Structure optimization method for four-flow tundish

Abstract

The invention discloses a structure optimization method of a four-flow tundish, which comprises the following steps: S1, acquiring real-time sensor data of molten steel, and installing a CCD (Charge Coupled Device) camera above the tundish to acquire real-time molten steel surface video image data; s2, the obtained real-time sensor data and real-time molten steel surface video image data are transmitted to a central control system to be preprocessed; s3, the central control system processes the real-time sensor data and the real-time molten steel surface video image data by adopting the trained convolutional neural network model, and outputs a turbulence suppressor type, a dam blocking position and an optimal turbulence suppressor parameter; and S4, the central control system carries out automatic adjustment by adopting an electric adjusting device according to the output turbulence suppressor type, the dam position and the optimal turbulence suppressor parameter. According to the central control system, the electric adjusting device is adopted for automatic adjustment according to the output turbulence suppressor type, the dam blocking position and the optimal turbulence suppressor parameter, and the casting blank quality is improved.

Description

Technical Field

[0001] The invention relates to the technical field of steel smelting, and in particular to a structure optimization method for a four-stream tundish. Background Art

[0002] In the modern steelmaking process, the tundish is one of the important equipment in the steelmaking continuous casting process. The tundish is not only the transmission channel for molten steel from the converter (or electric furnace) to the continuous casting machine, but also plays the role of balancing the temperature of the molten steel and removing inclusions. With the development of steelmaking technology, higher requirements are put forward for the optimization of the structure and function of the tundish, especially in the control of molten steel flow, in order to reduce the short-circuit flow phenomenon and improve the quality of the ingot. In the traditional tundish design, the commonly used turbulence suppressor structure mainly includes U-type and Y-type turbulence suppressors. In addition, there is a design that further optimizes the flow of molten steel by adding a dam on the basis of the U-type turbulence suppressor. However, these turbulence suppressors have certain limitations in practical applications. For example, it is difficult to adjust and adapt to different molten steel flow states in real time, resulting in uneven molten steel flow, which in turn affects the quality of the ingot.

[0003] In order to solve the above problems, in recent years, more and more research and applications have focused on the development of intelligent control systems. By introducing sensors, image recognition technology and advanced artificial intelligence algorithms, real-time monitoring and optimization of the flow of molten steel inside the tundish can be achieved. As a powerful image recognition and processing tool, convolutional neural network (CNN) has demonstrated its superior performance in many fields. In terms of tundish optimization, by training the convolutional neural network model, accurate analysis and prediction of the flow state of molten steel can be achieved, and the parameters of the turbulence suppressor can be adjusted in real time according to the analysis results, thereby optimizing the flow of molten steel and reducing short-circuit flow.

[0004] However, the existing convolutional neural network lacks flexibility in the selection of processing methods. Most processing technologies only adjust the intermediate package based on a single sensor data or a single image data, which cannot achieve the purpose of real-time adjustment. Furthermore, the existing machine learning method cannot take the historical usage of various types of turbulence suppressors as a factor for further adjustment, which makes the selection of turbulence suppressor types inaccurate, and the turbulence suppressor cannot be automatically and flexibly selected according to the state of the molten steel. In addition, the existing technology does not use an electric adjustment device to adjust the parameters and position of the turbulence suppressor in real time according to the instructions of the central control system to achieve the switching of different types of turbulence suppressors, thereby ensuring that the ideal molten steel flow effect can be obtained under different operating conditions. In view of the limitations of the existing molten steel processing, a new solution is urgently needed for a high-accuracy and high-efficiency automated processing method to improve the efficiency and accuracy of the processing, which is of great significance for improving user satisfaction and improving processing efficiency and accuracy. Summary of the invention

[0005] In view of the above problems mentioned in the prior art, the present invention provides a structural optimization method for a four-stream tundish, which first obtains real-time sensor data of molten steel and real-time sensor data of molten steel; secondly, the obtained real-time sensor data and real-time video image data of the surface of molten steel are transmitted to a central control system and preprocessed; thirdly, the central control system uses a trained convolutional neural network model to process the real-time sensor data and real-time video image data of the surface of molten steel, and outputs the type of turbulence suppressor, the position of the dam, and the optimal turbulence suppressor parameters; finally, the central control system is automatically adjusted by an electric regulating device according to the output turbulence suppressor type, the position of the dam, and the optimal turbulence suppressor parameters, so as to optimize the flow of molten steel, reduce the short-circuit flow phenomenon, and improve the quality of the ingot. The central control system of the present application is automatically adjusted by an electric regulating device according to the output turbulence suppressor type, the position of the dam, and the optimal turbulence suppressor parameters, so as to optimize the flow of molten steel, reduce the short-circuit flow phenomenon, and improve the quality of the ingot.

