Dry electrode calender control method and calendering system based on AI algorithm

Abstract

The present invention discloses a control method and a calendering system for a dry electrode calender based on an AI algorithm. The calendering control system includes a roll gap control subsystem, a belt speed control subsystem, a micro-tension control subsystem, and a deviation rectification subsystem. The control center of each subsystem uses the zebra optimization recurrent neural network algorithm (RNN-ZOA) to calculate signals, predict reasonable control quantities, and the obtained control quantities are used as the input vectors of the PID controller. The output quantity calculated by the PID controller is output to the corresponding actuator to adjust the corresponding components. Each control subsystem of the calendering system of the present invention introduces the zebra optimization recurrent neural network algorithm and deploys it in the PID controller, which improves the control performance of the system, enhances the stability and accuracy of the system, and also improves the production efficiency and product quality of the calender.

Description

Technical Field

[0001] The present invention relates to the technical field of calender control, and in particular to a control method and a calendering system for a dry electrode sheet calender based on an AI algorithm. Background Art

[0002] Calenders are applied in industries such as calendering, film stretching, and rolling. In the control system of a calender, multiple structural units involve feedback control: a roll gap adjustment unit, an electrode sheet deviation correction unit, and an electrode sheet tension control unit. Among them, the roll gap adjustment structure is used to adjust the roll gap between the main pressure rolls to complete the adjustment of the rolling force and the thickness of the electrode sheet; the electrode sheet deviation correction structure is used to adjust the deviation of the electrode sheet to avoid phenomena such as tower-shaped winding caused by the deviation of the electrode sheet, which affects the quality of the electrode sheet; the electrode sheet tension control system is used to maintain the stability of the electrode sheet tension, ensure the flatness of the rolled electrode sheet, and avoid phenomena such as waves caused by uneven elongation of the electrode sheet.

[0003] The control system of existing calenders usually adopts PID control (Proportional Integral Derivative control). The traditional PID control usually directly inputs the collected parameters or signals into the PID controller, and forms the control quantity output through linear combination according to proportion, integral, and derivative to control the controlled object. The output of the traditional PID control is composed of linear combination. When facing non-linear, time-varying or multi-variable systems, it has insufficient processing of complex dynamic performance and multi-variable coupling, and also lacks self-adaptability and robustness, resulting in poor control performance, affecting the control stability and accuracy of the system, and thus affecting the quality of the electrode sheet.

[0004] In the existing cases of using neural network algorithms, optimization algorithms are often used to optimize them to improve their convergence speed, such as Grey Wolf Optimization Algorithm (GWO), Genetic Algorithm (GA), Marine Predators Algorithm (MPA), Particle Swarm Optimization Algorithm (PSO), Tree Species Optimization Algorithm (TSA), Gravitational Search Algorithm (GSA), etc.

[0005] The conventional method of the electrode sheet tension control unit of the existing calender is to use a tension sensor to monitor the electrode sheet tension and adjust the tension according to the monitored tension value. This method has low regulation accuracy of the tension, and poor accuracy and stability. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing calender control system, the present applicant provides a reasonable dry electrode calender control method and calender system based on AI algorithms. The zebra optimization algorithm (ZOA) is introduced into the recurrent neural network (RNN) algorithm to obtain a faster convergence speed and more accurate output parameters. This method is called the zebra optimization recurrent neural network algorithm (RNN-ZOA). The zebra optimization recurrent neural network algorithm (RNN-ZOA) is deployed into the PID to improve the stability and accuracy of the system. The tension control system controls the actual tension of the electrode and the angle of the swing roller to improve the accuracy and stability of the tension control.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A dry electrode calender control method based on AI algorithms, the calender control system includes a roll gap control subsystem, a belt speed control subsystem, a micro-tension control subsystem and a deviation correction subsystem;

[0009] The roll gap control subsystem adjusts the roll gap distance between two film-forming calender rolls and / or two thinning calender rolls according to the roll gap signal and the diaphragm thickness signal;

[0010] The belt speed control subsystem adjusts the speed of the servo motor according to the speed signal of each servo motor;

[0011] The micro-tension control subsystem adjusts the rotation angle and speed of the swing roller according to the rotation angle signal of the swing roller and the tension signal of the diaphragm;

[0012] The deviation correction subsystem adjusts the frame of the take-up reel assembly according to the position signal of the diaphragm;

[0013] The control centers of the roll gap control subsystem, the belt speed control subsystem, the micro-tension control subsystem and the deviation correction subsystem use the zebra optimization recurrent neural network algorithm (RNN-ZOA) to calculate the signals, predict reasonable control quantities, and use the obtained control quantities as the input vector of the PID controller. The output quantity calculated by the PID controller is output to the corresponding actuator to adjust the corresponding components.

[0014] As a further improvement of the above technical solution:

[0015] The PID controller uses a discrete-time system; the transfer function of the PID is , s is the Laplace variable (s = α + jω), K p is the amplification coefficient, T I is the integral time, T D is the derivative time; the adjustment function of the PID is , e(t) is approximately used as the rectangular sum for integration , e’(t) is approximated as a straight line , the control expression of the PID is .

[0016] The calendering control system is embedded and deployed into the PID controller as a control module.

[0017] The control process of the roll gap control subsystem includes: the control center calculates the deviation rate one based on the actual thickness detected by the first thickness gauge assembly and the calibrated thickness here. If the deviation rate one is within the range of ±5%, the data is stored in the repository without control operations. If the deviation rate one is within the range of ±5% to ±10%, the control center adopts the first control module to adjust the roll gap of the thinning component. If the deviation rate one exceeds the range of ±10%, the control center adopts the second control module to adjust the roll gap of the film-forming component.

[0018] After the first control module adjusts the roll gap of the thinning component, the deviation film is thinned after being corrected by the thinning component. The control center calculates the deviation rate two based on the actual thickness detected by the second thickness gauge and the rated thickness here. When the deviation rate two is less than ±6%, the roll gap is not corrected. When the deviation rate two is greater than ±6%, based on the thickness signal of the second thickness gauge, the roll gap of the thinning component is adjusted until the deviation rate two is reduced to within ±6%. At the same time, the zebra optimized recurrent neural network algorithm (RNN-ZOA) for controlling the roll gap of the thinning component is optimized twice.

[0019] After the second control module adjusts the roll gap of the film-forming component so that the deviation rate one is within the range of ±10%, it switches to the first control module; during the time of control and adjustment by the second control module, the second thickness gauge monitors and records the timing data in real time, and removes the unqualified film during this period after winding is completed.

[0020] The control center of the belt speed control subsystem calculates the rotational speed deviation rate of each servo motor by taking the rotational speed of the servo motor of the thinning pressure roller as the reference and calculating the rotational speeds of the other servo motors and the reference rotational speed. When the deviation rate exceeds the normal range, commands are issued to the servo motors with the exceeded deviation rate to adjust their rotational speeds.

[0021] The micro-tension control subsystem collects the timing data of the tension sensor. If the timing tension data of a certain time exceeds the fluctuation range, after the control center retrieves the timing tension of the previous period of this data, it calculates and fits this set of timing data sets through the zebra optimized recurrent neural network algorithm (RNN-ZOA) to predict its tension trend. If this trend is outside the fluctuation range or it is predicted that it will exceed the fluctuation range subsequently, a correction value is calculated and a command is issued to the swing roller motor to adjust this section of tension by changing the output torque of the swing roller motor. When the adjustment of the output torque of the swing roller motor is completed, the tension sensor continuously monitors the tension of the film.

[0022] During the process of tension fluctuation correction, the angle signal of the swing roll motor will be synchronously sent to the control center. The control center constructs a function in advance by combining the angle information of the swing roll motor with the servo motor speed signal of the first tension isolation device, and outputs it to the servo motor of the first tension isolation device through the associated function calculation, so that it adjusts the speed to adjust the position of the swing arm and the angle of the swing roll motor.

[0023] The control center of the deviation correction subsystem judges the position offset of the diaphragm in real time. If the offset at a certain time sequence exceeds the fluctuation range, the control center immediately takes this time sequence offset as the reference, retrieves some time sequence data before this time period for fitting calculation, calculates the motor rotation angle of the deviation correction actuator, transmits this angle signal to the servo motor of the deviation correction actuator, and the servo motor rotates the corresponding angle to drive the movement rod of the deviation correction actuator to extend and retract, thereby pushing the take-up reel assembly to move, ensuring the neatness of the rolled edge after the diaphragm is wound.

