Aluminum plate calendering equipment and system based on production cumulative parameters

Abstract

This invention provides an aluminum plate rolling equipment and system based on cumulative production parameters, relating to the field of aluminum plate production technology. The aluminum plate rolling system includes a data acquisition module, a process fingerprint construction module, a real-time prediction and control module, a central data management module, and a global self-optimization module. This system establishes a full-cycle process fingerprint data model for the aluminum plate billet and, through a built-in microstructure evolution prediction model, proactively predicts the final microstructure state of the aluminum plate. If the prediction result deviates from the preset target, the real-time prediction and control module generates a collaborative control instruction set. The global self-optimization module performs offline analysis and data mining on this massive historical data and feeds back the resulting optimization knowledge to the real-time prediction and control module. This method solves the problem in traditional technologies where the cumulative effect of process parameters on the intrinsic quality of the final product cannot be effectively predicted and controlled, thus improving the pass rate and quality stability of high-end aluminum plate products.

Description

Technical Field

[0001] This invention relates to the field of aluminum plate production technology, specifically to an aluminum plate rolling equipment and system based on cumulative production parameters. Background Technology

[0002] According to Chinese Patent No. CN116851464A, a forming equipment for producing aluminum rolled materials relates to the field of aluminum plate production. It includes a base, pressure rollers, and a lifting seat. Multiple sets of pressure rollers are arranged inside the base and the lifting seat. Multiple sets of grooves are provided on the inner side of the base. Sliding grooves are provided on the sides of the grooves on the inner side of the base, and the length of the sliding grooves is the same as the length of the pressure rollers. A transmission mechanism is provided in the inner groove of the base. The transmission mechanism includes a rotating column, a connecting column, and a first bevel gear. The rotating column is located within the inner groove of the base, and the connecting column is located inside the rotating column. The end of the connecting column is fixedly installed to the lifting seat. The first bevel gear is located at the end of the connecting column. This invention, by providing a transmission mechanism and a limiting mechanism, ensures that the movement angle of the aluminum ingot does not change when it moves on the pressure rollers, thereby reducing the probability of wavy patterns on the aluminum plate surface and improving product quality.

[0003] Chinese Patent No. CN118162555A discloses a cold stamping and forming aluminum processing rolling equipment, which solves the problem of synchronous and identical feed of the conveyor belt during pressing. The equipment includes a base and a pressing block. A conveyor belt is rotatably connected inside the base, and a clamping box is fixedly installed on the outer wall of the conveyor belt. A frame is installed on the top of the base, and a hydraulic cylinder is installed on the top of the frame. The output end of the hydraulic cylinder is fixedly connected to the top of the pressing block, and a drive toothed plate is provided on the side wall of the pressing block. This invention enables the device to automatically drive the conveyor belt to move after pressing by linking the drive toothed plate and the conveyor belt. Simultaneously, the feed of the conveyor belt is the same each time, avoiding the phenomenon of the conveyor belt overfeeding or underfeeding the aluminum plate. This device ensures that the pressing depth and pressing position are the same each time, thereby ensuring that the feed length of the conveyor belt is the same each time.

[0004] The aforementioned patent documents and prior art have the following technical problems when used:

[0005] Problem 1: Most existing technologies focus on passive feedback control of macroscopic and instantaneous parameters during the rolling process, but they cannot effectively predict and regulate the complex impact of the cumulative effect of these process parameters on the final internal microstructure and mechanical properties of aluminum plates. This leads to significant fluctuations in the internal quality of the final product in the production of high-end aluminum plates, even if the process parameters are within the set range, making it difficult to guarantee the product qualification rate and consistency.

[0006] Problem 2: In the traditional production model, it is extremely difficult to trace the root cause when quality defects such as surface scratches and uneven performance occur. Due to the lack of structured recording and correlation analysis capabilities of historical data throughout the entire production process, engineers often have to rely on experience to troubleshoot, which is time-consuming and inaccurate. At the same time, the development of processes for new grades and the optimization of existing processes also heavily rely on long-term trial and error and manual summarization, lacking a data-driven and efficient continuous improvement mechanism. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an aluminum plate rolling equipment and system based on production cumulative parameters, which solves the problems mentioned in the background technology.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: an aluminum plate rolling system based on accumulated production parameters, wherein the aluminum plate rolling system includes a data acquisition module, a process fingerprint construction module, a real-time prediction and control module, a central data management module, and a global self-optimization module, wherein:

[0011] The data acquisition module is configured to connect to a multi-dimensional parameter sensor matrix on the aluminum plate rolling production line to acquire multi-dimensional instantaneous process parameters in real time.

[0012] The process fingerprint construction module, which is connected to the data acquisition module, is configured to receive multi-dimensional instantaneous process parameters from the aluminum plate rolling production line and generate a unique full-cycle process fingerprint data model for a single aluminum plate blank, representing its complete processing history.

[0013] The real-time prediction and control module, which is connected to the process fingerprint construction module, is configured to dynamically predict the final microstructure state of the aluminum plate blank based on the real-time evolution of the full-cycle process fingerprint data model, and generate collaborative control commands for real-time feedback control.

[0014] The central data management module is configured to store the full-cycle process fingerprint data model and its measured quality data, which are uniquely associated with each finished aluminum plate and generated by the process fingerprint construction module.

[0015] The global self-optimization module, which is connected to the central data management module, is configured to perform offline analysis on the massive amount of historical data stored therein, and feed the resulting optimization knowledge back to the real-time prediction and control module for iterative updates to its built-in prediction model or control strategy.

[0016] Preferably, the full-cycle process fingerprint data model includes at least one cumulative state variable among the cumulative equivalent plastic strain tensor, temperature-time integral thermal history, and total frictional work at the roller-sheet interface.

[0017] Preferably, the microstructure evolution prediction model built into the real-time prediction and control module is a recurrent neural network model. This model is trained with historical data to establish a nonlinear mapping relationship between the temporal characteristics of the full-cycle process fingerprint and the final grain size distribution, texture type and mechanical properties of the aluminum plate.

[0018] Preferably, the multidimensional instantaneous process parameters include, but are not limited to, pressing force, pressing temperature, pressing speed, plate thickness, and plate tension. The collaborative control instruction set includes combined and nonlinear synchronous adjustment instructions for subsequent passes, pressing speed curves, cooling water volume between passes, and post-pressing cooling paths.

[0019] Preferably, the global self-optimization module is configured to automatically perform multi-dimensional comparative analysis of its process fingerprint with the process fingerprint of historical high-quality batches when an abnormality in the quality of a certain batch of products is detected, so as to achieve rapid and accurate location of the root cause of the abnormal process.

[0020] Preferably, the aluminum plate rolling system further includes a digital twin module, which is configured to create a digital twin for each aluminum plate billet. The full-cycle process fingerprint serves as the core data driving the state evolution of the digital twin. The digital twin can virtually simulate and display the real-time changes in the internal microstructure of the billet throughout the entire production process.

[0021] Preferably, the aluminum plate rolling method of the aluminum plate rolling system includes the following steps:

[0022] Sp1: Data Acquisition and Identity Initialization: When an aluminum sheet blank enters the rolling production line, it is initialized with a unique identity identifier. At the same time, the data acquisition module is connected to the multi-dimensional parameter sensor matrix installed on the rolling line to start real-time acquisition of the multi-dimensional instantaneous process parameters of the aluminum sheet during the rolling process.