[0006] The present application provides a method for optimizing the structure of a four-stream tundish, comprising the steps of:

[0007] S1: Multiple sensors are installed in the tundish to obtain real-time sensor data of the molten steel, including the temperature, flow rate, flow rate and cleanliness parameters of the molten steel; a CCD camera is installed above the tundish to obtain real-time sensor data of the molten steel;

[0008] S2: The acquired real-time sensor data and real-time video image data of the molten steel surface are transmitted to the central control system and pre-processed;

[0009] S3: The central control system uses the trained convolutional neural network model to process the real-time sensor data and the real-time video image data of the molten steel surface, and outputs the turbulence suppressor type, dam position and optimal turbulence suppressor parameters. The optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor to optimize the flow of molten steel;

[0010] S4: The central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow and improve the quality of the ingot;

[0011] The turbulence suppressor types include the first type, the second type and the third type; the first type adopts a U-shaped turbulence suppressor with a dam-free design; the second type uses a Y-shaped turbulence suppressor with a dam-free design; the third type adds a dam in front of the proximal water outlet based on the U-shaped turbulence suppressor.

[0012] Preferably, the dimensions of the tundish are 1910 mm wide, 9970 mm long, 1150 mm deep, and 9.6 m3 in total. 3 ; The first type of U-shaped turbulence suppressor is a U-shaped structure, including two vertical side walls and a bottom connecting the side walls. It is installed in the middle ladle, has no dam design, and the opening faces upward; the second type of Y-type turbulence suppressor is a Y-shaped structure, including a vertical main wall and two obliquely forked side walls to form a Y shape. It is installed in the middle ladle, has no dam design, and the opening faces upward; the third type of U-shaped turbulence suppressor with a dam, the dam has a height of 250mm, and there are two holes on the dam, which are distributed on both sides with a hole spacing of 500mm, the hole inclination angle is 8°, and the dam is 220mm away from the near-end water outlet.

[0013] Preferably, a plurality of sensors are installed in the tundish, including a temperature sensor, a flow rate sensor, a flow rate sensor and a cleanliness sensor, so as to monitor the temperature, flow rate, flow rate and cleanliness parameters of the molten steel in real time.

[0014] Preferably, the preprocessing includes: first, noise filtering, using low-pass filtering to process the real-time sensor data, and using Gaussian filtering to remove noise in the real-time video image data of the molten steel surface; secondly, feature extraction, extracting the average temperature, average flow velocity, and average flow characteristics of the molten steel from multiple sensor data, and extracting the impurity distribution characteristics of the molten steel surface from the video image data; finally, the sensor data features and the video image data features are connected end to end to form a unified feature vector.

[0015] Preferably, the trained convolutional neural network model uses an improved Swish activation function It is expressed as follows:

[0016]

[0017] Where x is the activation function input, e is the base of the natural logarithm, are the historical usage frequencies of the first, second and third types of turbulence suppressor types respectively. The historical usage frequency of each type is the ratio of the historical usage times of the type to the total usage times of all types. γ1, γ2 and γ3 are the first, second and third trainable weight parameters respectively. β is the input data trainable weight parameter.

[0018] Preferably, the optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor, wherein the angle of the turbulence suppressor is the inclination angle between the turbulence suppressor and the bottom of the tundish; the position of the turbulence suppressor is the placement position of the turbulence suppressor in the tundish, including the distance between the turbulence suppressor and the side wall of the tundish and the height of the turbulence suppressor in the tundish; the aperture size of the turbulence suppressor is the diameter of the hole on the turbulence suppressor.

[0019] Preferably, the central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters, and realizes the switching of different types of turbulence suppressors through the central control system and the electric regulating device, which specifically includes the following steps: the central control system determines in real time whether the output turbulence suppressor type output by the convolutional neural network model is consistent with the turbulence suppressor type currently used in the ladle; when it is judged that they are inconsistent, the central control system issues an instruction to adjust the position and angle of the current turbulence suppressor through the electric regulating device to make it out of the molten steel flow path; activate the turbulence suppressor of the target type, move it to the molten steel flow path through the electric regulating device, and adjust its position, angle and aperture size to adapt to the current molten steel flow state.

[0020] Preferably, the inner wall of the tundish is provided with a high-alumina refractory brick lining to improve the high temperature resistance and service life of the tundish.

[0021] Preferably, the dam material is Inconel 718 alloy or Hastelloy C-276 alloy or molybdenum alloy to ensure the stability and durability of the dam in a high-temperature molten steel environment.

[0022] Preferably, the electric adjustment device includes a stepper motor and a linear actuator, which adjusts the angle, position and aperture size of the turbulence suppressor in real time through the instructions of the central control system, and ensures the accuracy and stability of its operation when switching different types of turbulence suppressors.

[0023] The present invention provides a structure optimization method for a four-stream tundish, which can achieve the following beneficial technical effects:

[0024] 1. The present invention first obtains the real-time sensor data of molten steel and the real-time sensor data of molten steel; secondly, the obtained real-time sensor data and real-time video image data of the surface of molten steel are transmitted to the central control system and preprocessed; thirdly, the central control system uses a trained convolutional neural network model to process the real-time sensor data and the real-time video image data of the surface of molten steel, and outputs the turbulence suppressor type, dam position and optimal turbulence suppressor parameters; finally, the central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow phenomenon, and improve the quality of ingots. The central control system of the present application uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow phenomenon, and improve the quality of ingots.