[0024] The control method of the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) model includes the following steps: 1) Start running; 2) Initialize the execution components of each control subsystem; 3) The Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) predicts the initial control quantity of the execution component according to the difference between the initialized actual parameter value and the set parameter value of the execution component of each control subsystem; 4) The actual parameter values detected in real time by the detection components of each control subsystem are input into the neural network model as lightweight input vectors; 5) The Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) model calculates the predicted real-time control quantity, and the real-time control quantity is used as a lightweight output vector; 6) The lightweight output vector of the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) model is input into the PID controller, and the PID controller adjusts the output quantity in real time according to the output vector; The adjustment function of the PID controller is: u'(t) = u(t) + f(Y), u(t) is a function based on the error e(t), including proportional, integral and differential terms, the error e(t) refers to the difference between the actual parameter value and the set parameter value, and f(Y) is an empirical adjustment function; 7) Judge whether the signal is in a stable state: If it is stable, execute the next step to complete the control; If it is not stable, return to the start of running and run again until the signal is stable.

[0025] Before the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model is embedded in the control system, it needs to be trained first. The training learns the mapping function X→Y, where X is the input vector and Y is the output vector. The Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model uses historical data for expert training. The historical data includes the roll gap width, diaphragm thickness, motor speed, diaphragm tension, swing roll angle, position, and other key time-series parameter data required by each sub-control system. The Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model is updated and adjusted regularly and retrained when the working conditions change. During the training process of the recurrent neural network model, optimization algorithms such as gradient descent and Adam are used.

[0026] The optimization parameter matrix of the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model has five variables: the number of network layers, the number of neurons, the learning rate, epoch, and batch size. Each variable is set with upper and lower limits. The minimized objective function for optimization is: , where t is the training time of the neural network, loss is the loss value of the neural network, and m and n represent weights. The process of the optimization algorithm is iterative. The termination conditions for iteration are reaching the upper limit of the number of iterative loops, the objective function value obtained by the optimal parameters being less than the ideal value, or the optimal parameters not being updated after m rounds of iteration, or one or more combinations of these conditions. After the iteration ends, the set of parameters with the minimum objective function is the optimal solution. The RNN model corresponding to the optimal solution is the optimal RNN model. Bringing the test set into the optimal RNN model can obtain the accuracy of the model on the test set.

[0027] The model in the first stage of zebra optimization is , , where r is a random number between [0,1], and I is a random value belonging to the set {1,2}. The model in the second stage of zebra optimization is , where t is the number of iterations, T is the maximum number of iterations, R is a constant of 0.01, Ps is the switching probability between the two strategies, and its value is a random number between [0,1], and AZ is the state of the attacked zebra.

[0028] A rolling system of a dry-type polar plate rolling machine is controlled by the above-mentioned control method of a dry-type polar plate rolling machine based on an AI algorithm, and includes a film forming component, a thinning component, and a winding component arranged in sequence along the film production direction. The film forming component, the thinning component, and the winding component are respectively provided with a micro-tension control unit.

[0029] As a further improvement of the above technical solution:

[0030] The micro-tension control unit includes a tension pendulum roller assembly and at least one tension sensor, and the tension sensor is located on the discharge side of the tension pendulum roller assembly; the tension pendulum roller assembly includes two symmetrically arranged pendulum rollers on the left and right. Both ends of the pendulum roller are respectively connected to the swing arms through the pendulum roller seats. A core shaft is arranged through the swing arms. The core shaft on one side is connected to the pendulum roller motor through a pedestal bearing, and an angle sensor is arranged on the core shaft on the same side as the pendulum roller motor. The angle sensor is used to detect the rotation angle of the pendulum roller; the pendulum roller motor adopts a servo motor; the pendulum roller motor, the angle sensor, and the tension sensor of the micro-tension control unit are electrically connected to the micro-tension control subsystem; Opposite sides of the pendulum roller seat are provided with setscrew holes. The central axes of the two setscrew holes pass through the center of the pendulum roller seat and are located in the same plane. The pendulum roller seats at both ends of the pendulum roller are axially offset by 90°, and the central axes of the setscrew holes of the pendulum roller seats at both ends are perpendicular to each other.

[0031] The film-forming component includes a pair of film-forming support rollers and a film-forming pressure roller. The film-forming pressure roller is located inside the film-forming support rollers. The film-forming pressure roller and the film-forming support rollers are eccentrically arranged. The film-forming pressure roller is connected to a servo motor. A first roll gap control actuator is arranged on the outer side of one of the film-forming support rollers. The first roll gap control actuator controls the displacement of the film-forming support roller through a control valve group to adjust the roll gap width between the film-forming pressure rollers. The film-forming support rollers, the film-forming pressure roller, and the first roll gap control actuator are arranged in the horizontal direction. A first roll gap detection sensor is arranged directly outside the nip of the two film-forming pressure rollers. A first bearing clearance elimination mechanism is arranged on the outer sides of the two film-forming support rollers; the thinning component includes a pair of thinning support rollers and a thinning pressure roller. The thinning pressure roller is located inside the thinning support rollers. A second roll gap control actuator is arranged on the outer side of one of the thinning support rollers. The thinning support rollers, the thinning pressure roller, and the second roll gap control actuator are arranged in the vertical direction. A second roll gap detection sensor is arranged outside the nip of the two thinning pressure rollers. The thinning pressure roller is connected to a servo motor. A first tension isolation device is arranged on the incoming material side of the thinning pressure roller. A first thickness gauge is arranged on the incoming material side of the first tension isolation device. A floating roller is arranged on the discharge side of the thinning pressure roller. A second bearing clearance elimination mechanism is arranged on the outer sides of the two thinning support rollers; the winding component includes a winding shaft assembly. The winding shaft of the winding shaft assembly is connected to a servo motor. A second tension isolation device, a second thickness gauge, and an ultrasonic sensor are sequentially arranged on the incoming material side of the winding shaft assembly along the film sheet conveying direction. A deviation rectification actuator is arranged on the frame of the winding shaft assembly; Guide rollers are respectively arranged on the film-forming component, the thinning component, and the winding component.

[0032] The roll gap control actuator of the film forming component and the thinning component is electrically connected to the roll gap detection sensor and the second thickness gauge of the winding component in the roll gap control subsystem. The roll gap control actuator adjusts the roll gap of the film forming roll according to the roll gap information detected by the roll gap detection sensor and the film thickness information detected by the second thickness gauge; the servo motors connected to the film forming roll, the thinning roll, and the winding shaft assembly, and the swing roll motor of the micro-tension control unit are electrically connected to the belt speed control subsystem. The belt speed control subsystem uses the rotation speed of the servo motor of one of the rolls as a reference to adjust the rotation speeds of the servo motors of the other rolls, the winding shaft, and the swing roll motor; the swing roll motor, the angle sensor, and the tension sensor of the micro-tension control unit are electrically connected to the micro-tension control subsystem. The swing roll motor adjusts the micro-tension of the film according to the rotation angle information detected by the angle sensor and the tension information detected by the tension sensor; the ultrasonic sensor of the winding component is electrically connected to the deviation rectification actuator in the deviation rectification subsystem. The deviation rectification actuator adjusts the position of the winding shaft assembly according to the position information detected by the ultrasonic sensor to perform deviation rectification.

[0033] The beneficial effects of the present invention are as follows:

[0034] Each control subsystem of the calendering system of the present invention introduces the zebra optimization recurrent neural network algorithm (RNN-ZOA) and deploys it in the PID controller. The PID controller automatically adjusts the output quantity in real time according to the optimal time series prediction control quantity of the RNN-ZOA, and can accurately control the roll gap, belt speed, micro-tension, and deviation rectification. When facing a non-linear, time-varying or multi-variable system, it can also accurately process complex dynamic performance and multi-variable coupling, with better self-adaptability and robustness, improving the control performance of the system, the stability and accuracy of the system, and also improving the production efficiency and product quality of the calender. Compared with optimization algorithms such as GWO, GA, MPA, PSO, TSA, and GSA, the zebra optimization algorithm (ZOA) has a faster convergence speed in the problem of optimizing RNN.

[0035] The present invention uses a discrete-time system in the PID controller. The discrete-time system can suppress noise and interference through techniques such as digital filtering, is more robust than the continuous-time system and is suitable for real-time control applications, and can implement complex control strategies through easy algorithms.

[0036] The present invention performs fitting calculations or analytical calculations on time series data. Time series data refers to data columns recorded in chronological order, mainly focusing on the trends over a period of time. Calculating time series data can more accurately reflect the periodic or quasi-periodic changes of the data compared to calculating data at specific time points, enabling more accurate and real-time monitoring and prediction based on periodicity or quasi-periodicity. The calculated control quantity is more precise, better meeting actual requirements, with a shorter analysis duration and faster feedback.

[0037] The present invention integrates the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) into devices and systems, making them more intelligent and autonomous. Compared with traditional centralized computing models, the embedded algorithm model embeds AI algorithm capabilities into the device itself, enabling it to process data locally and make decisions without relying on cloud servers. This can achieve faster and more real-time data analysis and response, reduce dependence on the network, improve the efficiency and performance of the system, quickly understand and respond to user needs, automatically adapt to environmental changes, and provide more intelligent and personalized services. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic structural diagram of the calendering system of the present invention.

