contains 31 eigenvalues. This high-dimensional vector comprehensively represents the overall state of the crystallizer flow field within this time window, providing a rich feature base for subsequent accurate judgment.
[0065] A hybrid model combining supervised and unsupervised learning is used for flow field evaluation. In a preferred embodiment of the invention, historical production data is collected, including data from normal periods and known periods of flow field anomalies. Based on process records (sudden changes in casting speed, nozzle replacement, and steel leakage) and billet quality inspection results, the historical data time periods are labeled with dual tags, including alarm level tags and diagnostic mode tags.
[0066] Alarm level labels: 0 - Normal; 1 - Level 1 Warning: Slight abnormal trend in the flow field, requires attention but no immediate action is needed; 2 - Level 2 Alarm: Clear abnormality in the flow field, operating parameters need to be adjusted as soon as possible; 3 - Emergency Alarm: Severe abnormality in the flow field, immediate action is required.
[0067] Diagnostic mode labels: 0-Normal; 1-Left deviation; 2-Right deviation; 3-Excessive fluctuation / eddy current; 4-Risk of slag entrainment; 5-Station blockage / abnormality.
[0068] like Figure 3 As shown, a multi-task learning model is used to output alarm level prediction and diagnostic modes. In a preferred embodiment of this invention, the multi-task learning model consists of a one-dimensional convolutional neural network and a long short-term memory network, with two branches as the output. The alarm levels include normal, level one warning, level two alarm, and emergency alarm. A level one alarm indicates a slight abnormal trend in the flow field, which requires attention but not immediate action. A level two alarm indicates a clear abnormality in the flow field, requiring adjustment of operating parameters as soon as possible. An emergency alarm indicates a severe abnormality in the flow field, requiring immediate action. The diagnostic modes include normal, left-side flow, right-side flow, excessive fluctuation / eddy current, slag entrainment risk, and nozzle blockage / abnormality.
[0069] One-dimensional convolutional neural networks (1D-CNNs) are adept at automatically extracting deeper feature patterns from local sequence data (temperature and vibration signals within a time window), which is more effective than manually extracted features; long short-term memory networks (LSTMs) are adept at handling long-term dependencies in time series and can capture the gradual trend and periodic characteristics of flow field states.
[0070] The invention also includes a human-computer interaction interface located in the main control room, which displays the flow field uniformity to the operator in a visual form.
[0071] Example 1
[0072] The online monitoring hardware consists of: temperature thermocouples from a steel plant's online system, with 6 rows and 11 columns on the wide side and 6 rows and 2 columns on the narrow side, totaling 156 thermocouples; vibration signals are detected by installing a triaxial high-frequency acceleration vibration sensor at the center of each of the four outer walls of the crystallizer; and friction force detection utilizes existing friction force sensors from the crystallizer's vibration device. The model is deployed on an industrial control computer. The implementation steps are as follows:
[0073] All thermocouple and vibration sensor signals and friction force are collected synchronously, and the pulling speed (S) and superheat (S) are read in real time through the basic automation system. Preliminary preprocessing is performed, including filtering (removing high-frequency noise interference), noise reduction (wavelet transform), signal amplification, and A / D conversion.
[0074] The acquired synchronous time-series data was sliced into fixed 30-second time windows. For the data within each time window, the following multidimensional features were extracted:
[0075] Temperature field characteristics: transverse temperature gradient, longitudinal temperature gradient, standard deviation of temperature distribution, and dominant frequency of temperature fluctuation.
[0076] Vibration field characteristics: vibration energy, vibration asymmetry index, and highest frequency component.
[0077] Friction feature extraction: mean friction and friction trend.
[0078] Take the average value within a 30-second time window and construct a feature vector from the above features. Each 30-second window is transformed into a feature vector V containing 33 dimensions.
[0079] A hybrid model combining supervised and unsupervised learning was used for flow field evaluation. Three months of historical data on the production of Q235B steel for automobiles were collected from the casting machine, and machine feature vectors of 6,000 data points with a time window of 30 seconds were extracted.
[0080] Labeling: Process experts labeled these 6,000 samples based on the process event logs of the same period's billet casting. Model Training: A CNN+LSTM multi-task model was used for training. 80% of the data was used for training, and 20% for testing. Hyperparameters were adjusted until the accuracy on the test set for both tasks reached over 90%.
[0081] Example 2
[0082] This embodiment takes a slight blockage at the water inlet causing rightward flow as an example.
[0083] With a stable pulling speed of 1.2 m / min and a superheat of 28°C, the operator noticed slight unstable fluctuations in the liquid level of the crystallizer, but could not determine the cause. The key characteristic values obtained from the raw data collected within a 30-second time window are shown in Table 1.
[0084] Table 1 Key characteristic values of rightward flow caused by slight blockage at the inlet
[0085]
[0086] After receiving the 33-dimensional feature vector, the model inputs it into the pre-trained multi-task network.