[0025] 2. The convolutional neural network model trained by the present invention adopts the improved Swish activation function. When the activation function processes the activation function input data, The historical usage frequencies of the first, second and third types of turbulence suppressor types are added to the calculation process of the activation function, which greatly improves the accuracy of turbulence suppressor type selection and the accuracy of turbulence suppressor type parameters. Among them, the historical usage frequency of each type is the ratio of the historical usage times of the type to the total usage times of all types. By optimizing the flow of molten steel according to the historical usage rates of different types, short-circuit flow phenomenon is reduced, and the quality of ingots is improved. The historical data is fully utilized and adjusted in real time. The calculation process is dynamically adjusted according to the historical usage, which greatly improves the accuracy and efficiency of molten steel regulation.

[0026] 3. The present invention realizes the switching of different types of turbulence suppressors through a central control system and an electric adjustment device. The central control system determines in real time whether the type of turbulence suppressor output by the convolutional neural network model is consistent with the type of turbulence suppressor currently used in the ladle; when it is judged that they are inconsistent, the central control system issues an instruction to adjust the position and angle of the current turbulence suppressor through the electric adjustment device to make it out of the flow path of the molten steel; activates the turbulence suppressor of the target type, moves it to the flow path of the molten steel through the electric adjustment device, and adjusts its position, angle and aperture size to adapt to the current flow state of the molten steel, thereby realizing intelligent automatic switching of various types of turbulence suppressors. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 It is a schematic diagram of the steps of a method for optimizing the structure of a four-stream tundish of the present invention;

[0029] Figure 2 It is a schematic diagram of the control system structure of a four-stream tundish of the present invention;

[0030] Figure 3 This is a schematic diagram of a four-stream tundish structure of the present invention;

[0031] Figure 4 It is a schematic diagram of the structure of the first type of U-shaped turbulence suppressor of the present invention;

[0032] Figure 5 is a schematic structural diagram of a second type of Y-type turbulence suppressor of the present invention;

[0033] Figure 6 It is a schematic diagram of the retaining dam structure of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Embodiment 1:

[0036] In view of the above problems mentioned in the prior art, in order to solve the above technical problems, as shown in the attached Figure 1 As shown: The present application provides a method for optimizing the structure of a four-stream tundish, comprising the steps of:

[0037] S1: Multiple sensors are installed in the tundish to obtain real-time sensor data of the molten steel, including the temperature, velocity, flow rate and cleanliness parameters of the molten steel; a CCD camera is installed above the tundish to obtain real-time sensor data of the molten steel; the structure of the tundish is as follows Figure 3As shown. In step S1, a plurality of sensors are installed in the tundish to obtain real-time sensor data of the molten steel, and these sensor data include the temperature, flow rate, flow rate and cleanliness parameters of the molten steel. In addition, a CCD camera is installed above the tundish to obtain real-time video image data of the surface of the molten steel. In one embodiment, the temperature sensor adopts a high-temperature thermocouple (such as a K-type thermocouple), and the temperature of the molten steel is monitored in real time at different heights at the bottom of the tundish and in the tundish to ensure that the temperature of the molten steel is within a suitable range. The flow rate sensor adopts an electromagnetic flowmeter at different positions in the tundish to obtain the flow rate of the molten steel at each position, monitor the flow rate of the molten steel in real time, and analyze the flow state of the molten steel. The flow sensor adopts an ultrasonic flowmeter, and the flow rate of the molten steel is monitored in real time at the inlet and outlet of the tundish to ensure the stability of the flow rate of the molten steel. The cleanliness sensor adopts a laser particle counter, and the cleanliness of the molten steel is monitored in real time at different positions in the tundish, especially near the outlet, to detect the number and size of inclusions in the molten steel. The CCD camera is installed above the tundish, looking down at the surface of the molten steel, obtaining the temperature distribution and flow state of the molten steel surface in real time, and analyzing the flow direction of the molten steel and the distribution of surface impurities through video image data.