[0023] Sp2: Real-time construction of full-cycle process fingerprint: Call the process fingerprint construction module to integrate and vectorize the real-time multi-dimensional instantaneous process parameters obtained by the data acquisition module, and continuously update them to build the full-cycle process fingerprint data model of the aluminum plate blank in real time.

[0024] Sp3: Proactive microstructure prediction: During the rolling process, the real-time updated full-cycle process fingerprint is input into the microstructure evolution prediction model built into the real-time prediction and control module to proactively calculate the final microstructure state of the aluminum billet under the current process path.

[0025] Sp4: Deviation Comparison and Decision: Compare the predicted final microstructure with the preset target microstructure of the product to determine whether there is a deviation between the two;

[0026] SP5: Closed-loop control and active intervention: If the comparison results show a deviation, the real-time prediction and control module generates a set of collaborative control instructions to compensate for the deviation in subsequent rolling passes, and drives the actuator of the rolling equipment to make adjustments, so as to achieve closed-loop control of the microstructure.

[0027] SP6: Historical Data Archiving and Association: After the aluminum billet completes all rolling passes and becomes a finished product, the final full-cycle process fingerprint data model is associated and bound with its measured quality data through the central data management module and stored in the central fingerprint database.

[0028] SP7: Global Optimization and Knowledge Discovery: Periodically or upon receiving a new production instruction, the global self-optimization module is invoked to perform offline analysis and data mining on the massive historical data stored in the central data management module. Based on the analysis results, the internal model of the real-time prediction and control module is updated, or the optimal process fingerprint for new products is generated.

[0029] Preferably, the equipment includes a worktable, a control system, a conveyor roller, a frame, a hydraulic cylinder, a mounting base, an upper calendering roller, a lower calendering roller, a motor, a monitoring structure, and a cooling structure. The control system is located on the side of the worktable, the conveyor roller is located on the surface of the worktable, the frame is located on the surface of the worktable, the hydraulic cylinder is located on the surface of the frame, the mounting base is located at the output end of the hydraulic cylinder, the motor is located on the side of the mounting base, the upper calendering roller is located at the output end of the motor, the lower calendering roller is located on the surface of the worktable, the monitoring structure is located on the surface of the worktable, and the cooling structure is located on the surface of the worktable.

[0030] Beneficial effects

[0031] This invention provides an aluminum sheet rolling equipment and system based on accumulated production parameters. It has the following beneficial effects:

[0032] 1. This invention achieves a leap from "passive process control" to "active shaping of results" by constructing a unique "full-cycle process fingerprint" data model for each aluminum plate and utilizing an artificial intelligence-based microstructure evolution prediction model. The system can proactively predict the final grain size and mechanical properties of the finished product based on the real-time evolving process fingerprint when the rolling process is in progress. Once the prediction result deviates from the target, the real-time prediction and control module will immediately generate and execute a set of coordinated, multi-parameter adjustment instructions to actively intervene and correct the microstructure evolution path of the material. This closed-loop control capability ensures that the intrinsic quality of the final product can accurately reach the preset target, thereby significantly improving the batch pass rate and quality stability of high-end products with stringent requirements for grain size and mechanical properties.

[0033] 2. This invention uses a central data management module to permanently associate and bind the final quality data of each finished aluminum plate with its unique full-cycle process fingerprint. This establishes a powerful "product gene" database. When any batch has a quality anomaly, the global self-optimization module can be activated immediately to conduct multi-dimensional comparative analysis of its process fingerprint with the "golden fingerprint" of historical high-quality batches. This enables rapid and accurate location of the root cause of the abnormal process. In addition, the module can also perform in-depth mining and learning of massive historical data in the database offline, continuously discovering deep-seated process-quality correlation patterns beyond single production. These newly discovered "knowledge" are used to continuously feed back and update the internal model of the real-time prediction and control module, enabling the entire system to have the ability to self-evolve, significantly improving the efficiency of process optimization and the success rate of new product development. Attached Figure Description

[0034] Figure 1 This is a system architecture diagram of the present invention;

[0035] Figure 2 This is a flowchart of the process of the present invention;

[0036] Figure 3 This is a structural diagram of the device of the present invention.

[0037] The components include: 1. Workbench; 2. Control system; 3. Conveyor roller; 4. Frame; 5. Hydraulic cylinder; 6. Mounting base; 7. Upper calendering roller; 8. Lower calendering roller; 9. Motor; 10. Monitoring structure; 11. Cooling structure. Detailed Implementation

[0038] 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. Specific Implementation Example 1:

[0040] like Figures 1 to 3 As shown, an aluminum plate rolling system based on accumulated production parameters is disclosed. The system includes a data acquisition module, a process fingerprint construction module, a real-time prediction and control module, a central data management module, and a global self-optimization module, wherein:

[0041] The data acquisition module is configured to connect to the multi-dimensional parameter sensor matrix on the aluminum sheet rolling production line to acquire multi-dimensional instantaneous process parameters in real time. These parameters include, but are not limited to, pressing force, pressing temperature, pressing speed, sheet thickness, and sheet tension. The core of this module is the multi-dimensional parameter sensor matrix, a set of precision hardware working collaboratively. This matrix includes, but is not limited to: a pressing force sensor (capturing pressure changes on the press frame in real time with millisecond accuracy using a high-precision force gauge); a temperature sensor (using a non-contact infrared thermometer to continuously monitor the temperature distribution on the sheet surface before entering the pressing rollers, after leaving the rollers, and between passes); a pressing speed sensor (providing real-time feedback of the roller rotation speed and sheet linear velocity using an encoder or laser tachometer); a sheet thickness sensor (using an X-ray or laser thickness gauge to perform high-frequency scanning of the sheet thickness at the pressing exit); and a tension sensor (continuously monitoring the sheet tension during the pressing process). These sensors transmit data to the control system 2 at high speed and precision, forming the original data source for the "full-cycle process fingerprint."

[0042] The process fingerprint construction module, connected to the data acquisition module, is configured to receive multi-dimensional instantaneous process parameters from the aluminum sheet rolling production line and generate a unique full-cycle process fingerprint data model representing the complete processing history of a single aluminum sheet billet. This full-cycle process fingerprint data model includes at least one cumulative state variable from the cumulative equivalent plastic strain tensor, temperature-time integral thermal history, and total frictional work at the roll-sheet interface. This module is the starting point for data intelligence; it transforms the original instantaneous process parameters into a full-cycle process fingerprint data model with metallurgical physical significance. This model is not merely a simple parameter accumulation but a state-space vectorized representation, encompassing temporal integration and feature extraction of the data. Specifically… Cumulative equivalent plastic strain tensor: This variable quantifies the degree and direction of overall plastic deformation experienced by the plate by continuously integrating the changes in plate thickness, friction between the pressure rollers, and the deformation history of the plate in each pass; Temperature-time integrated thermal history: This variable integrates the temperature changes experienced by the plate due to deformation heat generation and cooling between passes in each pass, and uses a single index to characterize its complete temperature history; Total frictional work at the pressure roller-plate interface: Calculated by integrating the pressing force and the pressure roller speed, this variable reflects the total energy dissipation between the plate and the pressure roller, which is closely related to the recrystallization and texture evolution of the material. These cumulative variables form a multi-dimensional dynamic data fingerprint, which fully records the unique "processing history" of each aluminum plate.

