A multi-stage cooperative high-precision constant tension control method and system for an electrode foil production line

Abstract

This invention discloses a multi-segment collaborative high-precision constant tension control method and system for electrode foil production lines. By deeply integrating three advanced technologies—multivariable dynamic feedforward decoupling control, adaptive neural fuzzy inference, and multi-source sensor data fusion—it fundamentally solves the inter-segment coupling problem, actively and dynamically eliminating coupling interference between multiple tension segments in the electrode foil production line, achieving truly segmented independent, high-precision closed-loop control. Furthermore, it constructs an intelligent architecture to realize tension progress control, process quality correlation analysis, predictive equipment maintenance, and parameter self-optimization, promoting the digital and intelligent upgrading of electrode foil production. This invention achieves a revolutionary improvement in control accuracy from ±5% to within ±1%, reducing foil breakage rate by over 90%, providing a feasible technical path for the stable production of ultra-high specific capacitance foil materials.

Description

Technical Field

[0001] This invention relates to the field of precision manufacturing equipment and industrial automation control technology for high-end metal foils, specifically to a multi-segment collaborative high-precision constant tension control method and intelligent control system for electrode foils used in aluminum electrolytic capacitors, particularly for etched foils and chemically formed foils in continuous production lines. Background Technology

[0002] Electrode foil (etched foil, formed foil) is a core material for aluminum electrolytic capacitors, and its production process has the following significant characteristics:

[0003] The material is extremely fragile: the corroded foil has a honeycomb-like porous structure with very low mechanical strength and is extremely sensitive to tension fluctuations. Even a slight over-tension can cause the foil to break, and uneven tension can lead to tensile deformation and damage to the porous structure, directly affecting the capacitor's key electrical properties such as capacitance, loss angle, and leakage current.

[0004] Multiple process segments with tight coupling: A typical electrode foil production line includes multiple continuous process segments such as unwinding, pretreatment, multi-stage etching, multi-stage formation, post-treatment and rewinding. Each segment forms a strongly coupled system through continuous foil materials.

[0005] Harsh production environment: The corrosion tank (acid solution) and formation tank (electrolyte solution) are subject to harsh environmental factors such as chemical corrosion and conductive ions.

[0006] Extremely high quality requirements: As the core dielectric material of capacitors, it has extremely high requirements for thickness uniformity, surface structure consistency, and electrical performance consistency.

[0007] The existing technology has the following problems: Currently, the industry generally adopts single-segment or segmented independent control schemes based on floating rollers (Dancer) and PID controllers, which have the following inherent defects:

[0008] Technical Issue 1: Severe Inter-Segment Dynamic Coupling Interference

[0009] There are strong mechanical and electrical couplings between the tension sections. Any tension adjustment or speed change in any section will act as a strong disturbance, instantly transmitted through the foil to the entire line, causing a systemic oscillation that has a ripple effect. Especially in the corrosion tank and formation tank, the foil vibration will seriously affect the uniformity of the chemical reaction, resulting in uneven corrosion pore size and inconsistent formation film thickness.

[0010] Technical Issue 2: Insufficient control precision leads to high foil breakage rate and low quality.

[0011] Traditional PID controllers struggle to handle highly nonlinear factors such as changes in roll diameter, time-varying material properties, and fluctuations in friction, resulting in tension control accuracy typically within ±5% or even worse. This is the primary reason for the high rate of foil production interruptions (an industry average of approximately 1-2 interruptions per 10,000 meters) and low product qualification rate (A-grade rate). Statistics show that when tension fluctuations exceed ±3%, the risk of foil breakage increases by more than 300%.

[0012] Technical problem 3: The system has a low level of intelligence, making debugging and maintenance difficult.

[0013] The parameters of multiple PID segments are mutually restrictive, and debugging relies on the experience of engineers, with adjustment cycles lasting several hours during production changes.

[0014] Lack of in-depth monitoring and quality correlation analysis capabilities for the tension process;

[0015] Predictive maintenance is not possible, and mechanical failures are often only discovered after foil breakage has occurred;

[0016] There is a lack of quantitative correlation between control parameters and final product quality (capacity CV value, leakage current).

[0017] Technical problem 4: Poor environmental adaptability. Traditional potentiometers and ordinary encoders have short lifespans, signal distortion, and require frequent maintenance in corrosive environments. Summary of the Invention

[0018] This invention aims to overcome the aforementioned deficiencies of the prior art and provide a multi-segment collaborative high-precision constant tension control method and system specifically for electrode foil production lines. Specific objectives include:

[0019] Fundamentally solves the inter-segment coupling problem: Through an innovative control architecture, it actively and dynamically eliminates coupling interference between multiple tension segments, achieving truly segmented independent, high-precision closed-loop control.

[0020] Achieving ultra-precision tension control: Improving the tension control accuracy of the entire production line from the traditional ±5% or more to within ±1%, laying a physical foundation for the stable production of ultra-high specific capacitance electrode foil.

[0021] Significantly improves production reliability: By suppressing tension mutations and oscillations, the foil breakage rate is reduced by more than 90%, significantly improving the overall efficiency of the equipment.

[0022] Construct an intelligent control system: realize tension "progress control", process quality correlation analysis, predictive equipment maintenance and parameter self-optimization, and promote the digital and intelligent upgrading of electrode foil production.

[0023] To achieve the above objectives, this invention employs the following core elements: deeply integrating three advanced technologies—multivariable dynamic feedforward decoupling control, adaptive neural fuzzy inference system (ANFIS), and multi-source sensor data fusion—to construct an integrated intelligent control system for electrode foil production, encompassing "perception-decision-execution-optimization." The system proactively predicts and counteracts inter-segment coupling disturbances, rather than responding passively; it replaces fixed PID algorithms with self-learning, adaptive intelligent controllers; and it achieves closed-loop self-calibration and advanced monitoring through multi-source data fusion.

[0024] This invention is achieved through the following technical solutions, tailored to specific application scenarios:

[0025] A multi-segment collaborative high-precision constant tension control method for an electrode foil production line includes the following steps:

[0026] Step S1: Collect multimodal sensing data from each process section of the electrode foil production line;

[0027] Step S2: Calculate the original tension error E_i of the foil in each process section based on the pre-processed multimodal sensing data, and convert the original tension error E_i into a set of decoupled errors Ed_i for independent closed-loop control of the foil tension.