[0039] Figure 2 It is a schematic structural diagram of the micro-tension control unit of the present invention.

[0040] Figure 3 It is a schematic structural diagram of the swing roll seat.

[0041] Figure 4 It is a control flow chart of the roll gap control subsystem.

[0042] Figure 5 It is a control flow chart of the belt speed control subsystem.

[0043] Figure 6 It is a control flow chart of the micro-tension control subsystem.

[0044] Figure 7 It is a control flow chart of the deviation correction subsystem.

[0045] Figure 8 It is the optimization process of the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model.

[0046] Figure 9 It is a neural network control flow chart of the control centers of each subsystem.

[0047] Figure 10 It is a flow chart from the neural network to the PID controller and then to the actuator.

[0048] In the figure:

[0049] 10. Micro-tension control unit; 1. Swing roller; 2. Swing arm; 3. Swing roller seat; 310. Set screw hole; 4. Mandrel; 5. Pillow block bearing; 6. Swing roller motor; 7. Angle sensor; 8. Tension sensor;

[0050] 100. Film forming component; 11. Film forming support roller; 12. Film forming pressure roller; 13. First roll gap control actuator; 14. First roll gap detection sensor; 15. First bearing backlash elimination mechanism; 16. First guide roller;

[0051] 200. Thinning component; 21. Thinning support roller; 22. Thinning pressure roller; 23. First tension isolation device; 24. Floating roller; 25. Second bearing backlash elimination mechanism; 26. Second guide roller; 27. First thickness gauge; 28. Second roll gap control actuator; 29. Second roll gap detection sensor;

[0052] 300. Rewinding component; 31. Rewinding shaft assembly; 32. Second tension isolation device; 33. Second thickness gauge; 34. Ultrasonic sensor; 35. Deviation rectification actuator; 36. Third guide roller;

[0053] 400. Diaphragm. Specific embodiments

[0054] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0055] As Figure 1 shown, the calendering system of the present invention includes a film forming component 100, a thinning component 200, and a rewinding component 300 arranged in sequence along the production direction of the diaphragm 400.

[0056] The film forming member 100 includes a pair of film forming support rollers 11 and a film forming pressure roller 12. The film forming pressure roller 12 is located inside the film forming support roller 11. The film forming pressure roller 12 and the film forming support roller 11 are eccentrically arranged. The film forming pressure roller 12 is connected to a servo motor and driven by the servo motor (not shown in the figure). The film forming pressure roller 12 can be adjusted up and down relative to the film forming support roller 11. A first roll gap control actuator 13 is provided on the outer side of one of the film forming support rollers 11. The film forming support roller 11, the film forming pressure roller 12, and the first roll gap control actuator 13 are arranged in the horizontal direction. The first roll gap control actuator 13 controls the displacement of the film forming support roller 11 through a control valve group to adjust the roll gap width between the film forming pressure rollers 12. A first roll gap detection sensor 14 is provided on the outer side of the nip of the two film forming pressure rollers 12 for real-time detection of the roll gap between the two film forming pressure rollers 12. A first bearing clearance elimination mechanism 15 is provided on the outer sides of the two film forming support rollers 11 for eliminating the bearing clearances of the film forming support rollers 11 and the film forming pressure rollers 12, avoiding the influence of the bearing clearances on the roll gap adjustment, and ensuring the adjustment accuracy of the first roll gap control actuator 13. The film forming member 100 is also provided with a plurality of first guide rollers 16 for guiding the film sheet 400.

[0057] The thinning member 200 includes a pair of thinning support rollers 21 and a thinning pressure roller 22. The thinning pressure roller 22 is located inside the thinning support roller 21. The thinning pressure roller 22 can be adjusted left and right relative to the thinning support roller 21. The thinning pressure roller 22 is connected to a servo motor and driven by the servo motor (not shown in the figure). A second roll gap control actuator 28 is provided on the outer side of one of the thinning support rollers 21. The thinning support roller 21, the thinning pressure roller 22, and the second roll gap control actuator 28 are arranged in the vertical direction. The second roll gap control actuator 28 controls the displacement of the thinning support roller 21 through a control valve group to adjust the roll gap width between the thinning pressure rollers 22. A second roll gap detection sensor 29 is provided on the outer side of the nip of the two thinning pressure rollers 22 for real-time detection of the roll gap between the two thinning pressure rollers 22. A first tension isolation device 23 is provided on the incoming material side of the thinning pressure roller 22 for isolating the tension of the film sheet 400, avoiding the mutual influence of the tensions of the film forming member 100 and the thinning member 200. A floating roller 24 is provided on the outgoing material side of the thinning pressure roller 24. The floating roller 24 tensions the film sheet 400 by gravity, and the floating roller 24 can be weighted according to the tension requirement of the film sheet 400. A second bearing clearance elimination mechanism 25 is provided on the outer sides of the two thinning support rollers 21 for eliminating the bearing clearances of the thinning support rollers 21 and the thinning pressure rollers 22, ensuring the accuracy of the roll gap adjustment. A first thickness gauge 27 is provided on the incoming material side of the first tension isolation device 23. The first thickness gauge 27 is used to detect the thickness of the formed film sheet 400. The thinning member 200 is also provided with a plurality of second guide rollers 26 for guiding the film sheet 400.

[0058] The winding component 300 includes a winding shaft assembly 31, and the winding shaft of the winding shaft assembly 31 is connected to a servo motor and driven by the servo motor (not shown in the figure). A second tension isolation device 32, a second thickness gauge 33, and an ultrasonic sensor 34 are sequentially arranged on the material-incoming side of the winding shaft assembly 31 along the conveying direction of the film 400. The second tension isolation device 32 is used to isolate the tension of the film 400 to prevent the tension of the thinning component 200 and the winding component 300 from affecting each other. The second thickness gauge 33 is used to detect the thickness of the film 400. The ultrasonic sensor 34 is used to detect the position of the film 400. A deviation correction actuator 35 is arranged on the frame of the winding shaft assembly 31, which is used to move the frame for deviation correction to ensure that the film 400 is accurately centered when winding. The winding component 300 is also provided with a plurality of third guide rollers 36 for guiding the film 400.

[0059] The film forming part 100, the thinning part 200 and the winding part 300 are respectively provided with a micro tension control unit 10. The micro tension control unit 10 of the film forming part 100 is arranged on the discharge side of the film forming roller 12; the micro tension control unit 10 of the thinning part 200 is arranged between the thinning roller 22 and the first tension isolation device 23, and between the thinning roller 22 and the floating roller 24; the micro tension control unit 10 of the winding part 300 is arranged between the winding shaft assembly 31 and the second thickness gauge 33.

[0060] like Figure 2 As shown, the micro tension control unit 10 includes a tension swing roller assembly and at least one tension sensor 8. The tension sensor 8 is located at the discharge side of the tension swing roller assembly and is used to detect the tension of the film 400. Figure 1 As shown, in this embodiment, the film forming part 100 and the winding part 300 are respectively provided with a tension sensor 8; the thinning part 200 is provided with two tension sensors 8, and the two tension sensors 8 are respectively arranged on the feeding side and the discharging side of the thinning roller 22. Figure 2 As shown, the tension swing roller assembly includes two swing rollers 1 that are symmetrical on both sides. The two ends of the swing rollers 1 are connected to the swing arm 2 through the swing roller seat 3. The swing arm 2 is provided with a core shaft 4. The core shaft 4 on one side is connected to the swing roller motor 6 through the seat bearing 5. The core shaft 4 on the same side as the swing roller motor 6 is provided with an angle sensor 7. The angle sensor 7 is used to detect the rotation angle of the swing roller 1. The swing roller motor 6 adopts a servo motor. Figure 1As shown in the figure, opposite sides of the swing roller seat 3 are provided with setscrew holes 310. The central axes of the setscrew holes 310 on both sides pass through the center of the swing roller seat 3 and are located in the same plane. The swing roller seats 3 at both ends of the swing roller 1 are axially staggered by 90°. The central axes of the setscrew holes 310 of the swing roller seats 3 at both ends are perpendicular to each other. One end of the swing roller seat 3 can be adjusted in two directions through its two setscrew holes 310, and the other end of the swing roller seat 3 can be adjusted in the other two directions through its two setscrew holes 310. Therefore, the positions of both ends of the swing roller 1 can be adjusted in a total of four directions. Without disassembling the machine, each swing roller 1 can be finely adjusted in position in four directions to ensure the symmetry of the swing roller 1 and its parallelism with each other, and to ensure the accuracy of tension adjustment.