[0043] The real-time prediction and control module, connected to the process fingerprint construction module, is configured to dynamically predict the final microstructure of the aluminum billet based on the real-time evolution of the full-cycle process fingerprint data model. It generates collaborative control commands for real-time feedback control. The microstructure evolution prediction model built into the real-time prediction and control module is a recurrent neural network-based model. This model, trained with historical data, establishes a nonlinear mapping relationship between the temporal characteristics of the full-cycle process fingerprint and the final grain size distribution, texture type, and mechanical properties of the aluminum plate. The collaborative control command set includes combined and nonlinear synchronous adjustment commands for subsequent passes' reduction rate allocation, pressing speed curves, inter-pass cooling water volume, and post-pressing cooling paths. The core of this module is the microstructure evolution prediction model, a physics-data hybrid model that integrates classical material constitutive equations with advanced recurrent neural networks or long-term short-term notation. The memory network model, trained on historical production data, learns and establishes a nonlinear mapping relationship between the complex temporal characteristics of the full-cycle process fingerprint and the final grain size distribution, texture type, and mechanical properties of the aluminum plate. When the real-time evolving process fingerprint is used as input, the model can proactively calculate the final microstructure of the plate if pressing continues along the current process path. This prediction is for the entire remaining processing, not just the next pass. The system compares the predicted final state with the preset target state. If there is a deviation, a set of collaborative control instructions is generated. These instructions are multi-parameter combined adjustments, that is, simultaneously adjusting the reduction rate distribution of subsequent passes, the pressing speed curve, the cooling water volume between passes, and the final post-pressing cooling path. These instructions drive actuators such as hydraulic cylinder 5, motor 9, and cooling valve to achieve closed-loop and active control of the microstructure.

[0044] The central data management module is configured to store the full-cycle process fingerprint data model and its measured quality data, which are uniquely associated with each finished aluminum plate and generated by the process fingerprint construction module. This module is responsible for the structured storage and management of all production data. Once an aluminum plate has completed all pressing, its complete full-cycle process fingerprint data model and the measured quality data obtained through offline testing will be associated and permanently stored, providing a valuable data foundation for subsequent offline analysis.

[0045] The global self-optimization module, connected to the central data management module, is configured to perform offline analysis of massive amounts of historical data and feed the resulting optimization knowledge back to the real-time prediction and control module for iterative updates to its built-in prediction models or control strategies. When an anomaly is detected in a batch of products, the global self-optimization module automatically performs multi-dimensional comparative analysis of its process fingerprint with that of historically high-quality batches, enabling rapid and accurate identification of the root cause of the anomaly. This module is responsible for extracting deep-level knowledge from massive amounts of historical data to achieve adaptive evolution of the system. Utilizing data mining and association analysis algorithms, it conducts in-depth research on the data in the central data management module in offline mode. When a strong correlation is found between certain specific fingerprint patterns and high-quality products, it feeds the resulting optimization knowledge back to the real-time prediction and control module. This learning process allows the system to continuously improve the accuracy of its predictions and the effectiveness of its control strategies, thus providing strong support for new product development or anomaly tracing.

[0046] The aluminum sheet rolling system also includes a digital twin module, configured to create a digital twin for each aluminum sheet billet. The full-cycle process fingerprint serves as the core data driving the evolution of this digital twin's state. The digital twin can virtually simulate and display the real-time changes in the internal microstructure of the billet throughout the entire production process. The full-cycle process fingerprint is the core data flow driving the evolution of this digital twin. Data input from the data acquisition module (pressing force, temperature, speed, etc.) is processed by the process fingerprint construction module to form a dynamic, cumulative fingerprint. This dynamic fingerprint is the data driver of the digital twin; it continuously uses real-world events as input to drive the physical simulation engine in the virtual model. Driven by the full-cycle process fingerprint, the digital twin can simulate and display in real time the changes in internal microstructure: the breakage and elongation of grains during pressing, and the recrystallization and growth processes during cooling between passes; and the stress and strain distribution: displaying the sheet material. The stress distribution in different internal regions is crucial for predicting defects such as delamination and cracks. Temperature field distribution: Simulating the cooling process on the surface of the sheet material, it visually demonstrates the control effect of cooling spray on the temperature field. It makes the abstract prediction results of the real-time prediction and control module visible. When the prediction model calculates that "if this continues, the grains will grow larger," the digital twin can visually display the virtual process of grain growth, helping engineers understand and verify the prediction results. When the system generates a set of collaborative control instructions, the digital twin can be used to simulate the possible effects of these instructions in advance. That is, before issuing the instruction to "increase the cooling water volume," this operation can be simulated in the digital twin to assess its impact on the temperature field and microstructure, ensuring that the adjustment is reasonable and effective. When quality defects occur in the finished product, the entire process fingerprint of the aluminum sheet can be traced back, and the entire processing process can be reproduced in the digital twin. This helps engineers quickly and intuitively locate the root cause of the defects in the process deviation.

[0047] The equipment includes a worktable 1, a control system 2, a conveyor roller 3, a frame 4, a hydraulic cylinder 5, a mounting base 6, an upper calendering roller 7, a lower calendering roller 8, a motor 9, a monitoring structure 10, and a cooling structure 11. The control system 2 is located on the side of the worktable 1. The conveyor roller 3 is located on the surface of the worktable 1. The frame 4 is located on the surface of the worktable 1. The hydraulic cylinder 5 is located on the surface of the frame 4. The mounting base 6 is located at the output end of the hydraulic cylinder 5. The motor 9 is located on the side of the mounting base 6. The upper calendering roller 7 is located at the output end of the motor 9. The lower calendering roller 8 is located on the surface of the worktable 1. The monitoring structure 10 is located on the surface of the worktable 1. The cooling structure 11 is located on the surface of the worktable 1. The worktable 1 and the conveyor roller 3 constitute the front pressing structure. The conveyor line, frame 4, upper calendering roll 7, lower calendering roll 8, motor 9, hydraulic cylinder 5, and mounting base 6 constitute the core of the calendering equipment. Among them, hydraulic cylinder 5 is the actuator for adjusting the reduction rate, motor 9 controls the rotation speed of upper calendering roll 7, monitoring structure 10 is the multi-dimensional parameter sensor matrix of the data acquisition module, which includes, but is not limited to, force sensors, thickness gauges, infrared thermometers, etc. Cooling structure 11 is the physical realization of the cooling water volume between passes and the cooling path after pressing, including nozzle matrix and coolant recovery system. Control system 2 is a data processing and control device, which is the operating carrier of software modules such as data acquisition module, process fingerprint construction module, real-time prediction and control module. Specific Implementation Example 2:

[0049] like Figures 1 to 3 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0050] The core models and algorithms in the above aluminum plate rolling system are mainly reflected in the following aspects:

[0051] In the process fingerprint construction module, the core model is the full-cycle process fingerprint data model. The full-cycle process fingerprint is not a simple number or table, but a dynamic, multi-dimensional digital archive with time-series characteristics. It creates a unique "processing history" for each aluminum plate throughout its entire processing lifecycle. This model achieves a complete depiction of the "linear" history. Traditional data records are usually instantaneous, such as the pressing force or temperature at a certain moment. These data are isolated. The core of the full-cycle process fingerprint lies in accumulating, integrating, and abstracting these discrete instantaneous data through mathematical and physical methods, transforming them into "cumulative state variables" with higher information density. The full-cycle process fingerprint data model consists of a series of key cumulative state variables, which are projections of the physical process into digital space, namely:

[0052] Cumulative equivalent plastic strain tensor: This is not simply a change in sheet thickness; it is a complex physical quantity. By integrating the pressing force, the rate of change in sheet thickness, and the slip ratio between the roller and the sheet, it precisely quantifies the total amount and direction of plastic deformation experienced by the sheet in every tiny region of three-dimensional space. This variable directly determines the degree of grain breakage and elongation within the material and is the fundamental basis for controlling grain size and texture type. Temperature-time integrated thermal history: This variable is the "energy record" of temperature changes throughout the entire processing history of the sheet. By integrating the residence time of the sheet at different temperatures in each pass, it precisely reflects the heat treatment history of the material. This thermal history directly affects the recrystallization, grain growth, and other processes. The formation of precipitates determines the final mechanical properties; Total frictional work at the roller-sheet interface: This variable is calculated by integrating the difference between the pressing force and the roller speed in each pressing process to determine the total energy consumed at the interface between the sheet and the roller. Changes in frictional work not only affect pressing efficiency and temperature, but are also key factors leading to surface defects such as scratches, abrasions, and black streaks; Inter-pass relaxation time series: This seemingly simple variable records the time the sheet stays between each pass. During hot pressing, this brief relaxation time is crucial for the dynamic recrystallization and grain growth inside the sheet. Using it as part of the fingerprint allows the model to capture the influence of time factors on the evolution of microstructure.

