A real-time feedback control system for melt mixing parameters in masterbatch production

Abstract

This invention relates to the field of control system technology, specifically a real-time feedback control system for melt blending parameters in masterbatch production. The invention obtains the corresponding process parameter baseline through a parameter baseline setting module; performs risk assessment through a grading module to obtain risk levels and handle them accordingly; constructs a global optimization model through a control sequence module; and initiates a target evolutionary algorithm based on the process parameter baseline. By analyzing the dynamic coupling correlation between various process parameters and their comprehensive impact on the whole, the adjustment amount of the process parameters is calculated and a control sequence is generated. Finally, the process parameters are adjusted through a closed-loop control module, forming a real-time closed-loop feedback control. Because the melt blending process of masterbatch has shortcomings in the real-time feedback and precise control of parameters, this invention proposes a real-time feedback control system for melt blending parameters in masterbatch production.

Description

Technical Field

[0001] This invention relates to the field of control system technology, specifically to a real-time feedback control system for melt mixing parameters in the production of color masterbatch. Background Technology

[0002] Currently, the feedback and control of melt blending parameters in masterbatch production are mainly achieved through a programmable logic controller (PLC). The PLC is essentially a programmable logic controller, whose core function is to execute pre-programmed control procedures. Before starting the PLC, process engineers pre-define the operational sequence, logical judgment conditions, and target values ​​for key process parameters in the control program. The PLC also integrates various sensors such as thermocouples, pressure transmitters, encoders, and precision weighing instruments for continuous monitoring and real-time data acquisition of actual process variables. This acquired real-time data forms the basis for the closed-loop feedback loop: some PLCs utilize integral-derivative control algorithms to compare real-time data with preset target values ​​in the control program and calculate the deviation. When the deviation exceeds a preset deviation range, an alarm is triggered, ensuring that the melt blending parameters in masterbatch production are controlled. The control program itself is programmed by process engineers based on material characteristics and product parameter requirements. When using the PLC, process engineers monitor the melt blending status of the entire masterbatch production process through a human-machine interface, and simultaneously calculate the parameter deviation between real-time and historical data, adjusting the parameters accordingly. Process technicians can also execute advanced operation commands such as starting, pausing, and terminating the control program. The program control system typically has data logging capabilities, automatically saving preset core operating parameters and production event data. This provides data support for subsequent problem analysis, problem tracing, and iterative optimization of the control program. Standard operating procedures are always formulated around the correct operation of the program control system, ensuring the stable execution of program instructions. Finally, the inspection results of the finished product quality serve as a judgment on the effectiveness of the control program's execution and are also important data guiding further improvement of the program control system.

[0003] The melt-blending process for color masterbatch suffers from deficiencies in real-time parameter feedback and precise control, leading to delayed process adjustments and fluctuations in product quality uniformity. Therefore, a real-time feedback control system for melt-blending parameters in color masterbatch production is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time feedback control system for melt blending parameters in masterbatch production. This invention obtains the corresponding process parameter baseline through a parameter baseline setting module; performs risk assessment through a grading module to obtain risk levels and handle them accordingly; constructs a global optimization model through a control sequence module; and initiates a target evolutionary algorithm based on the process parameter baseline. By analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole, the adjustment amount of the process parameters is calculated and a control sequence is generated. Finally, the process parameters are adjusted through a closed-loop control module, forming a real-time closed-loop feedback control. Because the melt blending process of masterbatch has shortcomings in the real-time feedback and precise control of parameters, this invention proposes a real-time feedback control system for melt blending parameters in masterbatch production.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The parameter baseline setting module acquires material characteristic data and performs hierarchical intelligent retrieval of historical material data through the process knowledge base and matching engine to determine the corresponding process parameter baseline.

[0007] The grading module acquires real-time data during the masterbatch production process; it calls the corresponding material prediction based on the material characteristic data to generate status data and prediction data; the status data is combined with the real-time data through a data assimilation algorithm to generate a real-time digital image; the real-time digital image is compared with the historical status to obtain deviation data; the deviation data is combined with the prediction data to conduct risk assessment, obtain the risk level, and perform grading processing.

