Molten metal casting method and system based on adaptive adjustment of liquid level

Abstract

The application relates to the technical field of image processing, and discloses a molten metal casting method and system based on liquid level self-adaptive adjustment. The method comprises the following steps: configuring image acquisition and controller parameters according to physical parameters of metal materials, identifying a liquid level through infrared and visible light image fusion, monitoring the liquid level in different zones and calculating an unevenness index, adopting a Smith predictor algorithm to compensate time lag and predict a liquid level change trend, superimposing a proportional integral differential control amount, an unevenness compensation amount and a feedforward control amount to generate a stopper target position and drive adjustment, and spraying protective gas to form a gas film isolation layer. The application improves the liquid level recognition accuracy, liquid level distribution uniformity control capability, control response speed and stopper continuous working life in a harsh environment.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and system for casting molten metal based on adaptive adjustment of liquid level height. Background Technology

[0002] In existing technologies, molten metal casting systems based on adaptive liquid level adjustment are mainly composed of four parts working together: a vision acquisition module, an image processing module, a control decision module, and an actuator. An industrial camera installed above the casting tank captures real-time images of the liquid level and transmits them to the image processing unit. The image processing module uses image recognition algorithms such as edge detection and threshold segmentation to extract the boundary features of the molten metal surface from the acquired images and calculates the actual height of the liquid level using a pre-calibrated reference coordinate system. The control decision module compares the measured liquid level height with a preset target value and generates corresponding control command parameters based on the magnitude and trend of the deviation using proportional-integral-derivative control algorithms or other advanced control strategies. These control commands drive a servo motor to precisely adjust the insertion depth of the stopper rod at the outlet of the casting ladle, thereby changing the outflow cross-sectional area and flow rate of the molten metal, so that the liquid level height gradually approaches and stabilizes at the target value. The entire process forms a continuous closed-loop feedback control loop that periodically executes the operation cycle of acquisition, recognition, calculation, and adjustment to achieve dynamic adaptive control of the liquid level height.

[0003] Existing molten metal casting methods based on adaptive liquid level adjustment suffer from the following technical problems. First, there is the issue of robustness in visual recognition. Harsh environments such as high temperatures, fumes, steam, and strong light at the casting site can severely interfere with image acquisition quality, leading to inaccurate or failed liquid surface boundary recognition. This is especially true when liquid level fluctuations are severe or lighting conditions change, making it difficult for image processing algorithms to stably extract liquid surface features. Second, there is the lag in control response. There is a certain time delay between image acquisition and processing and the action of the actuator. When the casting speed is fast or the liquid level changes drastically, this lag can lead to untimely control, resulting in liquid level overshoot or oscillation. Third, the system lacks adaptability. The physical properties of different metal materials, such as surface tension, fluidity, and reflectivity, vary greatly. Fixed image recognition parameters and control strategies are difficult to adapt to all working conditions, requiring parameter adjustments and algorithm optimization for different materials. In addition, the stopper rod, working in a high-temperature molten metal environment for a long time, is prone to problems such as adhesion and corrosion, affecting its action accuracy and service life, thereby reducing the control performance and reliability of the entire system. Summary of the Invention

[0004] This application provides a molten metal casting method and system based on adaptive liquid level adjustment. It uses dual-modal image fusion recognition, zoned liquid level monitoring, Smith predictor control, and gas film protection stopper rods to solve the problems of poor robustness of visual recognition, incomplete liquid level distribution monitoring, lag in control response, and short stopper rod life in the prior art. It improves the accuracy of liquid level recognition, the ability to control the uniformity of liquid level distribution, the control response speed, and the continuous working life of the stopper rod in harsh environments.

[0005] In a first aspect, this application provides a molten metal casting method based on adaptive adjustment of liquid level height, the molten metal casting method based on adaptive adjustment of liquid level height includes:

[0006] Step S1: Adjust the image acquisition parameters and controller parameters according to the physical property parameters of the metal material to be cast to obtain the material adaptation parameter set;

[0007] Step S2: Acquire infrared temperature images and visible light images of the liquid surface in the casting pool. Perform weighted fusion of the high-temperature area mask in the infrared temperature image and the edge features in the visible light image to extract the liquid surface contour coordinates and calculate the current liquid surface height.

[0008] Step S3: Divide the liquid surface contour coordinates into multiple monitoring zones according to spatial location, calculate the liquid surface height of each monitoring zone, and calculate the zone deviation of the liquid surface height of each monitoring zone relative to the current liquid surface height to obtain the liquid surface non-uniformity index.

[0009] Step S4: Combining the remaining metal mass of the casting ladle, the current insertion depth of the stopper rod, and the current liquid level height, the system time lag is compensated using the Smith predictor algorithm, and a predicted liquid level height sequence is calculated.

[0010] Step S5: Calculate the PID control quantity based on the deviation between the target liquid level height and the current liquid level height, superimpose the compensation control quantity generated by the liquid level non-uniformity index and the feedforward control quantity generated by the predicted liquid level height sequence, obtain the target insertion depth of the stopper rod, drive the stopper rod to move to the target position, and simultaneously spray protective gas onto the surface of the stopper rod to form a gas film isolation layer.

[0011] Secondly, this application provides a molten metal casting system based on adaptive adjustment of liquid level, the molten metal casting system based on adaptive adjustment of liquid level includes:

[0012] The adjustment module is used to adjust the image acquisition parameters and controller parameters according to the physical property parameters of the metal material to be cast, so as to obtain a material adaptation parameter set;

[0013] The weighting module is used to acquire infrared temperature images and visible light images of the liquid surface in the casting pool. It performs weighted fusion of the high-temperature area mask in the infrared temperature image and the edge features in the visible light image to extract the liquid surface contour coordinates and calculate the current liquid surface height.

[0014] The partitioning module is used to divide the liquid surface contour coordinates into multiple monitoring zones according to spatial location, calculate the partition liquid surface height of each monitoring zone, and calculate the partition deviation of the partition liquid surface height of each monitoring zone relative to the current liquid surface height to obtain the liquid surface non-uniformity index.

[0015] The calculation module is used to combine the remaining metal mass of the casting ladle, the current insertion depth of the stopper rod, and the current liquid level height, and to compensate for system time lag and calculate the predicted liquid level height sequence using the Smith predictor algorithm;

[0016] The moving module is used to calculate the PID control quantity based on the deviation between the target liquid level height and the current liquid level height, superimpose the compensation control quantity generated by the liquid level non-uniformity index and the feedforward control quantity generated by the predicted liquid level height sequence, obtain the target insertion depth of the stopper rod, drive the stopper rod to move to the target position, and simultaneously spray protective gas onto the surface of the stopper rod to form a gas film isolation layer.

[0017] Thirdly, a molten metal casting apparatus based on adaptive liquid level adjustment is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the molten metal casting apparatus based on adaptive liquid level adjustment to execute the aforementioned molten metal casting method based on adaptive liquid level adjustment.

[0018] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the above-described molten metal casting method based on adaptive adjustment of liquid level.

[0019] The technical solution provided in this application obtains a material adaptation parameter set by adjusting the image acquisition parameters and controller parameters based on the physical properties of the metal material to be cast before the casting operation begins. This solves the problem of poor adaptability to different metal materials caused by using fixed parameters in existing technologies. The system automatically calculates the exposure time of the visible light camera based on the surface reflectivity of the metal material to avoid image overexposure due to strong reflection from the molten metal surface. It sets the temperature compensation coefficient of the infrared thermal imaging camera based on the infrared emissivity to ensure accurate temperature measurement. It calculates the proportional gain parameter of the proportional-integral-derivative controller based on the flow viscosity coefficient and density parameters to match the control strategy with the material flow characteristics. This material-oriented parameter pre-configuration mechanism enables the same casting control system to quickly adapt to different materials. Switching between casting tasks using different metal materials eliminates the need for repeated manual parameter adjustments. This application acquires infrared temperature images and visible light images of the casting pool surface, weightedly fuses the high-temperature region mask from the infrared temperature image with the edge features from the visible light image to extract the liquid surface contour coordinates and calculate the current liquid level. Compared to existing single-vision methods, this dual-modal fusion recognition method combines the penetrability of infrared thermal radiation signals with the precision of visible light geometric edges. Even under harsh environmental interference such as smoke, steam, and strong light, the infrared high-temperature region mask can still accurately identify the liquid surface position, while the visible light edge features supplement the geometric details of the liquid surface contour. The weighted fusion strategy, with infrared high-temperature features as the primary modality and visible light geometric edges as a secondary modality, fully leverages the complementarity of the two modalities, significantly improving performance. To improve the robustness and accuracy of liquid level identification, this application divides the liquid level contour coordinates into multiple monitoring zones based on spatial location. The liquid level height of each zone is calculated separately, and the deviation of each zone's liquid level height relative to the current liquid level height is calculated to obtain a liquid level non-uniformity index. This zoned monitoring method overcomes the limitations of existing single-point liquid level monitoring technologies, enabling a comprehensive understanding of the liquid level distribution at different spatial locations in the casting pool. When the liquid level is higher in the inlet area due to metal inflow or lower in the outlet area due to metal outflow, the zoned monitoring method can promptly detect these liquid level tilts and local anomalies. The liquid level non-uniformity index quantifies the degree of fluctuation in the spatial distribution of the liquid level by calculating the root mean square of the deviation of each zone, providing a basis for subsequent... Uniform compensation control provides a quantitative basis. This application combines the remaining metal mass of the casting ladle, the current insertion depth of the stopper rod, and the current liquid level height. It uses the Smith predictor algorithm to compensate for system time lag and calculate the predicted liquid level height sequence. The Smith predictor is a predictive control method specifically designed for systems with pure time lag. This algorithm establishes a dynamic correlation model between flow rate and liquid level based on the continuity equation and Bernoulli equation. It predicts the future liquid level height evolution process in the time domain based on the current flow rate and historical liquid level change trends. In the prediction calculation, the lag compensation amount corresponding to the total system lag time is subtracted to achieve advance compensation for image processing delay and servo response delay. Compared with existing technologies that only use feedback control to passively adjust based on the current deviation, this approach offers a significant advantage.Predictive control can anticipate and adjust the liquid level before it actually deviates from the target value, effectively solving the problems of liquid level overshoot and oscillation caused by time lag.