[0053] The full-cycle process fingerprint data model is not static, but a real-time evolving process. First, data acquisition begins. Starting from the first pass, a multi-dimensional parameter sensor matrix collects raw data every millisecond. Then, fingerprint construction is performed. The process fingerprint construction module processes this data in real time, calculates and updates the aforementioned accumulated variables. After each pressing is completed, a new "timestamp" and a new set of accumulated data are added to the fingerprint. Finally, prediction and decision-making are performed. The neural network model of the real-time prediction and control module takes this dynamically evolving fingerprint as input. Since the model is trained on massive historical fingerprint data, it can identify the current fingerprint evolution trend and predict the final microstructure of the board based on this trend. That is, if the "temperature-time integral thermal history" in the fingerprint increases too quickly, the model will immediately determine that the final grains of this board will be too large, thereby generating corresponding collaborative control commands. In subsequent passes, this trend is corrected by increasing the cooling water volume. In this way, the full-cycle process fingerprint transforms the complex physical process into a digital language that can be understood by artificial intelligence, realizing a technological leap from "passive response" to "active shaping".

[0054] The core algorithm of this module is the data processing and feature extraction algorithm, which is responsible for transforming raw sensor data into meaningful cumulative state variables. This includes: a time-series integration algorithm, which synchronizes and aligns data streams collected from different sensors in milliseconds according to precise timestamps; a physical model algorithm, which uses the principles of materials mechanics and thermodynamics to transform raw data into cumulative variables. This can be achieved by using an integral algorithm to calculate the temperature-time integral thermal history and the total frictional work at the roller-sheet interface, and tensor calculation to calculate the cumulative equivalent plastic strain tensor. This requires processing multiple physical quantities such as pressing force, sheet thickness variation, and roller geometric parameters.

[0055] In the real-time prediction and control module, the core model is the microstructure evolution prediction model. This model is a physics-data hybrid model that combines traditional material constitutive relations with advanced artificial intelligence algorithms. The physics part refers to the material constitutive relations: traditional pressing process simulations mainly rely on physical formulas, such as constitutive equations describing material deformation resistance, recrystallization, and grain growth. These equations can describe how the microstructure of the material evolves under specific temperatures, strains, and strain rates. However, these physical models are often simplified and cannot accurately capture the complex, nonlinear coupling effects in actual production. The data part refers to the artificial intelligence algorithm: the model is based on recurrent neural networks and long short-term memory networks, suitable for processing time-series data, i.e., data that changes over time. The model includes an input layer, a hidden layer, and an output layer. The input layer receives real-time data from the process fingerprint construction module. This data is multi-dimensional, including cumulative strain, temperature-time integral thermal history, frictional work, etc., and is time-series. The input is not a single data point, but a series of historical data points; the hidden layer: this is the main body of the recurrent neural network and the long short-term memory network. It consists of multiple neuron layers and can capture the temporal features and nonlinear relationships in the input data. That is, it can "learn" how a sudden increase in the pressing force at a certain temperature in a certain pass will affect the grain growth in subsequent passes. The "memory units" of the long short-term memory network are particularly good at remembering such long-term correlations across passes, solving the "historical dependence" problem that traditional models cannot handle; the output layer: outputs the predicted results. These results are not simple numerical values, but the final microstructure of the board after all the remaining passes are completed. This includes: grain size distribution: predicting the final average grain size and the uniformity of grain size; texture type and strength: predicting the preferred orientation of crystals inside the board, which directly affects the anisotropic mechanical properties of the board; mechanical performance indicators: predicting the final tensile strength, yield strength, and elongation, etc.

[0056] The module's algorithms also include a prediction algorithm: this algorithm takes the real-time fingerprint input to the trained model, performs forward propagation calculations, and quickly outputs the prediction result of the future final state; and a cooperative control algorithm: after predicting the deviation, the system needs to generate an optimal cooperative control instruction set based on the magnitude and direction of the deviation. This is usually achieved through optimization algorithms, such as model predictive control and genetic algorithms. These algorithms can calculate the parameter combination that can most effectively eliminate the deviation under various constraints (i.e., equipment capabilities, safety limitations).

[0057] The global self-optimization module's core algorithm combines reinforcement learning and data mining / association analysis. Reinforcement learning is a method that allows the system to learn from trial and error and optimize its control strategy. Its state consists of all observable data on the production line, including current pressing parameters and historical process fingerprints. Actions are the collaborative control instruction set generated by the real-time prediction and control module. Rewards are given when the finished product rolls off the line and passes actual testing. If the quality data (thickness tolerance, grain size) meets the first-level standard, the system receives a positive reward; if the product is unqualified, it receives a negative reward. This module uses reinforcement learning to train a policy network. The goal of this network is to learn which action, given any state, will maximize future cumulative rewards. As production progresses, the system continuously accumulates experience (state-action-reward sequence) and uses this experience to update the policy network, making its control strategy increasingly precise.

[0058] Data mining and association analysis algorithms are used to find hidden, deep-seated patterns beyond single production runs in massive historical data. These include: clustering algorithms (K-means clustering) that cluster massive historical fingerprint data to discover different types of "process fingerprint patterns" and analyze the final product quality corresponding to each pattern; association rule mining algorithms (Apriori algorithm) that discover correlations between different parameters, such as "if the pressing force in the second pass exceeds a certain threshold and the cooling water volume between passes is lower than a certain value, the probability of surface scratches on the final product will increase significantly"; and feature importance analysis using algorithms such as decision trees or random forests to evaluate the impact of each process fingerprint variable on the final quality, helping the system better understand which parameters are most critical in the control process. Specific Implementation Example 3:

[0060] like Figures 1 to 3 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0061] The entire system operates smoothly. The following is the specific operation procedure for the aluminum sheet rolling method of the aluminum sheet rolling system:

[0062] Sp1: Data Acquisition and Identity Initialization: When an aluminum sheet enters the rolling production line, it is initialized with a unique identity. At the same time, the data acquisition modules installed on the production line are fully activated. These include the pressing force sensor on the press frame 4, the infrared thermometers before and after the press rollers, the X-ray thickness gauge at the press exit, and the tension sensor. They begin to capture all instantaneous process parameters related to the sheet in real time at an extremely high frequency, hundreds or even thousands of times per second. These parameters constitute the raw data for all subsequent intelligent judgments. The data acquisition module is connected to the multi-dimensional parameter sensor matrix installed on the rolling line to begin real-time acquisition of the multi-dimensional instantaneous process parameters of the aluminum sheet during the rolling process.