[0038] In one embodiment, the above structural optimization method is implemented in a four-stream tundish of a steel plant. Four K-type thermocouples are installed in the tundish, respectively located at the bottom of the tundish and three positions at different heights from the bottom, for real-time monitoring of the temperature of the molten steel. Four electromagnetic flowmeters are installed at different positions in the tundish for real-time monitoring of the flow rate of the molten steel at different positions. Two ultrasonic flowmeters are installed at the inlet and outlet of the tundish, respectively, for monitoring the flow rate of the molten steel. Two laser particle counters are installed at different positions in the tundish, especially near the outlet, for real-time monitoring of the cleanliness of the molten steel. An infrared CCD camera is installed above the tundish to overlook the surface of the molten steel, and is used to obtain video image data of the temperature distribution and flow state of the molten steel surface. Each sensor and CCD camera starts working to collect the temperature, flow rate, flow rate, cleanliness data and video image data of the molten steel in real time. All collected data are transmitted to the central control system through the data transmission module. The central control system preprocesses the collected sensor data and video image data. First, a low-pass filter is used to filter the noise of the real-time sensor data, and a Gaussian filter is used to remove the noise in the video image data. Secondly, feature extraction is performed to extract the average temperature, average flow velocity, and average flow characteristics of the molten steel from the sensor data, and the impurity distribution characteristics on the surface of the molten steel from the video image data. Finally, the sensor data features and the video image data features are connected end to end to form a unified feature vector. The trained convolutional neural network model processes the preprocessed data and outputs the optimal turbulence suppressor type, dam position, and optimal turbulence suppressor parameters (including the angle, position, and aperture size of the turbulence suppressor). According to the output results of the convolutional neural network model, the central control system automatically adjusts the type, position, angle, and aperture size of the turbulence suppressor through the electric adjustment device to optimize the flow of molten steel.

[0039] S2: The acquired real-time sensor data and real-time video image data of the molten steel surface are transmitted to the central control system and pre-processed;

[0040] S3: The central control system uses a trained convolutional neural network model to process real-time sensor data and real-time video image data of the molten steel surface, and outputs the turbulence suppressor type, dam position and optimal turbulence suppressor parameters. The optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor to optimize the flow of molten steel. The structure of the convolutional neural network (CNN) model is as follows: the input layer is used for sensor data input including vectors of molten steel temperature, flow rate, flow rate and cleanliness parameters; the video image data input includes real-time video images of the molten steel surface. The convolution layer includes multiple convolution layers for extracting spatial features of video image data, using a 3x3 convolution kernel, and the activation function is an improved Swish activation function. The pooling layer includes multiple maximum pooling layers, which are used to reduce the dimension of the feature map and retain important features. The fully connected layer is used to expand the feature maps extracted by the convolution layer and the pooling layer into a one-dimensional vector and connect it with the sensor data features; there are multiple fully connected layers, and the activation function is the improved Swish activation function; the output layer includes three outputs: turbulence suppressor type, dam position, and optimal turbulence suppressor parameters (including angle, position, and aperture size).

[0041] In one embodiment, a convolutional neural network model is applied to the four-stream tundish of a steel plant for structural optimization, including the following steps: Step 1: Data acquisition and processing, sensors are installed in the tundish to obtain the temperature, flow rate, flow rate and cleanliness parameters of the molten steel. An infrared CCD camera is installed above the tundish to obtain real-time video image data of the surface of the molten steel. Step 2: Data transmission and preprocessing, sensor data and video image data are transmitted to the central control system, data preprocessing is performed, noise is filtered and features are extracted to form a unified feature vector. Step 3: Model training, a large amount of historical data (sensor data, video image data and corresponding turbulence suppressor types and parameters) are collected to form a training data set, and the convolutional neural network model is trained using the training data set. Step 4: Real-time monitoring and optimization, the sensor data and video image data obtained in real time are input into the trained convolutional neural network model, the model processes the data, and outputs the optimal turbulence suppressor type, dam position and optimal turbulence suppressor parameters (angle, position and aperture size). The central control system adjusts the turbulence suppressor in real time through the electric adjustment device according to the model output to optimize the flow of molten steel. Step 5: Verify and adjust the results, detect the quality of the ingot and the flow state of the molten steel, and verify the optimization effect; based on the verification results, further adjust the model parameters and control strategy to ensure optimal system performance.

[0042] S4: The central control system uses an electric adjustment device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow phenomenon and improve the quality of the ingot; In step S4, the central control system uses an electric adjustment device to automatically adjust according to the turbulence suppressor type, dam position and optimal turbulence suppressor parameters (including angle, position and aperture size) output by the convolutional neural network model to optimize the flow of molten steel, reduce short-circuit flow phenomenon and improve the quality of the ingot. The following is a specific adjustment mechanism and an embodiment: The electric adjustment device comprises a stepper motor: used to accurately adjust the angle and position of the turbulence suppressor; a linear actuator: used to control the height and aperture size of the turbulence suppressor; a control module: receiving instructions from the central control system and coordinating the actions of the stepper motor and the linear actuator. During the automatic adjustment process, initialization is first performed, and the central control system calculates the optimal turbulence suppressor parameters through the convolutional neural network model based on real-time sensor data and video image data. Secondly, the instruction is sent, and the central control system sends an adjustment instruction to the control module, including the type, angle, position and aperture size of the turbulence suppressor to be adjusted. Finally, the adjustment is performed. The control module coordinates the stepper motor and linear actuator to accurately adjust the various parameters of the turbulence suppressor according to the instructions to ensure a smooth and fast adjustment process. After the adjustment is completed, the sensor and CCD camera continue to monitor the flow state of the molten steel in real time to ensure that the adjustment effect meets the expectations. If there is any deviation, further fine-tuning is performed.