[0028] Step S3: Construct an independent adaptive neural fuzzy inference network for each tension segment, taking Ed_i and its time rate of change ΔEd_i of each segment as input, and outputting a command Ui to characterize torque or current:

[0029] Step S4: Synchronously send the UI to each actuator: the magnetic powder brake of the unwinding section, the two servo motors SM-2A and SM-2B of the traction section, and the foil take-up servo motor, to perform multi-segment collaborative high-precision constant tension control;

[0030] In step S5, during the multi-segment collaborative high-precision constant tension control process, multimodal sensing data from each process segment are fused, calibrated online, and subjected to process monitoring and predictive maintenance analysis. The results are then visualized and presented to the user through human-computer interaction.

[0031] The multimodal sensing data includes:

[0032] The angle signal θ[1] of the first absolute magnetic encoder in the unwinding section and the angle signal θ[3] of the second absolute magnetic encoder in the winding section;

[0033] The direct tension value T_fb2_1 of the first non-contact ultrasonic tension meter in the unwinding section and the direct tension value T_fb2_2 of the second non-contact ultrasonic tension meter in the winding section;

[0034] The pressure value F_A of the third pressure sensor and the pressure value F_B of the fourth pressure sensor in the traction section;

[0035] The real-time roll diameter D(t) of the roll diameter sensor in the take-up section.

[0036] The data preprocessing includes:

[0037] Step S1201, Filtering: Use a Kalman filter to filter the multimodal sensing data to remove mechanical vibration and electrical noise;

[0038] Step S1202: Calculate the indirect tension feedback value T_fb1_1 of the unwinding section and the indirect tension feedback value T_fb1_3 of the winding section, using the following formula:

[0039] T_calc[i] = K[i]×θ[i] + C[i];

[0040] Where i=1 represents the unwinding section and i=3 represents the winding section; K[i] and C[i] represent the floating roller calibration parameters of the unwinding or winding area, which are obtained after pre-calibration;

[0041] Step S1203: Calculation of tension and error in the traction section;

[0042] Calculate the inlet tension T_in = F_A / (2·sinα);

[0043] Calculate the outlet tension T_out = F_B / (2·sinα);

[0044] Calculate the tension difference ΔT in the traction section: ΔT = T_out - T_in;

[0045] The output includes the indirect tension feedback value T_fb1_1 of the unwinding section, the indirect tension feedback value T_fb1_3 of the winding section, the tension difference ΔT within the traction section, the direct tension reference value T_fb2 of the three sections, and the real-time winding diameter D(t) of the winding diameter sensor in the winding section.

[0046] The original error of the foil in each process section is E_i = [E_1, E_2, E_3]^T, where i=1,2,3 correspond to the three tension sections, and ^T represents transposition. It is calculated and obtained as follows:

[0047] The tension error of the unwinding section is E_1 = T_sp_1 - T_fb1_1; where T_sp_1 is the target tension of the unwinding section, which is set by the process.

[0048] The tension difference error of the traction section is E_2 = E_diff = ΔT - ΔT_sp; where ΔT_sp is the target tension difference of the traction section, which is set by the process.

[0049] The take-up section employs a tapered control strategy with variable roll diameter and constant tension. This strategy involves real-time detection of the roll diameter and dynamic adjustment of torque commands to ensure constant take-up tension.

[0050] Based on the current real-time roll diameter D(t) and the taper formula T_sp(D)_3 = T0· (D0 / D)^k, calculate the target tension T_sp(D)_3 of the take-up section; calculate the tension error E_3 of the take-up section = T_sp(D)_3 - T_fb1_3; where T0 represents the initial tension setting value, D0 represents the initial roll diameter, ^k represents the exponentiation operation, k represents the taper coefficient, and 0≤k≤1.

[0051] In step S2, converting the original tension error E_i into a set of decoupled errors Ed_i for independent closed-loop control of foil tension specifically involves: obtaining a process coupling mathematical model G(s) beforehand through system identification. This model is used as input to convert the original tension error E_i, which has a strong coupling relationship, into a set of decoupled errors E_di for independent closed-loop control of foil tension; specifically including the following steps:

[0052] Ei is the original error, ΔUi is the control input; G(s) is a 3×3 coupling matrix;

[0053] The feedforward decoupling compensator D(s) adopts the diagonal matrix decoupling method and calculates the coupling compensation amount δE_ij= G(s)×ΔUi in real time;

[0054] The output error of the feedforward decoupler after decoupling is Ed_i = E_i - Σ(δE_ij);

[0055] After decoupling, the independent error vector Ed_i = [Ed_1, Ed_2, Ed_3]^T;

[0056] Where δE_ij is the estimated coupling effect of segment j on segment i; the output is the independent error Ed_i after eliminating coupling interference.

[0057] The adaptive neural fuzzy inference network is a five-layer feedforward network, including an input layer, a fuzzification layer, a rule layer, a normalization layer, a consequent calculation layer, and an output layer.

[0058] The input layer receives two precise quantities that characterize the error change trend of tension or tension difference in each segment: the decoupled tension error Ed of the current control cycle and the first-order difference ΔEd between the current error and the error of the previous cycle, and transmits them to the fuzzification layer.

[0059] The fuzzification layer defines five fuzzy subsets for each input variable: negative large NB, negative small NS, zero ZO, positive small PS, and positive large PB; each subset corresponds to a Gaussian membership function; for input variable E_d or ΔE_d, its membership degree μ(x) belonging to each fuzzy subset is calculated, and its value is between [0, 1]; this is then passed to the rule layer;

[0060] The rule layer, based on the five membership values ​​of each of the two input variables Ed_i and ΔEd_i, generates 25 rules using a full combination approach. Each rule takes the form: if Ed is A_m and ΔEd is B_n, then output f_mn; where A_m and B_n are the fuzzy antecedents (i.e., membership function values), and f_mn is the rule consequent function value, m=1,2……,5, n=1,2……,5. The trigger strength ω_k of this rule is calculated using a product method or a minimum method, where ω_k represents the degree of matching of the current input state to this rule. The 25 rule trigger strengths ω_1 ... ω_25 are output and passed to the normalization layer.

