A method, system and device for controlling production of a balling machine, and a storage medium

Abstract

This invention discloses a production control method, system, equipment, and storage medium for a pelletizing machine. The method collects multimodal data, including images and process parameters from the pelletizing machine. After preprocessing, standardization, and time alignment, a fusion feature is formed and input into a pre-trained control model for operating condition determination. For abnormal operating conditions, the model outputs corresponding process parameter adjustment strategies, which are ultimately executed by the pelletizing machine control system, achieving intelligent recommendation and closed-loop control of process parameters. This invention overcomes the limitations of existing automatic control systems that rely on single-sensor local perception. It achieves intelligent operating condition determination and accurate parameter recommendation based on a control model. The model can effectively identify abnormal and non-abnormal operating conditions, realizing standardized and intelligent pelletizing operating condition determination and parameter adjustment, significantly reducing reliance on human experience.

Description

Technical Field

[0001] This invention relates to the field of pelletizing technology, and in particular, to a pelletizing machine production control method, system, equipment, and storage medium. Background Technology

[0002] The disc pelletizer is a core piece of equipment in the iron and steel metallurgical pellet production process. During its operation, the quality of the pellets formed in the pelletizing disc directly determines the energy consumption of the subsequent roasting process, and has a key impact on the core indicators such as the strength and particle size of the final pellet product. Therefore, the precise control of the pelletizer process parameters is the key to achieving efficient and high-quality production in the pellet plant.

[0003] Currently, the operation and control of pelletizing machines in the industry still mainly relies on the manual experience of on-site operators. Operators observe the pelletizing state of the material in the pelletizing pan with their naked eyes and manually adjust process parameters such as the speed, water volume, and tilt angle of the pelletizing machine. The manual operation mode requires a high level of experience from the operators. It takes a long time to train an excellent operator who can accurately judge the pelletizing state and reasonably adjust the parameters, and the labor cost is high. Moreover, the human judgment standard is highly subjective and unstable. Different operators have different experience and cognition. The fluctuation of the same operator's state at different working times will also affect the parameter adjustment decision, which can easily lead to large fluctuations in the pelletizing production process and poor consistency of green pellet quality. In addition, the response of manual observation, judgment and operation is significantly delayed, making it difficult to quickly respond to changes in on-site working conditions such as sudden changes in raw material characteristics, which can easily lead to a decrease in pelletizing efficiency and an increase in defective products. Furthermore, manual operation cannot achieve continuous optimal control 24 / 7, making it difficult to maintain long-term stability in pelletizing production.

[0004] To improve the drawbacks of manual operation, some automatic control systems for pelletizing machines have emerged in the existing technology. These systems are mostly based on moisture data collected by a single sensor or use basic control models such as PID to adjust parameters. However, since the pelletizing state in the pelletizing disc is a complex and comprehensive state that includes particle size distribution, pellet roundness, material adhesion, moisture distribution, etc., a single sensor can only acquire local process data and cannot fully perceive and make comprehensive decisions on the complex pelletizing state. This results in limited control effect and makes it difficult to match the precise control requirements of actual production. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a production control method, system, equipment, and storage medium for a pelletizing machine.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A pelletizing machine production control method includes the following steps: S1, Collect multimodal data during the operation of the pelletizing machine. The multimodal data includes image data of the pelletizing area and the stable green pellet area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine. S2, preprocess the image data, identify and block the interference area formed by the ball-collecting robot's robotic arm, classify the image into fine powder area, qualified raw ball, and overly wet area, and obtain visual features; S3, standardize the real-time process parameter data to obtain process parameters; S4, align the visual features with the process parameters using timestamps to form multimodal fusion features; S5, the multimodal fusion features are input into the pre-trained control model, and the control model determines whether the current operating condition is an abnormal operating condition: if so, the adjustment strategy is determined and the adjustment signal of the process parameters is output. S6 outputs the process parameter adjustment signal to the pelletizing machine control system, and the pelletizing machine control system adjusts the corresponding process parameters according to the process parameter adjustment signal.

[0007] Further, determining whether the current working condition is an abnormal working condition includes at least one of the following: obtaining the rate of change of the proportion of pixels in the fine powder region within the effective analysis area and the rate of change of the proportion of pixels in the over-wet region within the effective analysis area; if one of the rates of change is greater than a first preset threshold, then the current working condition is determined to be a first abnormal working condition; determining whether there are one or more local areas that are over-wet abnormally rising local areas; if so, then the current working condition is determined to be a second abnormal working condition; identifying whether the number of extra-large pellets appearing in the edge area of ​​the pelletizing disc within a preset time period is greater than a second preset threshold, then the current working condition is determined to be a third abnormal working condition; the extra-large pellets are pellets with a particle size greater than the third preset threshold.

[0008] Furthermore, the determination of whether one or more local areas are abnormally wetted local areas: if so, it is determined to be a second abnormal operating condition, specifically including: S521 marks the effective analysis area of ​​the pelletizing area and the stable area of ​​the raw pellets on the pelletizing plate, assigns a pixel coordinate system to the effective area, and determines the horizontal and vertical pixel boundaries; S522, the effective analysis area after calibration is divided into multiple local units by proportionally dividing the grid; S523, in real time, counts the total number of pixels in the over-wet area of ​​each local unit, calculates the proportion of the over-wet area in each local unit, and obtains the global proportion of the over-wet area in the effective analysis area. S524, determine whether the number of pixels in the over-wet region of each local unit is greater than the minimum over-wet pixel count threshold: if yes, the local unit is determined to be in a valid analysis state; if no, the local unit is determined to be in an invalid analysis state. S525, determine whether each local element is in a valid analysis state within a preset time period: if yes, the local element is determined to be a valid analysis element; if no, the local element is determined to be an invalid analysis element. S526, determine whether the increase in the proportion of over-humidified areas of each effective analysis unit within a preset time period is greater than the first preset increase threshold: if yes, then determine that the effective analysis unit is a local area of ​​abnormally rising over-humidity; if no, then proceed to step S527. S527, determine whether the increase in the difference between the proportion of over-humidified areas of each effective analysis unit and the global proportion of over-humidified areas within a preset time period is greater than the second preset increase threshold: if so, determine that the effective analysis unit is a local area of ​​abnormally rising over-humidity.