[0043] The turbulence suppressor types include the first type, the second type, and the third type; wherein, Figure 4 The first type shown uses a U-shaped turbulence suppressor without a dam design; Figure 5 As shown in the figure, the second type uses a Y-type turbulence suppressor without a dam design; the third type is based on the U-type turbulence suppressor and adds a dam before the near end of the water outlet. Figure 6 shown.

[0044] In some embodiments, the dimensions of the tundish are 1910 mm wide, 9970 mm long, 1150 mm deep, and 9.6 m3 in total. 3 ; The first type of U-shaped turbulence suppressor is a U-shaped structure, including two vertical side walls and a bottom connecting the side walls. It is installed in the middle ladle, has no dam design, and the opening faces upward; the second type of Y-type turbulence suppressor is a Y-shaped structure, including a vertical main wall and two obliquely forked side walls to form a Y shape. It is installed in the middle ladle, has no dam design, and the opening faces upward; the third type of U-shaped turbulence suppressor with a dam, the dam has a height of 250mm, and there are two holes on the dam, which are distributed on both sides with a hole spacing of 500mm, the hole inclination angle is 8°, and the dam is 220mm away from the near-end water outlet.

[0045] In some embodiments, a plurality of sensors are installed in the tundish, and the sensors include temperature sensors, flow rate sensors, flow sensors and cleanliness sensors to monitor the temperature, flow rate, flow rate and cleanliness parameters of the molten steel in real time. In step S1, a plurality of sensors are installed in the tundish to monitor the temperature, flow rate, flow rate and cleanliness parameters of the molten steel in real time. These sensors include temperature sensors, flow rate sensors, flow rate sensors and cleanliness sensors. The temperature sensor uses a high-temperature thermocouple (e.g., a K-type thermocouple) at the bottom and different heights of the tundish (e.g., the bottom, 1 / 3 height, 2 / 3 height and top) to obtain comprehensive temperature distribution data; real-time monitoring of the temperature of the molten steel to ensure that the temperature of the molten steel is within a suitable range, which helps prevent the molten steel from overheating or uneven cooling. The flow rate sensor uses an electromagnetic flowmeter at different positions in the tundish (e.g., at the inlet, middle and outlet) to obtain the flow rate data of the molten steel at each position; real-time monitoring of the flow rate of the molten steel, analysis of the flow state of the molten steel, and ensuring the uniformity and stability of the flow of the molten steel. The flow sensor uses an ultrasonic flow meter at the inlet and outlet of the tundish to measure the flow of molten steel entering and leaving the tundish; real-time monitoring of the flow of molten steel ensures the stability of the flow of molten steel, which helps to adjust the supply of molten steel to avoid excessive or insufficient supply. The cleanliness sensor uses a laser particle counter installed at different positions in the tundish (especially near the outlet) to monitor the number and size of inclusions in the molten steel; real-time monitoring of the cleanliness of the molten steel, detection of inclusions in the molten steel, ensuring the quality of the molten steel, and improving the quality of the ingot.

[0046] In some embodiments, the preprocessing includes: first, noise filtering, low-pass filtering is used to process the real-time sensor data, and Gaussian filtering is used to remove noise in the real-time video image data of the molten steel surface; secondly, feature extraction is performed to extract the average temperature, average flow rate, and average flow characteristics of the molten steel from multiple sensor data, and the impurity distribution characteristics of the molten steel surface are extracted from the video image data; finally, the sensor data features and the video image data features are connected end to end to form a unified feature vector. Preprocessing includes the following main processes: noise filtering, feature extraction, and feature vector formation. The following are detailed steps and specific implementation methods: Low-pass filtering of sensor data is used to remove high-frequency noise in sensor data. Low-pass filtering reduces the impact of instantaneous spikes and other high-frequency noise by smoothing sensor data. Assuming that the data we obtain from the temperature sensor, flow rate sensor, and flow sensor may contain high-frequency interference, we can obtain more stable temperature, flow rate, and flow data through low-pass filtering. Gaussian filtering of video image data is used to remove noise in video image data. Gaussian filtering is an image processing technology that can smooth images and reduce random noise in images, thereby making the image data of the molten steel surface clearer and facilitating subsequent feature extraction. Feature extraction includes average temperature, etc. The average temperature of the molten steel is calculated from all temperature sensor data. For example, if multiple temperature sensors are installed in the tundish, the readings of each sensor are calculated and averaged to obtain the average temperature of the molten steel in the entire tundish. Average flow rate calculates the average flow rate of the molten steel from all flow rate sensor data. By combining the flow rate sensor data at different positions, the average flow rate of the molten steel in the tundish is obtained. Average flow rate calculates the average flow rate of the molten steel from all flow sensor data. The flow sensor data at different positions are averaged to obtain the average flow rate of the molten steel flowing into and out of the tundish. Impurity distribution characteristics of molten steel surface from video image data: The impurity distribution characteristics of the molten steel surface are extracted from video images using image processing algorithms. By analyzing the brightness, texture and edge features in the image, the impurity area on the surface of the molten steel is identified and extracted. Formation of feature vectors The extracted sensor data features (such as average temperature, average flow rate, average flow rate) and video image data features (such as impurity distribution on the surface of the molten steel) are connected end to end to form a unified feature vector. This feature vector will be used for subsequent convolutional neural network model processing.

