A method and system for synergistically controlling balling particle size

Abstract

This invention relates to the field of iron ore pelletizing control technology, and discloses a method and system for coordinated control of pelletizing particle size. The invention acquires images of the mother pellet area of ​​the pelletizing disc and extracts the pellet-to-powder ratio as a feedforward signal, while simultaneously acquiring images of the pelletizing area and calculating the average particle size as a feedback signal. Based on current operating parameters, a target pellet-to-powder ratio benchmark for dripping control is determined. The dripping rate in the mother pellet area is adjusted according to the deviation between the feedforward signal and the target benchmark, achieving feedforward control. The flow rate of the mist in the pelletizing area is adjusted according to the deviation between the feedback signal and the preset particle size target, achieving feedback diameter correction. When the feedback signal deviates from the preset target for a set time and the mist flow rate reaches its limit, a cascaded correction amount is calculated based on the deviation state, and the target pellet-to-powder ratio benchmark is dynamically corrected. This invention, through a two-stage coordinated feedforward and feedback mechanism and a cascaded correction mechanism, effectively solves the problems of frequent overshoot and oscillation in existing pelletizing control, significantly improving the green pellet particle size qualification rate and moisture stability.

Description

Technical Field

[0001] This invention relates to the field of iron ore pelletizing control technology, specifically to a method and system for coordinated control of pellet size. Background Technology

[0002] Iron ore pelletizing is a key process in iron and steel metallurgy. It involves mechanically rolling and using moisture to form green pellets from powdered raw materials, providing high-quality furnace feed for subsequent sintering or direct reduction. Precise particle size control is a crucial physical parameter determining the smooth operation of subsequent roasting processes and the quality of the finished pellets. Particle sizes that are too small or too large will disrupt the thermal equilibrium, significantly reducing the physical strength and metallurgical properties of the finished iron ore pellets.

[0003] Currently, mainstream intelligent pelletizing solutions in the industry generally adopt feedback control based on single-point vision, deploying visual detection only in the pelletizing area to monitor the average particle size and distribution. A video particle size analysis system is used to analyze the particle size of green pellets on the green pellet conveyor belt, controlling the dripping and atomized water volume based on the proportional relationship of green pellet size. Alternatively, a particle size detection device can be used to detect the pelletizing pass rate, adjusting the water addition or pelletizing disc rotation speed based on parameters such as changes in the pass rate and the slope of the small pellet ratio curve.

[0004] It is evident that the above solution can only rely on the detection results of the ball exit area for feedback adjustment. However, there is a pure lag of several minutes from water adjustment in the mother ball area to the particle size response in the ball exit area. By the time the system detects that the ball exit particle size deviates from the target, the abnormal working condition has already formed. This lag leads to frequent overshoot and oscillation in the control process, making it impossible to achieve stable and accurate particle size control. Summary of the Invention

[0005] This invention provides a method and system for coordinated control of pellet size, which solves the problem of frequent overshoot and oscillation in the control process of iron ore pelletizing, making it impossible to achieve stable and precise particle size control.

[0006] In a first aspect, the present invention provides a method for synergistic control of pellet size, the method comprising: Images of the mother ball area in the pelleting disc are acquired, and the ball-to-powder ratio in the mother ball area is extracted as a feedforward signal; Images of the pelletizing area of ​​the pelletizing disc are acquired, and the average particle size of the pelletizing area is calculated as a feedback signal; Based on the current operating parameters of the pelleting disc, the target pellet powder ratio benchmark for drip control is determined; The amount of water dripping into the mother ball area is adjusted based on the deviation between the feedforward signal and the target ball powder ratio benchmark. Based on the deviation between the feedback signal and the preset particle size target, the flow rate of the mist in the ball outlet area is adjusted; When the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, a cascade correction amount is calculated based on the deviation state of the feedback signal to dynamically correct the target sphere powder ratio benchmark.

[0007] In one optional implementation, the extraction of the ball-to-powder ratio in the mother ball region as a feedforward signal includes: The image of the mother ball area is input into a lightweight semantic segmentation network to identify the green ball area and powder area in the image; the lightweight semantic segmentation network is trained in advance using historical sample images of the balling tray and manual annotation results; The ratio between the pixel area of ​​the green ball region and the pixel area of ​​the powder region is calculated to obtain the ball-to-powder ratio as a feedforward signal.

[0008] In one optional implementation, calculating the average particle size of the spherical region as a feedback signal includes: Target detection is performed on the image of the ball-producing area of ​​the ball-making disc to identify individual unadhesive raw balls in the image, and the outer contour of each raw ball is extracted. Based on the circumscribed contour of each green ball, calculate the equivalent circle diameter of each green ball; Calculate the arithmetic mean of the equivalent circular diameters of all identified green pellets to obtain the average particle size of the pelletizing area and use it as a feedback signal.

[0009] In one optional implementation, the current operating parameters include the real-time feed rate and the pelletizing disc rotation speed; The determination of the target powder ratio for drip control based on the current operating parameters of the pelleting disc includes: The real-time feed rate and the pelletizing disc rotation speed are input into a pre-built online optimization model. The online optimization model outputs the optimal pellet-to-powder ratio that matches the current working conditions, which serves as the benchmark for the target pellet-to-powder ratio. The online optimization model is trained based on historical working condition data and ball-to-powder ratio data, and its parameters are updated online according to real-time production data.

[0010] In one optional implementation, adjusting the dripping amount in the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio reference includes: The deviation between the feedforward signal and the target ball powder ratio benchmark is input to the proportional-integral (PI) controller; The proportional-integral (PI) controller calculates the drip rate control value and adjusts the opening of the drip solenoid valve according to the drip rate control value.

[0011] In one optional implementation, adjusting the water flow rate in the ball-ejection zone based on the deviation between the feedback signal and the preset particle size target includes: The deviation between the feedback signal and the preset particle size target is input to the proportional-integral-derivative (PID) controller. The proportional-integral-derivative (PID) controller calculates the mist flow rate control value and adjusts the opening of the mist regulating valve according to the mist flow rate control value.

[0012] In one optional implementation, the calculation of the cascade correction amount based on the deviation state of the feedback signal includes: Determine the direction and degree of deviation of the average particle size from the preset particle size target; According to the deviation direction, the newly added correction value determined based on the current deviation degree is added to the historical cascade correction amount of the previous moment to obtain the cascade correction amount of the current moment.

[0013] In one optional implementation, the dynamic correction of the target ball powder ratio benchmark includes: The cascaded correction amount is linearly superimposed with the target ball powder ratio benchmark at the current moment to generate a corrected updated ball powder ratio benchmark, which is then used to replace the target ball powder ratio benchmark for drip volume adjustment.

[0014] Secondly, the present invention provides a pelletizing particle size synergistic control system, the system comprising: The feedforward acquisition module is used to acquire images of the mother ball area of ​​the ball-making disc and extract the ball-to-powder ratio of the mother ball area as the feedforward signal. The feedback acquisition module is used to acquire images of the pelletizing area of ​​the pelletizing disc and calculate the average particle size of the pelletizing area as a feedback signal. The benchmark determination module is used to determine the target powder ratio benchmark for drip control based on the current operating parameters of the pelleting disc; The feedforward control module is used to adjust the amount of water dripping in the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio benchmark. The feedback control module is used to adjust the mist flow rate in the ball outlet area based on the deviation between the feedback signal and the preset particle size target. The cascade correction module is used to calculate the cascade correction amount based on the deviation state of the feedback signal when the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, so as to dynamically correct the target sphere powder ratio benchmark.