[0020] This application calculates the proportional-integral-derivative (PID) control quantity based on the deviation between the target liquid level and the current liquid level, superimposes it with the compensation control quantity generated by the liquid level non-uniformity index, and the feedforward control quantity generated by the predicted liquid level sequence to obtain the target insertion depth of the stopper rod and drive the stopper rod to the target position. This multi-level control strategy integrates the correction effect of PID feedback control, the balancing effect of non-uniformity compensation control, and the advance effect of predictive feedforward control. The PID controller generates the basic control quantity based on the current liquid level deviation and its integral and derivative terms to achieve precise adjustment of the overall liquid level. The non-uniformity compensation layer addresses the spatial distribution of the liquid level. In a uniform situation, a compensation control quantity is generated. When the liquid surface non-uniformity index exceeds a threshold, a compensation mechanism is activated to adjust the stopper rod position and improve the uniformity of liquid surface distribution. The predictive feedforward layer calculates the prediction deviation for future moments based on the predicted liquid surface height sequence to generate a feedforward control quantity to proactively address liquid surface change trends. The total control quantity formed by the superposition of the three control quantities ensures precise control of the overall liquid surface height, takes into account the uniformity of the liquid surface spatial distribution, and achieves advanced adjustment of future trends. Compared with the existing single feedback control strategy, this multi-level control method significantly improves control accuracy and response speed. This application simultaneously drives the stopper rod to move to the target position. A protective gas is injected onto the stopper rod surface to form a gas film isolation layer. Argon gas is transported through the internal gas channels of the stopper rod to the micropores on the surface of the stopper rod, forming a gas protective film. The dynamic pressure scouring effect of the gas film continuously removes oxides and solidified metal particles adhering to the surface of the stopper rod. The isolation effect of the gas film prevents direct contact between the molten metal and the stopper rod matrix material, preventing chemical corrosion and adhesion. Due to the gas film protection, the surface temperature of the stopper rod is significantly lower than that of the molten metal, avoiding melting damage to the surface material of the stopper rod. The gas film protection technology fundamentally solves the problems of decreased operating accuracy and short service life of stopper rods caused by high-temperature corrosion and surface adhesion in existing technologies. The position control accuracy deviation of the stopper rod remains stable within a service cycle, extending the continuous working life of the stopper rod, reducing replacement frequency, production costs, and downtime. This application organically combines material adaptive parameter configuration, dual-modal image fusion recognition, zoned liquid level monitoring, Smith predictor, predictive control, multi-level control strategy superposition, and gas film protection for the stopper rod, etc., to construct a complete technical system from material adaptation to image acquisition, from liquid level recognition to state assessment, from trend prediction to multi-level control, from execution protection to closed-loop feedback, comprehensively improving the accuracy, stability, adaptability, and reliability of liquid level height control during molten metal casting. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments 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.

[0022] Figure 1 This is a schematic diagram of an embodiment of the molten metal casting method based on adaptive adjustment of liquid level height in this application.

[0023] Figure 2 This is a schematic diagram of one embodiment of the molten metal casting system based on adaptive adjustment of liquid level in this application.

[0024] Figure 3 This is a schematic block diagram of the molten metal casting equipment based on adaptive adjustment of liquid level height in an embodiment of the present invention. Detailed Implementation

[0025] This application provides a method and system for casting molten metal based on adaptive adjustment of liquid level height. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the molten metal casting method based on adaptive adjustment of liquid level height in this application includes:

[0027] Step S1: Adjust the image acquisition parameters and controller parameters according to the physical property parameters of the metal material to be cast to obtain the material adaptation parameter set;

[0028] Step S2: Acquire infrared temperature images and visible light images of the liquid surface in the casting pool. Perform weighted fusion of the high-temperature area mask in the infrared temperature image and the edge features in the visible light image to extract the liquid surface contour coordinates and calculate the current liquid surface height.

[0029] Step S3: Divide the liquid surface contour coordinates into multiple monitoring zones according to spatial location, calculate the liquid surface height of each monitoring zone, and calculate the zone deviation of the liquid surface height of each monitoring zone relative to the current liquid surface height to obtain the liquid surface non-uniformity index.

[0030] Step S4: Combining the remaining metal mass of the casting ladle, the current insertion depth of the stopper rod, and the current liquid level, the Smith predictor algorithm is used to compensate for system time lag and calculate the predicted liquid level sequence.

[0031] Step S5: Calculate the PID control quantity based on the deviation between the target liquid level height and the current liquid level height, superimpose the compensation control quantity generated by the liquid level non-uniformity index and the feedforward control quantity generated by the predicted liquid level height sequence, obtain the target insertion depth of the stopper rod, and drive the stopper rod to move to the target position. At the same time, spray protective gas onto the surface of the stopper rod to form a gas film isolation layer.

[0032] It is understood that the executing entity of this application can be a molten metal casting system based on adaptive adjustment of liquid level, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0033] Specifically, step S1 achieves adaptive parameter configuration by establishing a material property database. Before the casting operation begins, the operator selects the type of metal material to be cast through a human-machine interface. For example, after selecting copper, the system retrieves key physical property parameters from the database, such as surface reflectivity (0.85), infrared emissivity (0.15), flow viscosity coefficient (0.004 Pa·s), and density (8900 kg / m³). Then, the system automatically adjusts the temperature compensation coefficient of the infrared thermal imaging camera based on the infrared emissivity, sets the temperature measurement range to 1100 to 1400 degrees Celsius, and calibrates the temperature measurement accuracy to ±2 degrees Celsius. Simultaneously, it adjusts the temperature compensation coefficient based on the surface reflectivity. The exposure time of the visible light camera is calculated using a baseline exposure time of 5 milliseconds and the relationship with surface reflectivity. For copper, the calculated exposure time is 5 times 1 minus 0.85, which equals 0.75 milliseconds. This avoids overexposure of the image due to strong reflection from the molten metal surface. The system further calculates the proportional gain of the proportional-integral-derivative controller based on the flow viscosity coefficient and density. The empirical coefficient of 1200 is divided by the product of the flow viscosity coefficient and density. The calculated proportional gain is 1200 divided by 0.004 multiplied by 8900, which equals 33.71. The temperature compensation coefficient, the exposure time parameter of 0.75 milliseconds, and the proportional gain parameter of 33.71 are combined to form a material adaptation parameter set.

[0034] Step S2 achieves accurate liquid level identification through simultaneous acquisition and weighted fusion of dual-modal images. An infrared thermal imaging camera and a visible light camera, installed above the casting tank, are simultaneously activated at a frequency of 10 Hz. The infrared thermal imaging camera acquires a 640 x 480 pixel temperature distribution matrix, where each pixel value represents the temperature value at its corresponding location. The visible light camera acquires a 1920 x 1080 pixel color image. The system performs binarization on the infrared temperature image, setting a temperature threshold of 1250 degrees Celsius. When the temperature value at a pixel location in the image is greater than or equal to... At 1250 degrees Celsius, the pixel is set to 1 in the mask to identify a high-temperature liquid surface area; otherwise, it is set to 0 to identify a non-liquid surface area. This generates a high-temperature region mask to identify the molten metal surface area. For visible light images, the red, green, and blue three-channel image is first converted to a single-channel grayscale image. During conversion, the grayscale value is equal to 0.299 multiplied by the red channel value, plus 0.587 multiplied by the green channel value, plus 0.114 multiplied by the blue channel value. Then, an edge detection algorithm is applied to extract the liquid surface boundary. This algorithm first uses a Gaussian filter to smooth the grayscale image to reduce noise. Next, the image gradient intensity and direction are calculated. Non-maximum suppression is used to retain local maximum points in the gradient direction. Finally, dual threshold detection is used and a complete liquid surface boundary contour is formed by edge connection. The gradient threshold for edge detection is dynamically set to 80 to 120 according to the surface reflectivity parameter configured in step S1. The system performs coordinate registration and fusion of the infrared mask image and the visible light edge image. The fusion algorithm adopts a weighted superposition method. The value of each pixel in the fused image is equal to the weight coefficient 0.7 multiplied by the corresponding pixel value of the infrared mask plus the weight coefficient 0.3 multiplied by the corresponding pixel value of the visible light edge. This weight allocation reflects the fusion strategy of taking infrared high temperature features as the main feature and visible light geometric edges as the auxiliary feature. The highest point coordinate set of the liquid surface contour is extracted by connected component analysis in the fused image. The average ordinate of all the highest points is calculated as the pixel coordinate of the current liquid surface height. The current actual liquid surface height is calculated using the pre-calibrated pixel-to-actual height conversion relationship. In the conversion relationship, the actual height is equal to the calibration coefficient 0.5 mm per pixel multiplied by the baseline pixel coordinate 600 minus the average ordinate pixel value plus the baseline height 150 mm.

[0035] Step S3 assesses liquid surface non-uniformity through spatial partitioning analysis. The system uniformly divides the casting pool along its length into three monitoring zones: an inlet zone, a central zone, and an outlet zone. Each zone's width is one-third of the total length of the casting pool. From the set of coordinates of the highest point of the liquid surface profile extracted in Step S2, data points are assigned to the corresponding zones based on the range of their x-coordinates. The inlet zone contains data points with x-coordinates ranging from 0 to one-third of the total length; the central zone contains data points ranging from one-third to two-thirds of the total length; and the outlet zone contains data points ranging from two-thirds to the total length. The average liquid surface height within each of the three zones is calculated. The average height of the inlet zone is equal to the sum of all y-coordinate points within that zone divided by the number of points. The average height of the central zone is equal to the sum of all y-coordinate points within that zone divided by the number of points. The average height of the outlet zone is... The degree is equal to the sum of all vertical coordinate points in the area divided by the number of points. The overall average liquid level height is calculated as the sum of the average heights of the three zones divided by 3. The system further calculates the deviation of the liquid level height of each zone from the overall average. The deviation of the inlet zone is equal to the average height of the inlet zone minus the overall average height. The deviation of the center zone is equal to the average height of the center zone minus the overall average height. The deviation of the outlet zone is equal to the average height of the outlet zone minus the overall average height. The liquid surface non-uniformity index is calculated as the square root of the sum of the squares of the deviations of the three zones divided by 3. This index quantifies the uniformity of the liquid surface in spatial distribution. When the non-uniformity index is less than or equal to 2 mm, the liquid surface distribution is considered uniform. When it is greater than 2 mm but less than or equal to 5 mm, it is considered that there is slight non-uniformity that needs attention. When it is greater than 5 mm, it is considered that there is severe non-uniformity that requires immediate intervention.