[0009] Furthermore, the process parameter data includes at least the feed rate, atomized water volume, drip rate, pelletizer speed, pelletizer tilt angle, and main motor current.

[0010] Furthermore, the determination of the adjustment strategy specifically includes: if the current working condition is determined to be a first abnormal working condition, the adjustment strategy is to reduce the dripping water volume, increase the atomized water volume, and increase the pounder speed; if the current working condition is determined to be a second abnormal working condition, the adjustment strategy is to reduce the dripping water volume and increase the pounder speed for a limited time; if the current working condition is determined to be a third abnormal working condition, the adjustment strategy is a conservative recommended mode.

[0011] Furthermore, the adjustment strategy is to reduce the dripping water volume and increase the pelletizing machine speed within a limited time. Specifically, it includes: determining whether there are adjacent pre-set number of abnormally high local areas of excessive moisture: if so, the adjustment of the dripping water volume and the pelletizing machine speed is performed using a higher-order pre-set adjustment strategy; if not, the adjustment of the dripping water volume and the pelletizing machine speed is performed using a lower-order pre-set adjustment strategy; the adjustment amount of the dripping water volume and the pelletizing machine speed in the higher-order pre-set adjustment strategy is greater than the adjustment amount of the lower-order pre-set adjustment strategy.

[0012] Furthermore, the determination of whether the current working condition is an abnormal working condition also includes: obtaining the proportion of the over-wet area and the proportion of the fine powder area on the pelletizing tray surface; if the proportion of the over-wet area exceeds the fourth preset threshold or the proportion of the fine powder area exceeds the fifth preset threshold, then it is determined whether the deviation between the water volume setting and the feedback value is greater than the sixth preset threshold; if so, then the current working condition type is determined to be the fourth abnormal working condition, and the adjustment strategy corresponding to the fourth abnormal working condition is: to adjust the water volume and trigger a targeted alarm.

[0013] This invention also provides a pelletizing machine control system, comprising: a data acquisition module for acquiring multimodal data during the operation of the pelletizing machine, the multimodal data including image data of the pelletizing area and the stable green pellet area of ​​the pelletizing disc, and real-time process parameter data of the pelletizing machine; an image processing module for preprocessing the image data, identifying and shielding interference areas formed by the robotic arm of the pellet-collecting robot, classifying the images into fine powder areas, qualified green pellets, and over-wet areas, and obtaining visual features; a data processing module for standardizing the real-time process parameter data to obtain process parameters; a feature fusion module for aligning the visual features with the process parameters using timestamps to form multimodal fusion features; a model input / output module for inputting the multimodal fusion features into a pre-trained control model, the control model determining whether the current operating condition is an abnormal operating condition: if so, determining an adjustment strategy and outputting an adjustment signal for the process parameters; and an output adjustment module for outputting the adjustment signal for the process parameters to the pelletizing machine control system, the pelletizing machine control system adjusting the corresponding process parameters according to the adjustment signal for the process parameters.

[0014] The present invention also provides an electronic device, including a processor and a memory, wherein the memory is used to store program code and transmit the program code to the processor; the processor is used to execute the pelletizing machine production control method according to the instructions in the program code.

[0015] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the pelletizing machine production control method.

[0016] The present invention has the following beneficial effects: This system achieves fusion perception and precise analysis of multimodal data, overcoming the limitations of single-sensor local perception in existing automatic control systems. By collecting image data of the pelletizing area and the stable green pellet area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine, and combining image preprocessing, pixel classification, and process parameter standardization, it comprehensively extracts visual and process features during the pelletizing process. It also effectively shields the interference area of ​​the pellet-retrieval robot's robotic arm, ensuring the comprehensiveness, accuracy, and effectiveness of feature extraction, providing complete and reliable data support for process parameter control. Based on a control model, it achieves intelligent judgment of operating conditions and precise parameter recommendation. The model can effectively identify abnormal and non-abnormal operating conditions and output targeted process parameter adjustment signals, replacing traditional manual experience-based judgment. This solves the drawbacks of strong subjectivity and delayed response in manual operation, achieving standardization and intelligentization of pelletizing condition judgment and parameter adjustment, significantly reducing reliance on human experience and lowering labor costs. The adjustment signals output by the model can directly drive the pelletizing machine control system to complete parameter adjustments, realizing full-process automation from data acquisition, feature analysis, working condition judgment to parameter adjustment. This significantly improves the response speed of parameter control, enabling rapid adaptation to changes in working conditions during the pelletizing process and effectively suppressing production fluctuations. It enhances the stability of pelletizing production and the quality of green pellets. Through precise perception of the core pelletizing area of ​​the pelletizing pan and intelligent decision-making by a deep learning model, process parameter adjustments are always matched to the actual production conditions of pelletizing. This effectively improves the pelletizing state within the pelletizing pan, increases the proportion of qualified green pellets, and reduces problems such as decreased pelletizing efficiency and increased defective products caused by errors in working condition judgment and improper parameter adjustment. Simultaneously, it lays the foundation for low-energy, high-quality operation in subsequent roasting processes, achieving a dual improvement in overall production efficiency and product quality in the steel metallurgical pelletizing production process.

[0017] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall process of the method of the present invention; Figure 2 This is a detailed flowchart illustrating some steps of the method of the present invention; Figure 3 This is a schematic diagram of the operation of the disc pelletizer of the present invention. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0022] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0023] Please refer to Figure 1 , Figure 3 A preferred embodiment of the present invention provides a pelletizing machine production control method, comprising steps S1, S2, S3, S4, S5 and S6.