[0087] Alarm level: Output probability: P(2)=92%, P(3)=5%, P(1)=3%, P(0)=0%;
[0088] Conclusion: Level 2 alarm (0-normal, 1-Level 1 warning, 2-Level 2 warning, 3-Emergency alarm).
[0089] Diagnostic mode output: P(2)=88%, P(5)=10%, P(0)=2%, P(1)=0%, P(3)=0%, P(4)=0%;
[0090] Conclusion: Right deviation (0-normal, 1-left deviation, 2-right deviation, 3-excessive fluctuation / eddy current, 4-risk of slag entrainment, 5-bottom nozzle blockage / abnormality).
[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for rapid online detection of flow field inhomogeneity in a continuous casting mold, characterized in that, include: The signals from the hydraulic vibration device of the crystallizer are collected and converted into digital signals, which are used as time synchronization data and preprocessed. The signals from the hydraulic vibration device of the crystallizer include signals that reflect the frictional force between the crystallizer and the billet shell in real time, as well as the drawing speed and superheat. The preprocessing includes filtering, noise reduction and signal amplification. The time-series synchronization data is sliced into fixed time windows, and multi-dimensional features are extracted from the data within each time window to construct a feature vector; the multi-dimensional features include temperature field features, vibration field features, and friction force features. The frictional characteristics include the average frictional force and the frictional force trend; the average frictional force This is used to reflect the average frictional force level within a corresponding time window; the arithmetic average of the frictional force signals within the corresponding time window is calculated; the frictional force trend... This is used to reflect the changing trend of friction force within a time window; a linear fit is performed on the friction force within the corresponding time window, and the slope value is taken. A hybrid model combining supervised and unsupervised learning is used for flow field evaluation, specifically including: Collect historical production data, including data from normal periods and known periods of flow field anomalies. Based on process records and billet quality inspection results, the historical data time periods are labeled with dual tags, including alarm level tags and diagnostic mode tags. Utilize a multi-task learning model to output alarm level prediction and diagnostic patterns; The multi-task learning model consists of a one-dimensional convolutional neural network and a long short-term memory network, and its output is two branches. The alarm levels include normal, level one alarm, level two alarm, and emergency alarm; a level one alarm indicates a slight abnormal trend in the flow field, which requires attention but no immediate action is needed; a level two alarm indicates a clear abnormality in the flow field, requiring adjustment of operating parameters as soon as possible; an emergency alarm indicates a severe abnormality in the flow field, requiring immediate action. The diagnostic modes include normal, left-side flow, right-side flow, excessive fluctuation, eddy current, slag entrainment risk, and nozzle abnormality.
2. The method for online rapid detection of flow field non-uniformity in continuous casting molds according to claim 1, characterized in that, The temperature field characteristics include transverse temperature gradient, longitudinal temperature gradient, temperature distribution standard deviation, and dominant frequency of temperature fluctuation. The transverse temperature gradient Used to quantify the symmetry of heat distribution on the left and right sides of the same height layer, for each height layer of the wide face. The thermocouples are divided into left and right regions. The average temperature of all thermocouples in the left region is calculated within the corresponding time window. The average temperature of all thermocouples in the right region Then the height layer The transverse temperature gradient is: ； The longitudinal temperature gradient This is used to quantify the heat transfer intensity between different height layers, reflecting the intensity of the downward impact of molten steel and the growth of the solidified billet shell. Two height layers are combined into a single layer. Calculate combinations In the middle and upper layers, the average temperature of all thermocouples within the current time window. ; The average temperature of all thermocouples in the lower layer ; The standard deviation of the temperature distribution This is used to quantify the uniformity of temperature distribution within the same height layer; for each height layer of the wide face... Take the temperature values of all thermocouples in the corresponding row within the corresponding time window, and calculate the standard deviation of the temperature values; The temperature fluctuation main frequency This is used to characterize the periodic fluctuations of the temperature signal and the periodic oscillations of the flow stream; for each height layer of the wide surface... Take the temperature time series data of the thermocouple in the middle column of the corresponding row within the entire time window, perform a fast Fourier transform on it to obtain a spectrum, and extract the high-energy frequency components in the spectrum as the main frequency of the corresponding height layer.
3. The method for online rapid detection of flow field non-uniformity in continuous casting molds according to claim 1, characterized in that, The vibration field characteristics include vibration energy, vibration asymmetry index, and dominant vibration frequency; The vibration energy The vibration energy is used to characterize the vibration energy directly converted from the energy of the steel flow impacting the crystallizer wall. For the signal of each vibration sensor channel k, a fast Fourier transform is performed to obtain the spectrum, and the integral value of the spectrum energy is calculated as the vibration energy of the corresponding channel. The vibration asymmetry index The lateral asymmetry of the flow field is detected by directly comparing the vibration energy on the left and right sides; the current vibration energy is calculated from the sensors on the left and right sides of the wide face, and the ratio between the two is calculated: The dominant vibration frequency For each vibration sensor channel k, a fast Fourier transform is performed to extract the frequency component with the highest energy in the entire spectrum as the dominant vibration frequency.