[0061] The normalization layer and consequent calculation layer normalize the trigger strength of each rule as the normalized weight parameter bar{ω}_k = ω_k / Σ(ω_t); t=1,2……,25, k represents one of 25; for each rule, a linear function with respect to the input variable is defined, namely the adaptive consequent function f_k = p_k×E_d + q_k ×ΔE_d + r_k, where p_k, q_k, r_k are the connection weights of the neural network; for each rule, the product of the normalized weight and the consequent function is calculated by weighting: bar{ω}_k×f_k;

[0062] The output layer sums the weighted outputs of all rules to obtain a clear, unambiguous control quantity: U = Σ(bar{ω}_k×f_k) = Σ (bar{ω}_k×(p_k×E_d + q_k×ΔE_d + r_k));

[0063] Finally, when the input of E_di and its rate of change ΔE_di for each section is obtained, the control command Ui representing the torque or current of each section is obtained according to the fuzzy calculation formula, including: unwinding section braking current command U1, output traction section torque difference compensation amount U2, and winding section torque command U3.

[0064] The connection weights {p_k, q_k, r_k} of the neural network are obtained through training using the following adaptive learning mechanism:

[0065] The learning space is constructed using the network's adjustable parameter set {p_k, q_k, r_k} and the parameters of the membership function; the parameters of the membership function include the center and width of the Gaussian function.

[0066] By combining least squares (LSE) with backpropagation (BP), under the condition of a fixed antecedent membership function, LSE is used to optimize all consequent parameters {p_k, q_k, r_k} in one go; then, the BP algorithm is used to fine-tune the antecedent membership function.

[0067] Offline training is performed using historical production data to obtain an initial optimized set of adjustable parameters {p_k, q_k, r_k}, and then online fine-tuning is performed at an extremely low learning rate based on real-time performance metrics during operation to adapt to slow changes in operating conditions.

[0068] The unwinding section adopts a closed-loop control method of floating roller feedback + magnetic powder brake. U1 is sent to the current controller, the current controller outputs excitation current I_ex = K_c·U1, and the magnetic powder brake outputs braking torque T_brake = K_b·I_ex to control the foil speed in the unwinding section.

[0069] In the traction section, the torque difference compensation amount U2 is superimposed on the third intermediate traction servo motor SM-2A and the fourth intermediate traction servo motor SM-2B of the traction section to obtain torque commands U2A and U2B; where U2A = U_base -U2 / 2; U2B =U_base +U2 / 2; U_base is the given base torque command to maintain linear velocity;

[0070] In the winding section, the torque command U3 is corrected to obtain the compensated torque command U_ff = (T_sp(D)·D) / (2·η); the final output U3_cmd = U3 + U_ff is sent to the winding servo motor driver, which operates in torque control mode and drives the winding foil servo motor to perform winding.

[0071] The data fusion includes: performing time-stamp alignment processing on all collected data, and performing weighted fusion of the direct tension value T_fb2_i and the indirect tension feedback value T_fb1_i in the unwinding and rewinding sections to obtain the optimal tension estimate T_fused_i for the current section;

[0072] The online calibration involves automatically correcting the floating roller calibration parameters K[i] and C[i] of the current section when |T_fb2_i - T_fb1_i| > the threshold of the unwinding or rewinding section, based on the floating roller's "angle-tension" conversion model T_calc[i] = K[i]×θ[i] + C[i]; i=1 represents the unwinding section, and i=2 represents the rewinding section.

[0073] The process monitoring includes: calculating the mean, standard deviation, and process capability index CP / CPK in real time based on the estimated fusion tension value T_fused_i sequence of the unwinding or rewinding section; comparing the CPK value with the preset quality standard to draw an Xbar-R control chart; and generating a "process capability warning" if the CPK is continuously lower than the requirement or shows a trend of continuous increase / decrease of a preset number of points.

[0074] The predictive maintenance analysis includes: performing Fast Fourier Transform or Wavelet Analysis on the fused tension estimate T_fused_i signal of the unwinding or winding section to extract its frequency domain features; matching abnormal peaks in the spectrum with a pre-established mechanical fault feature library to identify faults and provide alerts.

[0075] A multi-segment collaborative high-precision constant tension control system for an electrode foil production line includes multi-modal sensors and data acquisition cards installed at the industrial site of the electrode foil production line, an industrial computer and a multi-segment ANFIS controller, and a multi-segment collaborative actuator.

[0076] The multimodal sensor includes: a first absolute magnetic encoder for the unwinding section and a second absolute magnetic encoder for the winding section, used to collect the floating angle signals θ[1] and θ[3] of the unwinding and winding sections respectively; a first non-contact ultrasonic tension meter and a second non-contact ultrasonic tension meter for the unwinding and winding sections, used to collect the direct tension values ​​T_fb2_1 and T_fb2_2 of the unwinding and winding sections respectively; a third pressure sensor and a fourth pressure sensor for the traction section, used to collect the pressure values ​​F_A and F_B of the two rollers in the traction section respectively; and a roll diameter sensor for the winding section, used to collect the real-time roll diameter D(t); the above collected data constitute the multimodal sensing data of each process section of the electrode foil production line.

[0077] The data acquisition card connects to the multimodal sensor and the industrial computer, and is used to acquire and upload data;

[0078] The industrial computer and the multi-segment ANFIS controller are equipped with a real-time data interface layer, an intelligent control algorithm layer, and a fusion monitoring and optimization layer; the industrial computer is equipped with a control backend and a front-end human-machine interface.

[0079] The real-time data interface layer is located in the industrial computer control backend, which collects multimodal sensing data from each process section of the electrode foil production line and performs data preprocessing.

[0080] The intelligent control algorithm layer includes a multivariable dynamic decoupling unit located in the industrial computer control backend and an adaptive neural fuzzy inference network unit located in the multi-segment ANFIS controller. The multivariable dynamic decoupling unit calculates the original tension error E_i of the foil in each process segment based on the preprocessed data, establishes a multivariable dynamic feedforward decoupling calculation network, and converts the original tension error E_i with strong coupling relationship into a set of decoupled errors Ed_i for independent closed-loop control of the foil tension. The adaptive neural fuzzy inference network unit is used to construct an independent adaptive neural fuzzy inference network for each tension segment. Through fuzzy inference and self-learning engine, it takes Ed_i and its rate of change ΔEd_i of each segment as input and outputs torque or current commands Ui: unwinding segment torque command U1, traction segment torque difference compensation U2, and winding segment torque command U3 to the multi-segment cooperative actuator.