[0024] S1 collects multimodal data during the pelletizing machine's operation. This multimodal data includes image data of the pelletizing area and the green pellet stabilization area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine. Specifically, image data of the core pelletizing area (pelletizing area + green pellet stabilization area) of the pelletizing disc can be collected at a frequency of 1Hz using a high-definition industrial color camera (e.g., resolution ≥ 2 million pixels, with protective cover and cleaning device). Simultaneously, the process operating parameters of the pelletizing machine are read in real-time from the DCS / PLC system via the OPCUA protocol. Both types of data carry timestamps, laying the foundation for subsequent data fusion. This achieves multi-dimensional, full-view data perception of the pelletizing process, providing a complete and original data source for subsequent intelligent analysis. Focusing on image collection in the pelletizing area and the green pellet stabilization area, meaningless regional data is eliminated, improving the targeting and efficiency of data collection.

[0025] like Figure 3As shown, the disc pelletizer includes a pelletizing disc, a pelletizing area, a pelletizing baffle, and a large green pellet conveyor belt. During operation, fine powdery material is fed into the pelletizing disc from the inlet point, where it rapidly nucleates and forms mother pellets under the action of dripping water. The mother pellets continuously roll as the pelletizing disc rotates, simultaneously receiving uniform wetting from atomized water, causing the fine powder to adhere layer by layer and gradually grow, ultimately forming qualified green pellets. During operation, a pellet-retrieval robot retrieves extra-large pellets in the stable pelletizing zone, ensuring the stability of the particle size distribution of the material on the disc surface.

[0026] S2. Image data is preprocessed to identify and mask interference areas caused by the robotic arm of the ball-collecting robot, classifying the image into fine powder areas, qualified unfinished balls, and overly wet areas to obtain visual features. First, Gaussian filtering and histogram equalization are performed on the image preprocessing. Then, the YOLO object detection algorithm is used to identify and mask the pixel areas of the robotic arm, eliminating equipment interference. Subsequently, the U-Net semantic segmentation model is used to classify the image into four categories. Based on the classification results, relevant data can be calculated, such as the proportion of fine powder areas, the proportion of overly wet areas, the particle size distribution of qualified unfinished balls, and the average roundness, among other core visual features. These can be integrated into visual features. It's important to understand that visual features such as particle size distribution, average roundness, the proportion of fine powder areas, and the proportion of overly wet areas are all percentage data, ranging from 0 to 1. Normalization is not required when inputting these into the control model. However, if the value range of the visual features is not between 0 and 1, normalization is necessary before using them as visual features. The shielding of the interference area formed by the robotic arm of the ball-collecting robot can effectively eliminate interference from on-site equipment and ensure the accuracy of image feature analysis; through semantic segmentation, the digitization and quantification of the ball-forming state are realized, transforming the blurry ball-forming state observed by the naked eye into machine-recognizable features, solving the problem of strong subjectivity in human judgment; visual features intuitively reflect the ball-forming effect of materials in the pan, providing the core basis for working condition judgment and parameter recommendation from a visual dimension.

[0027] S3 standardizes real-time process parameter data to obtain process parameters. The collected pelletizer process parameters (feed rate, atomized water volume, drip rate, rotational speed, tilt angle, main motor current, etc.) are Z-score standardized to eliminate the influence of dimensional differences and numerical ranges among different parameters, integrating the processed parameters into the overall process parameters. This addresses the model analysis bias caused by differences in units and numerical ranges of different process parameters, improving the accuracy and stability of subsequent deep learning model inference. The quantitative process parameters reflect the operating status of the pelletizer, providing core technological basis for condition determination and parameter adjustment.

[0028] S4 aligns visual features with process parameters using timestamps to form multimodal fusion features. Based on the image data acquisition timestamp, visual features and process parameters are spatiotemporally synchronized to ensure a one-to-one match between the sphere formation visual state and equipment operating parameters at the same time dimension, integrating them to form multimodal fusion features. This achieves deep fusion of visual features and process parameters, allowing the model to simultaneously combine sphere formation results and equipment operating causes for analysis and decision-making, overcoming the limitations of single-dimensional data analysis in existing technologies. Timestamp alignment ensures the spatiotemporal consistency of the fused data, providing a high-quality data foundation for accurate model inference. Multiple visual features can be integrated into a multidimensional vector, and multiple process parameters can also be integrated into a multidimensional vector, which can then be input into the control model. The multimodal fusion features can then be represented using these two multidimensional vectors.

[0029] S5, the multimodal fusion features are input into a pre-trained control model. The control model determines whether the current operating condition is abnormal. If so, it determines an adjustment strategy and outputs adjustment signals for process parameters. For abnormal operating conditions, based on the specific adjustment strategy and multimodal fusion features, the pre-trained control model outputs adjustment signals for process parameters (such as feed rate adjustment signal, atomized water rate adjustment signal, drip rate adjustment signal, rotation speed adjustment signal, and tilt angle adjustment signal). This distinguishes between abnormal and non-abnormal operating conditions and outputs targeted adjustment schemes, improving the accuracy and adaptability of parameter recommendations and avoiding blind adjustments. The control model can be a neural network deep learning model or other models that can be trained for input and output.

[0030] S6 outputs the process parameter adjustment signal to the pelletizing machine control system. The pelletizing machine control system adjusts the corresponding process parameters according to the adjustment signal. Specifically, the process parameter adjustment signal output from the model is sent to the pelletizing machine's DCS / PLC control system. The control system automatically adjusts parameters such as feed rate, water addition, rotation speed, and tilt angle of the pelletizing machine according to the adjustment signal. The adjustment amount is limited to ensure that parameter changes are within the safe range of the process. This achieves full automation from data acquisition, feature analysis, and operating condition determination to parameter adjustment, significantly improving the response speed of parameter control and enabling rapid response to changes in operating conditions such as sudden changes in raw material characteristics. It directly drives the equipment to adjust automatically, reducing manual intervention and dependence on operator experience, achieving standardized operation of pelletizing production; effectively suppressing production fluctuations and maintaining stable operation of the pelletizing process.