[0047] In some embodiments, the trained convolutional neural network model uses an improved Swish activation function It is expressed as follows:

[0048]

[0049] Where x is the activation function input, e is the base of the natural logarithm, are the historical usage frequencies of the first, second and third types of turbulence suppressor types respectively. The historical usage frequency of each type is the ratio of the historical usage times of the type to the total usage times of all types. γ1, γ2 and γ3 are the first, second and third trainable weight parameters respectively. β is the input data trainable weight parameter.

[0050] The trained convolutional neural network model adopts an improved Swish activation function, which not only takes into account the input data, but also introduces the historical usage frequency of the turbulence suppressor type, which is calculated by the ratio of the historical usage times of the type to the total usage times of all types. In addition, multiple trainable weight parameters are introduced in the activation function to better adapt to different input data and turbulence suppressor types. The input data includes real-time sensor data of molten steel and real-time video image data of the surface of molten steel. Historical usage frequency The historical usage frequency of a turbulence suppressor type is based on the ratio of its usage times in historical data to the total usage times of all types. For example, if the first type of turbulence suppressor is used 50 times, the second type is used 30 times, and the third type is used 20 times in past operations, then the historical usage frequency of the first type is 50%, the second type is 30%, and the third type is 20%. Trainable weight parameters These parameters are continuously adjusted through the training process to optimize the performance of the model. They are used to adjust the response of the activation function so that the model can better adapt to different operating conditions and input data. At the beginning of model training, the weight parameters are usually randomly initialized. During the model training process, these weight parameters are continuously adjusted through the back propagation algorithm to minimize the loss function. The loss function reflects the difference between the model prediction value and the actual value. As the training progresses, the prediction accuracy of the model continues to improve, and the corresponding weight parameters tend to stabilize. During the training process, the model is evaluated using the validation set data, and the weight parameters are further adjusted based on the evaluation results to prevent overfitting and underfitting. In one embodiment, the improved Swish activation function and convolutional neural network model are applied in a four-stream tundish of a steel plant. Multiple temperature, flow rate, flow rate and cleanliness sensors are installed in the tundish to monitor the parameters of the molten steel in real time. A CCD camera is installed above the tundish to obtain real-time video image data of the molten steel surface. The sensor data is low-pass filtered, and the video image data is Gaussian filtered to extract key features and form feature vectors. A large amount of historical data is collected, including the number of times different types of turbulence suppressors are used and the corresponding molten steel flow data; the historical frequency of use of each type of turbulence suppressor is calculated. For example, assuming that the first type of turbulence suppressor was used 60 times, the second type was used 25 times, and the third type was used 15 times in the historical data, the historical usage frequencies are 60%, 25%, and 15%, respectively. The weight parameters of the convolutional neural network model are randomly initialized, including the trainable weight parameters in the activation function. The model is trained using historical data, and the weight parameters are adjusted through the back propagation algorithm so that the model can accurately predict the flow state of the molten steel and output the optimal turbulence suppressor type and parameters. The sensor data and video image data collected in real time are input into the trained convolutional neural network model.The model predicts that the model processes data, outputs the optimal turbulence suppressor type, dam position and optimal turbulence suppressor parameters (angle, position and aperture size), and automatically adjusts the central control system according to the model output, automatically adjusts the turbulence suppressor through the electric adjustment device to optimize the flow of molten steel. Detect the quality of the ingot and the flow state of the molten steel to verify the optimization effect. The model adjustment further adjusts the model parameters and control strategy according to the verification results to ensure the long-term stable and efficient operation of the system.