[0081] The fusion monitoring and optimization layer is located in the industrial computer control backend. During the multi-segment collaborative high-precision constant tension control process, it performs data fusion, online calibration, process monitoring, and predictive maintenance analysis on the multimodal sensor data of each process segment.

[0082] The front-end human-machine interface is used to visually display to users the data acquisition, process parameters and command calculation results of multi-segment collaborative high-precision constant tension control, as well as early warning information of the process control process;

[0083] The multi-segment cooperative actuator includes the following actuators installed in the unwinding section, traction section, and winding section of the multi-segment cooperative high-precision constant tension system of the electrode foil production line: a magnetic powder brake for the unwinding section, two servo motors SM-2A and SM-2B for the traction section, and a foil winding servo motor, used to receive torque or current commands Ui and perform multi-segment cooperative high-precision constant tension control on each section of the foil.

[0084] The multi-segment collaborative high-precision constant tension system of the electrode foil production line includes three parts: an unwinding section, a traction section, and a winding section. The unwinding section is equipped with a first groove, the traction section is equipped with a second groove, and the winding section is equipped with a third groove. The first, second, and third grooves are either formation grooves or etching grooves according to actual needs. The electrode foil completes the etching or formation process through the unwinding section, traction section, and winding section.

[0085] The present invention has the following beneficial effects and advantages:

[0086] The core of this invention lies in its pioneering control architecture that deeply integrates "multi-variable dynamic feedforward decoupling" and "adaptive neuro-fuzzy inference," while also incorporating multi-source sensing and intelligent monitoring. This solution precisely overcomes the persistent industry challenges of strong tension coupling, high brittleness, and instability in multi-segment electrode foil production, achieving a revolutionary improvement in control accuracy from ±5% to within ±1%, reducing foil breakage rate by over 90%, and providing a feasible technical path for the stable production of ultra-high specific capacitance foil materials. Its technological combination possesses outstanding non-obviousness and brings unprecedented industrial value.

[0087] Furthermore, the calculation of the "multivariable dynamic feedforward decoupling" in this invention is used to convert the original tension error E_i, which has a strong coupling relationship, into a set of decoupled errors Ed_i = [Ed_1, Ed_2, Ed_3]^T for independent closed-loop control of foil tension, using the original tension error E_i with a strong coupling relationship as input. Each component Ed_i in this signal has eliminated coupling interference from other segments and known feedforward disturbances to the greatest extent. The ANFIS controllers of each segment receive such a "pure" error signal that only reflects the control requirements of its own segment, thus enabling independent and precise control and achieving true "segmented independent closed-loop control". The essential difference between this invention and traditional passive PID control is that traditional control corrects after coupling interference has occurred and caused consequences, while this invention uses "multivariable dynamic feedforward decoupling" to predict the impact of interference at the beginning of its generation and apply a countermeasure in advance. This "prediction-feedforward" active decoupling strategy is the fundamental reason why this invention can improve the tension control accuracy of the entire line to below ±1% and achieve correction without disturbing the tension.

[0088] Furthermore, the "adaptive neural fuzzy inference" of this invention uses the Ed_i and its time change rate ΔEd_i of each input segment to obtain the control command Ui representing the torque or current of each segment according to the fuzzy calculation formula, and sends it to the corresponding actuator, including: outputting the unwinding segment braking current command U1 to the current controller to indirectly control the magnetic powder brake; outputting the traction segment torque difference compensation amount U2 to the third intermediate traction servo motor SM-2A and the fourth intermediate traction servo motor SM-2B of the traction segment; and outputting the winding segment torque command U3 to the winding foil servo motor. In this invention, the ANFIS structure perfectly combines expert experience (initialized in the form of fuzzy rules) with data-driven learning (optimized through neural network parameters). It not only has the fuzzy inference capability to handle nonlinear and uncertain problems, but also enables the controller to continuously self-optimize and surpass the initial experience through the adaptive learning capability of the neural network. This is the fundamental reason why this invention can achieve an ultra-high control accuracy of ±1% and significantly reduce the dependence on debugging experience. This specific structure and its learning method are one of the core technical features claimed in this patent.

[0089] Furthermore, this invention enables: 1) High-precision benchmark: Initial calibration using a high-precision external benchmark establishes a reliable source of accuracy for the entire feedback system. 2) Closed-loop self-calibration: Automatic verification and parameter drift compensation during the production process are achieved using an online testing instrument, ensuring the long-term and continuous accuracy of the feedback system and overcoming accuracy degradation caused by mechanical wear, temperature drift, etc. 3) Data traceability: Complete calibration logs provide crucial equipment status data for product quality analysis, realizing the correlation between "equipment parameters" and "process quality". Attached Figure Description

[0090] Figure 1 The present invention relates to a multi-segment collaborative high-precision constant tension system for an electrode foil production line.

[0091] Figure 2 Hardware and software architecture block diagram of the intelligent core control module of this invention.

[0092] Figure 3 A schematic diagram of the multivariable dynamic decoupling calculation principle of this invention.

[0093] Figure 4 Detailed structural diagram of the ANFIS controller of the present invention (five-layer network).

[0094] Figure 5 The data fusion and monitoring logic flowchart of this invention.

[0095] Figure 6 The execution flowchart of the single control cycle core algorithm of this invention.

[0096] Among them, 1 is the foil feeding platform, 2 is the foil feeding servo motor, 3 is the foil feeding roll, 4 is the first foil feeding guide roller, 5 is the first non-contact ultrasonic tension meter, 6 is the second foil feeding guide roller, 7 is the third foil feeding guide roller, 8 is the foil feeding floating roller frame, 9 is the first foil feeding traction roller, 10 is the fourth foil feeding guide roller, 11 is the first pressure sensor, 12 is the second foil feeding traction roller, 13 is the fifth foil feeding guide roller, and 14 is the first non-contact sensor;

[0097] 15 is the main mounting frame of the intermediate traction roller group, 16 is the first intermediate traction roller, 17 is the first intermediate guide roller, 18 is the third pressure sensor, 19 is the second intermediate traction roller, 20 is the third intermediate traction roller, 21 is the second intermediate guide roller, 22 is the third intermediate guide roller, 23 is the fourth pressure sensor, 24 is the fourth intermediate guide roller, 25 is the fourth intermediate traction roller, and 26 is the third non-contact ultrasonic tension meter.