[0031] This invention provides a production control method for a pelletizing machine, achieving fusion perception and precise analysis of multimodal data. It overcomes the limitations of existing automatic control systems that rely on single-sensor local perception. By collecting image data of the pelletizing area and the stable green pellet area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine, and combining image preprocessing, pixel classification, and process parameter standardization, it comprehensively extracts visual and process features during the pelletizing process. Furthermore, it effectively shields the interference area of ​​the pellet-collecting robot's robotic arm, ensuring the comprehensiveness, accuracy, and effectiveness of feature extraction, providing complete and reliable data support for process parameter control. Relying on a control model, it achieves intelligent judgment of operating conditions and precise parameter recommendation. The model can effectively identify abnormal and non-abnormal operating conditions and output targeted process parameter adjustment signals, replacing traditional manual experience-based judgment. This solves the drawbacks of strong subjectivity and delayed response in manual operation, achieving standardization and intelligentization of pelletizing condition judgment and parameter adjustment, significantly reducing reliance on manual experience and lowering labor costs. By aligning visual features with process parameter timestamps to form multimodal fusion features, precise spatiotemporally synchronized data is provided for model inference. The adjustment signals output by the model can directly drive the pelletizing machine control system to complete parameter adjustments, realizing full-process automation from data acquisition, feature analysis, working condition judgment to parameter adjustment. This significantly improves the response speed of parameter control, enabling rapid adaptation to changes in working conditions during pelletizing production and effectively suppressing production fluctuations. It enhances the stability of pelletizing production and the quality of green pellets. Through precise perception of the core pelletizing area of ​​the pelletizing pan and intelligent decision-making by a deep learning model, process parameter adjustments are always matched to the actual production conditions of pelletizing. This effectively improves the pelletizing state within the pelletizing pan, increases the proportion of qualified green pellets, and reduces problems such as decreased pelletizing efficiency and increased defective products caused by errors in working condition judgment and improper parameter adjustment. Simultaneously, it lays the foundation for low-energy, high-quality operation in subsequent roasting processes, achieving a dual improvement in overall production efficiency and product quality in the steel metallurgical pelletizing production process. This method also has good versatility and adaptability. The process of image data preprocessing, feature extraction and model inference is standardized, which can be adapted to the production needs of different specifications of pelletizing machines. Furthermore, through continuous training and optimization of the model, the accuracy of working condition judgment and parameter recommendation can be continuously improved, providing technical support for the continuous optimization of pelletizing production.

[0032] It is understood that the control model needs to be pre-trained before applying this control method. This training process, based on historical successful operation data, specifically includes the following sub-steps: T1. Construct the training dataset: Obtain multiple sets of historical multimodal data from the historical database. For each set of historical multimodal data, record the changes in the pellet quality evaluation index within the subsequent time window. The pellet quality evaluation index includes one or more of the following: green pellet drop strength, compressive strength, and particle size distribution qualification rate. If the pellet quality significantly improves (e.g., drop strength increases by more than a preset threshold), then the set of historical multimodal data and its corresponding process parameter adjustment amount generated by expert operation are marked as a positive sample pair (input: historical multimodal data, label: expert adjustment amount); if the pellet quality does not improve or decreases, it is marked as a negative sample.

[0033] T2, Model Structure Design: Construct an initial deep learning model, which includes an input layer, a feature fusion layer, and an output layer. The input layer receives multimodal fused features, and the output layer outputs adjustments for each process parameter.

[0034] T3, Model Training: Using the positive sample pairs constructed in step T1, the initial deep learning model is trained with the optimization objective of minimizing the difference between the model's output adjustment and the expert's adjustment. The model parameters are iteratively updated using the backpropagation algorithm until the model converges, resulting in a trained control model. Optionally, negative samples can be introduced during training to enhance the model's ability to identify invalid operations.

[0035] In a specific embodiment of the present invention, step S5, determining whether the current operating condition is an abnormal operating condition, specifically includes at least one of the following steps: determining whether the current operating condition is an abnormal operating condition involves executing at least one of steps A1, A2, and A3: A1 obtains the rate of change of the percentage of pixels in the fine powder region and the percentage of pixels in the overly wet region within the effective analysis area. If either rate of change is greater than a first preset threshold, the current working condition is determined to be the first abnormal working condition. The time change rate of the percentage of pixels in the fine powder region and the percentage of pixels in the overly wet region in the visual features is calculated in real time and compared with the preset first threshold. If the rate of change of any indicator exceeds the standard, it is determined to be the first abnormal working condition, which can be regarded as a sudden change in material properties. This realizes the digital and accurate identification of working conditions with sudden changes in material properties, provides a clear basis for subsequent targeted adjustments, and improves the timeliness of handling abnormal working conditions.

[0036] It should be noted that the effective analysis area is the main area actually used for image acquisition and analysis. The effective analysis area includes the pelletizing area (effective pellet growing area) and the green pellet stabilization area (effective stabilization area) of the pelletizing disc. It does not cover the entire material movement area of ​​the pelletizing disc, but focuses on the core local area where material forming is most active and best reflects changes in working conditions. The effective analysis area (effective pellet growing area and effective stabilization area) can be manually defined according to the actual pelletizing process, or it can be automatically identified and calibrated using image recognition. Specifically, the pelletizing disc image is calibrated in step S521, assigning a pixel coordinate system to the effective area and determining the horizontal and vertical pixel boundaries, thereby locking in a main effective shooting and analysis area. After preprocessing to shield against interference from the pellet-collecting robot's robotic arm, the remaining image area is the effective analysis area. The total number of material pixels refers to the total number of pixels belonging to the material (including fine powder areas, qualified green pellets, and overly wet materials) within the effective analysis area, excluding background pixels. Based on this, the percentage of pixels in the fine powder region within the effective analysis region in this invention refers to the percentage of pixels classified as fine powder by the semantic segmentation model, which accounts for the total number of pixels of the material within the effective analysis region; the percentage of pixels in the overly wet region within the effective analysis region refers to the percentage of pixels classified as overly wet, which accounts for the total number of pixels of the material within the effective analysis region.