[0061] The present invention is monitored in real time and its parameters are adjusted in real time by a calendering control system to ensure the stability and accuracy of the calendering process. The calendering control system includes a roll gap control subsystem, a belt speed control subsystem, a micro-tension control subsystem, and a deviation correction subsystem.

[0062] The roll gap control subsystem is used to control the adjustment of the roll gaps of the film-forming roll 12 of the film-forming component 100 and the thinning roll 22 of the thinning component 200. The roll gap control actuators of the film-forming component 100 and the thinning component 200 are electrically connected to the roll gap control subsystem through the roll gap detection sensor, the first thickness gauge 27 of the thinning component 200, and the second thickness gauge 33 of the winding component 300. The roll gap control actuator adjusts the roll gaps of the film-forming roll 12 and the thinning roll 22 according to the roll gap information detected by the roll gap detection sensor and the film thickness information detected by the first thickness gauge 27 and the second thickness gauge 33.

[0063] The detailed steps are as follows: During the operation of the equipment, the first thickness gauge assembly 27 detects the thickness of the film 400 in real time and sends it to the control center. The control center calculates the deviation rate one between the actual thickness and the calibrated thickness at this location. The formula for the deviation rate one is

[0064]

[0065] where a is the calibrated thickness and x is the actual thickness

[0066] If the deviation rate one is within the normal fluctuation range (plus or minus 5%), the data is stored in the storage library without any control operation.

[0067] If the deviation rate one is within the rated range (plus or minus 5% to plus or minus 10%), the control center will adopt the first control module. First, based on the current sequential thickness data of the first thickness gauge 27, 10 sequential thickness data before this timestamp are retrieved from the storage library and combined with the current sequential thickness data to form a sequential data set {T1, T2, …, T 11}, and then the sequential data set {T1, T2, …, T11}Output parameter set {K after calculation by the input Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) P , T I , T D , tsd, Mpd, ess, e(t)} to the PID. After receiving the parameter set, the PID operates on e(t) to obtain a position adjustment command to the servo-hydraulic proportional valve of the second roll gap control actuator 28 of the thinning component 200. The time taken for this process is in milliseconds. The second roll gap control actuator 28 is controlled by the servo-hydraulic proportional valve to extend or retract to adjust the thinning support roll 21 and thus adjust the roll gap of the thinning component 200.

[0068] The deviation diaphragm thins after being corrected by the thinning component 200. The second thickness gauge 33 transmits the thickness of the finished diaphragm to the control center, and the control center calculates the deviation rate two between the actual thickness and the rated thickness of the finished diaphragm.

[0069] When the deviation rate two is less than plus or minus 6%, the roll gap is not corrected, and the time-series thickness data is stored in the storage library;

[0070] When the deviation rate two is greater than 6%, the time-series thickness data of the first thickness gauge 27 is masked. Based on the current time-series thickness data of the second thickness gauge 33, 10 time-series thickness data before this timestamp are retrieved from the storage library and combined with the current time-series thickness data to form a time-series data set {T1’, T2’, …, T 11 ’}. Subsequently, the time-series data set {T1’, T2’, …, T 11 ’} is input into the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) for calculation, and the output parameter set {K P , T I , T D, tsd, Mpd, ess, e(t)} is sent to the PID. After receiving the parameter set, the PID calculates e(t) to obtain a position adjustment instruction for the servo-hydraulic proportional valve of the second roll gap control actuator 28 of the thinning component 200. The time taken for this process is in milliseconds. The second roll gap control actuator 28 is controlled by the servo-hydraulic proportional valve to extend or retract to adjust the roll gap of the thinning component 200 until the deviation rate two is reduced within 6%. After the adjustment action is completed, the roll gap is detected by the second roll gap detection sensor 29 and fed back to the PID. Thereafter, the shielding of the first thickness gauge assembly 27 is removed, and adjustment is carried out based on the sequential thickness signal of the first thickness gauge assembly 27. At the same time, the zebra optimization recurrent neural network algorithm (RNN-ZOA) for controlling the roll gap of the thinning component 200 is optimized twice to ensure that the thickness deviation rate two of the finished diaphragm is within the set range. After the roll gap of the thinning component 200 is adjusted, the second thickness gauge 33 continuously monitors the thickness of the finished diaphragm 400 and sends it to the control center. If the thickness returns to the fluctuation range, this control is completed, and the second roll gap control actuator 28 stops operating.

[0071] If the deviation rate one exceeds the rated range (plus or minus 10%), this means that the second roll gap control actuator 28 needs a large adjustment or is no longer adjustable. Then the control center will adopt the second control module. First, based on the current sequential thickness data of the first thickness gauge 27, 10 sequential thickness data before this timestamp are retrieved from the storage library and combined with the current sequential thickness data to form a sequential data set {T1, T2, …, T 11}, and then the sequential data set {T1, T2, …, T 11} is input into the zebra optimization recurrent neural network algorithm (RNN-ZOA) for calculation and the output parameter set {K P , T I , T D , tsd, Mpd, ess, e(t)} is sent to the PID. After receiving the parameter set, the PID calculates e(t) to obtain a position adjustment instruction for the servo-hydraulic proportional valve of the first roll gap control actuator 13 of the film forming component 100. The time taken for this process is in milliseconds. The position instruction is sent through the servo-hydraulic proportional valve to control the first roll gap control actuator 13 to extend or retract to adjust the film forming support roll 11 and thus adjust the roll gap of the film forming component 100 until the thickness deviation rate one of the diaphragm 400 produced by the film forming component 100 is within plus or minus 10%. Then it switches to the first control module. During this period, the second thickness gauge 33 monitors and records the sequential data in real time, and removes the unqualified diaphragms 400 during this period after winding is completed.

[0072] During the roll gap adjustment process, the first roll gap detection sensor 14 synchronously monitors the actual roll gap and feeds the signal back to the control center to ensure the completion degree of the roll gap adjustment instruction. All the collected data are sequential data.

[0073] The ordinary PID control process is as follows: The first thickness gauge component 27 monitors the thickness of the diaphragm 400 and sends this signal to the PID. After the PID takes the average value of a dataset for a period of time, if the average value exceeds the fluctuation range compared with the set value, the output signal is adjusted to the roll gap between the film pressing rollers 12. Since the traditional PID control method measures after and adjusts before, there will inevitably be a section of the diaphragm 400 with unqualified thickness. In this way, the overall thickness uniformity of the diaphragm 400 will be poor. However, the control method based on the zebra optimization recurrent neural network algorithm (RNN-ZOA) can effectively reduce the length of the diaphragm 400 with unqualified thickness, making the overall thickness uniformity of the diaphragm 400 higher.

[0074] As Figure 4 shown, the control method of the roll gap control subsystem is as follows: 1) The first roll gap detection sensor 14 and the second roll gap detection sensor 29 transmit the real-time detected roll gap signals, and the first thickness gauge 27 and the second thickness gauge 33 transmit the real-time monitored thickness signals of the diaphragm 400 to the roll gap control center; 2) The roll gap control center performs fitting calculations on the roll gap signals and thickness signals based on the zebra optimization recurrent neural network algorithm (RNN-ZOA); 3) The roll gap control center sends the time series data obtained after the fitting calculation to the first roll gap control actuator 13 and the second roll gap control actuator 28; 4) The first roll gap control actuator 13 adjusts the displacements of the film-forming support roller 11 and the thinning support roller 21 through the control valve group according to the instruction to control the hydraulic system, thereby adjusting the displacements of the film pressing rollers 12 and the thinning pressing rollers 22, and further adjusting the roll gap distances between the two film pressing rollers 12 and the two thinning pressing rollers 22 to accurately adjust the thickness of the diaphragm 400.

[0075] The belt speed control subsystem is used to control the conveying speed of the diaphragm 400. The servo motors connected to the film pressing rollers 12, the thinning pressing rollers 22, the take-up reel assembly 31, the first tension isolation device 23, and the second tension isolation device 32 are electrically connected to the belt speed control subsystem. The belt speed control subsystem takes the rotation speed of the servo motor of one of the pressing rollers (here the thinning pressing roller 22) as a reference and adjusts the rotation speeds of the servo motors of the other pressing rollers, the take-up reel, and the swing roller motor 6.

[0076] The detailed steps are as follows:

[0077] The servo motors connected to the film-forming pressure roller 12, the thinning pressure roller 22, the winding shaft assembly 31, the first tension isolation device 23, and the second tension isolation device 32 transmit the timing data - rotational speed signals to the control center. The control center calculates the rotational speed deviation rates of the remaining servo motors with respect to the rotational speed of the servo motor of the thinning pressure roller 22 as a reference. When the deviation rate is within the normal range, the control center does not issue commands to other servo motors based on the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA). When the deviation rate exceeds the normal range, commands are issued to the servo motors with the exceeded deviation rate respectively to adjust their rotational speeds.