[0063] Sp2: Real-time construction of full-cycle process fingerprint: The process fingerprint construction module is called to integrate and vectorize the real-time multi-dimensional instantaneous process parameters obtained by the data acquisition module. Every deformation and every temperature fluctuation of the sheet is converted into cumulative variables. It will calculate in real time the cumulative equivalent plastic strain tensor, temperature-time integral thermal history and total friction work of the roller-sheet interface experienced by the sheet. These variables together constitute a dynamic and unique full-cycle process fingerprint data model. This fingerprint model grows continuously as the sheet moves and is processed in different passes, completely recording its "processing history" and continuously updating, thus constructing the full-cycle process fingerprint data model of the aluminum sheet blank in real time.

[0064] Sp3: Proactive Microstructure Prediction: During the rolling process, the real-time updated full-cycle process fingerprint is input into the microstructure evolution prediction model built into the real-time prediction and control module. The real-time prediction and control module receives the real-time updated full-cycle process fingerprint as input. This module has a built-in microstructure evolution prediction model based on recurrent neural networks and long short-term memory networks. By analyzing the evolution trend of fingerprint data, this model proactively calculates the final microstructure state of the aluminum plate after completing all remaining rolling passes, namely grain size distribution, texture type and final mechanical properties. This prediction process is carried out while the plate is still in a high-temperature ductile state, which buys time for subsequent intervention.

[0065] Sp4: Deviation Comparison and Decision: The predicted final microstructure is compared with the preset target microstructure of the product to determine whether there is a deviation between the two. This comparison is multi-dimensional, including not only the final thickness tolerance, but also the intrinsic quality indicators such as grain size and hydrogen content. The system will determine whether the final product will deviate from the quality requirements if processing continues according to the current trend. This step is a prerequisite for triggering active control.

[0066] SP5: Closed-loop control and active intervention: If the comparison results show a deviation, the real-time prediction and control module generates a set of collaborative control instructions to compensate for the deviation in subsequent rolling passes and drives the actuators of the rolling equipment to make adjustments to achieve closed-loop control of the microstructure. This set of instructions is not a single adjustment, but a combination and nonlinear synchronous adjustment of multiple rolling parameters. That is, in order to correct the problem of coarse grains, the instructions will simultaneously include: reducing the reduction rate of the next pass to reduce the generation of deformation heat; increasing the pressing speed to reduce the residence time of the plate in the high-temperature zone; and increasing the cooling water volume between passes to quickly reduce the plate temperature to the ideal range. These instructions are sent to the actuators of the rolling equipment, such as hydraulic cylinder 5, motor 9 and cooling nozzles, to achieve real-time and precise adjustment of the production process. The result of each adjustment is captured by the data acquisition module and integrated as new information into the full-cycle process fingerprint, forming a continuous and dynamic closed loop.

[0067] SP6: Historical Data Archiving and Association: After the aluminum billet completes all rolling passes and becomes a finished product, the final full-cycle process fingerprint data model is associated and bound with its measured quality data, namely the final grain size, mechanical properties and surface quality, through the central data management module, and stored in the central fingerprint database.

[0068] SP7: Global Optimization and Knowledge Discovery: The global self-optimization module periodically, or upon receiving a new production instruction, calls upon itself to perform offline analysis and data mining on the massive historical data stored in the central data management module. Based on the analysis results, it updates the internal model of the real-time prediction and control module or generates the optimal process fingerprint for new products. This module utilizes algorithms such as data mining and reinforcement learning to discover deep-seated patterns from massive amounts of data that go beyond a single production process, such as "Why is the quality of a certain batch of products particularly excellent?" or "A certain fingerprint pattern is strongly correlated with a certain type of defect." This discovered knowledge is used to update the micro-organization evolution prediction model in the real-time prediction and control module, making its predictive ability stronger and its control strategy more optimal. This ensures that the entire system can learn from past experience and provide more intelligent and efficient guidance for future production.

[0069] The data transmission in the above-mentioned operation process includes the following:

[0070] In Sp1-Sp2, the billet enters the production line, the data acquisition module starts working, and the sensors send massive amounts of raw data (once every millisecond) to the process fingerprint construction module. This module does not simply store the data, but processes the data in real time.

[0071] In Sp3-Sp5, the real-time updated "full-cycle process fingerprint" is used as input to enter the real-time prediction and control module. Data processing involves the prediction model within the module completing prediction calculations within milliseconds and comparing them with the target. Data output is that if there is a deviation, the module immediately generates a set of collaborative control instructions (adjusting the pressure roller gap, pressing speed, and cooling water volume). The instructions are sent to the actuators on the calendering equipment to complete the adjustment before entering the next pass.

[0072] In Sp6-Sp7, after all the processes of the board are completed, the process fingerprint construction module will generate the final fingerprint data. At the same time, the laboratory will conduct actual quality testing on the finished product. The central data management module will bind and store these two types of data. The global self-optimization module will periodically perform offline analysis on the massive amount of data in the central data management module. The analysis results will be used as new training data or parameters and fed back to the real-time prediction and control module to update its built-in prediction model. Specific Implementation Example 4:

[0074] like Figures 1 to 3 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0075] The specific application logic based on cumulative production parameters in the above aluminum plate rolling system is as follows:

[0076] In actual production, the initial state of each aluminum sheet (microstructure, hydrogen content, surface condition, etc.) varies slightly. Furthermore, the equipment itself experiences random fluctuations, including minor changes in roller temperature and slight differences in lubricant quantity. These minute differences accumulate over multiple rolling passes, causing the actual full-cycle process fingerprint to deviate from the ideal target process fingerprint. By analyzing historical data, the system has learned the deep correlation between these deviations and the final product quality. An abnormally high pressing force may indicate increased material deformation resistance, leading to uneven grain size and temperature variations. Excessive time-integrated thermal history may lead to abnormal grain growth, affecting the final grain size and mechanical properties. Excessive total frictional work at the roller-sheet interface may cause defects such as stripes and scratches on the sheet surface, and even affect surface quality and thickness tolerance. Excessive relaxation time between passes may cause the sheet temperature to drop too much, making subsequent pressing difficult and affecting thickness tolerance and sheet-to-sheet variation. Therefore, the fundamental purpose of parameter adjustment is to correct deviations in the intermediate process to ensure that all quality parameters of the final product, including thickness tolerance, grain size, hydrogen content, and appearance quality, meet the preset first-class standard.

[0077] The system's "microstructure prediction and control module" generates a collaborative control instruction set. Its adjustments are not isolated but rather combined and nonlinear synchronous adjustments to achieve optimal results. The reduction rate allocation is the primary means of controlling sheet thickness and plastic deformation. If the system predicts a larger final grain size, it may increase the reduction rate in subsequent passes to introduce more plastic strain energy and promote the recrystallization of fine grains. Conversely, if the predicted thickness may exceed the tolerance limit, the reduction rate will be reduced. The pressing speed curve, i.e., speed, is crucial for controlling deformation and temperature. If the sheet temperature is too high, the system may reduce the pressing speed to increase the time the sheet spends between the rollers, allowing more time for heat dissipation. Conversely, if the plate temperature is too low, the speed may be increased to utilize deformation heat to raise the temperature. The amount of cooling water between passes is a direct means of controlling the temperature. If it is predicted that the plate temperature is too high, which may lead to coarse grains, the system will increase the amount of cooling water to quickly reduce the plate temperature to the ideal recrystallization range. Conversely, if the temperature is too low, the amount of cooling water will be reduced or even the cooling will be turned off. The post-pressing cooling path is a key step in determining the final microstructure. The system will dynamically plan the optimal cooling curve based on the plate's full-cycle process fingerprint and prediction results. For products requiring high strength, the system may instruct the cooling system to perform rapid and uniform quenching to maintain fine grains and solid solution strengthening effect.