[0051] In some embodiments, the optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor, wherein the angle of the turbulence suppressor is the inclination angle between the turbulence suppressor and the bottom of the tundish; the position of the turbulence suppressor is the placement position of the turbulence suppressor in the tundish, including the distance between the turbulence suppressor and the side wall of the tundish and the height of the turbulence suppressor in the tundish; the aperture size of the turbulence suppressor is the diameter of the hole on the turbulence suppressor. The angle of the turbulence suppressor is the inclination angle between the turbulence suppressor and the bottom of the tundish. An angle sensor is used to measure the inclination angle of the turbulence suppressor relative to the bottom of the tundish. The angle sensor is installed on the turbulence suppressor. The sensor monitors the angle between the turbulence suppressor and the bottom of the tundish in real time and transmits the data to the central control system. The angle of the turbulence suppressor is adjusted by an electric adjustment device (such as a stepper motor). The central control system sends an adjustment instruction according to the optimal angle output by the model, and the stepper motor adjusts the inclination angle of the turbulence suppressor according to the instruction. The angle sensor feeds back the current angle in real time to ensure that the adjustment is in place. The position of the turbulence suppressor includes the placement position of the turbulence suppressor in the tundish, specifically the distance between the turbulence suppressor and the side wall of the tundish and the height of the turbulence suppressor in the tundish. Use a displacement sensor or a laser rangefinder to measure the position of the turbulence suppressor. Install a displacement sensor on the turbulence suppressor. The displacement sensor measures the distance between the turbulence suppressor and the side wall of the tundish and the height of the turbulence suppressor in the tundish in real time, and transmits the data to the central control system. The position of the turbulence suppressor is adjusted by an electric adjustment device (such as a linear actuator). The central control system sends an adjustment instruction based on the optimal position output by the model. The linear actuator adjusts the placement position of the turbulence suppressor according to the instruction to ensure that the distance and height from the side wall of the tundish meet the requirements. The displacement sensor feeds back the current position information in real time to ensure that the adjustment is in place. The aperture size of the turbulence suppressor refers to the diameter of the hole on the turbulence suppressor. Use an aperture sensor or a photoelectric diameter gauge to measure the aperture size of the turbulence suppressor. Install an aperture sensor around the hole of the turbulence suppressor. The aperture sensor measures the diameter of the hole of the turbulence suppressor in real time and transmits the data to the central control system. The aperture size of the turbulence suppressor is adjusted by an electric adjustment device. The central control system sends an adjustment command based on the optimal aperture size output by the model. The electric adjustment device adjusts the diameter of the turbulence suppressor hole according to the command to ensure that the aperture size meets the requirements. The aperture sensor provides real-time feedback on the current aperture size to ensure that the adjustment is in place.

[0052] In some embodiments, the central control system uses an electric adjustment device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters, and realizes switching of different types of turbulence suppressors through the central control system and the electric adjustment device, which specifically includes the following steps: the central control system determines in real time whether the output turbulence suppressor type output by the convolutional neural network model is consistent with the turbulence suppressor type currently used in the ladle; when it is judged that it is inconsistent, the central control system issues an instruction to adjust the position and angle of the current turbulence suppressor through the electric adjustment device to make it out of the molten steel flow path; activate the target type of turbulence suppressor, move it to the molten steel flow path through the electric adjustment device, and adjust its position, angle and aperture size to adapt to the current molten steel flow state.

[0053] In some embodiments, the inner wall of the tundish is lined with high-alumina refractory bricks to improve the high temperature resistance and service life of the tundish.

[0054] In some embodiments, the dam material is Inconel 718 alloy or Hastelloy C-276 alloy or molybdenum alloy to ensure the stability and durability of the dam in a high-temperature molten steel environment.

[0055] In some embodiments, the electric adjustment device includes a stepper motor and a linear actuator, which adjusts the angle, position and aperture size of the turbulence suppressor in real time through the instructions of the central control system, and ensures the accuracy and stability of its operation when switching different types of turbulence suppressors.

[0056] The present invention provides a structure optimization method for a four-stream tundish, which can achieve the following beneficial technical effects:

[0057] 1. The present invention first obtains the real-time sensor data of molten steel and the real-time sensor data of molten steel; secondly, the obtained real-time sensor data and real-time video image data of the surface of molten steel are transmitted to the central control system and preprocessed; thirdly, the central control system uses a trained convolutional neural network model to process the real-time sensor data and the real-time video image data of the surface of molten steel, and outputs the turbulence suppressor type, dam position and optimal turbulence suppressor parameters; finally, the central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow phenomenon, and improve the quality of ingots. The central control system of the present application uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow phenomenon, and improve the quality of ingots.

[0058] 2. The convolutional neural network model trained by the present invention adopts the improved Swish activation function. When the activation function processes the activation function input data, The historical usage frequencies of the first, second and third types of turbulence suppressor types are added to the calculation process of the activation function, which greatly improves the accuracy of turbulence suppressor type selection and the accuracy of turbulence suppressor type parameters. Among them, the historical usage frequency of each type is the ratio of the historical usage times of the type to the total usage times of all types. By optimizing the flow of molten steel according to the historical usage rates of different types, short-circuit flow phenomenon is reduced, and the quality of ingots is improved. The historical data is fully utilized and adjusted in real time. The calculation process is dynamically adjusted according to the historical usage, which greatly improves the accuracy and efficiency of molten steel regulation.

[0059] 3. The present invention realizes the switching of different types of turbulence suppressors through a central control system and an electric adjustment device. The central control system determines in real time whether the type of turbulence suppressor output by the convolutional neural network model is consistent with the type of turbulence suppressor currently used in the ladle; when it is judged that they are inconsistent, the central control system issues an instruction to adjust the position and angle of the current turbulence suppressor through the electric adjustment device to make it out of the flow path of the molten steel; activates the turbulence suppressor of the target type, moves it to the flow path of the molten steel through the electric adjustment device, and adjusts its position, angle and aperture size to adapt to the current flow state of the molten steel, thereby realizing intelligent automatic switching of various types of turbulence suppressors.