[0037] A2 determines whether there are one or more local areas where the humidity is abnormally rising. If so, the current operating condition is determined to be the second abnormal operating condition. This enables accurate identification of local humidity anomalies, solves the problem that the overall analysis cannot detect local moisture anomalies, realizes early warning of the snowball effect, and prevents problems before they occur. It also establishes a clear judgment logic for local humidity anomalies, improving the pertinence of anomaly handling.

[0038] A3 identifies whether the number of extra-large pellets appearing in the edge area of ​​the pelletizing tray within a preset time period exceeds a second preset threshold. If so, the current working condition is determined to be a third abnormal working condition. Extra-large pellets are those with a particle size larger than the third preset threshold. Machine vision identifies the particle size of pellets in the edge area of ​​the pelletizing tray and counts the number of extra-large pellets exceeding the third threshold within a preset time period. If the number exceeds the second preset threshold, it is determined to be a third abnormal working condition, i.e., the pellet retrieval robot is malfunctioning. The particle size threshold for extra-large pellets (e.g., ≥30mm) can be adjusted according to the on-site process. This achieves automatic identification of abnormal working conditions of the pellet retrieval robot, determining equipment malfunctions based on the number of extra-large pellets in the edge area. The determination criteria closely match the actual on-site production and are highly accurate. Timely identification of equipment malfunctions prevents the accumulation of extra-large pellets from affecting pelleting quality and production efficiency. The edge area can be manually defined or automatically defined by image recognition, with the area within a set distance from the boundary of the pelletizing tray designated as the edge area.

[0039] The system clearly categorizes and accurately identifies abnormal operating conditions, classifying common abnormal conditions in the pelletizing process into three main categories: sudden changes in material properties, abnormal increases in localized over-wetting, and abnormal operation of the pellet-collecting robot. Quantitative judgment criteria for each condition are established, achieving standardized, digitalized, and precise identification of abnormal conditions. This solves the problem that existing technologies cannot effectively identify and classify abnormal conditions within the pelletizing pan. Furthermore, the clear condition judgment logic provides a clear basis for the formulation of subsequent targeted adjustment strategies, avoiding blind parameter adjustments and improving the response efficiency and adaptability of parameter control under abnormal conditions.

[0040] Reference Figure 2 In a specific embodiment of the present invention, it is determined whether there are one or more local areas that are abnormally wet local areas: if so, it is determined to be a second abnormal working condition, specifically including steps S521, S522, S523, S524, S525, S526, and S527.

[0041] S521 marks the effective analysis areas of the pelletizing zone and the green pellet stabilization zone on the pelletizing disc, assigning a pixel coordinate system to the effective areas and determining the horizontal and vertical pixel boundaries. Accurately marking the effective analysis areas of the pelletizing zone and the green pellet stabilization zone establishes a pixel coordinate system for these areas, clearly defining the horizontal and vertical pixel start and end boundaries to ensure the uniqueness of the subsequent analysis range. Further refining the effective analysis area eliminates interference and invalid areas, ensuring the accuracy of local analysis; the pixel coordinate system provides a spatial reference for subsequent meshing and local unit positioning, achieving precise division of local areas. The pelletizing zone and the green pellet stabilization zone can be manually defined according to the actual pelletizing process, or they can be automatically identified using image recognition.

[0042] S522 divides the calibrated effective analysis area into multiple local units using a proportionally proportional rectangular grid. The grid size is an adjustable parameter (e.g., 8×8 pixels, 10×10 pixels), with each grid being an independent local analysis unit. There is no pixel overlap or omission of effective areas between units. Effects: By dividing the overall area into multiple local units, refined local analysis of the spherical state of the spherical disk is achieved, solving the problem of the overall analysis failing to detect local anomalies. The proportional division ensures fairness in the analysis of each local unit, and the adjustable grid size improves the method's adaptability to the field.

[0043] S523 performs real-time statistics on the total number of pixels in the over-wetted areas of each local unit, calculates the proportion of over-wetted areas in each local unit, and obtains the global proportion of over-wetted areas in the effective analysis area. It counts the number of pixels in the over-wetted areas of each local unit and calculates the over-wetted proportion of each unit (number of over-wetted pixels / total number of material pixels in the unit). The total number of material pixels in each local unit is not a fixed value because it may mask pixels from the robotic arm and other non-balling areas and green ball stabilization areas. This achieves quantitative statistics on the over-wetted state of each local unit and the overall area, providing accurate numerical basis for local anomaly detection; it also obtains the local and global over-wetted proportions, laying the data foundation for subsequent dual-increase determination.

[0044] S524, determine whether the number of pixels in the over-humidified area of ​​each local unit is greater than the minimum over-humidified pixel count threshold: if yes, the local unit is determined to be in a valid analysis state; if not, the local unit is determined to be in an invalid analysis state. Compare the number of over-humidified pixels in each local unit with the preset minimum over-humidified pixel count threshold. If the pixel count meets the threshold, it is a valid analysis state; otherwise, it is an invalid analysis state. This avoids significant fluctuations due to excessively low over-humidified pixel counts, which could lead to subsequent misjudgments, and avoids classifying meaningless pixel fluctuations as anomalies, thus improving the accuracy of local anomaly detection.

[0045] S525, determine whether each local unit is in a valid analysis state within a preset time period: if yes, the local unit is determined to be a valid analysis unit; otherwise, the local unit is determined to be an invalid analysis unit. The analysis state of each local unit is verified over time. If it remains in a valid analysis state for the preset time period, it is determined to be a valid analysis unit; otherwise, it is an invalid analysis unit. Valid analysis units are further filtered from a time perspective to ensure the stability and reliability of local anomaly detection.

[0046] S526, determine whether the increase in the proportion of over-humidified areas in each effective analysis unit within a preset time period is greater than a first preset increase threshold: if yes, determine that the effective analysis unit is a local area of ​​abnormally rising over-humidity; if no, proceed to step S527. Calculate the absolute increase in the proportion of over-humidity in the effective analysis unit within the preset time period and compare it with the first preset increase threshold. If the increase exceeds the threshold, it is directly determined as a local area of ​​abnormally rising over-humidity. This achieves the determination of the absolute increase of local over-humidity anomalies, covering typical scenarios where the proportion of local over-humidity rises rapidly. The determination logic is simple and the response is fast.