[0078] Through this proactive and collaborative adjustment, the system achieves results that are difficult to achieve with traditional control methods, improving product qualification rates. By implementing real-time prevention and correction, it can significantly reduce quality problems caused by process deviations, such as excessive hydrogen content, uneven grain size, and excessive thickness tolerances, thus significantly improving first-piece and batch qualification rates. It also optimizes internal quality by proactively shaping the internal structure, such as controlling grain size, morphology, and uniformity, reducing defects like granulation, delamination, and porosity, ensuring a dense internal structure. This directly improves the strength, toughness, and fatigue performance of the sheet material. Finally, it improves appearance quality with precise... Controlling the temperature of the pressure rollers and the tension between passes can effectively suppress the generation of surface defects. It enables on-demand customization; the system can generate the optimal target process fingerprint based on the specific requirements of different products (specific convexity, thickness tolerance, or grain size), and dynamically track and approximate it, thus realizing the concept of "proactive shaping." This allows the production line to flexibly meet the precise performance requirements of high-end customers. This intelligent control elevates "process control" to the level of "outcome prediction," thereby achieving comprehensive and refined control over product quality.

[0079] This intelligent system establishes a precise mapping relationship between physical product parameters and digital process data through a data acquisition module and a process fingerprint construction module. For dimensional parameters, including but not limited to thickness, width tolerance, and convexity, the system uses sensors such as laser or X-ray thickness gauges to perform high-frequency scanning at the press exit, generating continuous data curves. These instantaneous data are integrated into cumulative variables in the full-cycle process fingerprint, including the cumulative equivalent plastic strain tensor. With each pressing, the reduction in sheet thickness is precisely calculated and accumulated into this variable. The system's predictive model uses changes in this variable to determine whether the final thickness will exceed the tolerance range. Simultaneously, changes in instantaneous parameters such as pressing force and roller temperature are also recorded. The fingerprint is recorded because excessive pressing force can cause the pressure roller to deform, affecting the convexity of the board. Uneven temperature will cause thermal expansion, which will also affect the board shape. Assuming that the initial thickness of a blank is 300mm and the final finished product thickness is 10mm with a tolerance of ±0.02mm, when pressing for the 10th time, the thickness gauge of the press shows that the board thickness is 20mm. However, the full-cycle process fingerprint shows that the cumulative pressing force value of the past few times is abnormally high. The prediction model judges that if pressing continues according to the current trend, the final thickness may reach 10.03mm, which exceeds the tolerance. The system will immediately generate an instruction to reduce the pressing rate in subsequent times or adjust the cooling of the pressure roller to control the deformation of the pressure roller and ultimately control the thickness within the target range.

[0080] Regarding internal quality, including but not limited to hydrogen content and grain size, although the final data usually comes from offline laboratory analysis, its formation process is closely related to online data. The system's full-cycle process fingerprint records the temperature-time integral thermal history of the board, which is the most critical variable determining grain size. It records every temperature change and dwell time of the board during the pressing process. Excessive temperature or dwell time will lead to grain growth. By analyzing historical data, the global self-optimization module found that for products with a final grain size of level one, the temperature integral thermal history is controlled within a specific golden range. When the system detects a continuous upward trend in the temperature integral value of the board during real-time processing and predicts that it may eventually exceed the optimal range, resulting in coarse grains, it will immediately generate an instruction to increase the cooling water volume between passes, quickly reduce the board temperature to the ideal range, thereby actively shaping its internal grain structure and ensuring that the internal quality meets the standards.

[0081] Regarding appearance quality, including surface quality, the data comes from high-speed surface vision inspection systems, laser scanners, and final manual quality inspection data. The system can prevent surface defects by analyzing key variables in the entire process fingerprint. The cumulative value of the total friction work at the roller-sheet interface can directly reflect the degree of damage to the sheet surface. Historical data analysis shows that when the cumulative friction work exceeds a certain threshold, there is a strong correlation with severe surface scratches. In addition, improper tension control can cause the sheet to deviate, resulting in defects such as misalignment or pyramidal shapes. When the system detects abnormal friction work or tension data in the real-time fingerprint, it will react immediately by actively reducing friction by adjusting the pressing speed or the amount of lubricating oil, or by adjusting the tension control parameters, fundamentally preventing the generation of surface defects.

[0082] The entire system is a data-driven closed loop. Data is collected from the physical world, processed and analyzed in the digital world, and ultimately drives the adjustment of equipment in the physical world. A multi-dimensional parameter sensor matrix, including thickness gauges, thermometers, force gauges, speed gauges, and tension gauges, is the source of all data. They continuously deliver millisecond-level raw data streams. The process fingerprint construction module receives the raw data, performs mathematical operations and feature extraction on it, and generates a more metallurgically meaningful full-cycle process fingerprint. The real-time fingerprint enters the real-time prediction and control module. The deep learning model in the module uses its knowledge learned from historical data to proactively predict the final product quality. The prediction results are compared with the product parameter requirements (thickness tolerance, grain size range). If there is a deviation, the system generates a set of collaborative control instructions, including adjusting the pressing amount of hydraulic cylinder 5, changing the speed of motor 9, and controlling the opening of cooling water valve. These instructions are sent to the actuators of the rolling equipment, including hydraulic cylinder 5, motor 9, and cooling nozzles, to physically fine-tune the production process. Specific Implementation Example 5:

[0084] like Figures 1 to 3 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0085] The aforementioned aluminum plate rolling system not only relies on models and algorithms, but also involves hardware components and corresponding functions. The following is a detailed description of the hardware components of each module:

[0086] Data Acquisition Module: This module mainly consists of various high-precision sensors. Among them, the force sensor is usually a piezoelectric or strain gauge sensor installed on the press frame 4 to measure the pressing force. The infrared thermometer is a non-contact sensor installed before and after the pressure roller or between passes to measure the surface temperature of the sheet. The laser velocimeter measures the linear velocity of the sheet or pressure roller through the Doppler effect. The X-ray or laser thickness gauge performs a non-contact scan of the sheet at the press exit to accurately measure the thickness. The tension sensor is used to measure the tension of the sheet. These hardware components are responsible for converting analog signals such as force, temperature, velocity, thickness, and tension from the physical world into digital signals in real time and with high precision, providing the raw data stream for all subsequent modules.

[0087] Process fingerprint construction module: This module is responsible for processing and integrating data. The industrial computer has a high-performance processor and high-speed memory, which can process high-frequency data streams from multiple sensors simultaneously. A high-speed data acquisition card is installed in the industrial computer to receive and synchronize digital signals from the sensors. A large-capacity solid-state drive is used to cache and temporarily store large amounts of real-time data. The hardware configuration of this module ensures that it can receive and process massive amounts of data at extremely high speeds. It runs specific algorithms to perform mathematical operations and physical modeling on the raw data, and builds and updates the full-cycle process fingerprint data model of each aluminum plate in real time.