[0008] The control sequence module and control unit integrate the deviation data, the prediction data, and the risk level, and construct a global optimization model based on preset multiple optimization objectives; and combine the process parameter benchmark to start the target evolution algorithm, calculate the process parameter adjustment amount and generate the control sequence by analyzing the dynamic coupling relationship between each process parameter and its comprehensive impact on the whole.

[0009] The closed-loop control module uses the control sequence to adjust the process parameters, thus forming a real-time closed-loop feedback control.

[0010] Furthermore, the process of obtaining the corresponding process parameter reference includes:

[0011] The matching engine receives material characteristic data, performs precise matching on the material characteristic data, finds completely consistent historical material data, and retrieves the successful process parameters corresponding to the historical material data as the process parameter benchmark.

[0012] If the precise matching fails, fuzzy matching is initiated. By calculating the similarity between the material characteristic data and all historical material data, the closest cases are found. Based on the similarity between the cases and the material characteristic data, corresponding weights are assigned, and the historical successful process parameters corresponding to the cases are weighted and averaged to form preliminary recommended parameters. The machine learning model built into the matching engine is called to optimize the preliminary recommended parameters and generate the process parameter benchmark.

[0013] Furthermore, the process of generating state data and prediction data is as follows:

[0014] The prediction model library calls the corresponding prediction model based on material characteristic data; the prediction model estimates the internal physicochemical state parameters in the current melting and mixing process in real time through time series analysis algorithms, and extracts the feature data of the internal physicochemical state parameters to generate state data; the prediction model predicts the evolution trend of the internal physicochemical state parameters in a future preset time window based on the historical evolution patterns recorded internally, and generates prediction data.

[0015] Furthermore, the process of obtaining the risk level and performing classification is as follows:

[0016] The risk quantification unit adaptively adjusts the instantaneous amplitude, cumulative amount, and rate of change of the deviation data, as well as the weights of the future deterioration probability, evolution speed, and failure impact value of the predicted data, based on the trend of the rate of change of the deviation data and the fluctuation of the probability of deterioration of the predicted data. It then uses a nonlinear fusion algorithm to quantify the process risk and generate a quantified risk value.

[0017] The threshold optimization unit dynamically adjusts the threshold used to map the quantified risk value to the risk level based on the error situation of historical risk level classification, and establishes a dynamic evaluation threshold library.

[0018] The operation response procedure selection unit dynamically selects candidate operation response procedures from the operation response procedure library based on the risk level using a reinforcement learning algorithm.

[0019] The process of dynamically selecting candidate operation response procedures is as follows:

[0020] When a high-risk level is identified, an alarm is triggered immediately, and the melting and mixing equipment is switched to the predetermined safe operating mode.

[0021] When a medium-risk level is identified, the monitoring frequency of key process parameters is adjusted for more intensive observation, and early warning information and handling suggestions are pushed to the operation terminal.

[0022] When a low-risk level is identified, the current deviation data and risk assessment information are recorded and archived for process trend analysis, and the existing control strategy is maintained.

[0023] The candidate operation response procedure is submitted to the operation response procedure verification unit. Using a real-time updated digital twin model, the effect of the candidate operation response procedure under various production scenarios is simulated, and verification results are generated to evaluate its rationality. If the verification results show that the candidate operation response procedure is reasonable, the control unit executes the candidate operation response procedure to adjust the process parameters. If the verification results show that the candidate operation response procedure is unreasonable, an alternative operation response procedure is selected from the operation response procedure library and verified again to determine the final operation response procedure.

[0024] The risk assessment model is used to analyze the instantaneous amplitude, cumulative amount, and rate of change of deviation data, and to analyze the future deterioration probability, evolution speed, and failure impact value contained in the predicted data. Finally, a nonlinear fusion algorithm is used to quantify the risks existing in the process, obtaining a quantified risk value, which is then mapped to multiple predefined risk levels. Differentiated operational response procedures are triggered based on the different risk levels. These differentiated operational response procedures include:

[0025] When a high-risk level is identified, an alarm is triggered immediately, and the melting and mixing equipment is switched to the predetermined safe operating mode.

[0026] When a medium-risk level is identified, the monitoring frequency of key process parameters is adjusted for more intensive observation, and early warning information and handling suggestions are pushed to the operation terminal.

[0027] When a low-risk level is identified, the current deviation data and risk assessment information are recorded and archived for process trend analysis, and the existing control strategy is maintained.