[0047] S527, determine whether the increase in the difference between the proportion of over-humidified areas in each effective analysis unit and the global proportion of over-humidified areas within a preset time period is greater than a second preset increase threshold: if so, determine that the effective analysis unit is a local area of ​​abnormally rising over-humidity. Calculate the difference between the proportion of over-humidified areas in the effective analysis unit and the global proportion of over-humidity, and statistically analyze the increase in this difference within a preset time period. If the increase exceeds the second preset threshold, it is determined to be a local area of ​​abnormally rising over-humidity. This achieves the determination of the relative deviation increase of local over-humidity anomalies, covering hidden scenarios where the global proportion of over-humidity is normal but the local proportion of over-humidity is rapidly increasing, compensating for the shortcomings of absolute increase determination; the dual determination logic significantly improves the comprehensiveness and accuracy of the determination.

[0048] This paper proposes a refined and hierarchical method for identifying areas of localized excessive moisture. Through a complete process design including effective analysis area calibration, grid segmentation, effective analysis unit screening, and dual amplification threshold determination, it achieves pixel-level precise positioning and effective anomaly identification of localized excessive moisture anomalies. The grid segmentation solves the problem of the overall analysis of the pelletizing tray failing to detect localized moisture anomalies. Furthermore, the dual screening of minimum excessive moisture pixel number threshold and preset duration of effective analysis status eliminates invalid fluctuation misjudgments caused by image noise. Simultaneously, the dual determination logic of absolute amplification and relative global deviation amplification covers all scenarios of localized excessive moisture anomaly increases, making the determination results more accurate and more consistent with actual on-site production, providing a reliable basis for the early suppression of the "snowball" effect.

[0049] In a specific embodiment of the present invention, the process parameter data includes at least the feed rate, atomized water volume, drip rate, pelletizer speed, pelletizer tilt angle, and main motor current. The core acquisition dimensions of the real-time process parameters of the pelletizer are clearly defined, incorporating the feed rate, atomized water volume, drip rate, pelletizer speed, pelletizer tilt angle, and main motor current into the process parameter data acquisition scope. These six types of parameters are the core controllable process parameters affecting the pelletizing quality of the pelletizing disc, achieving comprehensive and core-dimensional perception of the pelletizer's process operation status. This solves the problem of existing technologies where sensors acquire only single parameters and cannot reflect the complete operating status of the pelletizer, providing comprehensive and crucial process parameter characteristics for the control model, ensuring the scientific nature of model inference and the effectiveness of parameter adjustment.

[0050] In a specific embodiment of the present invention, the adjustment strategy is determined as follows: if the current working condition is determined to be a first abnormal working condition, the adjustment strategy is to reduce the dripping water volume, increase the atomized water volume, and increase the pelletizer speed; reducing the dripping water volume avoids local over-wetting, increasing the atomized water volume achieves uniform moisture distribution, and increasing the speed enhances the rolling and dispersing effect of the material. A targeted composite adjustment strategy is formulated for sudden changes in material properties, which fits the characteristics of the on-site process and can quickly suppress pelleting abnormalities caused by changes in raw material properties, restoring pelleting stability; multi-parameter synergistic adjustment is more effective than single-parameter adjustment, effectively improving pelleting efficiency.

[0051] If the current operating condition is determined to be the second abnormal condition, the adjustment strategy is to reduce the dripping water volume and increase the pelletizer speed for a limited time. Reducing the dripping water volume controls moisture input at the source, while increasing the speed for a short period quickly breaks up locally adhered, excessively wet material through centrifugal force, preventing the snowball effect from spreading. This targeted approach addresses the localized excessive wetness problem, achieving early suppression of the snowball effect and preventing small-scale anomalies from developing into large-scale production failures. The short-term speed increase breaks up the adhesion while avoiding excessive impact on the overall pelletizing process, balancing anomaly handling and production stability. The limited-time increase in pelletizer speed can be understood as increasing the pelletizer speed for a preset period, followed by a return to the initial speed.

[0052] If the current operating condition is determined to be the third abnormal condition, the adjustment strategy is the conservative recommendation mode. For the third abnormal condition where the ball-collecting robot malfunctions, the model switches to the conservative recommendation mode, maintaining only minor adjustments to the existing process parameters of the ball-forming machine without making significant parameter adjustments, while simultaneously triggering an equipment malfunction alarm. This avoids further deterioration of ball quality due to significant parameter adjustments during equipment malfunctions, maximizing the basic stability of ball-forming production. The conservative mode also allows on-site personnel time to troubleshoot equipment faults, achieving coordination between process control and equipment maintenance. The conservative recommendation mode maintains the current operating parameters of the ball-forming machine and issues a warning message about the abnormal operation of the ball-collecting robot. It can be understood that this method first determines the adjustment strategy based on the type of abnormal condition. The adjustment strategy includes the type of parameters to be adjusted and the trend direction of parameter adjustment. The adjustment amount can be output based on the pre-trained control model and the input multimodal fusion features, outputting the adjustment amount of each required parameter. In other words, this method presets an adjustment strategy for the type of abnormal condition, avoiding blind parameter adjustments and enabling the control model to more accurately and quickly output the adjustment amount of each parameter, forming an adjustment signal for the process parameters. Of course, if it is determined to be a non-abnormal operating condition, then conventional control can be carried out based on the detected visual characteristics and process parameters. For example, when the proportion of the over-wet area is large, the amount of atomized water and the amount of dripping water can be reduced, and simple adjustment and control can be carried out.

[0053] The above steps make the adjustment of process parameters under abnormal conditions more professional and adaptable. The composite adjustment strategy for sudden changes in material properties, the rapid intervention strategy for local over-wetting, and the conservative control strategy for abnormalities in the ball-collecting robot are all in line with the process characteristics of each abnormal condition. They can quickly suppress the development of abnormal conditions and restore the normal production state of ball making. This solves the problems of existing technologies having no targeted abnormal handling strategies and poor parameter adjustment effects, and effectively reduces the impact of abnormal conditions on ball quality and production efficiency.