[0098] 27 is the foil take-up platform, 28 is the foil take-up servo motor, 29 is the foil take-up roll, 30 is the first foil take-up guide roller, 31 is the second non-contact ultrasonic tension meter, 32 is the second foil take-up guide roller, 33 is the third foil take-up guide roller, 34 is the foil take-up floating roller frame, 35 is the first foil take-up traction roller, 36 is the fourth foil take-up guide roller, 37 is the second pressure sensor, 38 is the second foil take-up traction roller, 39 is the fifth foil take-up guide roller, and 40 is the second non-contact sensor;

[0099] 001 is the first tank, 002 is the second tank, and 003 is the third tank. Detailed Implementation

[0100] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0101] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0102] like Figure 1 As shown, the industrial application scenarios of this invention are as follows:

[0103] A multi-segment collaborative high-precision constant tension system for an electrode foil production line includes three parts: an unwinding section, a traction section, and a winding section. The unwinding section is provided with a first tank 001, the traction section is provided with a second tank 002, and the winding section is provided with a third tank 003. The first tank 001, the second tank 002, and the third tank 003 can be a formation tank (electrolyte) or an etching tank (acid) according to actual needs.

[0104] The unwinding section includes a foil unwinding platform 1, a foil unwinding servo motor 2, a foil unwinding roll 3, a first foil unwinding guide roller 4, a first absolute magnetic encoder, a first non-contact ultrasonic tension meter 5, a second foil unwinding guide roller 6, a third foil unwinding guide roller 7, a foil unwinding floating roller frame 8, a first foil unwinding traction roller 9, a first foil unwinding traction servo motor, a fourth foil unwinding guide roller 10, a first pressure sensor 11, a second foil unwinding traction roller 12, a second foil unwinding traction servo motor, a fifth foil unwinding guide roller 13, a first non-contact sensor 14, and a first intelligent controller.

[0105] A foil-laying servo motor 2 is mounted on the foil-laying platform 1 and is used to drive the foil-laying roll 3 to unwind. A first absolute magnetic encoder is mounted on the first foil-laying guide roller 4. A first non-contact ultrasonic tension meter 5 is mounted near the first foil-laying guide roller 4. A first foil-laying traction servo motor is used to drive the first foil-laying traction roller 9 to assist in the traction of the electrode foil, and a second foil-laying traction servo motor is used to drive the second foil-laying traction roller 12 to assist in the traction of the electrode foil. A first pressure sensor 11 is used to detect the pressure applied by the electrode foil to the upper side of the fourth foil-laying guide roller 10. A foil-laying floating roller frame 8 is mounted in the first tank 001, and the foil-laying floating roller frame 8 is equipped with four floating directional rollers. The overall structure of the foil-laying floating roller frame 8 is set to a technology known in the art (for example, see patents such as CN210655505U). A first non-contact sensor 14 is used to detect the height position of the foil-laying floating roller frame 8 in the first tank 001. The first intelligent controller is connected to the foil-laying servo motor 2, the first absolute magnetic encoder, the first non-contact ultrasonic tension meter 5, the first foil-laying traction servo motor, the first pressure sensor 11, the second foil-laying traction servo motor, and the first non-contact sensor 14.

[0106] The traction section includes a main mounting frame 15 for the intermediate traction roller group, a first intermediate traction roller 16, a first intermediate traction servo motor SM-1A, a first intermediate guide roller 17, a third pressure sensor 18, a second intermediate traction roller 19, a second intermediate traction servo motor SM-1B, a third intermediate traction roller 20, a third intermediate traction servo motor SM-2A, a second intermediate guide roller 21, a third intermediate guide roller 22, a fourth pressure sensor 23, a fourth intermediate guide roller 24, a fourth intermediate traction roller 25, a fourth intermediate traction servo motor SM-2B, a third non-contact ultrasonic tension meter 26, and a second intelligent controller.

[0107] The main mounting frame 15 of the intermediate traction roller group is correspondingly disposed on the outer periphery of the second tank 002. The first intermediate traction roller 16, the first intermediate guide roller 17, the second intermediate traction roller 19, the third intermediate traction roller 20, the third intermediate guide roller 22, the fourth intermediate traction roller 25, and the third non-contact ultrasonic tension meter 26 are all disposed on the main mounting frame 15 of the intermediate traction roller group and located above the second tank 002. The second intermediate guide roller 21 and the fourth intermediate guide roller 24 are disposed within the second tank 002. The first intermediate traction servo motor is used to drive the first intermediate traction roller 16 for auxiliary traction of the electrode foil; the second intermediate traction servo motor is used to drive the second intermediate traction roller 19 for auxiliary traction of the electrode foil; the third intermediate traction servo motor is used to drive the third intermediate traction roller 20 for auxiliary traction of the electrode foil; and the fourth intermediate traction servo motor is used to drive the fourth intermediate traction roller 25 for auxiliary traction of the electrode foil. The third pressure sensor 18 is used to detect the pressure of the electrode foil on the upper side of the first intermediate guide roller 17; and the fourth pressure sensor 23 is used to detect the pressure of the electrode foil on the upper side of the third intermediate guide roller 22. The second intelligent controller is connected to the first intermediate traction servo motor, the third pressure sensor 18, the second intermediate traction servo motor, the third intermediate traction servo motor, the fourth pressure sensor 23, the fourth intermediate traction servo motor, and the third non-contact ultrasonic tension meter 26.

[0108] The take-up section includes a take-up platform 27, a take-up servo motor 28, a take-up roll 29, a first take-up guide roller 30, a second absolute magnetic encoder, a second non-contact ultrasonic tension meter 31, a second take-up guide roller 32, a third take-up guide roller 33, a take-up floating roller frame 34, a first take-up traction roller 35, a first take-up traction servo motor, a fourth take-up guide roller 36, a second pressure sensor 37, a second take-up traction roller 38, a second take-up traction servo motor, a fifth take-up guide roller 39, a second non-contact sensor 40, and a third intelligent controller.