[0078] When the material or process changes and the belt speed needs to be changed, the control center will simultaneously issue commands (including parameters such as acceleration, final speed, etc.) to the servo motors connected to the film-forming pressure roller 12, the thinning pressure roller 22, the winding shaft assembly 31, the first tension isolation device 23, and the second tension isolation device 32 based on the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) to increase or decrease their rotational speeds simultaneously to meet the requirements of speed change. After adjusting their speeds, the rotational speed of the servo motor of the thinning pressure roller 22 is still used as a reference to monitor the rotational speeds of the remaining servo motors to ensure that the deviation rates of the rotational speeds of each servo motor from the rotational speed of the servo motor of the thinning pressure roller 22 are within the normal range.

[0079] The Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) embedded PID is more accurate in output parameters compared to traditional PID control.

[0080] As Figure 5 shown, the control method of the belt speed control subsystem is as follows: 1) Each servo motor feeds back the rotational speed signal to the servo driver; 2) The servo driver transmits the received signal to the belt speed control center; 3) The belt speed control center, based on the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA), uses the rotational speed of the servo motor of one of the pressure rollers (reference roller) in the film-forming component 100 or the thinning component 200 as a reference to analyze and calculate other servo motors; 4) The belt speed control center sends the commands obtained after analysis and calculation to the servo driver; 5) The servo driver adjusts the rotational speed of the servo motor according to the commands to make the servo motor stable at the set rotational speed, avoiding problems such as belt breakage or insufficient compaction tightness caused by excessive differential speed between servo motors.

[0081] The micro-tension control subsystem is used to control the tension of the diaphragm 400 to prevent belt breakage, wrinkling, etc. of the diaphragm 400. The swing roller motor 6, the angle sensor 7, and the tension sensor 8 of the micro-tension control unit 10 are electrically connected to the micro-tension control subsystem. The swing roller motor 6 makes micro-tension adjustments to the diaphragm 400 according to the rotational angle information detected by the angle sensor 7 and the tension information detected by the tension sensor 8.

[0082] The detailed steps are as follows:

[0083] Taking the micro-tension control subsystem at the film-forming component 100 as an example, the logic and steps of the other micro-tension control subsystems are the same. Before the equipment runs, first place the micro-tension control unit 10 horizontally so that the swing arm 2 is horizontal and the swing roller 1 is at the same height. Subsequently, initialize the swing roller motor 6 and adjust it to the torque mode to output the rated torque. When the equipment is running, the tension sensor 8 will continuously monitor the tension of the diaphragm 400 and transmit the timing data at this time to the micro-tension control subsystem. The timing data tension is collected and input into the control center. If the tension timing data exceeds the fluctuation range at a certain time, after the control center retrieves the timing tension of the previous period of this data, it calculates and fits this set of timing data sets through the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) to predict its tension trend. If this trend is within the fluctuation range, the control center will not correct it; if this trend is outside the fluctuation range or it is predicted that it will exceed the fluctuation range subsequently, a correction value is calculated based on this AI algorithm and an instruction is sent to the swing roller motor 6 to adjust the tension of this section by changing the output torque of the swing roller motor 6. When the adjustment of the output torque of the swing roller motor 6 is completed, the tension sensor 8 continuously monitors the diaphragm tension. If the tension is within the normal fluctuation range at this time, this control is completed.

[0084] During the process of tension fluctuation correction, the rotation angle of the swing roller motor 6 will change, which will cause the swing arm 2 to break the dynamic balance state of the overall horizontal. This overall horizontal dynamic balance has an inestimable impact on the control of micro-tension and the smooth operation of the diaphragm 400 in the tension section. Therefore, the present invention also makes a correction design for the position of the swing arm 2 to ensure the dynamic balance state of the overall horizontal of the swing arm 2. During the process of tension fluctuation correction, the angle signal of the swing roller motor 6 will be synchronously sent to the control center. Based on the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA), the control center jointly constructs a function in advance with the angle information of the swing roller motor 6 and the servo motor speed signal of the first tension isolation device 23, and outputs it to the servo motor of the first tension isolation device 23 through the associated function calculation to adjust its speed to adjust the position of the swing arm 2 and the angle of the swing roller motor 6; as the servo motor speed of the first tension isolation device 23 increases quasi-statically, the roller speed of the first tension isolation device 23 will be faster than the roller speed of the film-forming pressure roller 12, and a speed difference will gradually be generated between the two. The diaphragm 400 will be gradually tightened, and at this time, the swing arm 2 will gradually tend to be horizontal, so as to adjust the angle deflection of the swing roller motor 6. When this adjustment process is completed, the angle information of the swing roller motor 6 is still continuously sent to the control center, and the control center judges this data. If it is already within the normal fluctuation range, this control is completed.

[0085] The traditional PID cannot construct an effective function when adjusting the overall horizontal dynamic balance of the swing arm 2, and is worse than the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) in terms of variability and adaptability to data changes. The Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) can output parameters more accurately and give a prediction based on this.

[0086] As Figure 6 shown, the control method of the micro-tension control subsystem is as follows: 1) The angle sensor 7 transmits the rotation angle signal of the swing roller 1 detected in real time, and the tension sensor 8 transmits the tension signal of the diaphragm 400 detected in real time to the micro-tension control center; 2) The micro-tension control center analyzes and calculates the rotation angle signal and the tension signal based on the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA); 3) The micro-tension control center sends the instruction obtained from the analysis and calculation to the swing roller motor 6; 4) The swing roller motor 6 adjusts the rotation angle and speed of the swing roller 1 according to the instruction, thereby adjusting the tension of the diaphragm 400, and the tension adjustment is more accurate. In addition to real-time monitoring of the tension of the diaphragm 400 through the tension sensor 8, the micro-tension control subsystem also detects the angle of the swing roller 1 in real time through the angle sensor 7. The angle sensor 7 can feedback the actual rotation angle of the swing roller 1 during tension adjustment in real time, providing more accurate parameters for control and making the tension adjustment more stable and accurate.

[0087] The deviation rectification subsystem is used to adjust the position of the take-up reel assembly 31 to ensure that the diaphragm 400 is accurately centered during winding. The ultrasonic sensor 34 of the winding component 300 is electrically connected to the deviation rectification execution mechanism 35 of the deviation rectification subsystem. The deviation rectification execution mechanism 35 adjusts the position of the take-up reel assembly 31 for deviation rectification according to the position information detected by the ultrasonic sensor 34.

[0088] The detailed steps are as follows: When the device is running, the ultrasonic sensor 34 continuously monitors the opposite sides of the diaphragm 400. Due to factors such as the tension and friction during the operation of the diaphragm 400, there will be a deviation in the relative position between the diaphragm 400 and the take-up reel assembly 31 at different times, which may lead to uneven hemming. Therefore, the ultrasonic sensor 34 monitors the opposite sides of the diaphragm and converts it into a position offset. At the same time, this offset signal is transmitted to the control center. The control center judges the offset in real time. If it does not exceed the fluctuation range, the control center does not issue an instruction. If the offset at a certain time sequence exceeds the fluctuation range, the control center immediately takes this time sequence offset as a reference, retrieves some time sequence data before this time period, and performs fitting calculations through the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) to predict its future trend. On this basis, the motor rotation angle of the deviation correction actuator 35 is calculated, and this angle signal is transmitted to the servo motor of the deviation correction actuator 35. The servo motor rotates by the corresponding angle to drive the movement rod of the deviation correction actuator 35 to extend and retract, thereby pushing the take-up reel assembly 31 to move, ensuring the neatness of the hemming after the diaphragm 400 is wound. When this step is completed, the ultrasonic sensor 34 still continuously monitors and sends signals. At the end, if the deviation amount is within the normal fluctuation range, the angle of the servo motor of the deviation correction actuator 35 no longer changes, and this deviation correction control is completed.

[0089] As Figure 7 shown, the control method of the deviation correction subsystem is as follows: 1) The ultrasonic sensor 34 transmits the position signal of the diaphragm 400 detected in real time to the deviation correction control center; 2) The deviation correction control center analyzes and calculates the position signal based on the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA); 3) The deviation correction control center sends the instruction obtained from the analysis and calculation to the deviation correction actuator 35; 4) The deviation correction actuator 35 controls the overall movement of the frame of the take-up reel assembly 31 according to the instruction, adjusts the position of the take-up reel assembly 31 to achieve deviation correction, and ensures winding centering.