[0028] Furthermore, the global optimization model is as follows:

[0029] The global optimization model includes a comprehensive objective function representing the overall performance of the melt blending system and a set of hard and soft constraints defining the safety boundaries of the process operation. The comprehensive objective function is used to integrate and quantify the optimal balance between product color difference index, pigment dispersion uniformity index, unit product energy consumption index, and equipment load balance index. The hard and soft constraints include the upper and lower limits of temperature in each section of the melt blending equipment, the screw speed range, and the melt pressure safety threshold.

[0030] Furthermore, the process of calculating the adjustment amount of the process parameters and generating the control sequence is as follows:

[0031] The objective evolution algorithm searches within the feasible solution space defined by the global optimization model, seeking the optimization direction by achieving a preset value for the comprehensive objective function, and strictly adhering to hard and soft constraints. It utilizes multi-generation iterative selection and crossover and mutation operations to calculate the adjustment amount of process parameters and generate control sequences by analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole.

[0032] Furthermore, the real-time closed-loop feedback control is as follows:

[0033] The control sequence is transmitted to the bottom control unit of the melt mixing equipment and drives the corresponding actuators to precisely adjust the set value of at least one process parameter in the production process. After the process parameter is adjusted, the grading module repeats the process of acquiring real-time data, generating status data and predicted data, generating real-time digital images, acquiring deviation data, and obtaining risk levels and performing grading. It then performs the operation of feeding back the deviation data, the predicted data, and the risk level to the control sequence module again to calculate a new round of process parameter adjustment and generate a new control sequence, thus forming a real-time closed-loop feedback control.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. This invention utilizes its real-time sensor data processing capabilities and employs a prediction model tailored to material characteristics for online state estimation and future trend prediction, generating a real-time digital image containing key process information. This mechanism overcomes the shortcomings of existing technologies, such as the lack of real-time online insight into the core quality status of production and the lag in control response. The system can further conduct forward-looking risk assessments based on this digital image and prediction data, and initiate corresponding graded processing procedures. This effectively anticipates and proactively intervenes in potential process fluctuations and possible product quality problems, significantly reducing batch production defects caused by untimely problem detection, ensuring the uniformity and stability of product quality, which is particularly significant for color masterbatch products with stringent quality requirements. It also compensates for the deficiencies in real-time parameter feedback and precise control in the color masterbatch melt mixing process.

[0036] 2. To address the problem in existing technologies where process parameter settings are often isolated and difficult to achieve global optimization, this invention employs a control engine for global optimization. First, the control engine integrates deviation data, prediction data, and risk level assessment results, and constructs a global optimization model based on multiple optimization objectives. Simultaneously, through the control engine's built-in goal evolution algorithm, it deeply analyzes the dynamic coupling relationships between various process parameters and their comprehensive impact on overall production performance, calculating the adjustment amounts of process parameters and generating control sequences. This collaborative control approach based on global considerations improves upon the previous piecemeal adjustment model, compensating for the deficiencies in real-time parameter feedback and precise control in the masterbatch melt mixing process.

[0037] 3. This invention addresses the over-reliance on operator experience in the melt-blending parameter control of color masterbatch production through hierarchical intelligent retrieval. During the hierarchical intelligent retrieval process, the system first receives material characteristic data through a matching engine, performs precise matching, and generates a baseline process parameter. If precise matching fails, fuzzy matching is initiated to form preliminary recommended parameters. Simultaneously, the machine learning model built into the matching engine is invoked to optimize the preliminary recommended parameters, generating the final process parameter baseline. Through hierarchical intelligent retrieval, this invention improves the scientific rigor and accuracy of process parameter settings, ensuring stable product quality reproduction under different operator conditions, and resolving the problem of traditional processes over-reliance on operator experience. It also provides technical support for the efficient production of new formulas and materials, shortening the process development cycle. Furthermore, it compensates for the deficiencies in real-time parameter feedback and precise control in the melt-blending process of color masterbatch. Attached Figure Description

[0038] Figure 1 A schematic diagram of a real-time feedback control system for melt mixing parameters in the production of color masterbatch, provided in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of a real-time feedback control system for melt mixing parameters in the production of color masterbatch, provided by an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the hierarchical intelligent retrieval process provided in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0042] Example 1

[0043] To improve the production of a high-gloss black polyethylene masterbatch, a factory adopted a real-time feedback control system for melt mixing parameters in masterbatch production. This system includes four modules: a parameter baseline setting module, a grading module, a control sequence module, and a closed-loop control module. This embodiment will combine... Figure 1 , Figure 2 , Figure 3 This describes how the system uses these four modules to perform specific operations.