[0088] Real-time prediction and control module: This module is responsible for intelligent decision-making and command issuance. In order to run complex deep learning models, the high-performance computing unit typically needs to be equipped with a dedicated graphics processor or field-programmable gate array to achieve parallel computing and accelerate the prediction and decision-making process. The industrial motherboard and central processing unit run the operating system and control logic. The hardware of this module can run complex prediction models at millisecond speeds, taking process fingerprints as input to predict the final product quality in a forward-looking manner. When a deviation is found in the prediction, it will quickly generate a set of collaborative control instructions and convert them into electrical signals, which will be sent to the actuators through the network.

[0089] Central Data Management Module: This module is responsible for data storage and management. The industrial-grade server features high reliability, redundant power supply, and high-performance processing capabilities, enabling it to handle large-scale data storage and retrieval. The enterprise-grade storage array employs redundant disk array technology to ensure data security and high availability. High-speed network switches ensure high-speed data transmission between the server and other modules on the production line. This module is responsible for receiving and permanently storing the complete fingerprint data of each aluminum plate generated by the process fingerprint building module, as well as the corresponding measured quality data. It provides data support for the global self-optimization module and is also the sole data source for production traceability and quality analysis.

[0090] Global Self-Optimization Module: This module is responsible for offline analysis and optimization. The high-performance server cluster typically consists of multiple servers with powerful parallel computing capabilities, used to run complex machine learning and data mining algorithms. It is a hardware coordination platform at the software level of the big data processing platform, capable of efficiently processing and analyzing massive amounts of historical data. This module does not directly intervene in production; its hardware is mainly used for offline computing. It performs in-depth mining of the data in the central data management module to find deep-seated process-quality patterns and uses this knowledge to update the prediction model of the real-time prediction and control module, enabling the entire system to continuously evolve.

[0091] The actuator is responsible for executing instructions and changing the physical state. The hydraulic pressing system consists of hydraulic cylinders 5, which are used to precisely adjust the gap between the pressure rollers and control the pressing rate. The variable frequency motor 9 driver controls the speed of the pressure roller motor 9, thereby adjusting the pressing speed. The proportional valve and solenoid valve control the water flow and direction of the cooling spray system to achieve precise regulation of the cooling water volume. These hardware components receive instructions from the real-time prediction and control module and convert them into specific physical actions, such as adjusting the gap between the pressure rollers, changing the pressing speed, or controlling the cooling water volume, thereby achieving active shaping of the microstructure of the aluminum plate. Specific Implementation Example Six:

[0093] like Figures 1 to 3 As shown, based on the content of the above specific embodiments, the following content is further disclosed:

[0094] To further verify the effectiveness of this system, the following are two application examples:

[0095] Case 1: Actively shaping the intrinsic quality (grain size) of high-end aerospace 7-series aluminum alloy thick plates;

[0096] Customer requirements: An aerospace manufacturing company needs a batch of 7-series aluminum alloy thick plates with extremely stringent requirements for the intrinsic quality of the material. Specifically, the finished product thickness must be 30mm, and the grain size must meet the first-class standard and be uniformly distributed to ensure the mechanical property stability and fatigue resistance during subsequent processing and service. Traditional processes have great difficulty in controlling the grain size of such thick plates, with a pass rate of only around 70%. The main pain point is that it is difficult to accurately control the hot deformation and recrystallization process.

[0097] Before the production task was assigned, engineers input "Grain size level 1" as the core quality target into the system. The global self-optimization module immediately started, performing in-depth analysis of all historical 7-series aluminum alloy production data stored in the central data management module. Through association rule mining algorithms, it discovered that all high-quality products with grain size reaching level 1 had their "temperature-time integral thermal history" strictly controlled within a very narrow "golden range," and the accumulation rate of the "cumulative equivalent plastic strain tensor" significantly improved in intermediate passes. Based on this discovery, the system generated an optimal target process fingerprint for this production task. This fingerprint not only defined the final target values ​​of the accumulated variables but also planned the appropriate values ​​for these variables in each pass. The ideal path for this "evolution" is as follows: a homogenized 7-series aluminum alloy billet enters the rolling production line, where the system assigns it a unique identifier. Starting from the first pass, high-precision infrared thermometers and pressing force sensors on the data acquisition module begin capturing data at millisecond-level frequencies. The process fingerprint construction module transforms this raw data into a dynamically evolving full-cycle process fingerprint in real time. When the pressing reaches the 5th pass, due to a slightly higher ambient temperature than usual, the microstructure evolution prediction model of the real-time prediction and control module, based on the current fingerprint evolution trend (especially the rapid increase in the "temperature-time integral thermal history"), predicts that if the current process continues, the final temperature integral will exceed the "golden zone". The upper limit of the "space" resulted in coarse grains, failing to meet the first-level standard. Upon detecting the risk of deviation, the real-time prediction and control module immediately generated and issued a set of coordinated control instructions. This set of instructions was not a simple single-parameter adjustment, but a combination: Instruction 1: Increase the reduction rate of the 6th pass by 5% to introduce more plastic strain energy and promote subsequent dynamic recrystallization nucleation; Instruction 2: Increase the cooling water volume between the 5th and 6th passes by 15% to quickly curb the excessive increase in temperature integral; Instruction 3: Slightly reduce the pressing speed of the 6th pass to extend the time the plate spends in the strengthening cooling zone and ensure cooling effect. Simultaneously with the issuance of control instructions, the digital twin module also operates synchronously, and the operator... Engineers can clearly see on the screen that without intervention, the simulated grains in the virtual aluminum plate become significantly coarser in subsequent passes. However, after applying collaborative control commands, the temperature field of the virtual aluminum plate is effectively controlled, the simulated recrystallization process is more complete, and a fine and uniform grain structure is ultimately formed. This visualization process greatly enhances engineers' confidence in the system's decision-making. Finally, after production, the batch of 7-series aluminum alloy thick plates showed a 98% product qualification rate, with all grain sizes meeting or exceeding the first-class standard. Through accurate prediction and proactive control of cumulative production parameters, the system successfully solved the pain points of traditional processes and provided strong technical support for the stable production of high-end products.

[0098] Case 2: Prevention and root cause tracing of appearance quality (surface scratches) of ultra-thin 5-series aluminum sheets for automobiles;

[0099] Customer requirements: A new energy vehicle manufacturer needs a large quantity of ultra-thin (0.8mm thick) 5-series aluminum alloy sheets for manufacturing body panels. The customer has a "zero defect" requirement for the surface quality of the sheets. Any visible scratches or streaks will result in the return of the entire roll. In traditional production, due to slight changes in the condition of the pressure rollers or fluctuations in lubrication conditions, sudden surface scratches may occasionally occur. Once this happens, it often leads to the scrapping of the entire roll, and it is difficult to trace the exact cause afterward.