[0090] The control centers of the roll gap control subsystem, the belt speed control subsystem, the micro-tension control subsystem, and the deviation correction subsystem all use the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) to perform fitting calculations or analytical calculations on time series data, predict reasonable control quantities based on the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA). The control quantity corresponds to the displacements of the film-forming support roll 11 and the thinning support roll 21 in the roll gap control subsystem, the rotational speed of the servo motor in the belt speed control subsystem; the rotation angle and rotational speed of the swing roll motor 6 in the micro-tension control subsystem; and the movement amount of the frame of the take-up reel assembly 31 in the deviation correction subsystem. The neural network structure includes an input layer, a hidden layer, and an output layer, and the neural network also has a suitable activation function. The Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) performs fitting calculations or analytical calculations on time series data. Time series data refers to data columns recorded in chronological order, and time series data mainly focuses on the trends in time periods (a period of time). Calculating time series data can more accurately reflect the periodic or quasi-periodic changes of the data compared to calculating data at specific time points, make more accurate and real-time monitoring and predictions based on periodicity or quasi-periodicity, calculate more accurate control quantities, better meet actual requirements, have a shorter analysis duration, and faster feedback.

[0091] The calendering control system of the present invention (using the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA)) is embedded and deployed as a control module into the PID controller. The control quantity obtained by the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) is used as the input vector of the PID controller, and the output quantity calculated by the PID controller is output to the corresponding actuator to adjust the corresponding components. The calendering control system of the present invention is embedded and deployed as a control module into the PID controller, concentrating the Recurrent Neural Network-Zebra Optimization Algorithm (RNN-ZOA) into the device and system, making it more intelligent and autonomous. Compared with the traditional centralized computing model, the embedded algorithm model embeds the AI algorithm ability into the device itself, enabling it to process data and make decisions locally without relying on a cloud server, achieving faster and more real-time data analysis and response, reducing dependence on the network, improving the efficiency and performance of the system, and also being able to quickly understand and respond to user needs, automatically adapt to environmental changes, and provide more intelligent and personalized services.

[0092] In the present invention, a discrete-time system is used in the PID controller. The discrete-time system can suppress noise and interference through techniques such as digital filtering, is more robust than the continuous-time system and suitable for real-time control applications, and can implement complex control strategies through easy algorithms. In the present invention, the PID uses the Euler transform discretization control algorithm: where P is the proportional band, and the smaller P is, the amplification value K of the error signal (e(t)) PThe larger it is; I represents the integral part of the control algorithm, which is used to eliminate the "residual error". Without this part, even in the steady state, the set value and the process value may be different. Finally, D represents the derivative part of the algorithm, which has two objectives. It makes the system more flexible to sudden changes or disturbances in the set value, but most importantly, it helps to stabilize the system.

[0093] The transfer function of PID is:

[0094]

[0095] where s is the Laplace variable (s = α + jω), and K p is the amplification coefficient, T I is the integral time, and T D is the derivative time. {K p , T I , T D} is the parameter set of PID, and this parameter set depends on the physical properties in the process.

[0096] The adjustment function of PID is:

[0097]

[0098] where the integral of e(t) can be approximated as a rectangular sum, that is

[0099]

[0100] where e’(t) can be approximated as a straight line, that is

[0101]

[0102] Therefore, when PID uses the Euler transform discrete control algorithm, the following expression is obtained:

[0103]

[0104] As Figure 10 shown, the present invention introduces the zebra algorithm to optimize the neural network model based on the Euler transform discrete control algorithm, and improves the accuracy of the PID input parameter set {K p , T I , T D} through the neural network model. At the same time, three basic characteristics of the dynamic response of neural network training {maximum overshoot (Mp), settling time (ts), steady-state error (ess)} are introduced to improve the accuracy.

[0105] Before the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) is embedded in the control system, it needs to be trained to learn the mapping function X→Y, where X is the input vector and Y is the output vector. The recurrent neural network model is trained by experts using historical data, which includes the roll gap width, diaphragm thickness, motor speed, diaphragm tension, swing roll angle, position, and other key time-series parameter data required by each sub-control system. The information collected by the sensor is preprocessed before being input into the recurrent neural network model. The preprocessing includes operations such as noise removal and data normalization, which is beneficial to improving the training effect of the neural network model. The recurrent neural network model can also be updated and adjusted regularly and retrained when the working conditions change to adapt to the changes in the working environment and the evolution of the system. The recurrent neural network model can also store multiple processes and perform self-learning to obtain a flexible self-regulation function. During the training process of the recurrent neural network model, the zebra optimization algorithm is used to continuously adjust the weights and biases of the neural network to minimize the error value between the predicted output and the actual output, and improve the accuracy and stability of the predicted output.

[0106] Figure 8 The optimization process of the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model is shown. In the problem of optimizing the RNN model, there are five variables in the parameter matrix to be optimized: the number of network layers, the number of neurons, the learning rate, the epoch, and the batch size. Each variable needs to set upper and lower limits.

[0107] The objective function to be minimized is:

[0108]

[0109] t is the training time of the neural network, and loss is the loss value of the neural network. m and n represent weights, which are user-defined parameters, enabling the user to selectively focus on the optimization goal.

[0110] After the optimization starts, the dataset needs to be divided into a training set and a test set first.

[0111] Next, build the RNN model. The variables to be optimized are not set for the time being and are to be determined.

[0112] Randomly initialize n sets of hyperparameters (n is a positive integer). These n sets of hyperparameters are called a population. Substitute the population into the RNN model and train the model. After the training is completed, the objective function value can be obtained.

[0113] Next, optimize and update the hyperparameters according to the zebra optimization algorithm. The optimization goal is to minimize the objective function. The parameter matrix to be optimized is abstracted as the position of the zebra.

[0114] In the zebra optimization algorithm, the first stage is the foraging stage. The individual that forages the most in the population (i.e., the set of hyperparameters that minimizes the objective function) is the pioneer zebra, and other individuals will move towards the position of the pioneer zebra. The mathematical model is as follows:

[0115]

[0116]

[0117] where r is a random number between [0, 1], and I is a random value belonging to the set {1, 2}.

[0118] The second stage is the stage of encountering predators. In this stage, for the encountered predators (determined by a random seed), according to the strength of the predators, individuals will choose to disperse or aggregate. Dispersing or aggregating will update the positions of the individuals. If the new position of an individual makes the objective function smaller, the position of that individual will be updated. The mathematical model is as follows:

[0119]

[0120]

[0121] where t is the number of iterations, T is the maximum number of iterations, R is a constant of 0.01, and Ps is the switching probability between the two strategies. Its value is a random number between [0, 1], and AZ is the state of the attacked zebra.

[0122] The process of the optimization algorithm is iterative. An end condition for the iteration needs to be set for the algorithm. The end conditions for the iteration can be: reaching the upper limit of the loop iteration times; the objective function value obtained by the optimal parameters is less than the ideal value; the optimal parameters have not been updated after m rounds of iteration... or one or more combinations of the above conditions.

[0123] After the iteration ends, find a set of parameters that minimizes the objective function. This set of parameters is the optimal solution. The RNN model corresponding to the optimal solution is the optimal RNN model. Bringing the test set into the optimal RNN model can obtain the accuracy of the model on the test set.

[0124] As Figure 9 shown, the model optimized and selected by the zebra optimization recurrent neural network algorithm (RNN-ZOA) is embedded in the control system. The specific control method of the control center of the system is as follows:

[0125] 1) The zebra optimization recurrent neural network algorithm (RNN-ZOA) starts to run;

[0126] 2) Initialization of the execution components of each control subsystem: The first roll gap control actuator 13 and the second roll gap control actuator 28 of the roll gap control subsystem are initialized to their initial positions; the servo driver of the belt speed control subsystem is initialized to a constant speed; the swing roll motor 6 of the micro-tension control subsystem is initialized to a constant torque; the deviation correction actuator 35 of the deviation correction subsystem is initialized to its initial position; the initial positions of each execution component are preset during installation;

[0127] 3) The Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model predicts the initial control quantity of the execution components based on the difference between the actual parameter values and the set parameter values of the execution components of each control subsystem;

[0128] 4) The actual parameter values detected in real time by the detection components of each control subsystem are input into the Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model as lightweight input vectors;

[0129] For the roll gap control subsystem, the input vector X1 = [W1_roll gap, W2_roll gap, T1_film, T2_film, timestamp] (W1 is the roll gap width between two film-forming rolls 12, W2 is the roll gap width between two thinning rolls 22, T1 is the thickness of the film sheet 400 before calendering (data detected by the first thickness gauge 27), T2 is the thickness of the film sheet 400 after calendering (data detected by the second thickness gauge 33), and timestamp is the time series);

[0130] For the belt speed control subsystem, the input vector X2 = [n1_motor] (n1 is the speed of each servo motor);

[0131] For the micro-tension control subsystem, the input vector X3 = [θ_actual, θ_set, v, a] (where θ_actual is the actual angle of the swing roll 1, θ_set is the set angle of the swing roll 1, v is the speed of the swing roll motor 6, and a is the acceleration of the swing roll motor 6);

[0132] For the deviation correction subsystem, the input vector X4 = [ΔL] (ΔL is the displacement amount of the film sheet 400 offset);