[0015] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform a pelletizing particle size collaborative control method described in the first aspect or any corresponding embodiment.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute a pelletizing particle size coordination control method as described in the first aspect or any corresponding embodiment thereof.

[0017] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute a pelletizing particle size coordination control method described in the first aspect or any corresponding embodiment.

[0018] This invention overcomes the visual perception blind spot in the mother pellet area by acquiring images of the nucleation zone and extracting the pellet-to-powder ratio as a feedforward signal, thus achieving quantitative monitoring of the initial nucleation state and solving the problem of lacking a feedforward signal at its source. By acquiring images of the nucleation zone and calculating the average particle size as a feedback signal, a two-stage closed-loop control system is constructed, where the feedforward determines the nucleation basis and the feedback corrects the final quality. This effectively overcomes the problem of minute-level pure lag in traditional single-loop feedback control, avoiding frequent overshoot and oscillations. Furthermore, through a cascaded correction mechanism, when the feedback signal continuously deviates and the mist adjustment becomes saturated, the feedforward target benchmark is dynamically corrected, achieving coordinated control of dripping and misting, solving the problem of strong coupling between the two, and fundamentally ensuring that the green pellet particle size meets the standards. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of a pelletizing particle size synergistic control method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the second process of a pelletizing particle size synergistic control method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a pelletizing and water control strategy according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the GhostNetV2 and DeepLabv3+ converged network architecture according to an embodiment of the present invention; Figure 6 This is a structural block diagram of a pelletizing particle size collaborative control system according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

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

[0022] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0023] As an optional application scenario of this invention, such as Figure 1 As shown, the pelletizing particle size collaborative control system of the present invention is specifically applied to the production site of a disc pelletizing machine. The physical structure of the system includes: a PLC control cabinet 101, a belt scale 102, an industrial camera A103, a drip valve 10, a mist valve 105, an industrial camera B106, and a disc pelletizing machine 107. The disc surface of the disc pelletizing machine 107 is divided into a mother ball area, a long ball area, and a ball outlet area along the material movement direction.

[0024] Specifically, belt scale 102 is located upstream of the pelletizing machine to detect the amount of raw material fed into the disc in real time and transmit the feeding signal to PLC control cabinet 101. Industrial camera A103 is installed above the mother pellet area to collect real-time images of the pellet powder mixing state in the mother pellet area; industrial camera B106 is installed above the pellet outlet area to collect real-time images of the finished green pellet area. Drip valve 104 is installed on the water supply pipe corresponding to the mother pellet area, and mist valve 105 is installed on the water supply pipe corresponding to the long pellet area. The opening degree of both is independently adjusted by PLC control cabinet 101 based on visual detection results.

[0025] In actual operation, industrial camera A103 captures images of the mother ball area and uploads them to PLC control cabinet 101. PLC control cabinet 101 extracts the ball-to-powder ratio as a feedforward signal through its built-in image processing module. Combined with the feed rate feedback from belt scale 102, it dynamically calculates the target ball-to-powder ratio benchmark, and then controls the opening of drip valve 104 to adjust the drip rate. Industrial camera B106 captures images of the ball area and uploads them. PLC control cabinet 101 calculates the average particle size as a feedback signal, and adjusts the opening of mist valve 105 accordingly to control the mist flow. When the opening of mist valve 105 reaches its limit and the particle size still does not meet the standard, PLC control cabinet 101 automatically triggers a cascaded correction algorithm to adjust the target benchmark for drip control in reverse, achieving coordinated control of drip and mist flow.

[0026] It should be noted that, Figure 1 This is merely an exemplary application scenario of the present invention. In actual deployment, the number and installation position of the industrial camera B106 and control valve can be adaptively adjusted according to the size of the pelletizing plate and the on-site working conditions. All such adjustments fall within the protection scope of the present invention.

[0027] According to an embodiment of the present invention, a method for coordinated control of pellet size is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] This embodiment provides a method for coordinated control of pellet size, which can be used in the aforementioned PLC control cabinet 101. Figure 2 This is a first flowchart of a pelletizing particle size synergistic control method according to an embodiment of the present invention, as follows: Figure 2 As shown, the process includes the following steps: Step S201: Acquire an image of the mother ball area of ​​the ball-making disc and extract the ball-to-powder ratio of the mother ball area as a feedforward signal.

[0029] Furthermore, the pelletizing disc is the core equipment for iron ore pelletizing. Its surface is divided into a mother pellet area, a pellet growing area, and a pellet exit area according to the green pellet growth stage. The mother pellet area is where green pellets are initially formed. Here, the material is in the nucleation stage, and the pellet-to-powder ratio directly affects the subsequent growth capacity of the green pellets. Industrial camera A is installed above the mother pellet area to acquire images of this area in real time. A lightweight semantic segmentation network (pre-trained with historical samples) identifies the green pellet area (formed pellets) and the powder area (un-pelleted fine material) in the image, calculates the ratio of their pixel areas, and obtains the pellet-to-powder ratio parameter. This pellet-to-powder ratio serves as a feedforward signal, reflecting the pelleting status in the mother pellet area in advance. If the pellet-to-powder ratio is too low, it indicates that there is too much powder and the pelleting speed is slow; if it is too high, it indicates that there are too many green pellets and insufficient powder. The function of the feedforward signal is to allow the system to predict the pelleting quality before the green pellets grow, providing a basis for subsequent drip rate adjustment and avoiding the lag in end-feedback adjustment.

[0030] Step S202: Acquire an image of the pelletizing area of ​​the pelletizing disc and calculate the average particle size of the pelletizing area as a feedback signal.

[0031] Furthermore, the pelleting zone is where the green pellets complete their growth, and the particle size of these green pellets is a core indicator of the final product quality. Industrial camera B is installed above the pelleting zone to capture images of this area in real time. A target detection algorithm identifies individual green pellets without adhesion in the image (eliminating stacking interference), extracts the circumscribed contour of each green pellet, and calculates its equivalent circle diameter (equivalent to the diameter of a circle for irregular green pellet contours). The arithmetic mean of all identified equivalent circle diameters is taken to obtain the average particle size of the pelleting zone. This average particle size serves as a feedback signal, directly reflecting whether the final green pellet particle size meets the standard. If the average particle size is smaller than the preset target, it indicates that the green pellets are too small; if it is larger than the target, it indicates that the green pellets are too large. The feedback signal verifies the front-end pelleting effect, provides a basis for adjusting the end-end mist flow rate, and ensures that the final product meets process requirements.

[0032] Step S203: Based on the current operating parameters of the pelleting disc, determine the target pellet powder ratio benchmark for drip control.