[0036] Step S4 uses the Smith predictor algorithm to calculate the predictive trend of liquid level change. The Smith predictor is a predictive control algorithm specifically designed to compensate for pure time delay in a system. The system first collects the remaining molten metal mass in the ladle in real time using a weighing sensor, obtains the current depth of the stopper rod inserted into the ladle through feedback from the servo motor encoder, and collects historical data sequences of liquid level height over the past second, with a sampling interval of 0.1 seconds and a total of 11 data points. The system establishes a dynamic correlation model between flow rate and liquid level based on the continuity equation and Bernoulli's equation. The ladle outlet flow rate is equal to the flow coefficient multiplied by the flow cross-sectional area multiplied by twice the gravitational acceleration multiplied by the square root of the height difference, where the flow coefficient is taken as 0.62. The flow cross-sectional area is equal to pi multiplied by the square of the ladle outlet radius minus the square of the stopper rod radius, where the ladle outlet radius is 50 mm. The stopper rod radius is calculated based on the stopper rod insertion depth, which is equal to 45 minus the insertion depth divided by 2. The force acceleration is taken as 9.8 m / s², and the height difference is 800 mm, which is the height difference between the liquid surface in the casting ladle and the outlet. The rate of change of liquid level in the casting pool is calculated according to the law of conservation of mass. The rate of change of liquid level is equal to the difference between the flow rate at the outlet of the casting ladle and the flow rate at the outlet of the casting pool, divided by the metal density and multiplied by the cross-sectional area of ​​the casting pool. The system uses an improved Smith predictor algorithm to predict the liquid level at future moments. The prediction time domain is set to 3 seconds, the prediction step size is 0.1 seconds, and a total of 30 prediction steps are made. The core idea of ​​the Smith predictor is to compensate for the pure lag time of the system. The pure lag time of this system includes an image acquisition and processing delay of 0.15 seconds and a servo motor response delay of 0.25 seconds, for a total delay of 0.4 seconds. In the predictor model, the predicted liquid level at a certain future moment is equal to the current liquid level plus the sum of the sums from 1 to the number of prediction steps. The sum of each step is the corresponding time minus the lag time and multiplied by the average rate of change of liquid level. The average rate of change of liquid level is calculated from the historical data sequence.

[0037] Step S5 generates a stopper rod position adjustment command through a multi-level control strategy. Based on the liquid surface non-uniformity assessment results obtained in step S3 and the predicted liquid surface change trend calculated in step S4, the system employs a three-level control strategy. The first-level main control layer uses the proportional-integral-derivative (PID) controller configured in step S1. Taking the deviation between the overall average liquid surface height and the target liquid surface height of 160 mm as input, the basic control quantity is calculated as proportional gain multiplied by the deviation, plus the integral term of proportional gain divided by the integral time constant multiplied by the deviation, plus the derivative term of proportional gain multiplied by the derivative time constant multiplied by the deviation. Substituting the proportional gain of 33.71, the integral time constant of 0.0199 seconds, and the derivative time constant of 0.004... The calculation takes 98 seconds. The second non-uniformity compensation layer generates a compensation control quantity for the non-uniformity of the liquid surface detected in step S3. It is activated when the non-uniformity is greater than 2 mm. The compensation quantity is equal to the non-uniformity compensation gain coefficient multiplied by the non-uniformity of the liquid surface multiplied by the sign of the maximum partition deviation. The non-uniformity compensation gain coefficient is set to 15. The third predictive feedforward layer generates a leading control quantity based on the future liquid surface height trend predicted in step S4. It extracts the predicted liquid surface height for the next second in the prediction sequence and calculates the prediction deviation, which is equal to the target liquid surface height minus the predicted liquid surface height. The feedforward control quantity is equal to the feedforward gain coefficient multiplied by the prediction deviation. The feedforward gain coefficient is set to 20. The system then superimposes the three control quantities to obtain the final result. The total control quantity equals the basic control quantity plus the non-uniform compensation quantity plus the feedforward control quantity. The total control quantity is converted into the target position increment of the stopper rod through a mapping relationship between the control quantity and the stopper rod position. In this mapping relationship, the position increment equals the total control quantity divided by the control quantity position conversion coefficient, which is set to 8 per millimeter. The system limits the position increment amplitude, restricting the maximum position change in a single adjustment to within ±60 millimeters to prevent excessive stopper rod movement from causing severe fluctuations in the liquid surface. The system calculates the new target position of the stopper rod, which equals the current stopper rod position plus the position increment. The system sends the calculated target position of the stopper rod to the servo motor driver via the communication bus. The driver controls the servo motor to drive the stopper rod according to a trapezoidal speed curve. The stopper moves from its current position to the target position. During the movement of the stopper, the compressed gas supply system inside the stopper is activated simultaneously. Argon gas is delivered through the axial internal channel of the stopper to 12 evenly distributed micro-pores on the surface of the stopper. The pores are 0.8 mm in diameter and are evenly distributed at 30-degree intervals along the circumference of the stopper. The argon gas flow rate is automatically adjusted according to the metal material temperature parameters configured in step S1. For copper material with a casting temperature of 1300 degrees Celsius, the argon gas flow rate is set to the base flow rate of 15 liters per minute multiplied by the casting temperature divided by the base temperature of 1200 degrees Celsius to the power of 0.5. After the argon gas is ejected from the pores, it forms a gas protective film with a thickness of about 0.5 to 1 mm on the surface of the stopper. The dynamic pressure of this gas film is equal to 0.5 times the argon density multiplied by the square of the gas injection velocity, where the gas injection velocity is calculated based on the pore area and total flow rate, the airflow generated by dynamic pressure continuously removes oxides and solidified metal particles adhering to the stopper rod surface. Simultaneously, the gas film's isolating effect prevents direct contact between the molten copper and the stopper rod substrate material, thus preventing chemical corrosion and adhesion. After the stopper rod completes its position adjustment, the system continues the cyclic control process from steps S2 to S5. The control cycle is 0.1 seconds; every 0.1 seconds, the system re-acquires dual-modal images, identifies the liquid level height, assesses liquid level uniformity, predicts liquid level trends, generates control commands, and executes the stopper rod adjustment action, forming a continuous closed-loop adaptive control loop.

[0038] In one specific embodiment, step S1 includes:

[0039] Retrieve the surface reflectance, infrared emissivity, flow viscosity coefficient, and density parameters of the metal material to be cast from the metal material physical property database;

[0040] The temperature compensation coefficient of the infrared thermal imaging camera is set according to the infrared emissivity, and the exposure time parameters of the visible light camera are calculated according to the surface reflectivity.

[0041] Calculate the proportional gain parameter of the PID controller based on the flow viscosity coefficient and density parameter;

[0042] By summarizing the temperature compensation coefficient, exposure time parameter, and proportional gain parameter, a material adaptation parameter set is obtained.

[0043] Specifically, during the process of retrieving the surface reflectivity, infrared emissivity, flow viscosity coefficient, and density parameters of the metal material to be cast from the metal material property database, the operator selects the specific metal material type through a human-computer interaction interface. The system then reads the characteristic parameters of the material from a pre-established database based on the selection result. The database stores the key physical property parameters of different metal materials such as copper, aluminum, and steel in tabular form. For example, when selecting copper, the system retrieves a surface reflectivity value of 0.85, an infrared emissivity value of 0.15, a flow viscosity coefficient of 0.004 Pa·s, and a density value of 8900 kg / m³. These parameters reflect the optical and flow characteristics of copper in its high-temperature molten state. Surface reflectivity indicates the ability of the molten metal surface to reflect visible light; the higher the value, the stronger the reflection. Infrared emissivity indicates the ability of the molten metal surface to radiate infrared heat energy. Flow viscosity coefficient indicates the internal resistance of the molten metal during flow, and density indicates the mass of molten metal per unit volume. When setting the temperature compensation coefficient of the infrared thermal imaging camera based on infrared emissivity, the system inputs the retrieved infrared emissivity value of 0.15 into the parameter configuration module of the infrared camera. The camera internally calculates the relationship between the radiation power of the object surface and the temperature according to the Stefan-Boltzmann law. Since different materials have different infrared emissivity, the infrared energy radiated at the same temperature will also be different. The camera needs to be compensated according to the emissivity in order to accurately measure the temperature. The system sets the temperature measurement range of the camera to 1100 to 1400 degrees Celsius based on the emissivity of 0.15 and calibrates the temperature measurement accuracy to ±2 degrees Celsius to ensure accurate measurement of the liquid surface temperature distribution during the copper material casting process. When calculating the exposure time parameters of a visible light camera based on surface reflectivity, the system substitutes the retrieved surface reflectivity value of 0.85 into the exposure time calculation formula. The exposure time equals the baseline exposure time multiplied by 1 minus the surface reflectivity. The baseline exposure time is set to 5 milliseconds, which is the recommended exposure time for non-reflective materials under standard lighting conditions. For copper, the calculated exposure time is equal to 5 multiplied by 1 minus 0.85, which equals 5 multiplied by 0.15, which equals 0.75 milliseconds. Since the reflectivity of copper liquid surface is as high as 0.85, it means that 85% of the incident light is reflected back. If the standard 5-millisecond exposure time is used, the camera sensor will receive too much reflected light, causing the image to be overexposed and unable to identify the liquid surface boundary. Shortening the exposure time to 0.75 milliseconds reduces the exposure of the camera sensor to 15% of the original, which exactly matches the actual reflectivity of the copper liquid surface and avoids image overexposure.When calculating the proportional gain parameter of the proportional-integral-derivative controller based on the flow viscosity coefficient and density parameters, the system substitutes the flow viscosity coefficient of 0.004 Pa·s and the density of 8900 kg / m³ into the proportional gain calculation formula. The proportional gain equals the empirical coefficient divided by the flow viscosity coefficient multiplied by the density. The empirical coefficient of 1200 is the optimal control parameter benchmark value obtained from a large amount of casting experimental data. The calculated proportional gain is equal to 1200 divided by 0.004 multiplied by 8900, which equals 1200 divided by 35.6, which equals 33.71. This proportional gain parameter determines the controller's response strength to liquid level deviation. The larger the flow viscosity coefficient, the greater the flow resistance of the molten metal and the slower the liquid level change. In this case, a smaller proportional gain is needed to avoid over-control. The larger the density, the greater the mass and inertia of the same volume of metal, and the slower the liquid level response, which also requires a smaller proportional gain. Copper has a relatively small flow viscosity coefficient and a relatively large density. Therefore, the calculated proportional gain of 33.71 is at a moderate level, which can respond quickly to liquid level deviation without causing over-adjustment. When compiling the temperature compensation coefficient, exposure time parameter, and proportional gain parameter to obtain the material adaptation parameter set, the system packages the infrared emissivity (0.15) corresponding to the infrared camera's temperature compensation coefficient, the exposure time parameter (0.75 ms) of the visible light camera, the proportional gain parameter (33.71) of the proportional-integral-derivative controller, and the integral time constant (0.0199 s) and derivative time constant (0.00498 s) calculated based on the liquid surface fluctuation frequency range into a complete parameter configuration file. Each parameter in this parameter set is optimized for the physical properties of copper. The system stores this parameter set in the configuration file of the current casting task and directly calls it during subsequent image acquisition and control processes. These parameters do not need to be recalculated. When other materials are changed, such as aluminum, the system retrieves the physical property parameters of aluminum from the database and recalculates to generate a new set of material-adaptive parameters. Aluminum has a higher surface reflectivity (about 0.92), a lower infrared emissivity (about 0.08), a smaller flow viscosity coefficient (about 0.0012 Pascals / second), and a smaller density (about 2700 kg / m³). The recalculated exposure time of aluminum is about 0.4 milliseconds, and the proportional gain is about 370. This material-oriented parameter adaptive configuration mechanism allows the same casting control system to adapt to the casting process requirements of different metal materials, avoiding the problem of manually and repeatedly adjusting parameters when switching between different materials using the traditional fixed parameter method.