[0133] 5) The Zebra Optimization Recurrent Neural Network Algorithm (RNN-ZOA) model optimizes the data and calculates the predicted real-time control quantity, and the real-time control quantity is used as the lightweight output vector;

[0134] For the roll gap control subsystem, the output vector Y1 = [P] (P is the pressure execution instruction of the oil cylinders of the first roll gap control actuator 13 and the second roll gap control actuator 28);

[0135] For the belt speed control subsystem, the output vector Y2 = [n_motor] (n is the rotational speed execution instruction of the motor of the reference roller);

[0136] For the micro-tension control subsystem, the output vector Y3 = [y_1, y_2,...] (where y_1 is the speed execution instruction of the swing roller motor 6, and y_2 is the acceleration execution instruction of the swing roller motor 6);

[0137] For the deviation rectification subsystem, the output vector Y4 = [α] (α is the position execution instruction of the actuator of the deviation rectification actuator 35);

[0138] 6) The lightweight output vector of the zebra optimization recurrent neural network algorithm (RNN-ZOA) model is input into the PID controller, and the PID controller adjusts the output quantity in real time according to the output vector (time series prediction control quantity) of the neural network model;

[0139] The adjustment function of the PID controller is: u'(t) = u(t) + f(Y) (where u(t) is usually a function based on the error e(t), including proportional, integral, and differential terms, and the error e(t) refers to the difference between the actual parameter value and the set parameter value; f(Y) is the empirical adjustment function).

[0140] 7) Judge whether the signal is in a stable state from the values detected by the detection components of each control subsystem: If it is stable, perform the next step to complete the control; if it is not stable, return to the start of operation and run again until the signal is stable.

[0141] Each control subsystem of the calendering system of the present invention introduces the combination of the zebra optimization recurrent neural network algorithm (RNN-ZOA) and the PID controller. The PID controller automatically adjusts the output quantity in real time according to the real-time prediction control quantity of the zebra optimization recurrent neural network algorithm (RNN-ZOA), and can accurately control the roll gap, belt speed, micro-tension, and deviation rectification. When facing a non-linear, time-varying or multi-variable system, it can also precisely process complex dynamic performance and multi-variable coupling, with better self-adaptability and robustness, improving the control performance of the system, the stability and accuracy of the system, and also improving the production efficiency and product quality of the calender.

[0142] The above description is an explanation of the present invention, not a limitation of the present invention. Without departing from the spirit of the present invention, the present invention can be modified in any form.

Claims

1. A control method for a dry electrode calender based on an AI algorithm, characterized in that: The calendering control system includes a roll gap control subsystem, a tape speed control subsystem, a micro-tension control subsystem, and a deviation rectification subsystem; The roll gap control subsystem adjusts the roll gap distance between the two film-forming press rolls (12) and / or the two thinning press rolls (22) according to the roll gap signal and the film thickness signal of the diaphragm (400); The tape speed control subsystem adjusts the rotational speed of the servo motors according to the rotational speed signals of the respective servo motors; the servo motors connected to the film-forming press rolls (12), the thinning press rolls (22), the take-up reel assembly (31), the first tension isolation device (23), and the second tension isolation device (32) are connected to the tape speed control subsystem. The tape speed control subsystem uses the rotational speed of the servo motor of one of the press rolls as a reference to adjust the rotational speeds of the servo motors of the other press rolls, the take-up reel, and the swing roll motor (6); The micro-tension control subsystem adjusts the rotational angle and speed of the swing roll (1) according to the rotational angle signal of the swing roll (1) and the tension signal of the diaphragm (400); the micro-tension control subsystem collects the timing data of the tension sensor (8). If the timing tension data at a certain time exceeds the fluctuation range, after the control center retrieves the timing tension of the previous period of this data, it calculates and fits this set of timing data sets through the zebra optimization recurrent neural network algorithm to predict its tension trend. If this trend is outside the fluctuation range or it is predicted that it will exceed the fluctuation range subsequently, a correction value is calculated and an instruction is sent to the swing roll motor (6) to adjust the tension of this section by changing the output torque of the swing roll motor (6). When the adjustment of the output torque of the swing roll motor (6) is completed, the tension sensor (8) continuously monitors the diaphragm tension; During the process of tension fluctuation correction, the angle signal of the swing roll motor (6) is synchronously sent to the control center. The control center constructs a function by combining the angle information of the swing roll motor (6) and the rotational speed signal of the servo motor of the first tension isolation device (23) in advance based on the zebra optimization recurrent neural network algorithm, and outputs it to the servo motor of the first tension isolation device (23) through the associated function calculation to adjust its rotational speed to adjust the position of the swing arm (2) and the angle of the swing roll motor (6); as the rotational speed of the servo motor of the first tension isolation device (23) increases quasi-statically, the roll speed of the first tension isolation device (23) will be faster than the roll speed of the film-forming press roll (12), and a speed difference will gradually occur between the two. The diaphragm (400) will be gradually tightened, and at this time, the swing arm (2) will gradually tend to be horizontal to adjust the angle deflection of the swing roll motor (6); The deviation rectification subsystem adjusts the frame of the take-up reel assembly (31) according to the position signal of the diaphragm (400); The control center of the roll gap control subsystem, the tape speed control subsystem, the micro-tension control subsystem, and the deviation rectification subsystem uses the zebra optimization recurrent neural network algorithm to calculate the signals, predicts reasonable control quantities, and uses the obtained control quantities as the input vector of the PID controller. The output quantity calculated by the PID controller is output to the corresponding actuator to adjust the corresponding components.

2. The control method of the dry electrode calender based on the AI algorithm according to claim 1, wherein: The PID controller uses a discrete-time system; the transfer function of PID is where s is the Laplace variable, (s = α + jω), and K p is the amplification factor, T I is the integral time, and T D is the derivative time; The adjustment function of the PID is The integral of e(t) is approximately the sum of rectangles, ∫e(t)dt≈∑ i e(i)×T S =T s ∑ i e(i), e’(t) is approximately a straight line The control expression of the PID is 3. The control method of the dry electrode calender based on the AI algorithm according to claim 1, characterized in that: The calendering control system is embedded and deployed into the PID controller as a control module.

4. The dry electrode calender control method based on the AI algorithm according to claim 1, characterized in that: The control process of the roll gap control subsystem includes: the control center calculates the first deviation rate based on the actual thickness detected by the first thickness gauge assembly (27) and the calibrated thickness here. When the first deviation rate is within the range of ±5%, the data is stored in the storage repository without control operations. When the first deviation rate is within the range of ±5% to ±10%, the control center adopts the first control module to adjust the roll gap of the thinning component (200). When the first deviation rate exceeds the range of ±10%, the control center adopts the second control module to adjust the roll gap of the film forming component (100).

5. The control method of the dry electrode calender based on the AI algorithm according to claim 4, wherein: After the first control module adjusts the roll gap of the thinning component (200), the deviation film is thinned after being corrected by the thinning component (200). The control center calculates the second deviation rate based on the actual thickness detected by the second thickness gauge (33) and the rated thickness here. When the second deviation rate is less than ±6%, the roll gap is not corrected. When the second deviation rate is greater than ±6%, based on the thickness signal of the second thickness gauge (33), the roll gap of the thinning component (200) is adjusted until the second deviation rate is reduced to within ±6%. At the same time, the zebra optimization recurrent neural network algorithm for controlling the roll gap of the thinning component (200) is optimized twice.

6. The control method of the dry electrode calender based on the AI algorithm according to claim 4, characterized in that: After the second control module adjusts the roll gap of the film forming component (100) so that the first deviation rate is within the range of ±10%, it switches to the first control module; during the time of control and adjustment by the second control module, the second thickness gauge (33) monitors and records the timing data in real time, and removes the unqualified film (400) during this period after winding is completed.

7. The control method of the dry electrode calender based on the AI algorithm according to claim 1, characterized in that: The control center of the belt speed control subsystem calculates the rotational speeds of the other servo motors and the reference rotational speed based on the rotational speed of the servo motor of the thinning pressure roller (22), and calculates the rotational speed deviation rate of each servo motor respectively. When the deviation rate exceeds the normal range, commands are sent to the servo motors with the exceeded deviation rate to adjust their rotational speeds respectively.

8. The control method of the dry electrode calender based on the AI algorithm according to claim 1, characterized in that: The control center of the deviation correction subsystem judges the position offset of the film (400) in real time. If the offset at a certain timing exceeds the fluctuation range, the control center immediately takes this timing offset as the reference, retrieves some of the timing data before this period for fitting calculation, calculates the motor rotation angle of the deviation correction actuator (35), and transmits this angle signal to the servo motor of the deviation correction actuator (35). The servo motor rotates by the corresponding angle to drive the moving rod of the deviation correction actuator (35) to extend and retract, thereby pushing the winding shaft assembly (31) to move, ensuring the neatness of the rolled edge of the film (400) after winding.