[0033] Furthermore, the pelleting effect during the pelleting process is affected by various operating parameters, primarily including the real-time feed rate (the amount of material added to the pelleting disc per unit time) and the pelleting disc rotation speed (the speed at which the disc rotates). A larger feed rate requires more moisture in the mother pellet zone to promote pelleting; a faster rotation speed results in a shorter residence time of the material in the mother pellet zone, necessitating more precise moisture control. The system inputs the real-time feed rate and rotation speed into a pre-built online optimization model (trained based on historical operating data and pellet-to-powder ratio data, and capable of updating parameters according to real-time production data). The model outputs the optimal pellet-to-powder ratio matching the current operating conditions, serving as the benchmark for the target pellet-to-powder ratio for drip control. This benchmark represents the expected value for drip adjustment in the mother pellet zone. For example, when the feed rate increases, the model outputs a higher target pellet-to-powder ratio, requiring more green pellets and less powder in the mother pellet zone to ensure sufficient nuclei for subsequent growth.

[0034] Step S204: Adjust the amount of water dripping into the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio benchmark.

[0035] Furthermore, the deviation between the feedforward signal (actual ball-to-powder ratio) and the target ball-to-powder ratio benchmark (expected ball-to-powder ratio) reflects the gap between the ball-forming state in the mother ball area and the expectation. For example, if the actual ball-to-powder ratio is 1.2 (high proportion of unformed balls), and the target benchmark is 1.5 (expected higher proportion of unformed balls), then the deviation is -0.3 (actual is lower than expected). The system inputs this deviation into a proportional-integral (PI) controller. The controller responds quickly to the deviation through the proportional element (adjusting significantly if the deviation is large), and eliminates steady-state error through the integral element (ensuring the final target is achieved). The controller outputs a drip rate control value to adjust the opening of the drip solenoid valve above the mother ball area. If the deviation is negative (actual ball-to-powder ratio is low), the valve opening is increased to increase the drip rate and promote ball-forming; if the deviation is positive (actual ball-to-powder ratio is high), the valve opening is decreased to reduce the drip rate and prevent the unformed balls from becoming too wet. The core of this step is feedforward adjustment, which controls the ball-forming quality from the source by intervening in the moisture of the mother ball area in advance.

[0036] Step S205: Based on the deviation between the feedback signal and the preset particle size target, adjust the mist flow rate of the ball outlet area.

[0037] Furthermore, the deviation between the feedback signal (actual average particle size) and the preset particle size target (the green pellet size required by the process, such as 12mm) reflects whether the particle size of the final product meets the standard. For example, if the actual average particle size is 11mm (smaller than the target), the deviation is -1mm; if it is 13mm (larger than the target), the deviation is +1mm. The system inputs this deviation into a proportional-integral-derivative (PID) controller, where the proportional element responds quickly to the deviation, the integral element eliminates steady-state errors, and the derivative element predicts the trend of deviation changes (e.g., if the particle size continues to be too small, increase the misting in advance). The controller outputs a misting flow rate control value to adjust the opening of the misting regulating valve above the pellet outlet area. If the deviation is negative (particle size too small), the valve opening is increased to increase the misting flow rate and promote green pellet growth; if the deviation is positive (particle size too large), the valve opening is decreased to reduce misting and prevent the green pellets from becoming too large. The core of this step is feedback regulation, which corrects the final particle size deviation through end-point moisture intervention.

[0038] Step S206: When the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, the cascade correction amount is calculated based on the deviation state of the feedback signal to dynamically correct the target particle powder ratio benchmark.

[0039] Furthermore, when the feedback signal (actual average particle size) deviates from the preset target for more than a set duration (e.g., particle size is too small for 5 consecutive minutes), and the mist flow rate has reached its limit (e.g., even with the mist valve fully open, particle size cannot be improved), it indicates that the end-feedback adjustment has failed. At this point, simply increasing the mist flow rate cannot solve the particle size problem; the root cause lies in the poor pelleting state of the front-end mother pellet area (e.g., a low pellet-to-powder ratio leading to insufficient pellet nuclei). The system then triggers a cascade correction mechanism: first, it determines the direction of deviation (e.g., particle size is too small) and the degree of deviation (e.g., the magnitude of the deviation); then, based on the direction of deviation, it adds the newly added correction value determined based on the current degree of deviation (e.g., the larger the deviation, the larger the correction value) to the historical cascade correction amount from the previous moment, obtaining the current cascade correction amount. Finally, it linearly superimposes the cascade correction amount with the current target pellet-to-powder ratio benchmark to generate a corrected target pellet-to-powder ratio benchmark, replacing the original benchmark. The core of this step is cascade correction, which reverses the end-feedback quality deviation to the front-end, solving problems that cannot be addressed at the root when end-feedback adjustment fails.

[0040] In summary, this embodiment overcomes the visual perception blind spot in the mother ball area by acquiring images of the mother ball region and extracting the ball-to-powder ratio as a feedforward signal, thus achieving quantitative monitoring of the initial nucleation state and solving the problem of lacking a feedforward signal at its source. By acquiring images of the mother ball region and calculating the average particle size as a feedback signal, a two-stage closed-loop control system is constructed, where the feedforward determines the basis of nucleation and the feedback corrects the final quality. This effectively overcomes the problem of pure time delay of several minutes in traditional single-loop feedback control and avoids frequent overshoot and oscillation. Furthermore, through a cascaded correction mechanism, when the feedback signal continuously deviates and the mist adjustment is saturated, the feedforward target benchmark is dynamically corrected, achieving coordinated control of dripping and misting, solving the problem of strong coupling between the two, and fundamentally ensuring that the green ball particle size meets the standard.

[0041] This embodiment provides a method for coordinated control of pellet size, which can be used in the aforementioned PLC control cabinet 101. Figure 3 This is a second flowchart of a pelletizing particle size synergistic control method according to an embodiment of the present invention, as follows: Figure 3 As shown, the process includes the following steps: Step S301: Acquire an image of the mother ball area of ​​the ball-making disc and extract the ball-to-powder ratio of the mother ball area as a feedforward signal.

[0042] In one optional implementation, step S301 includes: The image of the mother ball region is input into a lightweight semantic segmentation network to identify the green ball region and powder region in the image; this lightweight semantic segmentation network is trained in advance using historical sample images of the pelleting disc and manual annotation results; The ratio between the pixel area of ​​the green ball region and the pixel area of ​​the powder region is calculated to obtain the ball-to-powder ratio as a feedforward signal.

[0043] Furthermore, this embodiment specifically illustrates how to extract core features from complex industrial images using a deep learning model. Addressing issues such as dust interference, motion blur, and powder-ball adhesion present at the pelletizing site, this embodiment employs a pre-trained lightweight semantic segmentation network (e.g., a GhostNetV2 + DeepLabv3 + fusion model) to process the acquired mother pellet area image. This lightweight semantic segmentation network consists of a backbone network composed of multiple Ghost modules, responsible for extracting low-redundancy, high-representational-power powder-ball features. Combined with a multi-scale feature extraction module (ASPP), it achieves feature capture of targets at different scales, thereby accurately identifying the green pellet area and powder area in the image, achieving pixel-level precise segmentation. In this way, the system can effectively extract the pellet and powder states in the mother pellet area under strong dust conditions, solving the problem that related solutions struggle to distinguish between pellets and powder particles with similar colors and adhered edges. After obtaining the green pellet area mask and powder area mask through the semantic segmentation network, the system performs pixel statistics on the corrected masks, obtaining the total number of pixels in the green pellet area and the total number of pixels in the powder area, respectively. Based on the refined segmentation mask, the system calculates the ball-to-powder ratio (BPR), which is calculated as the number of sphere pixels divided by the number of powder pixels. This quantitative indicator directly reflects the nucleation state of the mother sphere region. The concept of ball-to-powder ratio is established as the only quantifiable feedforward signal for the nucleation state, enabling the control system to directly perceive the nucleation quality. This solves the problem in related technologies where there is a lack of effective quantitative characterization indicators and reliance on manual visual inspection is necessary.