[0109] A foil-taking servo motor 28 is mounted on the foil-taking platform 27 and is used to drive the foil-taking roll 29 to take up the foil. A second absolute magnetic encoder is mounted on the first foil-taking guide roller 30. A second non-contact ultrasonic tension meter 31 is mounted near the first foil-taking guide roller 30. A first foil-taking traction servo motor is used to drive the first foil-taking traction roller 35 to assist in the traction of the electrode foil, and a second foil-taking traction servo motor is used to drive the second foil-taking traction roller 38 to assist in the traction of the electrode foil. A second pressure sensor 37 is used to detect the pressure of the electrode foil on the upper side of the fourth foil-taking guide roller 36. A foil-taking floating roller frame 34 is mounted in the third tank 003. The foil-taking floating roller frame 34 is equipped with four floating directional rollers, and the overall structure of the foil-taking floating roller frame 34 is based on technology known in the art (for example, see patents such as CN210655505U). A second non-contact sensor 40 is used to detect the height position of the foil-taking floating roller frame 34 in the third tank 003. The third intelligent controller is connected to the foil-taking servo motor 28, the second absolute magnetic encoder, the second non-contact ultrasonic tension meter 31, the first foil-taking traction servo motor, the second pressure sensor 37, the second foil-taking traction servo motor, and the second non-contact sensor 40.

[0110] Working principle: The electrode foil released from foil roll 3 sequentially passes through the first foil release guide roller 4, the second foil release guide roller 6, the third foil release guide roller 7, the first floating reversing roller of foil release floating roller frame 8, the first foil release traction roller 9, the second floating reversing roller of foil release floating roller frame 8, the fourth foil release guide roller 10, the third floating reversing roller of foil release floating roller frame 8, the second foil release traction roller 12, the fourth floating reversing roller of foil release floating roller frame 8, the fifth foil release guide roller 13, the first intermediate traction roller 16, the first intermediate guide roller 17, the second intermediate traction roller 19, and the third intermediate traction roller 10. The foil includes a guide roller 20, a second intermediate guide roller 21, a third intermediate guide roller 22, a fourth intermediate guide roller 24, a fourth intermediate traction roller 25, a fifth take-up guide roller 39, a first floating deflector roller of a take-up floating roller frame 34, a second take-up traction roller 38, a second floating deflector roller of a take-up floating roller frame 34, a fourth take-up guide roller 36, a third floating deflector roller of a take-up floating roller frame 34, a third take-up guide roller 33, a fourth floating deflector roller of a take-up floating roller frame 34, a second take-up guide roller 32, and a first take-up guide roller 30, which are then wound up by a take-up roll 29.

[0111] like Figure 2 The diagram shown illustrates the hardware and software architecture of the intelligent core control system, illustrating the hardware platform, three-layer software architecture, and data flow. The intelligent core control module of this invention employs a hardware platform based on an industrial real-time computer and a layered, modular software architecture to achieve a complete closed loop from data acquisition to intelligent decision-making.

[0112] Hardware configuration: Industrial computer: Intel i7 processor, QNX real-time operating system; control cycle: 1ms (hard real-time); fieldbus: EtherCAT, synchronization cycle 1ms; storage: solid-state drive, storing model parameters and historical data.

[0113] Hardware Platform Parameters: The core of the hardware platform is an industrial computer, configured to meet the demands of high real-time performance and complex calculations. The central processing unit (CPU) employs a multi-core high-performance processor, with at least one core dedicated to running a real-time operating system, ensuring determinism and high reliability of the control cycle. The data acquisition interface is equipped with a high-speed, multi-functional data acquisition card for receiving analog and digital signals from devices such as absolute magnetic encoders and non-contact ultrasonic tension meters, achieving low-latency, high-synchronization data acquisition. The real-time communication interface integrates an EtherCAT master card. This fieldbus, with its extremely high synchronization performance and bandwidth, connects all high-speed real-time devices such as servo drives, forming a nanosecond-level synchronized "motion control network." The storage unit is equipped with a large-capacity solid-state drive for storing system software, the initial rule base and training parameters of the fuzzy neural network, historical process data, and system operation logs.

[0114] This intelligent core control system is the "intelligent brain" of the entire system. It is responsible for fusing and processing multi-source data collected from industrial application scenarios, executing core algorithms, and outputting high-precision control commands.

[0115] Software architecture:

[0116] Layer 1: Real-time data interface layer (data synchronization and acquisition, Kalman filtering);

[0117] Layer 2: Intelligent control algorithm layer (multivariable decoupled computation, ANFIS inference);

[0118] Layer 3: Integrated monitoring and optimization layer (data fusion, SPC monitoring, parameter learning).

[0119] Detailed description of software architecture functions

[0120] The software architecture is divided into three layers from top to bottom, with clear data flow and functional isolation.

[0121] First layer: Data interface and preprocessing layer (corresponding to step 1)

[0122] Function: Serves as an interface for interacting with the physical world. It synchronously reads all input signals within each fixed control cycle (e.g., 1ms).

[0123] Core processing: The original signal is digitally filtered (e.g., by a Kalman filter) to remove measurement noise; and engineering unit conversion is performed to convert signals such as voltage and pulses into standard physical quantities such as angle, tension, and position, providing "clean and reliable" input data for the upper-level control algorithm.

[0124] Second layer: Intelligent decoupling and control layer (algorithm layer, corresponding to steps 2-3)

[0125] Function: Executes the core innovative algorithm of this invention to achieve high-precision, decoupled tension control.

[0126] Multivariable dynamic decoupling computational unit:

[0127] Receive the tension error of each segment after preprocessing.

[0128] It incorporates a process coupling mathematical model obtained through system identification.

[0129] Run the feedforward decoupling algorithm to calculate the compensation amount for inter-segment coupling interference in real time.

[0130] Output a set of decoupled, independent tension error signals.

[0131] ANFIS controller array:

[0132] Configure an independent adaptive neuro-fuzzy inference system for each tension segment on the production line.

[0133] It receives the decoupled error signal and performs parallel inference computation through its five-layer network structure (input, fuzzification, rule, normalization, output).

[0134] The parameters of each ANFIS controller (such as membership functions and rule consequent parameters) can be optimized online or offline by the parameter self-learning engine of the third layer.

[0135] Output precise control values ​​(such as torque or current setpoints).

[0136] The third layer: Integrated monitoring and optimization layer (intelligent decision-making layer, corresponding to steps 4 and 5).

[0137] Functions: Enables the system to perform "progress control", adaptive optimization and health management.

[0138] Multi-source data fusion unit:

[0139] By integrating tension information from the direct detection module (non-contact ultrasonic tension meter) and the indirect feedback module (floating roller), a more reliable and accurate "system tension state estimate" is obtained through data fusion algorithms (such as weighted averaging or Bayesian estimation).