[0054] In a specific embodiment of the present invention, the adjustment strategy is to reduce the amount of water dripped and increase the speed of the pelletizing machine for a limited time, specifically including: Determine if there are adjacent areas of abnormally high humidity levels (a preset number). If so, the drip rate and pelletizer speed are adjusted using a higher-order preset adjustment strategy; otherwise, the drip rate and pelletizer speed are adjusted using a lower-order preset adjustment strategy. The adjustment amounts for drip rate and pelletizer speed under the higher-order preset adjustment strategy are greater than those under the lower-order preset adjustment strategy.

[0055] The system determines whether a preset number (e.g., 3 or more) of abnormal units are spatially adjacent. It quantifies the impact range of localized over-wetness anomalies, distinguishing between small-scale and large-scale localized anomalies, providing a basis for tiered adjustments. If a preset number of adjacent abnormal units exist (e.g., 3 connected areas of rapidly increasing over-wetness), it is determined to be a large-scale localized over-wetness, and a higher-order adjustment (a larger reduction in drip rate and a larger increase in rotation speed) is executed. If it is a single abnormal unit, it is determined to be a small-scale localized over-wetness, and a lower-order adjustment (a small reduction in drip rate and a small increase in rotation speed) is executed. The adjustment range is a field-configurable parameter. This enables tiered and gradient control of localized over-wetness anomalies, solving the problem of over-adjustment for small-scale anomalies and under-adjustment for large-scale anomalies with a single adjustment range. It ensures that the parameter adjustment range matches the anomaly's impact range, rapidly dispersing over-wet, sticky materials and suppressing the snowball effect while maximizing the stability of the overall process state of the pelletizing disc, further improving the precision and field adaptability of parameter control.

[0056] In a specific embodiment of the present invention, determining whether the current operating condition is an abnormal operating condition further includes: obtaining the proportion of the over-wet area and the proportion of the fine powder area on the pelletizing disc surface; if the proportion of the over-wet area exceeds a fourth preset threshold or the proportion of the fine powder area exceeds a fifth preset threshold, then determining whether the deviation between the water volume setting and the feedback value (the water volume feedback value can be obtained by detecting the water volume through an electromagnetic flowmeter installed on the water supply pipe) is greater than a sixth preset threshold; if so, then the current operating condition type is determined to be the fourth abnormal operating condition, and the adjustment strategy corresponding to the fourth abnormal operating condition is: to adjust the water volume and trigger a targeted alarm. Water volume adjustment is usually performed within a short preset time period (e.g., within 3 seconds). When the proportion of the fine powder area is too large, the water volume can be quickly increased; when the over-wet area is large, the water volume can be quickly reduced or even the water supply can be stopped first.

[0057] By employing a dual-judgment logic that distinguishes between moisture changes caused by raw material characteristics and moisture imbalances caused by equipment malfunctions in the water supply system, the system effectively differentiates between moisture variations caused by raw material characteristics and moisture imbalances caused by equipment malfunctions in the water supply system. Furthermore, the system incorporates strong intervention measures for water supply system imbalances and targeted alarm handling strategies (such as "Please check the nozzle or water pressure"), enabling both emergency intervention of process parameters and timely alerts to on-site personnel to troubleshoot equipment malfunctions. This addresses water supply system imbalances from both process adjustment and equipment maintenance perspectives, preventing continuous abnormalities in pelletizing production caused by equipment failures.

[0058] The present invention also provides a pelletizing machine control system, comprising: The data acquisition module collects multimodal data during the pelletizing machine's operation. This multimodal data includes image data of the pelletizing area and the green pellet stabilization area on the pelletizing disc, as well as real-time process parameter data of the pelletizing machine. It integrates a high-definition industrial camera and an OPCUA data acquisition interface. The camera is responsible for acquiring image data, and the data interface is responsible for reading process parameter data. The module adds second-level timestamps to all acquired data, achieving unified acquisition and storage of multimodal data. The image processing module preprocesses the image data, identifies and filters out interference areas caused by the pellet-collecting robot's robotic arm, and classifies the images into fine powder areas, qualified green pellets, and over-wet areas to obtain visual features. The data processing module standardizes the real-time process parameter data to obtain the process parameters. It incorporates a Z-score standardization algorithm to automatically eliminate dimensions and normalize the acquired process parameters, integrating the processed parameters into the output process parameters. The feature fusion module is used to align visual features with process parameters using timestamps to form multimodal fusion features. The model input-output module is used to input the multimodal fusion features into a pre-trained control model, which determines whether the current operating condition is abnormal. If so, it determines the adjustment strategy and outputs the adjustment signal of the process parameters. The output adjustment module is used to output the adjustment signal of the process parameters to the pelletizing machine control system, which adjusts the corresponding process parameters according to the adjustment signal.

[0059] The present invention also provides an electronic device, including a processor and a memory, wherein the memory is used to store program code and transmit the program code to the processor; the processor is used to execute a pelletizing machine production control method according to the instructions in the program code.

[0060] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements a pelletizing machine production control method.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A production control method for a pelletizing machine, characterized in that, Includes the following steps: S1, Collect multimodal data during the operation of the pelletizing machine. The multimodal data includes image data of the pelletizing area and the stable green pellet area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine. S2, preprocess the image data, identify and block the interference area formed by the ball-collecting robot's robotic arm, classify the image into fine powder area, qualified raw ball, and overly wet area, and obtain visual features; S3, standardize the real-time process parameter data to obtain process parameters; S4, align the visual features with the process parameters using timestamps to form multimodal fusion features; S5, the multimodal fusion features are input into the pre-trained control model, and the control model determines whether the current operating condition is an abnormal operating condition: if so, the adjustment strategy is determined and the adjustment signal of the process parameters is output. S6 outputs the process parameter adjustment signal to the pelletizing machine control system, and the pelletizing machine control system adjusts the corresponding process parameters according to the process parameter adjustment signal.