[0044] Step S302: Acquire an image of the pelletizing area of ​​the pelletizing disc and calculate the average particle size of the pelletizing area as a feedback signal.

[0045] In one optional implementation, step S302 includes: Target detection was performed on the image of the ball-producing area of ​​the ball-making plate to identify individual unadhesive raw balls in the image and extract the outer contour of each raw ball; Based on the circumscribed contour of each green sphere, calculate the equivalent circle diameter of each green sphere; Calculate the arithmetic mean of the equivalent circular diameters of all identified green pellets to obtain the average particle size of the pelletizing area and use it as a feedback signal.

[0046] Furthermore, this embodiment describes the specific implementation method of particle size detection in the ball-eating area. The feedback mist control module uses the particle size detection result in the ball-eating area as the core feedback signal, and utilizes the YOLOv8 model to capture images of the ball-eating area in real time, generating a green ball detection box. The system performs target detection on the acquired ball-eating area image, aiming to identify individual green balls without adhesion in the image and extract the circumscribed contour of each green ball, thereby providing accurate geometric data for subsequent particle size calculation. Compared with related technologies, this deep learning-based target detection method can adapt to complex industrial environments and accurately identify the position and contour of green balls. After extracting the circumscribed contour of each green ball, the system converts the actual physical diameter using the pixel size of the detection box and calibration coefficients, calculating the equivalent circular diameter of each green ball. Subsequently, the system calculates the arithmetic mean of the equivalent circular diameters of all identified green balls to obtain the average particle size of the ball-eating area as feedback information. This feedback signal reflects the final product quality after feedforward adjustment and mist control in the growing area, and is crucial for identifying and correcting deviations in the dripping and misting processes, and can be used to identify and correct deviations in the process.

[0047] Step S303: Based on the current operating parameters of the pelleting disc, determine the target pellet powder ratio benchmark for drip control.

[0048] In one optional implementation, the current operating parameters include the real-time feed rate and the pelletizing disc rotation speed; step S303 includes: The real-time feed rate and the pelletizing disc rotation speed are input into a pre-built online optimization model. The online optimization model outputs the optimal pellet-to-powder ratio that matches the current working conditions, which serves as the benchmark for the target pellet-to-powder ratio. The online optimization model is trained based on historical working condition data and ball-to-powder ratio data, and its parameters are updated online according to real-time production data.

[0049] Furthermore, the system constructs an online optimization model (gradient boosting tree) to dynamically predict the optimal target value for the pellet powder ratio under the current operating conditions, i.e., the target pellet powder ratio benchmark. This online optimization model uses features such as real-time feed rate, pelletizing disc rotation speed, recent moving average of pellet proportion, and historical opening of the drip valve as inputs. Through supervised learning, it establishes a nonlinear mapping relationship between operating parameters and the optimal nucleation state. Real-time feed rate and pelletizing disc rotation speed are chosen as core operating parameters because these two parameters directly determine the material's movement state and nucleation opportunity within the pelletizing disc, accurately reflecting the current real-time operating characteristics. By inputting the real-time feed rate and pelletizing disc rotation speed into the pre-constructed online optimization model, the model can learn online based on real-time operating conditions and historical best production data, and output the current optimal target value for the pellet powder ratio. Through this mechanism, the online optimization model can dynamically adapt to operating conditions such as raw material batch fluctuations and equipment wear, continuously outputting the optimal pellet powder ratio setpoint that matches the current conditions, thereby determining the target pellet powder ratio benchmark for drip control and ensuring the optimal nucleation state.

[0050] Step S304: Adjust the amount of water dripping into the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio benchmark.

[0051] In one optional implementation, step S304 includes: The deviation between the feedforward signal and the target ball powder ratio reference is input to the proportional-integral (PI) controller; The proportional-integral (PI) controller calculates the drip rate control value and adjusts the opening of the drip solenoid valve according to the drip rate control value.

[0052] Furthermore, this embodiment utilizes a proportional-integral (PI) controller to eliminate the deviation between the actual state and the dynamic optimal target. The system compares the real-time calculated feedforward signal (actual ball-to-powder ratio) with the target ball-to-powder ratio benchmark to obtain the deviation signal, which is then input to the PI controller. The PI controller parameters consist of a proportional gain coefficient and an integral gain coefficient, which respectively determine the controller's immediate response strength to the current deviation and the correction strength for the cumulative historical deviation. This control method can effectively eliminate steady-state errors, ensuring that the nucleation state of the mother ball area can quickly and stably track the optimal target value. The drip rate control value calculated by the PI controller reflects the flow rate adjustment required to eliminate the ball-to-powder ratio deviation. The system adjusts the opening of the drip solenoid valve according to this control value, thereby changing the drip rate in the mother ball area and achieving precise control of the initial nucleation stage. At the same time, to ensure process safety, the final drip rate must be limited within a preset safety range to prevent the occurrence of dry powder or excessively wet agglomeration. This execution process constitutes a complete feedforward control closed loop, ensuring the accurate implementation of the control intention.

[0053] Step S305: Based on the deviation between the feedback signal and the preset particle size target, adjust the mist flow rate of the ball outlet area.

[0054] In one optional implementation, step S305 includes: The deviation between the feedback signal and the preset particle size target is input to the proportional-integral-derivative (PID) controller. The proportional-integral-derivative (PID) controller calculates the mist flow rate control value and adjusts the opening of the mist regulating valve according to the mist flow rate control value.

[0055] Furthermore, this embodiment employs a proportional-integral-derivative (PID) controller to precisely regulate the mist flow rate. The system compares the average particle size calculated in real-time in the pelletizing zone with the preset particle size target to obtain a deviation signal, which is then input to the PID controller. The PID controller includes proportional, integral, and derivative coefficients, corresponding to the amplification factors of the deviation signal, cumulative deviation, and deviation change rate, respectively, determining the controller's immediate response strength to the current deviation, the cumulative correction strength of historical deviations, and the response strength to deviation change trends. This control algorithm enables comprehensive regulation of particle size deviation, improving the system's response speed and stability. The system calculates the mist flow rate control value through the PID controller and adjusts the opening of the mist regulating valve according to this value, thereby changing the mist flow rate in the growth zone. The mist flow rate directly affects the growth rate and the density of the green pellets. By adjusting the mist flow rate, the system can correct the particle size deviation in the pelletizing zone, keeping it within the target range. In addition, to ensure process safety, the mist flow rate must be limited within a preset safety range to prevent over-wetting and agglomeration. This step achieves precise end-point correction of the finished pellet quality.