[0140] Process monitoring and predictive maintenance unit:

[0141] Based on the fused tension data, advanced statistical process control is performed to calculate the CPK value, draw control charts, and achieve "progress control" and over-limit alarms.

[0142] Analyze the spectral characteristics of tension signals to provide early warnings of mechanical component failures (such as roller eccentricity and bearing wear).

[0143] Parameter self-learning engine:

[0144] Based on the results of process monitoring and historical high-quality production data, continuously optimize the parameters of the ANFIS controller and / or the parameters of the decoupling model.

[0145] The management system's automatic calibration process triggers the calibration of the floating roller's calibration parameters when a systematic deviation occurs between direct detection and indirect feedback data.

[0146] Data flow and collaborative work:

[0147] As shown by the arrow, the data flow forms a complete closed loop:

[0148] Forward control flow: Sensor signal → Preprocessing → Decoupling calculation → ANFIS control → Drive actuator.

[0149] Feedback optimization flow: Direct detection signal and indirect feedback signal → data fusion and monitoring → generation of calibration instructions and optimization targets → feedback to decoupling unit and ANFIS controller to realize system self-calibration, self-adaptation and self-optimization.

[0150] The specific method steps of the present invention include the following five steps: S1~S5;

[0151] The control method of the present invention is executed within each defined control cycle (e.g., 1 ms), and includes the following steps:

[0152] Step S1: Synchronous acquisition and intelligent preprocessing of multi-source signals;

[0153] Step S11: Synchronous reading;

[0154] The angle signal θ[1] of the first absolute magnetic encoder in the unwinding section and the angle signal θ[3] of the second absolute magnetic encoder in the winding section;

[0155] The direct tension value T_fb2_1 of the first non-contact ultrasonic tension meter 5 in the unwinding section, the direct tension value T_fb2_2 of the second non-contact ultrasonic tension meter 31 in the winding section, and the direct tension value T_fb2_3 of the third non-contact ultrasonic tension meter 26 in the traction section.

[0156] The pressure value F_A of the third pressure sensor and the pressure value F_B of the fourth pressure sensor in the traction section;

[0157] A roll diameter sensor is installed on the take-up shaft of the take-up section to detect the real-time roll diameter D(t).

[0158] Step S12: Intelligent preprocessing;

[0159] Step S1201, Filtering: Use a Kalman filter to filter all the above original signals to remove mechanical vibration and electrical noise;

[0160] Step S1202: Calculate the indirect tension feedback value T_fb1_1 of the unwinding section and the indirect tension feedback value T_fb1_3 of the winding section, using the following formula:

[0161] T_calc[i] = K[i]×θ[i] + C[i];

[0162] Where i=1 represents the unwinding section and i=3 represents the winding section; K[i] and C[i] represent the floating roller calibration parameters of the unwinding or winding area, which are obtained after pre-calibration;

[0163] Step S1203: Calculation of tension and error in the traction section;

[0164] Calculate the inlet tension T_in = F_A / (2·sinα);

[0165] Calculate the outlet tension T_out = F_B / (2·sinα);

[0166] Calculate the tension difference ΔT within the traction section: ΔT = T_out - T_in;

[0167] Output: Indirect tension feedback value T_fb1_1 of the clean unwinding section, indirect tension feedback value T_fb1_3 of the winding section, tension difference ΔT in the traction section, direct tension reference value T_fb2 of the three sections, and real-time winding diameter D(t) of the winding diameter sensor in the winding section.

[0168] Step S2: Multivariable dynamic feedforward decoupling calculation;

[0169] like Figure 3 As shown, the core function of the multivariable dynamic decoupling calculation unit of this invention is to convert the original tension error with strong coupling into a set of decoupled errors that can be used for independent closed-loop control. Its principle is based on a feedforward decoupling strategy.

[0170] Step S21: Calculate the original error E_i, where i=1,2,3 corresponds to the three tension segments; the specific process is as follows:

[0171] Step S211, Input signal: The original tension error vector E_i = [E_1, E_2, E_3]^T is calculated from the set tension and the measured values ​​from the main feedback elements (absolute magnetic encoder and pressure sensor) of each section. Wherein, E_1, E_2, and E_3 are the tension error of the unwinding section, the tension difference error of the traction section, and the tension error of the winding section, respectively, and ^T represents transpose.

[0172] Step S211_a: Calculate the original error E_1 of the unwinding section;

[0173] E_1 = T_sp_1 - T_fb1_1; where T_sp_1 is the target tension of the unwinding section, which is set by the process, and in this embodiment it is taken as 65-85N.

[0174] Step S211_b: Calculate the original error E_2 of the tension difference in the traction section;

[0175] E_2 = E_diff = ΔT - ΔT_sp; where ΔT_sp is the target tension difference of the traction section, which is set by the process, and in this embodiment, it is taken as 5-15N. Furthermore, for example, an increase in E_2 indicates that the actual measured tension of the traction section is lower than the set value.

[0176] Step S211_c: Calculate the original error E3 of the take-up section;

[0177] Because the winding zone adopts a variable diameter constant tension control strategy, the winding tension is kept constant by detecting the winding diameter in real time and dynamically adjusting the torque command.

[0178] Control principle: The winding tension T is controlled by the output torque U3 of the servo motor. The tension T and the torque U3 satisfy the following relationship:

[0179] T = (2·η·U3) / D;

[0180] Where T represents foil tension (N), η represents transmission efficiency coefficient (including reduction ratio, friction loss, etc., in this embodiment η≈0.85-0.95), U3 represents servo motor torque command (N·m), and D represents real-time roll diameter (m).

[0181] As the winding shaft rotates, the foil diameter D gradually increases as it winds. To maintain a constant tension T, the torque command U3 must increase accordingly with the increase in the coil diameter.

[0182] Taper control strategy:

[0183] To accommodate the winding requirements of different roll diameters, this invention employs the following taper formula (exponential taper control):

[0184] T_sp(D)_3 = T0· (D0 / D)^k;

[0185] Where: T0 represents the initial tension setting value (N), which is set to 70-95N in this embodiment; D0 represents the initial roll diameter (empty shaft diameter, mm), which is about 150mm in this embodiment; D represents the real-time detected roll diameter (mm); ^k represents exponentiation; k represents the taper coefficient, which is 0≤k≤1 and is 0.3-0.6 in this embodiment.