9. The control method of the dry electrode calender based on the AI algorithm according to claim 1, characterized in that: The control method of the zebra optimization recurrent neural network algorithm model includes the following steps: 1) Start running; 2) Initialize the execution components of each control subsystem; 3) The zebra optimization recurrent neural network algorithm predicts the initial control amount of the execution component based on the difference between the initialized actual parameter value and the set parameter value of the execution component of each control subsystem; 4) The actual parameter values detected in real time by the detection components of each control subsystem are input into the neural network model as lightweight input vectors; 5) The prediction real-time control quantity is calculated by the zebra optimization recurrent neural network algorithm model, and the real-time control quantity is used as the lightweight output vector; 6) The lightweight output vector of the zebra optimization recurrent neural network algorithm model is input into the PID controller, and the PID controller adjusts the output quantity in real time according to the output vector; The adjustment function of the PID controller is: u'(t) = u(t) + f(Y), where u(t) is a function based on the error e(t), including proportional, integral, and differential terms. The error e(t) refers to the difference between the actual parameter value and the set parameter value, and f(Y) is an empirical adjustment function; 7) Determine whether the signal is in a stable state: If it is stable, execute the next step to complete the control; If it is unstable, return to the start of operation and re-run until the signal is stable.

10. The dry electrode calender control method based on the AI algorithm according to claim 1, characterized in that: Before the zebra optimization recurrent neural network algorithm model is embedded in the control system, it needs to be trained first to learn the mapping function X→Y, where X is the input vector and Y is the output vector; The zebra optimization recurrent neural network algorithm model uses historical data for expert training. The historical data includes the roll gap width, diaphragm thickness, motor speed, diaphragm tension, swing roll angle, position, and other key time-series parameter data required by each sub-control system. The zebra optimization recurrent neural network algorithm model is updated and adjusted regularly and re-trained when the working conditions change; During the training process of the recurrent neural network model, optimization algorithms such as gradient descent and Adam are used.

11. The control method of the dry electrode calender based on the AI algorithm according to claim 1, characterized in that: The optimized parameter matrix of the zebra optimization recurrent neural network algorithm model has five variables: the number of network layers, the number of neurons, the learning rate, epoch, and batch size. Each variable is set with upper and lower limits; The minimized objective function for optimization is: f(t, loss) = m * t + n * loss, where t is the training time of the neural network, loss is the loss value of the neural network, and m and n represent weights; The process of the optimization algorithm is iterative. The iteration end conditions are one or more combinations of reaching the upper limit of the number of loop iterations, the objective function value obtained by the optimal parameters being less than the ideal value, and the optimal parameters not being updated after m rounds of iteration; After the iteration ends, the set of parameters with the minimum objective function is the optimal solution, and the RNN model corresponding to the optimal solution is the optimal RNN model. Bringing the test set into the optimal RNN model can obtain the accuracy of the model on the test set.

12. The control method of the dry-type pole piece calender based on the AI algorithm according to claim 1, characterized in that: The model in the first stage of zebra optimization is r is a random number between [0, 1], and I is a random value belonging to the set {1, 2}; the model in the second stage of zebra optimization is t is the number of iterations, T is the maximum number of iterations, R is a constant of 0.01, Ps is the switching probability between the two strategies, whose value is a random number between [0, 1], and AZ is the state of the attacked zebra.

13. A calendering system for a dry electrode sheet calender, characterized in that: The control is carried out by using the dry-type pole piece calender control method based on the AI algorithm described in claim 1, including a film forming component (100), a thinning component (200), and a winding component (300) arranged in sequence along the production direction of the diaphragm (400); The film forming component (100), the thinning component (200), and the winding component (300) are each provided with a micro-tension control unit (10).

14. The calendering system of the dry electrode calender according to claim 13, characterized in that: The micro-tension control unit (10) includes a tension pendulum roller assembly and at least one tension sensor (8), and the tension sensor (8) is located on the discharging side of the tension pendulum roller assembly; the tension pendulum roller assembly includes two symmetrically arranged pendulum rollers (1) on the left and right. Both ends of the pendulum roller (1) are respectively connected to the swing arm (2) through the pendulum roller seat (3). A mandrel (4) is passed through the swing arm (2). One of the mandrels (4) is connected to the pendulum roller motor (6) through a pedestal bearing (5), and an angle sensor (7) is arranged on the mandrel (4) on the same side as the pendulum roller motor (6). The angle sensor (7) is used to detect the rotation angle of the pendulum roller (1); the pendulum roller motor (6) uses a servo motor; the pendulum roller motor (6), the angle sensor (7), and the tension sensor (8) of the micro-tension control unit (10) are electrically connected to the micro-tension control subsystem; on the opposite sides of the pendulum roller seat (3), there are set screw holes (310). The central axes of the two side set screw holes (310) pass through the center of the pendulum roller seat (3) and are located in the same plane. The pendulum roller seats (3) at both ends of the pendulum roller (1) are axially staggered by 90°. The central axes of the set screw holes (310) of the pendulum roller seats (3) at both ends are perpendicular to each other.

15. The calendering system of the dry electrode sheet calender according to claim 13, characterized in that: The film forming component (100) includes a pair of film forming support rollers (11) and a film forming pressure roller (12). The film forming pressure roller (12) is located inside the film forming support rollers (11). The film forming pressure roller (12) and the film forming support rollers (11) are eccentrically arranged. The film forming pressure roller (12) is connected to a servo motor. A first roll gap control actuator (13) is provided on the outer side of one of the film forming support rollers (11). The first roll gap control actuator (13) controls the displacement of the film forming support roller (11) through a control valve group to adjust the roll gap width between the film forming pressure rollers (12). The film forming support rollers (11), the film forming pressure rollers (12), and the first roll gap control actuator (13) are arranged in the horizontal direction. A first roll gap detection sensor (14) is provided directly outside the nip of the two film forming pressure rollers (12). A first bearing backlash elimination mechanism (15) is provided on the outer sides of the two film forming support rollers (11); The thinning component (200) includes a pair of thinning support rollers (21) and a thinning pressure roller (22). The thinning pressure roller (22) is located inside the thinning support rollers (21). A second roll gap control actuator (28) is provided on the outer side of one of the thinning support rollers (21). The thinning support rollers (21), the thinning pressure roller (22), and the second roll gap control actuator (28) are arranged in the vertical direction. A second roll gap detection sensor (29) is provided outside the nip of the two thinning pressure rollers (22). The thinning pressure roller (22) is connected to a servo motor. A first tension isolation device (23) is provided on the incoming material side of the thinning pressure roller (22). A first thickness gauge (27) is provided on the incoming material side of the first tension isolation device (23). A floating roller (24) is provided on the outgoing material side of the thinning pressure roller (22). A second bearing backlash elimination mechanism (25) is provided on the outer sides of the two thinning support rollers (21); The winding component (300) includes a winding shaft assembly (31). The winding shaft of the winding shaft assembly (31) is connected to a servo motor. A second tension isolation device (32), a second thickness gauge (33), and an ultrasonic sensor (34) are sequentially arranged on the incoming material side of the winding shaft assembly (31) along the conveying direction of the film sheet (400). A deviation correction actuator (35) is provided on the frame of the winding shaft assembly (31); The film forming component (100), the thinning component (200), and the winding component (300) are respectively provided with guide rollers.

16. The calendering system of the dry electrode calender according to claim 15, characterized in that: The roll gap control actuator of the film forming component (100) and the thinning component (200), the roll gap detection sensor, and the second thickness gauge (33) of the winding component (300) are electrically connected to the roll gap control subsystem. The roll gap control actuator adjusts the roll gap of the film forming press roll (12) according to the roll gap information detected by the roll gap detection sensor and the thickness information of the film sheet (400) detected by the second thickness gauge (33); the film forming press roll (12), the thinning press roll (22), the servo motors connected to the winding shaft assembly (31), and the swing roll motor (6) of the micro-tension control unit (10) are electrically connected to the belt speed control subsystem. The belt speed control subsystem adjusts the speeds of the servo motors of the other press rolls, the winding shaft, and the swing roll motor (6) with the speed of one of the press roll servo motors as the reference; the swing roll motor (6), the angle sensor (7), and the tension sensor (8) of the micro-tension control unit (10) are electrically connected to the micro-tension control subsystem. The swing roll motor (6) makes micro-tension adjustments to the film sheet (400) according to the rotation angle information detected by the angle sensor (7) and the tension information detected by the tension sensor (8); the ultrasonic sensor (34) of the winding component (300) and the deviation rectifying actuator (35) are electrically connected to the deviation rectifying subsystem. The deviation rectifying actuator (35) adjusts the position of the winding shaft assembly (31) for deviation rectification according to the position information detected by the ultrasonic sensor (34).

Images