[0056] Step S306: When the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, determine the deviation direction and degree of the average particle size from the preset particle size target.

[0057] Furthermore, if the feedback signal (actual average particle size) from the ball-out area deviates from the preset particle size target for more than a set duration (e.g., particle size is too small for 3 consecutive minutes), and the mist flow rate has reached its limit (e.g., even with the mist valve opened to its maximum, particle size cannot be improved), it indicates that the end-feedback adjustment has failed. At this point, simply increasing the mist flow rate cannot solve the particle size problem; the root cause lies in the poor ball-forming state of the front-end mother ball area. The system then needs to analyze the direction of deviation (e.g., whether the particle size is too small or too large) and the degree of deviation (e.g., the absolute value of the deviation). The direction of deviation determines the direction of subsequent correction (e.g., if the particle size is too small, the proportion of green balls at the front end needs to be increased; if it is too large, it needs to be decreased); the degree of deviation determines the magnitude of the correction (e.g., the larger the deviation, the greater the correction force). This step is the core prerequisite for cascaded correction, providing a clear basis for subsequent calculations of the correction amount.

[0058] Step S307: According to the deviation direction, the newly added correction value determined based on the current deviation degree is added to the historical cascade correction amount of the previous moment to obtain the cascade correction amount of the current moment.

[0059] Furthermore, based on the deviation direction determined in step S306 and the current degree of deviation, the system generates a new correction value. If the deviation direction is due to a smaller particle size (requiring more front-end green pellets), the new correction value is positive; if it is due to a larger particle size (requiring less front-end green pellets), the new correction value is negative. The magnitude of the new correction value is positively correlated with the degree of deviation (e.g., the greater the degree of deviation, the larger the absolute value of the correction value). Subsequently, the new correction value is added to the historical cascaded correction amount from the previous moment to obtain the cascaded correction amount at the current moment. The accumulation mechanism of historical cascaded correction amounts allows the system to remember previous deviation adjustments, avoiding control shocks caused by single corrections and achieving a smooth transition of the reference offset (e.g., for multiple consecutive small deviations, the correction amount is gradually accumulated, rather than a sudden large adjustment).

[0060] Step S308: Linearly superimpose the cascaded correction amount with the target ball powder ratio benchmark at the current moment to generate the corrected updated ball powder ratio benchmark, and replace the target ball powder ratio benchmark for drip volume adjustment.

[0061] Furthermore, in this embodiment, the cascaded correction amount at the current moment is linearly superimposed (i.e., directly added) with the target ball powder ratio benchmark (the expected value of the front-end dripping amount adjustment) to generate a corrected updated ball powder ratio benchmark. For example, if the original target benchmark is 1.5 and the cascaded correction amount is +0.2, then the new benchmark is 1.7, requiring an increase in the proportion of green balls in the mother ball area. Subsequently, the updated benchmark replaces the original benchmark and participates in the subsequent dripping amount adjustment (i.e., the feedforward adjustment in step S304). The core of this step is to reverse the quality deviation at the end to the front end, fundamentally improving the ball formation basis by adjusting the front-end benchmark and solving the problem of end-end adjustment failure.

[0062] In summary, please refer to Figure 4The diagram illustrates the ball-making and water control strategy. This embodiment employs a fusion model of GhostNetV2 and DeepLabv3+ (a specific implementation of a lightweight semantic segmentation network) to achieve pixel-level accurate segmentation of the ball powder. To address noise issues such as dust interference and motion blur in the ball-making site images, this embodiment designs a multi-step preprocessing procedure to enhance the ball powder features and improve the extraction accuracy and convergence efficiency of the subsequent semantic segmentation model. First, a contrast-limited adaptive histogram equalization (CLAHE) algorithm is used to increase the contrast between the dark powder area and the ball edge without amplifying local noise. Then, a non-local means denoising algorithm is used to effectively suppress motion blur noise while fully preserving the circular edge texture features of the balls. Finally, the preprocessed image pixel values ​​are linearly mapped to the [0,1] interval to eliminate model training bias caused by differences in pixel value magnitudes. During the training phase, the images were randomly rotated, mirrored, and aliased to simulate the visual state of powder and sphere mixing under different humidity levels, and labeled into three categories: background (label=0, non-material area), powder (label=1, loose material with a particle size of less than 2mm), and spherical particles (label=2, approximately spherical particles with a particle size of greater than or equal to 2mm).

[0063] Please see Figure 5 The diagram shows a fusion network architecture of GhostNetV2 and DeepLabv3+. In terms of segmentation model construction, to achieve efficient and accurate segmentation of powder and spheres while adapting to the computing power constraints of industrial edge devices, this embodiment designs a fusion model structure of a lightweight backbone network and a high-precision segmentation head. Specifically, GhostNetV2 is selected as the backbone network, and DeepLabv3+ is selected as the segmentation head. The backbone feature extraction module in the DeepLabv3+ encoder is replaced with a GhostNetV2 network, which consists of multiple Ghost modules and is responsible for extracting low-redundancy, high-representational powder and sphere features, and outputting two types of key features: high-resolution low-level features (preserving details such as sphere edges and fine powder texture) and low-resolution high-level features (containing semantic category information of spheres, powder, and background). The ASPP module is used at the encoder end to capture multi-scale features through convolutional kernels with different dilation rates: small dilation rate convolution captures the fine texture of powder, and large dilation rate convolution captures the shape of large spheres. The fusion of these two methods reduces misclassification of small spheres and powder. The decoder section integrates low-level features from GhostNetV2 (detailed information about the edges of spheres and grains and the adhesion regions), and with simple upsampling operations, accurately restores pixel-level edges, ensuring that the outlines of spheres and grains are not overly smoothed, and improves the segmentation precision of adhesion regions.

[0064] To address the common issues of adhesion and blurred edges in spherical particles in real-world images, this embodiment performs morphological post-processing and connected component analysis on the initial mask output from semantic segmentation to refine the mask. For the spherical region, an elliptical structuring element is used to perform a closing operation (dilation followed by erosion) on the initial mask to fill the holes and defects inside the spherical particles caused by specular reflection. Then, connected component analysis is performed on the refined spherical region. An area threshold T is set, and connected components with an area less than T are identified as misjudged powder noise and forcibly corrected to the powder category, thus filtering noise in the spherical region. For the powder region, a rectangular structuring element is used to perform an opening operation (erosion followed by dilation) on the initial mask to remove isolated noise points. Pixel-level logical operations are used to exclude overlapping areas between the powder mask and the refined spherical mask, avoiding interference from powder categories covering the spherical region. Subsequently, the masks are merged according to the semantic priority of "spherical > powder > background," and a polygon approximation algorithm or Gaussian smoothing algorithm is used to optimize the spherical segmentation boundary, making the boundary conform to the physical morphological characteristics of the spherical particles and reducing the impact of jagged edge defects on subsequent state quantization.