[0044] In one specific embodiment, step S2 includes:

[0045] Infrared temperature images and visible light images of the liquid surface in the casting pool are acquired simultaneously using an infrared thermal imaging camera and a visible light camera.

[0046] The infrared temperature image is binarized by setting a temperature threshold to generate a high-temperature region mask that identifies the molten metal surface area. At the same time, the visible light image is converted to grayscale and the Canny edge detection algorithm is applied to extract the liquid surface boundary to obtain edge features.

[0047] The high-temperature region mask and edge features are registered in coordinates and then weighted and superimposed according to preset weight coefficients to obtain a fused image.

[0048] Extract the set of coordinates of the highest point of the liquid surface contour from the fused image, calculate the average value of the ordinate of each point in the set of coordinates of the highest point, and combine it with the pre-calibrated pixel-actual height conversion relationship to obtain the current liquid surface height.

[0049] Specifically, when simultaneously acquiring infrared temperature images and visible light images of the casting pool surface using an infrared thermal imaging camera and a visible light camera, the two cameras are mounted on a fixed bracket approximately 3 meters above the casting pool surface. The infrared thermal imaging camera and the visible light camera are synchronously activated via a hardware trigger signal, with the trigger frequency set to 10 Hz, meaning 10 frames are acquired per second. The image acquired by the infrared thermal imaging camera is a 640 x 480 pixel temperature distribution matrix. Each pixel value in the matrix is ​​not color information but the actual temperature value corresponding to that location. For example, a pixel value of 1285 in the 100th row and 200th column of the matrix indicates that the temperature at that location is 1285 degrees Celsius. The image acquired by the visible light camera is a 1920 x 1080 pixel color image, with each pixel containing red, green, and blue color channels ranging from 0 to 255. The acquisition timestamps of the two cameras are completely identical to ensure that the liquid surface state is captured at the same moment. The acquired image data is transmitted to the image processing module of the industrial control computer via gigabit Ethernet.

[0050] When generating a high-temperature region mask to identify the molten metal surface area by binarizing an infrared temperature image with a temperature threshold, the system sets the temperature threshold to 1250 degrees Celsius. It traverses each pixel in the infrared temperature image matrix, reads the temperature value stored at that pixel location, and determines that the location belongs to the molten metal surface area when the temperature value is greater than or equal to 1250 degrees Celsius, assigning the pixel a value of 1 in the mask image. When the temperature value is less than 1250 degrees Celsius, it determines that the location belongs to the background area, such as the edge of the casting pool, smoke, or solidified metal slag, assigning the pixel a value of 0 in the mask image. After pixel-by-pixel traversal processing, a 640 x 480 pixel binary image, i.e., the high-temperature region mask, is generated. The area formed by the pixels with a value of 1 in the mask is the location of the molten metal surface. Since the infrared thermal imaging camera captures thermal radiation signals, even if there is smoke or dust obstructing the casting site, the infrared signal can still penetrate the smoke and dust to detect the high-temperature characteristics of the liquid surface. Therefore, the high-temperature region mask can stably identify the liquid surface location in harsh environments.Simultaneously, when extracting edge features of the liquid surface boundary using an edge detection algorithm after converting the visible light image to grayscale, the system first converts the 1920 x 1080 pixel color image into a single-channel grayscale image. During the conversion, the red, green, and blue channel values ​​of each pixel are weighted and summed. The grayscale value is equal to 0.299 multiplied by the red channel value, plus 0.587 multiplied by the green channel value, plus 0.114 multiplied by the blue channel value. These three weighting coefficients conform to the human eye's perception characteristics of different color brightness. Each pixel in the converted grayscale image has only one grayscale value from 0 to 255. Then, the edge detection algorithm is applied to extract the liquid surface boundary. This algorithm consists of four steps. The first step is to smooth the grayscale image using a Gaussian filter. The Gaussian filter is a 5 x 5 convolution kernel with the center weight being the largest and the edge weight decreasing. This convolution kernel is slid across the grayscale image, and a weighted average is performed on each pixel and its surrounding pixels. The smoothed image has reduced noise but retains edge information. The second step is to calculate the image gradient intensity and direction. The Sobel operator is applied to the smoothed grayscale image to calculate the horizontal and vertical gradients respectively. The gradients are calculated in the horizontal and vertical directions. The horizontal gradient reflects the rate of change of pixel brightness in the horizontal direction, while the vertical gradient reflects the rate of change in the vertical direction. The gradient strength is equal to the square root of the sum of the squares of the horizontal and vertical gradients. The gradient direction is equal to the arctangent of the vertical gradient divided by the horizontal gradient. The third step uses non-maximum suppression to retain local maxima in the gradient direction. The gradient strength of each pixel is compared with that of its neighboring pixels along the gradient direction. If the gradient strength of a pixel is not a local maximum, it is suppressed to 0. The retained pixels form refined edge candidate points. The fourth step uses dual threshold detection and edge connection to form a complete liquid surface boundary contour. The high threshold is set to 120 and the low threshold to 80. Pixels with a gradient strength greater than 120 are directly marked as strong edge points, pixels with a gradient strength between 80 and 120 are marked as weak edge points, and pixels with a gradient strength less than 80 are discarded. Then, the 8-neighborhood of each weak edge point is checked. If there is a strong edge point in the neighborhood, the weak edge point is also marked as an edge point to achieve edge connection. Finally, a continuous liquid surface boundary contour, i.e., edge features, is obtained.

[0051] When merging the high-temperature region mask and edge features by coordinate registration and weighted superposition according to preset weighting coefficients to obtain the fused image, coordinate registration is required first because the infrared thermal imaging camera has a resolution of 640 x 480 pixels while the visible light camera has a resolution of 1920 x 1080 pixels. The system scales the visible light edge feature image from 1920 x 1080 pixels to 640 x 480 pixels to align with the size of the infrared mask image. The scaling uses bilinear interpolation. For each pixel coordinate in the scaled image, its corresponding floating-point coordinate position in the original image is calculated. The scaled pixel value is obtained by weighted interpolation based on the values ​​of the four integer coordinate pixels surrounding the floating-point coordinate. After coordinate registration, both the infrared high-temperature region mask and the visible light edge feature image are 640 x 480 pixels in size, and the corresponding pixel positions represent the same physical position of the casting pool liquid surface. Then, weighted superposition is performed according to preset weighting coefficients. The value of each pixel in the fused image is equal to the weighting coefficient 0.7 multiplied by the value of the corresponding image of the high-temperature region mask. The prime value is added to a weighting coefficient of 0.3 and multiplied by the pixel value corresponding to the edge feature. For example, in the fused image at the 100th row and 200th column pixel position, if the high-temperature region mask value at this position is 1, indicating a high-temperature liquid surface region, and the edge feature value at this position is 1, indicating an edge has been detected, then the fused image value at this position is equal to 0.7 multiplied by 1 plus 0.3 multiplied by 1, which equals 1. If the high-temperature region mask value at this position is 1 but the edge feature value is 0, then the fused image value is equal to 0.7 multiplied by 1 plus 0.3 multiplied by 0, which equals 0.7. If the value is 0 but the edge feature value is 1, then the fused image value is equal to 0.7 multiplied by 0 plus 0.3 multiplied by 1, which equals 0.3. The allocation of weight coefficients 0.7 and 0.3 reflects the fusion strategy of prioritizing infrared high-temperature features and supplementing with visible light geometric edges. This is because the anti-interference ability of infrared signals is stronger than that of visible light. In the case of smoke and dust obscuring the image, the infrared mask can still accurately mark the liquid surface, while the visible light edge may be blurred or broken. The fused image combines the advantages of both modes, preserving the robustness of the infrared signal and supplementing the geometric accuracy of the visible light edge.

[0052] When extracting the highest point coordinate set of the liquid surface contour from the fused image, calculating the average of the ordinates of each point in the highest point coordinate set, and combining this with the pre-defined pixel-to-actual height conversion relationship to obtain the current liquid surface height, the system performs connected component analysis on the fused image. A connected component refers to a region of pixels in the image whose values ​​are greater than a certain threshold and are interconnected. Setting the threshold to 0.5, the system traverses the fused image, marking pixels with values ​​greater than 0.5 as foreground pixels. The 8-connectivity principle is used to determine whether adjacent pixels belong to the same connected component. An 8-connectivity component refers to a pixel's eight neighboring pixels in the top, bottom, left, right, and four diagonal directions. If adjacent pixels... Pixels that are all foreground pixels are grouped into the same connected component. After connected component analysis, several independent liquid surface regions are obtained. The connected component with the largest area is selected as the main liquid surface region. The contour boundary is extracted from this connected component. The contour boundary is the outermost set of pixels in the connected component. All pixels on the contour boundary are traversed, and the x and y coordinates of each point are read. In the image coordinate system, the smaller the y coordinate value, the higher the position of the point in the image, and the higher the actual height of the liquid surface in the casting pool. All pixels with the smallest y coordinates on the contour boundary are found to form the highest point coordinate set. Assuming that this set contains 120 pixels, the coordinates of these 120 pixels are calculated. The average ordinate value of 20 points is calculated by summing the 120 ordinate values ​​and dividing by 120. For example, if the result is 542 pixels, the actual height of the liquid level is calculated by combining this average with the pre-calibrated pixel-to-actual-height conversion relationship. The calibration process is performed during the installation and commissioning of the casting system. A standard calibration plate with precise height markings is placed in the casting pool. Images of the calibration plate are captured by a camera, recording the ordinate pixel values ​​corresponding to different actual heights. A conversion relationship is obtained by fitting multiple sets of calibration data. The actual height is equal to the calibration coefficient multiplied by the baseline pixel coordinates. Subtracting the current vertical coordinate pixel value and adding the reference height, with a calibration coefficient of 0.5 mm per pixel, indicates that each pixel in the image corresponds to a distance of 0.5 mm in actual space. The reference line pixel coordinate is 600 pixels, corresponding to a fixed reference height of the casting pool, which is 150 mm. Substituting the average vertical coordinate pixel value of 542 into the conversion formula, the current actual height of the liquid surface is equal to 0.5 multiplied by 600 minus 542 plus 150, which equals 0.5 multiplied by 58 plus 150, which equals 29 plus 150, which equals 179 mm. This value indicates that the current vertical height of the liquid surface from the bottom of the casting pool is 179 mm.