[0044] First, material characteristic data is obtained through the parameter benchmark setting module, and the historical material data is intelligently retrieved in a hierarchical manner through the process knowledge base and matching engine to determine the corresponding process parameter benchmark.

[0045] The hierarchical intelligent retrieval process mainly consists of:

[0046] The matching engine receives material characteristic data, performs precise matching on the material characteristic data, finds completely consistent historical material data, and retrieves the successful process parameters corresponding to the historical material data as the process parameter benchmark.

[0047] If the precise matching fails, fuzzy matching is initiated. By calculating the similarity between the material characteristic data and all historical material data, the closest cases are found. Based on the similarity between the cases and the material characteristic data, corresponding weights are assigned, and the historical successful process parameters corresponding to the cases are weighted and averaged to form preliminary recommended parameters. The machine learning model built into the matching engine is called to optimize the preliminary recommended parameters and generate the process parameter benchmark.

[0048] The parameter baseline setting module optimizes the process of acquiring process parameter baselines through hierarchical intelligent retrieval. Hierarchical intelligent retrieval is mainly divided into two levels: precise matching and fuzzy matching. Precise matching enables rapid retrieval of mature operating conditions, ensuring the stability of the production process. Fuzzy matching is initiated when precise matching fails. Fuzzy matching first generates preliminary recommended parameters, which are then optimized using a machine learning model to generate reliable process parameter baselines. Hierarchical intelligent retrieval enables adaptive adjustment to raw material fluctuations, reducing the cost of manual trial and error.

[0049] Subsequently, the grading module is used to acquire real-time data during the masterbatch production process; based on the material characteristic data, the corresponding material prediction is called to generate status data and prediction data; the status data is combined with the real-time data through a data assimilation algorithm to generate a real-time digital image; the real-time digital image is compared with the historical status to obtain deviation data, and the deviation data is combined with the prediction data to conduct risk assessment, obtain the risk level, and perform grading processing.

[0050] The process of generating state data and prediction data is as follows:

[0051] The prediction model library calls the corresponding prediction model based on material characteristic data; the prediction model estimates the internal physicochemical state parameters in the current melting and mixing process in real time through time series analysis algorithms, and extracts the feature data of the internal physicochemical state parameters to generate state data; the prediction model predicts the evolution trend of the internal physicochemical state parameters in a future preset time window based on the historical evolution patterns recorded internally, and generates prediction data.

[0052] By invoking a predictive model adapted to the material properties and employing time series analysis algorithms to estimate the difficult-to-measure internal state in real time, the system gains a "penetrating" ability to understand the melting and mixing process. Predicting future trends based on historical patterns transforms the control system from a passive response to an active prediction, buying time for preventative adjustments and thus more effectively suppressing potential quality fluctuations and improving the consistency of the final product. While the predictive model can predict trends based on its internally recorded historical evolution patterns, the patterns it describes are relatively fixed, lacking a mechanism for online correction of the prediction logic based on real-time production data. In actual production, this can lead to a decrease in model accuracy due to factors such as equipment aging and environmental changes. To overcome this deficiency, an online correction step can be added: comparing the real-time mirrored state of the masterbatch production process in the grading module with the previous predicted state of the masterbatch production process, obtaining the prediction error, and then using a recursive least squares algorithm to update the relevant parameters of the internally recorded historical evolution patterns online, improving the accuracy of the prediction.

[0053] The process of obtaining risk levels and classifying them is as follows:

[0054] The risk quantification unit adaptively adjusts the instantaneous amplitude, cumulative amount, and rate of change of the deviation data, as well as the weights of the future deterioration probability, evolution speed, and failure impact value of the predicted data, based on the trend of the rate of change of the deviation data and the fluctuation of the probability of deterioration of the predicted data. It then uses a nonlinear fusion algorithm to quantify the process risk and generate a quantified risk value.