2. The pelletizing machine production control method according to claim 1, characterized in that, The determination of whether the current operating condition is abnormal includes at least one of the following situations: The rate of change of the percentage of pixels in the fine powder area and the rate of change of the percentage of pixels in the overly wet area are obtained. If one of the rates of change is greater than the first preset threshold, the current working condition is determined to be the first abnormal working condition. Determine if there are one or more local areas that are abnormally wet and rising: if so, determine the current operating condition type as the second abnormal operating condition; The system identifies whether the number of extra-large pellets appearing in the edge area of ​​the pelletizing disc within a preset time period is greater than a second preset threshold. If so, the current working condition is determined to be a third abnormal working condition. The extra-large pellets are pellets with a particle size greater than the third preset threshold.

3. The pelletizing machine production control method according to claim 2, characterized in that, The determination of whether one or more local areas are abnormally wet and rising: if so, it is determined to be the second abnormal operating condition, specifically including: S521 marks the effective analysis area of ​​the pelletizing area and the stable area of ​​the raw pellets on the pelletizing plate, assigns a pixel coordinate system to the effective area, and determines the horizontal and vertical pixel boundaries; S522, the effective analysis area after calibration is divided into multiple local units by proportionally dividing the grid; S523, in real time, counts the total number of pixels in the over-wet area of ​​each local unit, calculates the proportion of the over-wet area in each local unit, and obtains the global proportion of the over-wet area in the effective analysis area. S524, determine whether the number of pixels in the over-wetted area of ​​each local unit is greater than the minimum over-wetted pixel count threshold: If so, the local unit is determined to be in a valid analysis state; If not, the local element is determined to be in an invalid analysis state; S525, determine whether each local element is in a valid analysis state within a preset time period: If so, the local cell is determined to be a valid analysis cell; If not, the local element is determined to be an invalid analysis element; S526, determine whether the increase in the proportion of over-humidified areas of each effective analysis unit within a preset time period is greater than a first preset increase threshold: If so, the effective analysis unit is determined to be the local area of ​​abnormally high humidity. If not, proceed to step S527; S527, determine whether the increase in the difference between the proportion of over-wetted areas in each effective analysis unit and the global proportion of over-wetted areas within a preset time period is greater than a second preset increase threshold: If so, the effective analysis unit is determined to be the local area of ​​abnormally high humidity.

4. The pelletizing machine production control method according to claim 3, characterized in that, The process parameter data includes at least the feed rate, atomized water volume, drip rate, pelletizer speed, pelletizer tilt angle, and main motor current.

5. The pelletizing machine production control method according to claim 4, characterized in that, The determination of the adjustment strategy specifically includes: If the current operating condition is determined to be the first abnormal operating condition, the adjustment strategy is to reduce the dripping water volume, increase the atomized water volume, and increase the pelletizer speed. If the current operating condition is determined to be the second abnormal operating condition, the adjustment strategy is to reduce the dripping water volume and increase the pelletizing machine speed for a limited time. If the current working condition is determined to be the third abnormal working condition, the adjustment strategy is the conservative recommendation mode, which maintains the current operating parameters of the ball-making machine and issues an abnormal warning message for the ball-collecting robot.

6. The pelletizing machine production control method according to claim 5, characterized in that, The adjustment strategy involves reducing the drip rate and increasing the pelletizing machine speed for a limited time, specifically including: Determine if there are any adjacent areas of abnormally high humidity levels (preset number): If so, the adjustment of the drip rate and the speed of the pelletizing machine will follow a high-order preset adjustment strategy; If not, the adjustment of the drip rate and the speed of the pelletizing machine will follow a low-order preset adjustment strategy. The adjustment amounts of the higher-order preset adjustment strategy for the drip rate and the pelletizing machine speed are greater than those of the lower-order preset adjustment strategy.

7. The pelletizing machine production control method according to claim 2, characterized in that, The determination of whether the current operating condition is abnormal includes: obtaining the proportion of the over-wet area and the proportion of the fine powder area on the pelletizing tray; if the proportion of the over-wet area exceeds the fourth preset threshold or the proportion of the fine powder area exceeds the fifth preset threshold, then determining whether the deviation between the water volume setting and the feedback value is greater than the sixth preset threshold; if so, then determining the current operating condition type as the fourth abnormal operating condition, and the adjustment strategy corresponding to the fourth abnormal operating condition is: if the proportion of the over-wet area exceeds the fourth preset threshold, then reduce the dripping water volume; if the proportion of the fine powder area exceeds the fifth preset threshold, then increase the dripping water volume; and simultaneously triggering a targeted alarm containing fault location information.

8. A pelletizing machine production control system, characterized in that, include: The acquisition module is used to acquire multimodal data during the operation of the pelletizing machine. The multimodal data includes image data of the pelletizing area and the green pellet stabilization area of ​​the pelletizing disc, as well as real-time process parameter data of the pelletizing machine. The image processing module is used to preprocess the image data, identify and block the interference area formed by the robotic arm of the ball-collecting robot, classify the image into fine powder area, qualified raw ball, and excessively wet area, and obtain visual features. The data processing module is used to standardize the real-time process parameter data to obtain process parameters; The feature fusion module is used to align the visual features with process parameters using timestamps to form multimodal fusion features; The model input / output module is used to input the multimodal fusion features into a pre-trained control model. The control model determines whether the current operating condition is an abnormal operating condition. If so, it determines the adjustment strategy and outputs the adjustment signal of the process parameters. The output adjustment module is used to output the adjustment signal of the process parameters to the pelletizing machine control system, and the pelletizing machine control system adjusts the corresponding process parameters according to the adjustment signal.

9. An electronic device, characterized in that, The device includes a processor and a memory, wherein the memory is used to store program code and transmit the program code to the processor; the processor is used to execute the pelletizing machine production control method according to any one of claims 1-7 according to the instructions in the program code.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the pelletizing machine production control method as described in any one of claims 1 to 7.

Images