[0065] Based on the refined powder-sphere segmentation mask, the state parameters of the mother sphere region are quantitatively calculated through pixel statistics. This embodiment proposes a sphere-to-powder ratio (BPR) index (i.e., feedforward signal), which is calculated as follows: The total number of pixels of each category in the corrected mask is statistically analyzed, and the ratio of the number of sphere pixels to the number of powder pixels is calculated, i.e., BPR = number of sphere pixels / number of powder pixels. This index quantifies the nucleation state of the mother sphere region, solving the problem that related technologies can only rely on manual visual inspection and cannot quantify it. Addressing technical challenges such as blurred boundaries due to similar features in the mother sphere region and the high degree of adhesion of small targets, this embodiment employs CLAHE and non-local mean filtering for feature enhancement preprocessing, a GhostNetV2+DeepLabv3+ fusion network combined with morphological post-processing to separate adhesion, and proposes the concept of sphere-to-powder ratio to quantify the nucleation state through pixel-level statistics.

[0066] When monitoring particle size in the exit zone, the particle size detection results in the exit zone serve as the core feedback signal. Specifically, images of the exit zone are captured in real time, and a target detection model (specifically, a YOLOv8 network) is used to generate green particle detection boxes. The actual physical diameter is calculated by converting the pixel size of the detection boxes with the camera calibration coefficients, and the arithmetic mean of the diameters of all detected green particles is calculated within a preset time window to obtain the average particle size (D). avg This serves as a feedback signal (i.e., the average particle size feedback signal). This signal is used to compare with the preset particle size target, providing a basis for subsequent mist flow rate adjustment.

[0067] In addition, such as Figure 4As shown, this embodiment introduces a collaborative adjustment mechanism (i.e., cascaded correction) between the feedforward control of the mother ball area and the feedback control of the ball outlet area. Specifically, the system uses a deviation calculator (ΔD) to calculate in real time the difference between the average particle size and the preset particle size target (i.e., particle size deviation). This deviation value is not only used for PID adjustment of the mist flow rate but also serves as the basis for triggering cascaded correction. Simultaneously, the system uses a sliding window parameter to statistically process the recently detected green ball particle size: calculating the average particle size within a preset time window to eliminate instantaneous noise interference; this sliding window parameter can also be used in the online optimization model to update the recent sliding average of the ball-to-powder ratio, thereby dynamically adjusting the target ball-to-powder ratio benchmark. When the particle size deviation persists and the mist flow rate reaches the flow limit (i.e., the mist valve opening has reached the upper or lower limit), the system triggers collaborative adjustment, that is, it reverses the target ball-to-powder ratio benchmark in the mother ball area through cascaded correction, realizing cross-loop collaborative control of the drip valve and the mist valve.

[0068] In the specific implementation of semantic segmentation networks, such as Figure 5 As shown, this embodiment employs the GhostNetV2 backbone network (a lightweight feature extractor, i.e., a lightweight semantic segmentation network). This network consists of multiple Ghost module stages connected in series: the input image sequentially passes through Ghost module stages 1, 2, 3, and 4, progressively extracting features from low to high levels, ultimately outputting a high-level feature map. In the encoder stage, the aforementioned high-level feature map is fed into the ASPP module (Atrous Spatial Pyramid Pooling, i.e., a multi-scale feature extraction module). This module captures multi-scale contextual information through dilated convolutions with different dilation rates, specifically including a 1×1 convolution, three 3×3 convolutions (with dilation rates of 6, 12, and 18, respectively), and an image pooling followed by a 1×1 convolution. The outputs of each branch are concatenated and then subjected to a 1×1 convolution for feature fusion. In the decoder stage, the feature map output by the encoder is first upsampled by a factor of 4, then concatenated with the corresponding low-level features output by the GhostNetV2 backbone network (feature fusion). Next, a 1×1 convolution is applied for channel dimensionality reduction, followed by edge refinement using a 3×3 convolutional block. Finally, it is upsampled again by a factor of 4 to output a segmentation mask with the same resolution as the input image. This decoder design ensures precise restoration of the spherical edges.

[0069] Furthermore, this embodiment constructs a two-level closed-loop control system that uses feedforward to determine the nucleation basis and feedback to correct the final quality (i.e., the feedforward signal adjusts the drip volume and the feedback signal adjusts the mist flow rate).

[0070] For feedforward dripping control in the mother pellet area, the core lies in abandoning fixed thresholds and empirical strategies, and constructing an intelligent feedforward control system that seeks online optimization. This system uses a lightweight online learning model (such as a gradient boosting tree, i.e., an online optimization model) to learn and output the current optimal target value for the pellet powder ratio based on real-time operating conditions and historical best production data. Then, it adjusts the dripping amount to make the actual pellet powder ratio converge towards this target value.

[0071] The lightweight online learning model constructed in this embodiment uses features such as real-time feed rate, pelletizing disc rotation speed, recent moving average of pellet percentage, and historical opening of drip valve as input. Through supervised learning, it establishes a nonlinear mapping relationship between operating parameters and the optimal nucleation state, dynamically predicting the optimal pellet-to-powder ratio target value under the current operating conditions. The model first undergoes offline pre-training using data from historical high-quality production periods (corresponding to high pellet yield and high moisture stability) to acquire preliminary predictive capabilities. During continuous system operation, whenever a production cycle ends and the pellet size qualification rate exceeds 95%, the system uses the average operating data and corresponding average pellet-to-powder ratio of the current cycle as new high-quality samples, incorporating them into the training set in real time for incremental learning. This allows the model to dynamically adapt to operating conditions such as raw material batch fluctuations and equipment wear, continuously outputting the optimal pellet-to-powder ratio setting value that matches the current conditions.

[0072] The feedforward dripping control strategy utilizes a proportional-integral (PI) controller to eliminate the deviation between the actual state and the dynamic optimal objective. The formula for calculating the dripping control amount is as follows:

[0073] in, The drip rate at time t; The current ball-to-powder ratio is calculated in real time (feedforward signal); The optimal target value (target ball powder ratio benchmark) is predicted in real time by the dynamic target optimization model. It indicates the deviation of the ball-to-powder ratio at the current moment, reflecting the degree of deviation between the actual nucleation state and the optimal target; This is the cumulative sum of the deviations. The proportional gain coefficient determines the instantaneous response of the controller to the current deviation. The integral gain coefficient determines the strength of the controller's correction for accumulated historical deviations. and These constitute the parameters of the PI controller.

[0074] The final drip rate is: Δ ; To ensure process safety, They must be confined to a preset safe zone: ; in, Minimize safe flow rate to prevent dry powder; To maximize the allowable flow rate and prevent excessive moisture and clumping.

[0075] The mist control in the ball-ejection zone uses the particle size detection results in the ball-ejection zone as the core feedback signal. The average particle size D is calculated in real-time based on a target detection model (specifically, a YOLOv8 network, i.e., a single-stage target detection network based on anchor frames). avg The deviation ΔD between the particle size and the target particle size (e.g., 12.5 mm) is: ΔD = - 12.5mm; A negative ΔD indicates that the particle size is too small, while a positive ΔD indicates that the particle size is too large. The closer ΔD is to 0, the closer the particle size is to the acceptable standard. The mist control uses a proportional-integral-derivative (PID) control algorithm, and the specific implementation of the mist control amount is as follows: ; in, This is the current mist control volume; = ΔD t = ( 12.5) represents the deviation between the current particle size and the target particle size (the negative sign is due to the control logic: if the particle size is too large, water needs to be reduced, so the output is negative when the deviation is positive). For all deviations from the start of control to the current time. The algebraic sum; This is the difference between the current time deviation and the previous time deviation; for The amplification factor of the deviation signal determines the instantaneous response strength of the controller to the current deviation; Cumulative deviation sum The amplification factor determines the cumulative correction strength of the controller for historical deviations; Rate of change of deviation The amplification factor determines the controller's response strength to the trend of deviation change; , , These respectively constitute the proportional, integral, and derivative coefficients for mist control.