[0055] The threshold optimization unit dynamically adjusts the threshold used to map the quantified risk value to the risk level based on the error situation of historical risk level classification, and establishes a dynamic evaluation threshold library.

[0056] The operation response procedure selection unit dynamically selects candidate operation response procedures from the operation response procedure library based on the risk level using a reinforcement learning algorithm.

[0057] The process of dynamically selecting candidate operation response procedures is as follows:

[0058] When a high-risk level is identified, an alarm is triggered immediately, and the melting and mixing equipment is switched to the predetermined safe operating mode.

[0059] When a medium-risk level is identified, the monitoring frequency of key process parameters is adjusted for more intensive observation, and early warning information and handling suggestions are pushed to the operation terminal.

[0060] When a low-risk level is identified, the current deviation data and risk assessment information are recorded and archived for process trend analysis, and the existing control strategy is maintained.

[0061] The candidate operation response procedure is submitted to the operation response procedure verification unit. Using a real-time updated digital twin model, the effect of the candidate operation response procedure under various production scenarios is simulated, and verification results are generated to evaluate its rationality. If the verification results show that the candidate operation response procedure is reasonable, the control unit executes the candidate operation response procedure to adjust the process parameters. If the verification results show that the candidate operation response procedure is unreasonable, an alternative operation response procedure is selected from the operation response procedure library and verified again to determine the final operation response procedure.

[0062] This design enables dynamic quantification of potential risks, avoiding production disruptions caused by overreacting to low-risk events while ensuring timely and effective handling of medium- and high-risk events, thereby significantly enhancing the overall robustness, safety, and operational stability of the production process.

[0063] Then, the control sequence module is invoked, and the control unit integrates the deviation data, the prediction data, and the risk level to construct a global optimization model based on preset multiple optimization objectives. The target evolution algorithm is then initiated in conjunction with the process parameter benchmark. By analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole, the process parameter adjustment amount is calculated and a control sequence is generated.

[0064] The global optimization model is as follows:

[0065] The global optimization model includes a comprehensive objective function representing the overall performance of the melt blending system and a set of hard and soft constraints defining the safety boundaries of the process operation. The comprehensive objective function is used to integrate and quantify the optimal balance between product color difference index, pigment dispersion uniformity index, unit product energy consumption index, and equipment load balance index. The hard and soft constraints include the upper and lower limits of temperature in each section of the melt blending equipment, the screw speed range, and the melt pressure safety threshold.

[0066] The objective function can integrate and balance multiple, even conflicting, performance indicators such as product quality, energy consumption, and equipment load, overcoming the one-sidedness of traditional single-objective optimization. At the same time, the explicit hard and soft constraints ensure that all optimization adjustments are made within the boundaries of safe equipment operation, making the pursuit of the "global optimum" of the production process both clear in direction and practically feasible.

[0067] The process of calculating the adjustment amount of process parameters and generating the control sequence is as follows:

[0068] The objective evolution algorithm searches within the feasible solution space defined by the global optimization model, seeking the optimization direction by achieving a preset value for the comprehensive objective function, and strictly adhering to hard and soft constraints. It utilizes multi-generation iterative selection and crossover and mutation operations to calculate the adjustment amount of process parameters and generate control sequences by analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole.

[0069] The application of the objective evolutionary algorithm enables it to efficiently search a broad feasible solution space and handle the nonlinear dynamic coupling relationships between various process parameters. This method can discover and utilize the synergistic effects between parameters to find optimal parameter combinations that surpass traditional experience, ensuring that the generated control sequences are optimal and synergistic, thereby driving the system towards a better overall performance state in an efficient manner.

[0070] This method can be further improved by finding a non-dominated solution in the Pareto optimal solution set among multiple conflicting objective functions, thereby ensuring that the algorithm always searches in the direction that optimizes the overall system performance. This makes the pursuit of "global optimum" in the production process both clear in direction and practically feasible.

[0071] Finally, a closed-loop control module is used to adjust the process parameters using the control sequence, thereby forming a real-time closed-loop feedback control.