[0060] The above is a detailed introduction to a structural optimization method for a four-stream tundish. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the ideas and methods of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for optimizing the structure of a four-stream tundish, characterized in that: Includes steps: S1: Multiple sensors are installed in the tundish to obtain real-time sensor data of the molten steel, including the temperature, velocity, flow rate and cleanliness parameters of the molten steel; a CCD camera is installed above the tundish to obtain real-time video image data of the molten steel surface; S2: The acquired real-time sensor data and real-time video image data of the molten steel surface are transmitted to the central control system and pre-processed; S3: The central control system uses the trained convolutional neural network model to process the real-time sensor data and the real-time video image data of the molten steel surface, and outputs the turbulence suppressor type, dam position and optimal turbulence suppressor parameters. The optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor to optimize the flow of molten steel; S4: The central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters to optimize the flow of molten steel, reduce short-circuit flow and improve the quality of the ingot; The turbulence suppressor types include the first type, the second type and the third type; the first type adopts a U-shaped turbulence suppressor with a dam-free design; the second type uses a Y-shaped turbulence suppressor with a dam-free design; the third type adds a dam in front of the proximal water outlet based on the U-shaped turbulence suppressor.

2. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: The dimensions of the tundish are 1910 mm wide, 9970 mm long, 1150 mm deep, and a total volume of 9.6 m 3 ; The first type of U-shaped turbulence suppressor is a U-shaped structure, including two vertical side walls and a bottom connecting the side walls. It is installed in the middle ladle, has no dam design, and the opening faces upward; the second type of Y-type turbulence suppressor is a Y-shaped structure, including a vertical main wall and two obliquely forked side walls to form a Y shape. It is installed in the middle ladle, has no dam design, and the opening faces upward; the third type of U-shaped turbulence suppressor with a dam, the dam has a height of 250mm, and there are two holes on the dam, which are distributed on both sides with a hole spacing of 500mm, the hole inclination angle is 8°, and the dam is 220mm away from the near-end water outlet.

3. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: A plurality of sensors are installed in the tundish, including a temperature sensor, a flow rate sensor, a flow rate sensor and a cleanliness sensor, so as to monitor the temperature, flow rate, flow rate and cleanliness parameters of the molten steel in real time.

4. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: The preprocessing includes: firstly, filtering noise, processing real-time sensor data by low-pass filtering, and removing noise in real-time video image data of molten steel surface by Gaussian filtering; secondly, extracting features, extracting average temperature, average flow velocity, and average flow characteristics of molten steel from multiple sensor data, and extracting impurity distribution characteristics on the surface of molten steel from video image data; finally, connecting sensor data features and video image data features head to tail to form a unified feature vector.

5. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: The trained convolutional neural network model uses an improved Swish activation function It is expressed as follows: Where x is the activation function input, e is the base of the natural logarithm, are the historical usage frequencies of the first, second and third types of turbulence suppressor types respectively. The historical usage frequency of each type is the ratio of the historical usage times of the type to the total usage times of all types. γ1, γ2 and γ3 are the first, second and third trainable weight parameters respectively. β is the input data trainable weight parameter.

6. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: The optimal turbulence suppressor parameters include the angle, position and aperture size of the turbulence suppressor, wherein the angle of the turbulence suppressor is the inclination angle between the turbulence suppressor and the bottom of the tundish; the position of the turbulence suppressor is the placement position of the turbulence suppressor in the tundish, including the distance between the turbulence suppressor and the side wall of the tundish and the height of the turbulence suppressor in the tundish; the aperture size of the turbulence suppressor is the diameter of the hole on the turbulence suppressor.

7. The structural optimization method of a four-stream tundish according to claim 1, characterized in that: The central control system uses an electric regulating device to automatically adjust according to the output turbulence suppressor type, dam position and optimal turbulence suppressor parameters, and realizes the switching of different types of turbulence suppressors through the central control system and the electric regulating device, which specifically includes the following steps: the central control system determines in real time whether the output turbulence suppressor type output by the convolutional neural network model is consistent with the turbulence suppressor type currently used in the ladle; when the judgment is inconsistent, the central control system issues an instruction to adjust the position and angle of the current turbulence suppressor through the electric regulating device to make it out of the flow path of the molten steel; activate the turbulence suppressor of the target type, move it to the flow path of the molten steel through the electric regulating device, and adjust its position, angle and aperture size to adapt to the current flow state of the molten steel.

8. The method for optimizing the structure of a four-stream tundish according to claim 1, characterized in that: The inner wall of the tundish is lined with high-alumina refractory bricks to improve the high-temperature resistance and service life of the tundish.

9. The method for optimizing the structure of a four-stream tundish according to claim 1, characterized in that: The dam material is Inconel 718 alloy or Hastelloy C-276 alloy or molybdenum alloy to ensure the stability and durability of the dam in a high-temperature molten steel environment.

10. The method for optimizing the structure of a four-stream tundish according to claim 1, characterized in that: The electric adjustment device includes a stepper motor and a linear actuator, which adjusts the angle, position and aperture size of the turbulence suppressor in real time through the instructions of the central control system, and ensures the accuracy and stability of its operation when switching different types of turbulence suppressors.

Images