[0053] In one specific embodiment, step S3 includes:

[0054] The casting pool is divided into three monitoring zones along its length: the inlet zone, the center zone, and the outlet zone. The liquid surface contour coordinates are assigned to the corresponding monitoring zones according to the range of the abscissa of each point in the liquid surface contour coordinates.

[0055] Calculate the average ordinate of the liquid surface contour coordinate points in each monitoring zone to obtain the liquid surface height of each monitoring zone;

[0056] Calculate the difference between the liquid level in each monitoring zone and the current liquid level to obtain the zone deviation for each monitoring zone;

[0057] The root mean square of the sum of the squares of the deviations in each monitoring zone is used to obtain the liquid surface non-uniformity index.

[0058] Specifically, the casting pool is divided into three monitoring zones along its length: an inlet zone, a central zone, and an outlet zone. When assigning the liquid surface contour coordinates to the corresponding monitoring zones based on the range of the abscissas of each point in the liquid surface contour coordinate system, the system first obtains the total length parameter of the casting pool. In the image coordinate system, the abscissa range corresponding to the length of the casting pool is from 0 pixels to 640 pixels. The width of each monitoring zone is one-third of the total length. The inlet zone corresponds to an abscissa range of 0 to 213 pixels, covering the area of ​​the casting pool starting from the metal inflow end. The central zone corresponds to an abscissa range of 213 to 426 pixels, covering the middle area of ​​the casting pool. The outlet zone corresponds to an abscissa range of... The range is 426 to 640 pixels, covering the area near the metal outflow end of the casting pool. The system reads the x-coordinate value of each point from the set of coordinates of the highest point of the liquid surface contour extracted in step S2, determines which range the x-coordinate belongs to, and then assigns the coordinate point to the data set of the corresponding partition. After traversing all the liquid surface contour coordinate points, the coordinate point sets of the three partitions are obtained. This partitioning method can reflect the liquid surface height distribution characteristics of different spatial locations in the casting pool. The liquid surface in the inlet area near the metal inflow position may be higher due to metal impact, the liquid surface in the outlet area near the metal outflow position may be lower due to metal loss, and the liquid surface in the central area is relatively stable.

[0059] When calculating the average ordinate of the coordinate points of the liquid surface contour in each monitoring zone to obtain the zone liquid level height, the system extracts the ordinate value of each point from the coordinate point set of the inlet zone, sums them, and divides them by the number of coordinate points to obtain the average ordinate of the inlet zone. In the image coordinate system, the smaller the ordinate value, the higher the position and the higher the actual liquid level height. Substituting the average ordinate of the inlet zone into the pixel-to-actual height conversion relationship, the zone liquid level height of the inlet zone is calculated. In the conversion formula, the calibration coefficient is 0.5 mm per pixel, the baseline pixel coordinate is 600 pixels, and the baseline height is 150 mm. The same method is used to calculate the zone liquid level height of the central zone and the outlet zone. The ordinate of each zone coordinate point is summed and divided by the number of points to obtain the average ordinate, which is then substituted into the conversion formula. The liquid level height data of the three zones reflects the spatial distribution characteristics of the casting pool liquid level. The liquid level in the inlet zone is usually higher than the average level due to the impact of metal inflow, the liquid level in the outlet zone is lower than the average level due to metal outflow, and the liquid level in the central zone is relatively close to the overall average value.

[0060] To calculate the zoning deviation of each monitoring zone, the system first calculates the overall average liquid level. The three zone liquid level heights are then added together and divided by 3 to obtain the overall average value, which is used as the representative value of the current liquid level. Next, the deviation of each zone's liquid level relative to the overall average liquid level is calculated. The inlet zone deviation equals the inlet zone's liquid level height minus the overall average liquid level; a positive value indicates that the inlet zone's liquid level is higher than the overall average. The center zone deviation equals the center zone's liquid level height minus the overall average. The outlet zone deviation equals the outlet zone's liquid level height minus the overall average; a negative value indicates that the liquid level in that area is lower than the overall average. These three zone deviation values ​​reflect the fluctuation distribution of the liquid level along the length of the casting pool. When the absolute values ​​of all zone deviations are small, it indicates a uniform liquid level distribution. When the absolute value of a zone deviation is large, it indicates that the liquid level in that area deviates significantly from the average level, indicating a local anomaly.

[0061] When calculating the root mean square (RMS) of the squared deviations of each monitoring zone to obtain the liquid surface non-uniformity index, the system squares the deviations of the three zones separately to eliminate the influence of positive and negative signs. The sum of the squared deviations of the inlet zone, center zone, and outlet zone is then obtained. This sum is divided by 3 to obtain the average squared deviation. Taking the square root of the average squared deviation yields the liquid surface non-uniformity index. This RMS calculation method effectively quantifies the degree of fluctuation in the spatial distribution of the liquid surface. A small non-uniformity index value indicates that the liquid surface heights of the three zones are close to the overall average, resulting in a uniform liquid surface distribution. A large non-uniformity index value indicates that the liquid surface heights of some zones significantly deviate from the overall average. Uneven liquid surface distribution is detected by the system, which determines the liquid surface condition based on the non-uniformity index. When the non-uniformity index is less than or equal to 2 mm, the liquid surface distribution is considered uniform and no intervention is required. When the non-uniformity index is greater than 2 mm but less than or equal to 5 mm, slight non-uniformity is detected and attention is needed. When the non-uniformity index is greater than 5 mm, severe non-uniformity is detected and the compensation control layer needs to be activated immediately to adjust the stopper position and correct the liquid surface distribution. Compared with the traditional single-point liquid surface monitoring method, this zoned monitoring can comprehensively grasp the spatial distribution of the liquid surface in the casting pool, avoiding product quality defects caused by local liquid surface anomalies. In particular, it has significant detection capabilities for liquid surface tilting and local fluctuations along the length of large casting pools.

[0062] In one specific embodiment, step S4 includes:

[0063] The remaining metal mass of the casting ladle is collected by a weighing sensor, the current insertion depth of the stopper rod is obtained by a servo motor encoder, and historical data of the liquid level height within a preset time period are collected.

[0064] The flow cross-sectional area of ​​the ladle outlet is calculated based on the current insertion depth of the stopper rod. Combined with the liquid level difference in the ladle corresponding to the remaining metal mass, the ladle outlet flow rate is calculated based on the continuity equation and Bernoulli equation.

[0065] Substitute the outlet flow rates of the casting ladle and the casting pool into the law of conservation of mass to calculate the rate of change of the liquid level at the current moment.

[0066] The rate of change of liquid level and historical data of liquid level are input into the Smith predictor algorithm for lag time compensation. The predicted liquid level sequence at each time in the future preset time domain is obtained through iterative calculation.

[0067] Specifically, when collecting the remaining metal mass of the casting ladle using a weighing sensor, the weighing sensor is installed on the support platform at the bottom of the casting ladle to measure the total weight of the casting ladle and its internal molten metal in real time. The system reads the current total weight value from the sensor and subtracts the empty weight of the casting ladle itself to obtain the remaining metal mass. When obtaining the current insertion depth of the stopper rod through the servo motor encoder, the encoder is directly connected to the servo motor shaft to record the rotation angle of the motor. The system calculates the current insertion depth of the stopper rod into the casting ladle based on the transmission ratio between the motor rotation angle and the linear displacement of the stopper rod. When collecting historical data of liquid level height within a preset time period, the system extracts the historical data of the past second from the liquid level height data continuously collected in step S2. Since the control cycle is 0.1 seconds, the past second contains 11 liquid level height data points. These data are stored in a loop buffer and arranged in chronological order. The remaining metal mass data is used to estimate the current liquid level height difference inside the casting ladle, the stopper rod insertion depth data is used to calculate the actual flow cross-sectional area of ​​the casting ladle outlet, and the historical liquid level height data is used to extract the trend characteristics of liquid level changes.

[0068] When calculating the flow cross-sectional area of ​​the ladle outlet based on the current insertion depth of the stopper rod, the ladle outlet is a circular opening with a fixed radius of 50 mm, and the stopper rod is a cylindrical structure with a maximum radius of 45 mm. When the stopper rod is fully inserted into the ladle outlet, an annular gap is formed between the stopper rod radius of 45 mm and the outlet radius of 50 mm, serving as a metal flow channel. When the stopper rod is partially withdrawn, its radius changes linearly with the insertion depth. The stopper rod radius is equal to 45 mm minus the insertion depth divided by 2. The flow cross-sectional area is equal to pi multiplied by the square of the outlet radius minus the square of the stopper rod radius. Considering the remaining metal mass and the corresponding liquid level difference within the ladle, the liquid level difference refers to the vertical distance from the molten metal surface in the ladle to the center of the outlet. This distance gradually decreases as metal flows out. The volume of the remaining metal is obtained by dividing the mass of the remaining metal by the density of the metal. Then, the current liquid level is calculated based on the geometric dimensions of the inner cavity of the casting ladle. The height difference between the liquid level and the center of the outlet is equal to the total height of the casting ladle minus the current liquid level. When calculating the outlet flow rate of the casting ladle based on the continuity equation and Bernoulli's equation, the continuity equation describes the mass conservation relationship when the fluid flows in the pipe, and the Bernoulli equation describes the conversion relationship between pressure energy, kinetic energy and potential energy of the fluid at different positions. The outlet flow rate of the casting ladle is equal to the square root of the flow coefficient multiplied by the cross-sectional area of ​​the flow, twice the gravitational acceleration, and the height difference. The flow coefficient reflects the energy loss in the actual flow and is taken as 0.62. The gravitational acceleration is taken as 9.8 m / s². The height difference is the vertical distance from the liquid level inside the casting ladle to the outlet.

[0069] When calculating the rate of change of liquid level at the current moment by substituting the ladle outlet flow rate and the casting pool outlet flow rate into the law of conservation of mass, the law of conservation of mass states that the rate of change of the mass of molten metal in the casting pool is equal to the inflow mass flow rate minus the outflow mass flow rate. The ladle outlet flow rate is the inflow flow rate of the casting pool. The casting pool outlet flow rate is determined by the billet drawing speed of the continuous casting process. The billet drawing speed refers to the linear velocity at which the solidified metal billet is pulled away from the casting pool. The casting pool outlet flow rate is equal to the billet drawing speed multiplied by the cross-sectional area of ​​the billet. The rate of change of liquid level is equal to the difference between the ladle outlet flow rate and the casting pool outlet flow rate, divided by the metal density and then divided by the horizontal cross-sectional area of ​​the casting pool. When the inflow flow rate is greater than the outflow flow rate, the difference is positive, indicating that the liquid level is rising. When the inflow flow rate is less than the outflow flow rate, the difference is negative, indicating that the liquid level is falling. When the inflow flow rate equals the outflow flow rate, the difference is zero, and the liquid level remains stable. The unit of the rate of change of liquid level is millimeters per second, which directly reflects the speed at which the liquid level rises and falls.