[0076] The final mist flow rate is: ; To avoid frequent adjustments near the target value, this embodiment sets a control dead zone. δ,+δ] (take δ=0.5 mm), when ∣ΔDt When |≤δ, maintain the current mist flow rate constant, i.e., Δu 2,t =0.

[0077] To ensure process safety, u 2,t They must be confined to a preset safe zone: ; in, To ensure the minimum water mist flow rate required to guarantee minimal wetting of the pellet surface; To maximize the allowable flow rate and prevent excessive moisture and clumping.

[0078] Under normal operating conditions, the mother ball zone feeds forward to control the number of nuclei, while the growth zone provides feedback to adjust particle size; the two operate in tandem. When the exiting particle size remains consistently low and the mist level has reached its upper limit without improvement, it indicates a deviation in the nucleation stage, with too many mother balls and excessively small particle sizes. In this case, a cascaded correction mechanism must be used to reversely lower the control target in the mother ball zone, reducing the nucleation density and ensuring sufficient powder coating space for each nucleus during the growth stage. Conversely, when the particle size remains consistently large and the mist level has dropped to its lower limit without improvement, it indicates too few nuclei or excessively large initial nuclei. In this case, the target particle-to-powder ratio in the mother ball zone needs to be increased to increase the nucleation density and disperse the raw materials. The cascaded correction mechanism designed in this embodiment uses the long-term trend of particle size deviation in the exiting zone and the saturation state of the mist regulating valve as trigger conditions to dynamically correct the target particle-to-particle ratio value in the mother ball zone, effectively solving the problem of strong coupling between dripping and mist control.

[0079] The system monitors the following two states in real time to determine whether to enter correction mode: Forward cascade triggering (small particle size): when the average particle size <12.5-δ (particle size continues to be small), and the mist flow rate has reached its upper limit. Duration exceeds ( If the time is set to 3 minutes, it indicates that even with maximum hydration, the green bulbs still cannot grow. This is because excessive water dripping into the mother bulb area leads to an excessive number of nuclei, resulting in insufficient dry powder per nucleus. In this case, the target powder-to-nucleus ratio in the mother bulb area needs to be reduced to decrease the nucleation density.

[0080] Reverse cascade triggering (larger particle size): when the average particle size >12.5+δ (particle size continues to be large), and the mist flow rate has dropped to the lower limit. Duration exceeds ( If the timer is set to 3 minutes, it indicates that the green pellets are still too large even after water replenishment is stopped. This is because the number of nuclei in the mother pellet area is too small or the initial nuclei are too large, causing the snowball effect to occur too quickly. In this case, it is necessary to increase the target proportion of pellet powder in the mother pellet area to increase the nucleation density and disperse the raw materials.

[0081] After the cascading correction condition is triggered, the system no longer relies solely on the aforementioned optimization model, but instead introduces a cascading correction operator with memory. (That is, the cascaded correction amount). The specific correction formula is as follows: ; Among them, cascaded correction terms Using recursive integral form: = +λ sgn ( D); in, It is the cascaded correction amount of the target value of the cue ball area at the current moment, and it has memory properties, which can continuously retain the influence of the historical control direction; The correction amount from the previous moment forms the basis for the integral recursion; λ is the cascade correction step size coefficient, used to control the magnitude of a single correction, ensuring the smooth evolution of the target value and avoiding system oscillations. sgn ( D) is the sign function of particle size deviation, defined as: ; To achieve a smooth and stable correction of the cue ball area target value, this embodiment progressively adds the historical cumulative effect of the ball exit area deviation to the cue ball area target value. The corrected cue ball area is then set as the final target value. From model predictions and cascaded correction cumulative amount The decision is made jointly. Through this mechanism, the system can automatically find the most suitable nucleation density of the mother pellets under the current raw material characteristics, fundamentally solving the particle size control problem, rather than just making ineffective remedies at the end.

[0082] In summary, this embodiment overcomes the visual perception blind spot in the mother ball area by acquiring images of the mother ball region and extracting the ball-to-powder ratio as a feedforward signal, thus achieving quantitative monitoring of the initial nucleation state and solving the problem of lacking a feedforward signal at its source. By acquiring images of the mother ball region and calculating the average particle size as a feedback signal, a two-stage closed-loop control system is constructed, where the feedforward determines the nucleation basis and the feedback corrects the final quality. This effectively overcomes the problem of several-minute-level pure lag in traditional single-loop feedback control and avoids frequent overshoot and oscillation. Furthermore, through a cascaded correction mechanism, when the feedback signal continuously deviates and the mist adjustment is saturated, the feedforward target benchmark is dynamically corrected, achieving coordinated control of dripping and misting, solving the problem of strong coupling between the two, and fundamentally ensuring that the green ball particle size meets the standard.

[0083] This embodiment also provides a pelletizing particle size coordination control system, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0084] This embodiment provides a pelletizing particle size coordinated control system, such as Figure 6 As shown, it includes: The feedforward acquisition module 601 is used to acquire images of the mother ball area of ​​the ball-making disc and extract the ball-to-powder ratio of the mother ball area as a feedforward signal. The feedback acquisition module 602 is used to acquire images of the pelletizing area of ​​the pelletizing disc and calculate the average particle size of the pelletizing area as a feedback signal. The benchmark determination module 603 is used to determine the target powder ratio benchmark for drip control based on the current operating parameters of the pelleting disc; The feedforward control module 604 is used to adjust the dripping amount of the mother ball area according to the deviation between the feedforward signal and the target ball powder ratio reference. The feedback control module 605 is used to adjust the mist flow rate of the ball outlet area based on the deviation between the feedback signal and the preset particle size target. The cascade correction module 606 is used to calculate the cascade correction amount based on the deviation state of the feedback signal when the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, so as to dynamically correct the target particle powder ratio benchmark.

[0085] In some alternative implementations, the feedforward acquisition module 601 is used for: The image of the mother ball region is input into a lightweight semantic segmentation network to identify the green ball region and powder region in the image; this lightweight semantic segmentation network is trained in advance using historical sample images of the pelleting disc and manual annotation results; The ratio between the pixel area of ​​the green ball region and the pixel area of ​​the powder region is calculated to obtain the ball-to-powder ratio as a feedforward signal.

[0086] In some alternative implementations, the feedback acquisition module 602 includes: Target detection was performed on the image of the ball-producing area of ​​the ball-making plate to identify individual unadhesive raw balls in the image and extract the outer contour of each raw ball; Based on the circumscribed contour of each green sphere, calculate the equivalent circle diameter of each green sphere; Calculate the arithmetic mean of the equivalent circular diameters of all identified green pellets to obtain the average particle size of the pelletizing area and use it as a feedback signal.