[0100] In the early stages of system operation, the global self-optimization module analyzed historical data and discovered a strong correlation rule through the decision tree algorithm: when the cumulative value of the "total friction work at the roller-sheet interface" suddenly exceeds a certain threshold in the second half of a roller's service cycle, the probability of surface scratches on the finished product will rise sharply from 2% to 85%. The system solidifies the threshold of the friction work growth slope into a key early warning indicator in the real-time prediction and control module. A 5-series aluminum alloy sheet is undergoing its final few passes of precision pressing. The data acquisition module and process fingerprint construction module are building its full-cycle process fingerprint in real time. Suddenly, the real-time prediction and control module detects an abnormal surge in the instantaneous value of "total friction work at the roller-sheet interface" within the fingerprint, approaching the early warning threshold. The system determines that this is due to a slight aluminum adhesion phenomenon on the roller surface, leading to a sharp increase in friction. Without intervention, continuous scratches will soon appear on the sheet surface. The system immediately triggers an active intervention program, generating collaborative control commands: Command 1: Immediately reduce the pressing speed of the current pass by 8%; Command 2: Instantly increase the lubrication flow rate in the corresponding area of ​​the emulsion spraying system by 20% to stronger flush and repair the lubrication film. These adjustments are completed within milliseconds, effectively suppressing further deterioration of friction work and preventing the formation of macroscopic scratches. Although some areas of the aluminum sheet have experienced friction... An anomaly occurred during the cleaning process, but due to timely intervention, the final product passed surface quality inspection. Although the coil passed inspection, the system recorded this near-accident. The engineer on duty immediately reviewed the full-cycle process fingerprint of the aluminum coil, and the anomaly diagnosis function of the global self-optimization module was activated. By comparing this fingerprint with the "golden fingerprint" of historical high-quality batches, the problem was precisely located in the middle area of ​​the upper pressure roller on the 4th stand. The timing perfectly matched the time when the friction work abnormally spiked. The system automatically generated a maintenance suggestion: "The upper pressure roller on the 4th stand needs to be inspected and polished immediately after the current shift, as there is a risk of early aluminum adhesion." The traditional "blind" roller replacement was transformed into precise predictive maintenance. The automaker's order achieved a 100% surface quality pass rate. More importantly, the system not only prevented defects from occurring but also transformed potential quality hazards into precise equipment maintenance instructions, fundamentally improving the stability of the production process, reducing unplanned equipment downtime, and significantly reducing the scrap rate.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0102] 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. An aluminum plate rolling system based on production cumulative parameters, characterized in that: The aluminum plate rolling system includes a data acquisition module, a process fingerprint construction module, a real-time prediction and control module, a central data management module, and a global self-optimization module, wherein: The data acquisition module is configured to connect to a multi-dimensional parameter sensor matrix on the aluminum plate rolling production line to acquire multi-dimensional instantaneous process parameters in real time. The process fingerprint construction module, which is connected to the data acquisition module, is configured to receive multi-dimensional instantaneous process parameters from the aluminum plate rolling production line and generate a unique full-cycle process fingerprint data model for a single aluminum plate blank, representing its complete processing history. The real-time prediction and control module, which is connected to the process fingerprint construction module, is configured to dynamically predict the final microstructure state of the aluminum plate blank based on the real-time evolution of the full-cycle process fingerprint data model, and generate collaborative control commands for real-time feedback control. The central data management module is configured to store the full-cycle process fingerprint data model and its measured quality data, which are uniquely associated with each finished aluminum plate and generated by the process fingerprint construction module. The global self-optimization module, which is connected to the central data management module, is configured to perform offline analysis on the massive amount of historical data stored therein, and feed the resulting optimization knowledge back to the real-time prediction and control module for iterative updates to its built-in prediction model or control strategy.

2. The aluminum plate rolling system based on production cumulative parameters according to claim 1, characterized in that: The full-cycle process fingerprint data model includes at least one cumulative state variable among the cumulative equivalent plastic strain tensor, temperature-time integral thermal history, and total frictional work at the roller-sheet interface.

3. The aluminum plate rolling system based on production cumulative parameters according to claim 1, characterized in that: The microstructure evolution prediction model built into the real-time prediction and control module is a recurrent neural network model and a long short-term memory network model. This model is trained with historical data to establish a nonlinear mapping relationship between the temporal characteristics of the full-cycle process fingerprint and the final grain size distribution, texture type and mechanical properties of the aluminum plate.

4. The aluminum plate rolling system based on production cumulative parameters according to claim 1, characterized in that: The multidimensional instantaneous process parameters include, but are not limited to, pressing force, pressing temperature, pressing speed, plate thickness, and plate tension. The collaborative control instruction set includes combined and nonlinear synchronous adjustment instructions for subsequent passes, pressing speed curves, cooling water volume between passes, and post-pressing cooling paths.

5. The aluminum plate rolling system based on production cumulative parameters according to claim 1, characterized in that: The global self-optimization module is configured to automatically perform multi-dimensional comparative analysis of the process fingerprint of a batch of products with the process fingerprint of historical high-quality batches when an abnormality is detected, so as to achieve rapid and accurate location of the root cause of the abnormal process.

6. The aluminum plate rolling system based on production cumulative parameters according to claim 1, characterized in that: The aluminum plate rolling system also includes a digital twin module, which is configured to create a digital twin for each aluminum plate billet. The full-cycle process fingerprint serves as the core data driving the state evolution of the digital twin. The digital twin can virtually simulate and display the real-time changes in the internal microstructure of the billet throughout the entire production process.

7. An aluminum plate rolling system based on production cumulative parameters according to any one of claims 1-6, characterized in that: The aluminum plate rolling method of the aluminum plate rolling system includes the following steps: Sp1: Data Acquisition and Identity Initialization: When an aluminum sheet blank enters the rolling production line, it is initialized with a unique identity identifier. At the same time, the data acquisition module is connected to the multi-dimensional parameter sensor matrix installed on the rolling line to start real-time acquisition of the multi-dimensional instantaneous process parameters of the aluminum sheet during the rolling process. Sp2: Real-time construction of full-cycle process fingerprint: Call the process fingerprint construction module to integrate and vectorize the real-time multi-dimensional instantaneous process parameters obtained by the data acquisition module, and continuously update them to build the full-cycle process fingerprint data model of the aluminum plate blank in real time. Sp3: Proactive microstructure prediction: During the rolling process, the real-time updated full-cycle process fingerprint is input into the microstructure evolution prediction model built into the real-time prediction and control module to proactively calculate the final microstructure state of the aluminum billet under the current process path. Sp4: Deviation Comparison and Decision: Compare the predicted final microstructure with the preset target microstructure of the product to determine whether there is a deviation between the two; SP5: Closed-loop control and active intervention: If the comparison results show a deviation, the real-time prediction and control module generates a set of collaborative control instructions to compensate for the deviation in subsequent rolling passes, and drives the actuator of the rolling equipment to make adjustments, so as to achieve closed-loop control of the microstructure. SP6: Historical Data Archiving and Association: After the aluminum billet completes all rolling passes and becomes a finished product, the final full-cycle process fingerprint data model is associated and bound with its measured quality data through the central data management module and stored in the central fingerprint database. SP7: Global Optimization and Knowledge Discovery: Periodically or upon receiving a new production instruction, the global self-optimization module is invoked to perform offline analysis and data mining on the massive historical data stored in the central data management module. Based on the analysis results, the internal model of the real-time prediction and control module is updated, or the optimal process fingerprint for new products is generated.

8. The equipment for an aluminum plate rolling system based on production cumulative parameters according to any one of claims 1-7, characterized in that: The equipment includes a workbench (1), a control system (2), a conveyor roller (3), a frame (4), a hydraulic cylinder (5), a mounting base (6), an upper calendering roller (7), a lower calendering roller (8), a motor (9), a monitoring structure (10), and a cooling structure (11). The control system (2) is provided on the side of the workbench (1), the conveyor roller (3) is provided on the surface of the workbench (1), the frame (4) is provided on the surface of the workbench (1), the hydraulic cylinder (5) is provided on the surface of the frame (4), the mounting base (6) is provided at the output end of the hydraulic cylinder (5), the motor (9) is provided on the side of the mounting base (6), the upper calendering roller (7) is provided at the output end of the motor (9), the lower calendering roller (8) is provided on the surface of the workbench (1), the monitoring structure (10) is provided on the surface of the workbench (1), and the cooling structure (11) is provided on the surface of the workbench (1).

Images