[0070] When the liquid level change rate and historical liquid level data are input into the Smith predictor algorithm for lag time compensation, and the predicted liquid level sequence for each moment within the preset future time domain is obtained through iterative calculation, the Smith predictor algorithm is a predictive control method specifically designed for systems with pure lag time. Pure lag time refers to the time delay between the issuance of the control command and the actual effect. The pure lag time of this system includes an image acquisition and processing delay of 0.15 seconds and a servo motor response delay of 0.25 seconds, which are added together to obtain a total lag time of 0.4 seconds. The core idea of ​​the Smith predictor is to establish a dynamic model of the system to predict the future state and compensate for the impact of lag time in advance. The system sets the prediction time domain to 3 seconds and the prediction step size to 0.1 seconds, requiring the prediction of liquid level height for 30 future moments. The liquid level height values ​​of each sampling moment in the past second are extracted from the historical liquid level height data. The difference in liquid level height between adjacent sampling points is calculated and divided by the sampling interval to obtain the historical liquid level change rate for each moment. These historical change rates are summed and divided by the number of historical data points to obtain the average historical change rate. The predictor uses the liquid level change at the current moment. The comprehensive change rate for prediction is obtained by weighting the rate of change with the average historical change rate. The weighting coefficients are determined based on the timeliness of the data, with the current change rate having a larger weight and the historical change rate having a smaller weight. The prediction calculation starts from the current liquid level height and progresses step by step in the future with a prediction step size of 0.1 seconds. The first step of predicting the liquid level height is equal to the current liquid level height plus the comprehensive change rate multiplied by the prediction step size minus the lag compensation amount. The lag compensation amount is equal to the comprehensive change rate multiplied by the total lag time of 0.4 seconds. The introduction of this compensation amount allows the prediction results to reflect the dynamic characteristics of the system in advance. The second step of prediction is based on the prediction results of the first step and continues to accumulate the comprehensive change rate multiplied by the step size increment. The calculation is iterated until 30 prediction steps are completed to obtain the predicted liquid level height sequence every 0.1 seconds in the next 3 seconds. This sequence depicts the trend curve of the liquid level height evolution over time under the current flow conditions. When the prediction sequence shows that the liquid level will continue to rise above the target height, the control system needs to reduce the insertion depth of the stopper rod in advance to reduce the inflow flow. When the prediction sequence shows that the liquid level will continue to fall below the target height, the control system needs to increase the insertion depth of the stopper rod in advance to increase the inflow flow.

[0071] In one specific embodiment, the rate of change of liquid level and historical liquid level data are input into the Smith predictor algorithm for lag time compensation, including:

[0072] The total system lag time is calculated by summing the image acquisition and processing delay and the servo motor response delay.

[0073] Extract the historical liquid level change rate sequence from the historical liquid level height data, calculate the average value of the historical liquid level change rate sequence, and obtain the average historical change rate;

[0074] Set the prediction time domain and prediction step size, take the current liquid level as the prediction starting point, and weight the liquid level change rate with the average historical change rate to obtain the single-step predicted change.

[0075] The predicted liquid level height is calculated step by step by iteratively accumulating the single-step predicted change and subtracting the lag compensation corresponding to the total system lag time. The predicted liquid level height is then summarized to obtain the predicted liquid level height sequence.

[0076] Specifically, when calculating the total system lag time by summing the image acquisition and processing delay and the servo motor response delay, the image acquisition and processing delay includes the camera exposure time, image data transmission time, and the computation time of the image processing algorithm. The exposure time required for the infrared thermal imaging camera and the visible light camera to simultaneously acquire one frame of image is determined according to the exposure parameters set in step S1. The time for image data to be transmitted from the camera to the industrial control computer via gigabit Ethernet depends on the image resolution and network bandwidth. The image processing algorithm includes binarization processing of the infrared temperature image, grayscale conversion and edge detection of the visible light image, coordinate registration and fusion of the dual-modal image, and liquid surface contour extraction. The total computation time required for these algorithms to be executed sequentially in the industrial control computer is averaged to 0.15 seconds based on actual test statistics. The servo motor response delay includes the communication delay of the control command being transmitted from the industrial control computer to the motor driver via the communication bus, the driver parsing the command, and the time for the servo motor to respond. The processing delay of the starting motor and the dynamic response time of the motor accelerating from a standstill to the set speed are considered. The communication bus uses the Controller Area Network protocol with a transmission rate of 1 megabit per second. The control command data packet size is 8 bytes, and the transmission time is extremely short. The sum of the driver processing delay and the motor dynamic response time is 0.25 seconds, which is obtained by comparing motor performance parameters and actual tests. The image acquisition and processing delay of 0.15 seconds is added to the servo motor response delay of 0.25 seconds to obtain the total system lag time of 0.4 seconds. This lag time indicates that there is a 0.4-second time delay between the start of image acquisition and the actual movement of the stopper rod to the target position. During the high-speed casting process, the liquid level continues to change according to the original trend during the 0.4-second delay. If the control algorithm does not consider this lag time and only adjusts according to the current deviation, the liquid level state will have changed by the time the stopper rod actually moves, resulting in a mismatch in the adjustment amount and causing liquid level overshoot or oscillation.

[0077] To extract the historical liquid level change rate sequence from historical liquid level data and calculate the average historical change rate, the historical liquid level data is stored in a circular buffer. Liquid level values ​​are recorded chronologically, with samples taken every 0.1 seconds within the past second. Since the sampling interval is 0.1 seconds, there are 11 liquid level data points within the past second. These 11 data points are arranged chronologically to form a historical data sequence. When extracting the historical liquid level change rate, the difference in liquid level height between two adjacent data points is calculated and divided by the sampling interval of 0.1 seconds to obtain the liquid level change rate for that time period. The first change rate equals the liquid level height of the second data point minus the liquid level height of the first data point, divided by 0.1 seconds. The rate of change is equal to the liquid level height at the third data point minus the liquid level height at the second data point, divided by 0.1 seconds. This process is repeated until the tenth rate of change is equal to the liquid level height at the eleventh data point minus the liquid level height at the tenth data point, divided by 0.1 seconds. This yields a series of 10 historical liquid level height change rate values. To calculate the average of this series, the 10 rate of change values ​​are summed and divided by 10 to obtain the average historical rate of change. The average historical rate of change reflects the overall trend of liquid level height change over the past second. A positive average historical rate of change indicates an overall upward trend in liquid level, a negative average historical rate of change indicates an overall downward trend in liquid level, and an average historical rate of change close to zero indicates a relatively stable liquid level.

[0078] When setting the prediction time domain and prediction step size, and using the current liquid level as the prediction starting point, and weighting the rate of change of liquid level with the average historical rate of change to obtain the single-step predicted change, the prediction time domain is set to 3 seconds, indicating that the liquid level change needs to be predicted within the next 3 seconds. The prediction step size is set to 0.1 seconds, meaning that the liquid level value at each prediction moment is calculated every 0.1 seconds. Therefore, from the current moment to the next 3 seconds, a total of 30 moments' liquid level heights need to be predicted to form a predicted liquid level height sequence. Using the current liquid level as the prediction starting point means that the first value of the prediction sequence is equal to the liquid level height actually measured in step S2 at the current moment. When weighting the rate of change of liquid level with the average historical rate of change, the current liquid level change rate is based on the step... The real-time liquid level change rate obtained from the flow rate calculation in S4 is the average historical change rate, which is the average change rate statistically obtained from the historical data of the past 1 second. The weighted combination uses a weighted average method to multiply the current change rate by a weight coefficient and add it to the average historical change rate multiplied by another weight coefficient. The weight coefficient of the current change rate is set to 0.7 to reflect the dominant role of the current flow state in future liquid level changes, and the weight coefficient of the average historical change rate is set to 0.3 to reflect the corrective role of historical trends in future changes. The comprehensive change rate obtained after weighted combination is used as the basic change amount for prediction calculation. The single-step predicted change amount is equal to the comprehensive change rate multiplied by the prediction step size of 0.1 seconds. This value represents the expected change amount of liquid level height in 0.1 seconds under the current flow conditions.

[0079] When iteratively accumulating the single-step predicted change and subtracting the lag compensation corresponding to the total system lag time to progressively calculate the predicted liquid level height values ​​at each future prediction step, the iterative accumulation refers to advancing one prediction step forward each time from the prediction starting point and accumulating the single-step predicted change. The first prediction time is the current time plus 0.1 seconds, and the first predicted liquid level height equals the current liquid level height plus the single-step predicted change. The second prediction time is the current time plus 0.2 seconds, and the second predicted liquid level height equals the first predicted liquid level height plus the single-step predicted change. The third step predicts the liquid level height, which equals the second step's predicted liquid level height plus the single-step predicted change. This process is iterated and accumulated until the thirtieth step, where the predicted time is the current time plus 3 seconds. The lag compensation is the core compensation mechanism of the Smith predictor algorithm. The lag compensation equals the overall rate of change multiplied by the total system lag time of 0.4 seconds. This compensation represents the cumulative amount of liquid level change at the current rate of change within the 0.4-second lag time. Subtracting the lag compensation from the predicted liquid level height calculated in each iteration achieves advance compensation for the system's lag characteristics. The predicted liquid level height after subtracting the lag compensation reflects the consideration of... The actual liquid level evolution trend after the control system response delay is calculated. After 30 iterative calculations, the predicted liquid level height values ​​at all predicted times are arranged in chronological order to form a predicted liquid level height sequence. This sequence depicts the liquid level height prediction curve every 0.1 seconds from the current time to the next 3 seconds. The control algorithm extracts the predicted liquid level height value for the next second from this sequence and compares it with the target liquid level height to calculate the prediction deviation. Based on the prediction deviation, a feedforward control quantity is generated and superimposed on the proportional-integral-derivative control quantity and the non-uniformity compensation control quantity to achieve advance adjustment. When the prediction sequence shows that the liquid level height for the next second will be... When the target value is exceeded, the feedforward control variable becomes negative, driving the stopper rod to increase the insertion depth and reduce the inflow rate. When the prediction sequence shows that the liquid level will be lower than the target value in the next second, the feedforward control variable becomes positive, driving the stopper rod to decrease the insertion depth and increase the inflow rate. The Smith predictor adjusts the control action before the liquid level actually deviates from the target value by predicting the future liquid level state and compensating for the system lag time in advance. Compared with the passive method of traditional feedback control that starts adjusting only after the liquid level deviation occurs, predictive control can advance the control response time by 0.4 seconds, effectively suppressing the liquid level overshoot and oscillation caused by lag.