[0087] In some optional implementations, the current operating parameters include the real-time feed rate and the pelletizing disc rotation speed; the reference determination module 603 is also used for: The real-time feed rate and the pelletizing disc rotation speed are input into a pre-built online optimization model. The online optimization model outputs the optimal pellet-to-powder ratio that matches the current working conditions, which serves as the benchmark for the target pellet-to-powder ratio. The online optimization model is trained based on historical working condition data and ball-to-powder ratio data, and its parameters are updated online according to real-time production data.

[0088] In some alternative implementations, the feedforward control module 604 is further configured to: The deviation between the feedforward signal and the target ball powder ratio reference is input to the proportional-integral (PI) controller; The proportional-integral (PI) controller calculates the drip rate control value and adjusts the opening of the drip solenoid valve according to the drip rate control value.

[0089] In some alternative implementations, the feedback control module 605 is further configured to: The deviation between the feedback signal and the preset particle size target is input to the proportional-integral-derivative (PID) controller. The proportional-integral-derivative (PID) controller calculates the mist flow rate control value and adjusts the opening of the mist regulating valve according to the mist flow rate control value.

[0090] In some alternative implementations, the cascaded correction module 606 is further configured to: Determine the direction and degree of deviation of the average particle size from the preset particle size target; Based on the direction of deviation, the newly added correction value determined according to the current degree of deviation is added to the historical cascade correction amount of the previous moment to obtain the cascade correction amount of the current moment.

[0091] In some alternative implementations, the cascaded correction module 606 is further configured to: The cascaded correction amount is linearly superimposed with the target ball powder ratio benchmark at the current moment to generate a corrected updated ball powder ratio benchmark, which is then used to replace the target ball powder ratio benchmark for drip volume adjustment.

[0092] The pelletizing particle size coordination control system provided in this embodiment of the invention can execute the pelletizing particle size coordination control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0093] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0094] The following is a detailed reference. Figure 7 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to bus 704.

[0095] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0096] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the pelletizing particle size collaborative control method of the embodiments of the present invention.

[0097] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0098] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the pelletizing particle size collaborative control method shown in the above embodiments.

[0099] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0100] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the invention.

Claims

1. A method for synergistic control of pellet size, characterized in that, The method includes: Images of the mother ball area in the pelleting disc are acquired, and the ball-to-powder ratio in the mother ball area is extracted as a feedforward signal; Images of the pelletizing area of ​​the pelletizing disc are acquired, and the average particle size of the pelletizing area is calculated as a feedback signal; Based on the current operating parameters of the pelleting disc, the target pellet powder ratio benchmark for drip control is determined; The amount of water dripping in the mother ball area is adjusted based on the deviation between the feedforward signal and the target ball powder ratio benchmark. Based on the deviation between the feedback signal and the preset particle size target, the flow rate of the mist in the ball outlet area is adjusted; When the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, a cascade correction amount is calculated based on the deviation state of the feedback signal to dynamically correct the target sphere powder ratio benchmark.

2. The method according to claim 1, characterized in that, The extraction of the ball-to-powder ratio in the mother ball region as a feedforward signal includes: The image of the mother ball area is input into a lightweight semantic segmentation network to identify the green ball area and powder area in the image; the lightweight semantic segmentation network is trained in advance using historical sample images of the balling tray and manual annotation results; The ratio between the pixel area of ​​the green ball region and the pixel area of ​​the powder region is calculated to obtain the ball-to-powder ratio as a feedforward signal.

3. The method according to claim 1, characterized in that, The calculated average particle size of the spherical region is used as a feedback signal, including: Target detection is performed on the image of the ball-producing area of ​​the ball-making disc to identify individual unadhesive raw balls in the image, and the outer contour of each raw ball is extracted. Based on the circumscribed contour of each green ball, calculate the equivalent circle diameter of each green ball; Calculate the arithmetic mean of the equivalent circular diameters of all identified green pellets to obtain the average particle size of the pelletizing area and use it as a feedback signal.

4. The method according to claim 1, characterized in that, The current operating parameters include the real-time feed rate and the pelletizing disc rotation speed; The determination of the target powder ratio for drip control based on the current operating parameters of the pelleting disc includes: The real-time feed rate and the pelletizing disc rotation speed are input into a pre-built online optimization model. The online optimization model outputs the optimal pellet-to-powder ratio that matches the current working conditions, which serves as the benchmark for the target pellet-to-powder ratio. The online optimization model is trained based on historical working condition data and ball-to-powder ratio data, and its parameters are updated online according to real-time production data.

5. The method according to claim 1, characterized in that, The step of adjusting the dripping amount in the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio benchmark includes: The deviation between the feedforward signal and the target ball powder ratio benchmark is input to the proportional-integral (PI) controller; The proportional-integral (PI) controller calculates the drip rate control value and adjusts the opening of the drip solenoid valve according to the drip rate control value.

6. The method according to claim 1, characterized in that, The step of adjusting the water flow rate in the ball outlet area based on the deviation between the feedback signal and the preset particle size target includes: The deviation between the feedback signal and the preset particle size target is input to the proportional-integral-derivative (PID) controller. The proportional-integral-derivative (PID) controller calculates the mist flow rate control value and adjusts the opening of the mist regulating valve according to the mist flow rate control value.

7. The method according to any one of claims 1 to 6, characterized in that, The calculation of the cascade correction amount based on the deviation state of the feedback signal includes: Determine the direction and degree of deviation of the average particle size from the preset particle size target; According to the deviation direction, the newly added correction value determined based on the current deviation degree is added to the historical cascade correction amount of the previous moment to obtain the cascade correction amount of the current moment.

8. The method according to claim 7, characterized in that, The dynamic correction of the target ball powder ratio benchmark includes: The cascaded correction amount is linearly superimposed with the target ball powder ratio benchmark at the current moment to generate a corrected updated ball powder ratio benchmark, which is then used to replace the target ball powder ratio benchmark for drip volume adjustment.

9. A pelletizing particle size synergistic control system, characterized in that, The system includes: The feedforward acquisition module is used to acquire images of the mother ball area of ​​the ball-making disc and extract the ball-to-powder ratio of the mother ball area as the feedforward signal. The feedback acquisition module is used to acquire images of the pelletizing area of ​​the pelletizing disc and calculate the average particle size of the pelletizing area as a feedback signal. The benchmark determination module is used to determine the target powder ratio benchmark for drip control based on the current operating parameters of the pelleting disc; The feedforward control module is used to adjust the dripping amount in the mother ball area based on the deviation between the feedforward signal and the target ball powder ratio benchmark. The feedback control module is used to adjust the mist flow rate in the ball outlet area based on the deviation between the feedback signal and the preset particle size target. The cascade correction module is used to calculate the cascade correction amount based on the deviation state of the feedback signal when the feedback signal deviates from the preset particle size target for a set time and the mist flow rate reaches the flow limit value, so as to dynamically correct the target sphere powder ratio benchmark.

10. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected and the memory stores computer instructions. The processor executes the computer instructions to perform a pelletizing particle size control method according to any one of claims 1 to 8.

Images