[0072] The control sequence is transmitted to the bottom control unit of the melt mixing equipment and drives the corresponding actuators to precisely adjust the set value of at least one process parameter in the production process. After the process parameter is adjusted, the grading module repeats the process of acquiring real-time data, generating status data and predicted data, generating real-time digital images, acquiring deviation data, and obtaining risk levels and performing grading. It then performs the operation of feeding back the deviation data, the predicted data, and the risk level to the control sequence module again to calculate a new round of process parameter adjustment and generate a new control sequence, thus forming a real-time closed-loop feedback control.

[0073] Through this continuous dynamic tracking and optimization, the system successfully maintained the color difference and dispersion of the high-gloss black masterbatch within the target range, effectively addressing the potential impact of raw material fluctuations. This ensures the system can instantly assess the control effect and respond quickly to any new disturbances or changes during the process, making the control behavior a dynamic process of continuous self-correction and optimization, thereby guaranteeing the stability and precision of the process state throughout the entire production cycle.

[0074] First, this design completely ignores the system's significant inertia and time delay. Second, it employs a single, synchronous control rhythm, failing to account for the vastly different time scales of fast variables like melt pressure and slow variables like barrel temperature, leading to untimely risk responses or control jitter in slow processes. Most critically, this mechanism is essentially a purely reactive "post-hoc" control mechanism, reliant on the occurrence of deviations. For masterbatch production, once quality deviations such as color difference occur, the product is already defective. This control logic cannot meet the demands of high-quality production and lacks the foresight and adaptive capability to handle disturbances.

[0075] A digital twin based on rheological and heat transfer physical models is constructed as a real-time "soft measurement" system to infer the "latent states" inside the melt mixing mill barrel that cannot be directly measured, such as the shear stress field, material residence time distribution, and melt viscosity distribution. This method enables the closed-loop control module to directly manage the physical nature of the production process, thereby proactively adapting to raw material fluctuations.

[0076] Example 2

[0077] To better produce a pearlescent white PETG masterbatch for cosmetic packaging, a factory adopted a real-time feedback control system for melt mixing parameters in masterbatch production. This embodiment will combine... Figure 1 , Figure 2 , Figure 3 A real-time feedback control system for melt mixing parameters in the production of color masterbatch is described in detail.

[0078] First, material characteristic data is obtained through the parameter benchmark setting module, and the historical material data is intelligently retrieved in a hierarchical manner through the process knowledge base and matching engine to determine the corresponding process parameter benchmark.

[0079] The hierarchical intelligent retrieval process mainly consists of:

[0080] The matching engine receives material characteristic data, performs precise matching on the material characteristic data, finds completely consistent historical material data, and retrieves the successful process parameters corresponding to the historical material data as the process parameter benchmark.

[0081] If the precise matching fails, fuzzy matching is initiated. By calculating the similarity between the material characteristic data and all historical material data, the closest cases are found. Based on the similarity between the cases and the material characteristic data, corresponding weights are assigned, and the historical successful process parameters corresponding to the cases are weighted and averaged to form preliminary recommended parameters. The machine learning model built into the matching engine is called to optimize the preliminary recommended parameters and generate the process parameter benchmark.

[0082] Subsequently, the grading module is used to acquire real-time data during the masterbatch production process; based on the material characteristic data, the corresponding material prediction is called to generate status data and prediction data; the status data is combined with the real-time data through a data assimilation algorithm to generate a real-time digital image; the real-time digital image is compared with the historical status to obtain deviation data, and the deviation data is combined with the prediction data to conduct risk assessment, obtain the risk level, and perform grading processing.

[0083] The process of generating state data and prediction data is as follows:

[0084] The prediction model library calls the corresponding prediction model based on material characteristic data; the prediction model estimates the internal physicochemical state parameters in the current melting and mixing process in real time through time series analysis algorithms, and extracts the feature data of the internal physicochemical state parameters to generate state data; the prediction model predicts the evolution trend of the internal physicochemical state parameters in a future preset time window based on the historical evolution patterns recorded internally, and generates prediction data.

[0085] The process of obtaining risk levels and classifying them is as follows:

[0086] The risk quantification unit adaptively adjusts the instantaneous amplitude, cumulative amount, and rate of change of the deviation data, as well as the weights of the future deterioration probability, evolution speed, and failure impact value of the predicted data, based on the trend of the rate of change of the deviation data and the fluctuation of the probability of deterioration of the predicted data. It then uses a nonlinear fusion algorithm to quantify the process risk and generate a quantified risk value.