[0080] In one specific embodiment, step S5 includes:

[0081] Calculate the liquid level deviation between the target liquid level and the current liquid level, and substitute the liquid level deviation and its integral and derivative terms into the PID controller to obtain the PID control quantity;

[0082] Multiply the liquid surface non-uniformity index by the non-uniformity compensation gain coefficient, and determine the sign direction based on the partition deviation with the largest absolute value among the partition deviations of each monitoring partition to obtain the compensation control quantity;

[0083] Extract the predicted liquid level height value for a future preset time from the predicted liquid level height sequence, calculate the prediction deviation between the target liquid level height and the predicted liquid level height value, and multiply the prediction deviation by the feedforward gain coefficient to obtain the feedforward control quantity;

[0084] The total control quantity is obtained by superimposing the PID control quantity, the compensation control quantity, and the feedforward control quantity. After the total control quantity is limited, it is converted by the control quantity-position conversion coefficient. Combined with the current insertion depth of the stopper rod, the target insertion depth of the stopper rod is calculated. The servo motor is driven to adjust the stopper rod to the target insertion depth. At the same time, protective gas is injected into the micro-pores on the surface of the stopper rod through the gas channel inside the stopper rod to form a gas film isolation layer.

[0085] Specifically, when calculating the liquid level deviation between the target liquid level and the current liquid level, and substituting the liquid level deviation, its integral term, and derivative term into the proportional-integral-derivative (PID) controller to obtain the PID control quantity, the target liquid level is a pre-set casting process requirement value, typically set to 160 mm for continuous casting of copper materials. The current liquid level is the real-time measurement value obtained in step S2 through dual-modal image fusion recognition. The liquid level deviation equals the target liquid level minus the current liquid level. A positive deviation indicates that the current liquid level is lower than the target height, requiring an increase in metal inflow; a negative deviation indicates that the current liquid level is higher than the target height, requiring a decrease in metal inflow. The PID controller is a classic closed-loop feedback control algorithm containing three control links. The proportional link generates a control action based on the magnitude of the current deviation. The proportional control quantity equals the proportional gain multiplied by the liquid level deviation. The proportional gain is determined in step S1 based on the metal... The viscosity coefficient and density of the material are pre-calculated. The integral stage generates control based on the cumulative effect of the deviation to eliminate steady-state error. The integral term is equal to the sum of all historical deviation values ​​multiplied by the sampling time from the start of casting to the current moment. The integral control quantity is equal to the proportional gain divided by the integral time constant and then multiplied by the integral term. The integral time constant is pre-set in step S1 according to the frequency range of liquid surface fluctuation. The derivative stage generates control based on the rate of change of the deviation to improve the system response speed. The derivative term is equal to the liquid surface height deviation at the current moment minus the liquid surface height deviation at the previous moment divided by the sampling time. The derivative control quantity is equal to the proportional gain multiplied by the derivative time constant and then multiplied by the derivative term. The derivative time constant is one-quarter of the integral time constant. The proportional control quantity, integral control quantity, and derivative control quantity are added together to obtain the proportional-integral-derivative control quantity. This control quantity reflects the control strength of feedback adjustment based on the current liquid surface deviation.

[0086] When multiplying the liquid surface non-uniformity index by the non-uniformity compensation gain coefficient and determining the sign direction based on the partition deviation with the largest absolute value among the partition deviations of each monitoring zone to obtain the compensation control quantity, the liquid surface non-uniformity index is the value obtained in step S3 by calculating the root mean square of the partition deviations of the three monitoring zones. This index reflects the degree of fluctuation of the liquid surface in spatial distribution. The non-uniformity compensation gain coefficient is an empirical parameter with a value of 15, determined based on the geometry of the casting pool and the metal flow characteristics. Multiplying the liquid surface non-uniformity index by the non-uniformity compensation gain coefficient yields the amplitude of the compensation quantity. The partition deviations of each monitoring zone include three values: inlet zone deviation, center zone deviation, and outlet zone deviation. The absolute values ​​of these three deviations are compared to find the partition deviation with the largest absolute value. The sign direction of this largest partition deviation determines the sign of the compensation control quantity. When the maximum partition deviation is positive, it indicates that the liquid level in that area is significantly higher than the overall average level, and the metal inflow in that area needs to be reduced. The compensation control quantity takes a negative value to drive the stopper rod to increase the insertion depth and reduce the overall inflow flow. When the maximum partition deviation is negative, it indicates that the liquid level in that area is significantly lower than the overall average level, and the metal inflow in that area needs to be increased. The compensation control quantity takes a positive value to drive the stopper rod to decrease the insertion depth and increase the overall inflow flow. The compensation control quantity is equal to the non-uniformity compensation gain coefficient multiplied by the liquid surface non-uniformity index and then multiplied by the sign function of the maximum partition deviation. The sign function outputs a positive 1 when the input is positive and a negative 1 when the input is negative. The compensation control quantity is only activated when the liquid surface non-uniformity index is greater than 2 mm. When the non-uniformity index is less than or equal to 2 mm, the liquid surface distribution is determined to be uniform and the compensation control quantity is set to zero.

[0087] When extracting the predicted liquid level height value at a future preset time from the predicted liquid level height sequence, calculating the prediction deviation between the target liquid level height and the predicted liquid level height value, and multiplying the prediction deviation by the feedforward gain coefficient to obtain the feedforward control quantity, the predicted liquid level height sequence is the predicted liquid level height value calculated every 0.1 seconds within the next 3 seconds using the Smith predictor algorithm in step S4. The future preset time is usually chosen as 1 second in the future, i.e., the current time plus 1 second. The predicted liquid level height value corresponding to that time is extracted from the prediction sequence. The prediction deviation is equal to the target liquid level height minus the predicted liquid level height value. The prediction deviation reflects the liquid level height in the next 1 second relative to the target value if the current control strategy is not changed. The degree of deviation is determined by the prediction deviation. A positive prediction deviation indicates that the future liquid level will be lower than the target height, while a negative prediction deviation indicates that the future liquid level will be higher than the target height. The feedforward gain coefficient is a control parameter with a value of 20, determined based on the system's dynamic response characteristics and the prediction time domain. The feedforward control quantity is equal to the feedforward gain coefficient multiplied by the prediction deviation. This control quantity embodies the control idea of ​​making advance adjustments based on the future liquid level trend. When the prediction deviation is positive, the feedforward control quantity is positive, driving the stopper rod to reduce the insertion depth and increase the metal inflow in advance to prevent the liquid level from dropping. When the prediction deviation is negative, the feedforward control quantity is negative, driving the stopper rod to increase the insertion depth and reduce the metal inflow in advance to prevent the liquid level from rising.

[0088] The total control quantity is obtained by superimposing the proportional-integral-derivative (PID) control quantity, the compensation control quantity, and the feedforward control quantity. After limiting the total control quantity, the target insertion depth of the stopper is calculated by combining the position conversion coefficient of the control quantity with the current insertion depth of the stopper. The servo motor is then driven to adjust the stopper to the target insertion depth. Simultaneously, protective gas is injected into the micro-pores on the surface of the stopper through the internal gas channel to form a gas film isolation layer. At this point, the total control quantity is equal to the PID control quantity plus the compensation control quantity plus the feedforward control quantity. This total control quantity integrates the correction effect of feedback control, the balancing effect of non-uniform compensation, and the advance effect of predictive feedforward. Limiting the total control quantity constrains its numerical range to prevent excessive control quantity from causing excessive stopper movement and violent fluctuations in the liquid surface. The range is set based on the mechanical structure of the stopper rod and the dynamic response characteristics of the liquid surface. When the absolute value of the total control quantity exceeds the limit threshold, it is truncated to the threshold boundary. The control quantity position conversion coefficient is a mapping parameter that converts the dimensionless control quantity value into the stopper rod position increment, with a value of 8 per millimeter. The stopper rod position increment is equal to the total control quantity divided by the control quantity position conversion coefficient. The target insertion depth of the stopper rod is equal to the current insertion depth of the stopper rod plus the position increment. The current insertion depth of the stopper rod is obtained in real time through feedback from the servo motor encoder. The calculated target insertion depth of the stopper rod is sent to the servo motor driver through the controller area network communication bus. The driver controls the servo motor to move from the current position to the target position according to the trapezoidal speed curve, which includes an acceleration stage, a constant speed stage, and a deceleration stage. The acceleration is set to 500 mm / s², and the maximum speed is set to 200 mm / s. During the movement of the stopper rod, the compressed gas supply system inside the stopper rod is activated simultaneously. Argon gas is delivered through the axial internal channel of the stopper rod to 12 micro-holes evenly distributed along the circumference on the surface of the stopper rod. The diameter of the micro-holes is 0.8 mm, and they are spaced at 30-degree intervals. The argon gas flow rate is automatically adjusted according to the metal material temperature parameters configured in step S1. For copper material with a casting temperature of 1300 degrees Celsius, the argon gas flow rate is set to the base flow rate multiplied by the casting temperature divided by the base temperature to the power of 0.5. After the argon gas is ejected from the micro-holes, it forms a gas protective film with a thickness of approximately 0.5 to 1 mm on the surface of the stopper rod. The dynamic pressure of the gas film is equal to 0.5 multiplied by the argon gas density multiplied by the square of the gas injection velocity. The gas injection velocity is rooted by the... The calculation is based on the argon flow rate divided by the total pore area. The dynamic pressure scouring effect generated by the gas film continuously removes oxides and solidified metal particles adhering to the surface of the stopper rod. The isolation effect of the gas film blocks direct contact between the molten copper liquid and the stopper rod matrix material, preventing chemical corrosion and adhesion. The surface temperature of the stopper rod is maintained at 950 to 1050 degrees Celsius, which is 250 to 350 degrees Celsius lower than the temperature of the molten copper liquid, avoiding melting damage to the surface material of the stopper rod. The gas film protection technology extends the continuous working life of the stopper rod from 8 hours in the traditional technology to 72 hours. Throughout the entire service cycle, the position control accuracy deviation of the stopper rod is kept within ±0.5 mm. After the stopper rod completes the position adjustment, the controller continues to execute the cyclic control process from step S2 to step S5 to form a continuous closed-loop adaptive control loop.

[0089] The above describes the molten metal casting method based on adaptive liquid level adjustment in the embodiments of this application. The following describes the molten metal casting system based on adaptive liquid level adjustment in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the molten metal casting system based on adaptive liquid level adjustment in this application includes:

[0090] The adjustment module is used to adjust the image acquisition parameters and controller parameters according to the physical property parameters of the metal material to be cast, so as to obtain a material adaptation parameter set;

[0091] The weighting module is used to acquire infrared temperature images and visible light images of the liquid surface in the casting pool. It performs weighted fusion of the high-temperature area mask in the infrared temperature image and the edge features in the visible light image to extract the liquid surface contour coordinates and calculate the current liquid surface height.

[0092] The partitioning module is used to divide the liquid surface contour coordinates into multiple monitoring zones according to spatial location, calculate the partition liquid surface height of each monitoring zone, and calculate the partition deviation of the partition liquid surface height of each monitoring zone relative to the current liquid surface height to obtain the liquid surface non-uniformity index.