[0087] The threshold optimization unit dynamically adjusts the threshold used to map the quantified risk value to the risk level based on the error situation of historical risk level classification, and establishes a dynamic evaluation threshold library.

[0088] The operation response procedure selection unit dynamically selects candidate operation response procedures from the operation response procedure library based on the risk level using a reinforcement learning algorithm.

[0089] The process of dynamically selecting candidate operation response procedures is as follows:

[0090] When a high-risk level is identified, an alarm is triggered immediately, and the melting and mixing equipment is switched to the predetermined safe operating mode.

[0091] When a medium-risk level is identified, the monitoring frequency of key process parameters is adjusted for more intensive observation, and early warning information and handling suggestions are pushed to the operation terminal.

[0092] When a low-risk level is identified, the current deviation data and risk assessment information are recorded and archived for process trend analysis, and the existing control strategy is maintained.

[0093] The candidate operation response procedure is submitted to the operation response procedure verification unit. Using a real-time updated digital twin model, the effect of the candidate operation response procedure under various production scenarios is simulated, and verification results are generated to evaluate its rationality. If the verification results show that the candidate operation response procedure is reasonable, the control unit executes the candidate operation response procedure to adjust the process parameters. If the verification results show that the candidate operation response procedure is unreasonable, an alternative operation response procedure is selected from the operation response procedure library and verified again to determine the final operation response procedure.

[0094] Then, the control sequence module is invoked, and the control unit integrates the deviation data, the prediction data, and the risk level to construct a global optimization model based on preset multiple optimization objectives. The target evolution algorithm is then initiated in conjunction with the process parameter benchmark. By analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole, the process parameter adjustment amount is calculated and a control sequence is generated.

[0095] The global optimization model is as follows:

[0096] The global optimization model includes a comprehensive objective function representing the overall performance of the melt blending system and a set of hard and soft constraints defining the safety boundaries of the process operation. The comprehensive objective function is used to integrate and quantify the optimal balance between product color difference index, pigment dispersion uniformity index, unit product energy consumption index, and equipment load balance index. The hard and soft constraints include the upper and lower limits of temperature in each section of the melt blending equipment, the screw speed range, and the melt pressure safety threshold.

[0097] The process of calculating the adjustment amount of process parameters and generating the control sequence is as follows:

[0098] The objective evolution algorithm searches within the feasible solution space defined by the global optimization model, seeking the optimization direction by achieving a preset value for the comprehensive objective function, and strictly adhering to hard and soft constraints. It utilizes multi-generation iterative selection and crossover and mutation operations to calculate the adjustment amount of process parameters and generate control sequences by analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole.

[0099] Finally, a closed-loop control module is used to adjust the process parameters using the control sequence, thereby forming a real-time closed-loop feedback control.

[0100] The control sequence is transmitted to the bottom control unit of the melt mixing equipment and drives the corresponding actuators to precisely adjust the set value of at least one process parameter in the production process. After the process parameter is adjusted, the grading module repeats the process of acquiring real-time data, generating status data and predicted data, generating real-time digital images, acquiring deviation data, and obtaining risk levels and performing grading. It then performs the operation of feeding back the deviation data, the predicted data, and the risk level to the control sequence module again to calculate a new round of process parameter adjustment and generate a new control sequence, thus forming a real-time closed-loop feedback control.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time feedback control system for melt mixing parameters in masterbatch production, characterized in that, include: The parameter baseline setting module acquires material characteristic data and performs hierarchical intelligent retrieval of historical material data through the process knowledge base and matching engine to determine the corresponding process parameter baseline. The grading module acquires real-time data during the masterbatch production process; Based on the material characteristic data, the corresponding material prediction is invoked to generate status data and prediction data; The state data is combined with real-time data using a data assimilation algorithm to generate a real-time digital image; The real-time digital mirror is compared with the historical state to obtain deviation data. The deviation data is combined with the prediction data to conduct risk assessment, obtain the risk level, and perform grading processing. The control sequence module and control unit integrate the deviation data, the prediction data, and the risk level, and construct a global optimization model based on preset multiple optimization objectives; and combine the process parameter benchmark to start the target evolution algorithm, calculate the process parameter adjustment amount and generate the control sequence by analyzing the dynamic coupling relationship between each process parameter and its comprehensive impact on the whole. The closed-loop control module uses the control sequence to adjust the process parameters, thus forming a real-time closed-loop feedback control.

2. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The process of obtaining the corresponding process parameter baseline includes: The matching engine receives material characteristic data, performs precise matching on the material characteristic data, finds completely consistent historical material data, and retrieves the successful process parameters corresponding to the historical material data as the process parameter benchmark. If the precise matching fails, fuzzy matching is initiated. By calculating the similarity between the material characteristic data and all historical material data, the closest cases are found. Based on the similarity between the cases and the material characteristic data, corresponding weights are assigned, and the historical successful process parameters corresponding to the cases are weighted and averaged to form preliminary recommended parameters. The machine learning model built into the matching engine is called to optimize the preliminary recommended parameters and generate the process parameter benchmark.

3. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The process of generating state data and prediction data is as follows: The prediction model library calls the corresponding prediction model based on material characteristic data; the prediction model estimates the internal physicochemical state parameters in the current melting and mixing process in real time through time series analysis algorithms, and extracts the feature data of the internal physicochemical state parameters to generate state data; the prediction model predicts the evolution trend of the internal physicochemical state parameters in a future preset time window based on the historical evolution patterns recorded internally, and generates prediction data.

4. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The process of obtaining the risk level and performing tiered processing is as follows: The risk quantification unit adaptively adjusts the instantaneous amplitude, cumulative amount, and rate of change of the deviation data, as well as the weights of the future deterioration probability, evolution speed, and failure impact value of the predicted data, based on the trend of the rate of change of the deviation data and the fluctuation of the probability of deterioration of the predicted data. It then uses a nonlinear fusion algorithm to quantify the process risk and generate a quantified risk value. The threshold optimization unit dynamically adjusts the threshold used to map the quantified risk value to the risk level based on the error situation of historical risk level classification, and establishes a dynamic evaluation threshold library. The operation response procedure selection unit dynamically selects candidate operation response procedures from the operation response procedure library based on the risk level using a reinforcement learning algorithm. The candidate operation response procedure is submitted to the operation response procedure verification unit. A real-time updated digital twin model is used to simulate the use of the candidate operation response procedure in various production scenarios and generate verification results to evaluate its rationality. If the verification result indicates that the candidate operation response procedure is reasonable, the candidate operation response procedure is executed by the control unit to adjust the process parameters; if the verification result indicates that the candidate operation response procedure is unreasonable, an alternative operation response procedure is selected from the operation response procedure library and verified again to determine the final operation response procedure.

5. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The global optimization model is as follows: The global optimization model includes a comprehensive objective function representing the overall performance of the melt blending system and a set of hard and soft constraints defining the safety boundaries of the process operation. The comprehensive objective function is used to integrate and quantify the optimal balance between product color difference index, pigment dispersion uniformity index, unit product energy consumption index, and equipment load balance index. The hard and soft constraints include the upper and lower limits of temperature in each section of the melt blending equipment, the screw speed range, and the melt pressure safety threshold.

6. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The process of calculating the adjustment amount of process parameters and generating the control sequence is as follows: The objective evolution algorithm searches within the feasible solution space defined by the global optimization model, seeking the optimization direction by achieving a preset value for the comprehensive objective function, and strictly adhering to hard and soft constraints. It utilizes multi-generation iterative selection and crossover and mutation operations to calculate the adjustment amount of process parameters and generate control sequences by analyzing the dynamic coupling relationship between various process parameters and their comprehensive impact on the whole.

7. The real-time feedback control system for melt mixing parameters in masterbatch production according to claim 1, characterized in that, The real-time closed-loop feedback control is as follows: The control sequence is transmitted to the bottom control unit of the melt mixing equipment and drives the corresponding actuators to precisely adjust the set value of at least one process parameter during the production process; After the process parameters are adjusted, the grading module repeats the process of acquiring real-time data, generating status data and prediction data, generating real-time digital images, acquiring deviation data, and obtaining risk levels and performing grading. The process then repeats the operation of feeding back the deviation data, the predicted data, and the risk level to the control sequence module, performing a new round of process parameter adjustment calculations and generating a new control sequence to form a real-time closed-loop feedback control.

Images