[0093] The calculation module is used to combine the remaining metal mass of the casting ladle, the current insertion depth of the stopper rod, and the current liquid level height, and to compensate for system time lag and calculate the predicted liquid level height sequence using the Smith predictor algorithm;

[0094] The moving module is used to calculate the PID control quantity based on the deviation between the target liquid level height and the current liquid level height, superimpose the compensation control quantity generated by the liquid level non-uniformity index and the feedforward control quantity generated by the predicted liquid level height sequence, obtain the target insertion depth of the stopper rod, drive the stopper rod to move to the target position, and simultaneously spray protective gas onto the surface of the stopper rod to form a gas film isolation layer.

[0095] above Figure 2 The molten metal casting system based on adaptive liquid level adjustment in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The molten metal casting equipment based on adaptive liquid level adjustment in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0096] Reference Figure 3 This invention also provides a molten metal casting device based on adaptive liquid level adjustment. This molten metal casting device can be a server, and its internal structure can be as follows: Figure 3As shown, the molten metal casting equipment based on adaptive liquid level adjustment includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the molten metal casting equipment based on adaptive liquid level adjustment includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the molten metal casting equipment based on adaptive liquid level adjustment stores the data corresponding to this embodiment. The network interface of the molten metal casting equipment based on adaptive liquid level adjustment is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0097] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the molten metal casting equipment based on adaptive liquid level adjustment applied thereto.

[0098] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the molten metal casting method based on adaptive adjustment of liquid level height.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0100] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a molten metal casting device (which can be a personal computer, server, or network device, etc.) based on adaptive adjustment of liquid level to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A molten metal casting method based on self-adapting regulation of a liquid level height, characterized by, The method comprises: Step S1: adjusting image acquisition parameters and controller parameters according to physical parameters of the metal material to be cast to obtain a material adaptation parameter set; Step S2: acquiring an infrared temperature image and a visible light image of the liquid surface of the casting pool, performing weighted fusion of a high-temperature area mask in the infrared temperature image and edge features in the visible light image, extracting liquid surface contour coordinates and calculating a current liquid surface height; Step S3: dividing the liquid surface contour coordinates into a plurality of monitoring sub-zones according to spatial positions, respectively calculating sub-zone liquid surface heights of the monitoring sub-zones, and calculating sub-zone deviations of the sub-zone liquid surface heights of the monitoring sub-zones relative to the current liquid surface height to obtain a liquid surface unevenness index; Step S4: combining a residual metal mass of the casting ladle, a current insertion depth of the stopper, and the current liquid surface height, compensating for system time lag through a Smith predictor algorithm and calculating a predicted liquid surface height sequence; Step S5: calculating a PID control amount according to a deviation between a target liquid surface height and the current liquid surface height, superimposing a compensation control amount generated by the liquid surface unevenness index and a feedforward control amount generated by the predicted liquid surface height sequence, obtaining a target insertion depth of the stopper, and driving the stopper to move to a target position while spraying a protective gas on the surface of the stopper to form an air film isolation layer.

2. The molten metal casting method based on adaptive adjustment of the liquid level height according to claim 1, characterized in that, The step S1 comprises: retrieving surface reflectivity, infrared emissivity, flow viscosity coefficient, and density parameters corresponding to the metal material to be cast from a metal material physical characteristic database; setting a temperature measurement compensation coefficient of an infrared thermal imaging camera according to the infrared emissivity, and calculating an exposure time parameter of a visible light camera according to the surface reflectivity; calculating a proportional gain parameter of a PID controller based on the flow viscosity coefficient and the density parameters; summarizing the temperature measurement compensation coefficient, the exposure time parameter, and the proportional gain parameter to obtain the material adaptation parameter set.

3. The molten metal pouring method based on self-adapting adjustment of the liquid level height according to claim 1, characterized in that, The step S2 comprises: synchronously acquiring the infrared temperature image and the visible light image of the liquid surface of the casting pool through an infrared thermal imaging camera and a visible light camera; setting a temperature threshold for the infrared temperature image to perform binaryzation processing, generating the high-temperature area mask for identifying the molten metal liquid surface area, and simultaneously performing gray scale conversion on the visible light image and applying a Canny edge detection algorithm to extract a liquid surface boundary to obtain the edge features; performing coordinate registration on the high-temperature area mask and the edge features, performing weighted superposition processing according to a preset weight coefficient to obtain a fusion image; extracting a highest point coordinate set of the liquid surface contour from the fusion image, calculating an average value of the longitudinal coordinates of each point in the highest point coordinate set, and combining a pre-labeled pixel-actual height conversion relationship to obtain the current liquid surface height.

4. The molten metal pouring method based on self-adapting adjustment of the liquid level height according to claim 1, characterized in that, The step S3 comprises: dividing the casting pool into three monitoring sub-zones, namely an inlet zone, a center zone, and an outlet zone along a length direction, and distributing the liquid surface contour coordinates to the corresponding monitoring sub-zones according to the horizontal coordinate range of each point in the liquid surface contour coordinates; respectively calculating the average value of the longitudinal coordinates of the liquid surface contour coordinate points in each monitoring sub-zone to obtain the sub-zone liquid surface heights of the monitoring sub-zones; Calculate the difference between the partition liquid level and the current liquid level of each monitoring partition to obtain the partition deviation of each monitoring partition; Square sum the partition deviations of each monitoring partition and take the root mean square to obtain the liquid level unevenness index.

5. The molten metal pouring method based on self-adapting adjustment of the liquid level height according to claim 1, characterized in that, The step S4 comprises: Collecting the remaining metal mass of the casting ladle through a weighing sensor, obtaining the current insertion depth of the stopper through a servo motor encoder, and collecting liquid level historical data in a past preset time period; Calculating the flow area of the casting ladle outlet according to the current insertion depth of the stopper, combining the liquid level difference corresponding to the remaining metal mass in the casting ladle, and calculating the casting ladle outlet flow rate based on the continuity equation and Bernoulli equation; Substituting the casting ladle outlet flow rate and the casting pool outlet flow rate into the law of conservation of mass to calculate the liquid level change rate at the current time; Inputting the liquid level change rate and the liquid level historical data into the Smith predictor algorithm for lag time compensation, and calculating the predicted liquid level sequence at each time in a future preset time domain through iteration.

6. The molten metal pouring method based on the self-adaptive adjustment of the liquid level height according to claim 5, characterized in that, The inputting the liquid level change rate and the liquid level historical data into the Smith predictor algorithm for lag time compensation comprises: Cumulatively calculating the system total lag time according to image acquisition processing delay and servo motor response delay; Extracting a historical liquid level change rate sequence from the liquid level historical data, calculating the average value of the historical liquid level change rate sequence to obtain an average historical change rate; Setting a prediction time domain and a prediction step, taking the current liquid level as a prediction starting point, and combining the liquid level change rate and the average historical change rate to obtain a single-step prediction change amount; Iteratively accumulating the single-step prediction change amount and subtracting a lag compensation amount corresponding to the system total lag time to gradually calculate the predicted liquid level value at each prediction step in the future, and the predicted liquid level sequence is obtained by summarizing.

7. The molten metal pouring method based on adaptive adjustment of the liquid level height according to claim 1, characterized in that, The step S5 comprises: Calculating the liquid level deviation between the target liquid level and the current liquid level, inputting the liquid level deviation, its integral term and differential term into a PID controller to obtain the PID control amount; Multiplying the liquid level unevenness index by an unevenness compensation gain coefficient, and determining the sign direction according to the partition deviation with the largest absolute value among the partition deviations of the monitoring partitions to obtain the compensation control amount; Extracting the predicted liquid level value at a future preset time from the predicted liquid level sequence, calculating the prediction deviation between the target liquid level and the predicted liquid level value, multiplying the prediction deviation by a feedforward gain coefficient to obtain the feedforward control amount; and Multiplying the liquid level unevenness index by an unevenness compensation gain coefficient, and determining the sign direction according to the partition deviation with the largest absolute value among the partition deviations of the monitoring partitions to obtain the compensation control amount; Extracting the predicted liquid level value at a future preset time from the predicted liquid level sequence, calculating the prediction deviation between the target liquid level and the predicted liquid level value, multiplying the prediction deviation by a feedforward gain coefficient to obtain the feedforward control amount; and The PID control amount, the compensation control amount and the feedforward control amount are superimposed to obtain a total control amount, the total control amount is subjected to amplitude limiting processing and then converted through a control amount-position conversion coefficient, the target insertion depth of the stopper is calculated in combination with the current insertion depth of the stopper, a servo motor is driven to adjust the stopper to the target insertion depth of the stopper, and the protective gas is sprayed to the micro pores on the surface of the stopper through the gas channel in the stopper to form the gas film isolation layer.

8. A molten metal pouring system based on adaptive regulation of liquid level height, characterized by, The molten metal casting method based on liquid level adaptive adjustment comprises a molten metal casting system based on liquid level adaptive adjustment as claimed in any one of claims 1-7, and the molten metal casting system based on liquid level adaptive adjustment comprises: An adjusting module is configured to adjust image acquisition parameters and controller parameters according to physical parameters of the metal material to be cast, so as to obtain a material adaptation parameter set; A weighting module is configured to acquire an infrared temperature image and a visible light image of the liquid surface of the casting pool, to perform weighted fusion of a high-temperature region mask in the infrared temperature image and an edge feature in the visible light image, to extract liquid surface contour coordinates and to calculate a current liquid level; A division module is configured to divide the liquid surface contour coordinates into a plurality of monitoring sub-zones according to spatial positions, to calculate sub-zone liquid levels of the monitoring sub-zones respectively, and to calculate sub-zone deviations of the sub-zone liquid levels of the monitoring sub-zones relative to the current liquid level, so as to obtain a liquid surface unevenness index; A calculation module is configured to combine a residual metal mass of the casting ladle, a current insertion depth of a stopper and the current liquid level, to compensate for system time lag through a Smith predictor algorithm and to calculate a predicted liquid level sequence; A moving module is configured to calculate a PID control amount according to a deviation between a target liquid level and the current liquid level, to superimpose a compensation control amount generated by the liquid surface unevenness index and a feedforward control amount generated by the predicted liquid level sequence, to obtain a target insertion depth of the stopper and to drive the stopper to move to a target position, and to spray a protective gas to the surface of the stopper to form a gas film isolation layer.

9. A molten metal pouring apparatus based on self-adapting regulation of a liquid level, characterized by The computer program is stored in the memory and can be run on the processor, and the processor implements the molten metal casting method based on liquid level adaptive adjustment as claimed in any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when being run on the processor, causes the processor to execute the molten metal casting method based on liquid level adaptive adjustment as claimed in any one of claims 1-7.

Images