[0186] Control logic:

[0187] When k=0, constant torque control is applied, and the tension decreases as the roll diameter increases.

[0188] When k=1, constant tension control is applied, and the torque increases linearly with the roll diameter.

[0189] In this embodiment, k=0.5 is chosen to balance constant tension and prevention of internal loosening.

[0190] Real-time control process:

[0191] Step S211_c1: Real-time detection of roll diameter;

[0192] The DS-1 roll diameter sensor collects the distance to the surface of the take-up spool every 100ms, calculates the real-time roll diameter D(t) based on the spool center position, and outputs it.

[0193] Step S211_c2: Target tension calculation;

[0194] Based on the current real-time roll diameter D(t) and taper formula, calculate the target tension T_sp(D)_3 of the take-up section;

[0195] Step S211_c3, Error Calculation;

[0196] Calculate the original error of the take-up section: E_3 = T_sp(D)_3 - T_fb1_3.

[0197] Step S22: Feedforward decoupling based on the coupling model G(s): Ed_i = E_i - Σ(δE_ij); where δE_ij is the estimated coupling effect of segment j on segment i; output the independent error Ed_i with coupling interference eliminated.

[0198] Given a feedforward disturbance, ΔU, originating from changes in control inputs within the controller (such as an increase in the servo torque command during the winding segment). This change itself is the source of subsequent disturbances.

[0199] Core calculation process

[0200] The decoupling unit maintains a key mathematical model of the coupling process, G(s). This model is an n×n transfer function matrix (3x3 in this example), obtained through prior system identification. The off-diagonal elements g_ij (i≠j) in G(s) quantitatively describe the dynamic impact of disturbances in the j-th segment (such as changes in control input) on the tension in the i-th segment, where i=1,2,3 represent the unwinding segment, traction segment, and rewinding segment, respectively.

[0201] Disturbance prediction

[0202] Real-time monitoring of ΔU. The feedforward decoupling compensator D(s) immediately calculates the estimated coupling effect of these known disturbances on the tension of each segment of the entire line based on the G(s) model.

[0203] Formulated representation (taking control disturbance as an example): If a change in control quantity ΔU3 is detected in the take-up section, the feedforward decoupling compensator calculates its impact on the preceding section: [δE1, δE2]^T = [g 13 , g 23 ]^T×ΔU3. Where δE1 and δE2 are the predicted additional errors that the winding action will cause to the unwinding and traction sections.

[0204] proactive compensation

[0205] The feedforward decoupling compensator D(s) converts the calculated estimated coupling effect δE_i =[δE1, δE2, δE3]^T into a corresponding feedforward compensation signal. This signal is then superimposed on the original error signal Ei at the summation node ∑.

[0206] Compensation logic: Ed_i = E_i - ΣδE_i (for the i-th affected segment)

[0207] Physical meaning: Before the actual coupling disturbance δE2 caused by ΔU3 actually affects the tension of the traction section, the system subtracts this estimated disturbance value from the control error of the traction section in advance. In this way, when the actual disturbance δE2 is transmitted through the physical process, it is just neutralized by the reverse control command issued by the controller to cancel -δE2, thus realizing "the disturbance is actively canceled, rather than passively responded to".

[0208] Output signal: Decoupled independent error vector Ed_i = [Ed_1, Ed_2, Ed_3]^T: Each component Ed_i in this signal has been minimized by eliminating coupling interference from other segments and known feedforward disturbances. The subsequent ANFIS controllers of each segment receive such a "pure" error signal that reflects only the control requirements of its own segment, thus enabling independent and precise control and achieving true "segmented independent closed-loop control".

[0209] Advantages: This step clearly demonstrates the fundamental difference from traditional passive PID control: traditional control corrects after coupling disturbances have occurred and caused consequences, while this invention predicts the impact of disturbances at their inception and applies countermeasures in advance. This "predictive-feedforward" active decoupling strategy is the fundamental reason why this invention can improve the overall tension control accuracy to below ±1% and achieve tension correction without disturbing the tension.

[0210] In summary, in this method, the 3×3 coupling matrix G(s) is identified using the pseudo-random binary sequence excitation method:

[0211] G(s) =

[0212] E_i is the original error, ΔUi is the control input; G(s) is a 3×3 coupling matrix;

[0213] The feedforward decoupling compensator D(s) adopts the diagonal matrix decoupling method and calculates the coupling compensation amount δE_ij= G(s)×ΔUi in real time;

[0214] The output error of the feedforward decoupler after decoupling is Ed_i = E_i - Σ(δE_ij);

[0215] After decoupling, the independent error vector Ed_i = [Ed_1, Ed_2, Ed_3]^T;

[0216] Where δE_ij is the estimated coupling effect of segment j on segment i; the output is the independent error Ed_i after eliminating coupling interference.

[0217] Step S3: Control of the adaptive neural fuzzy reasoning system;

[0218] For each tension segment, an independent Adaptive Neural Fuzzy Inference System (ANFIS) controller is configured;

[0219] Input: Ed_i and its rate of change ΔEd_i for each segment;

[0220] Five-layer feedforward network inference: Input layer → Fuzzification layer → Rule layer → Normalization layer → Output layer;

[0221] Output: High-precision control quantity Ui (torque or current command); that is, unwinding section torque command U1, traction section torque difference compensation quantity U2, and winding section torque command U1.

[0222] ANFIS controller design:

[0223] Network structure: five-layer feedforward, [2-10-25-25-1] nodes;

[0224] Input fuzzification: Ed and ΔEd are each assigned 5 Gaussian membership functions;

[0225] Rule base: 25 rules for full combinations;

[0226] Learning algorithm: hybrid least squares + backpropagation, learning rate 0.01.

[0227] The detailed working principle of the five-layer structure of the ANFIS controller is explained below:

[0228] like Figure 4 As shown, the ANFIS controller of this invention employs a five-layer feedforward network structure, deeply integrating the intuitive reasoning ability of fuzzy logic with the autonomous learning ability of neural networks. The functions of each layer and the data flow are detailed below:

[0229] Layer 1: Input Layer;

[0230] Function: Receives a clear signal from the upstream decoupling unit, after coupling interference has been eliminated.

[0231] Inputs include two precise quantities: the decoupled tension error Ed of the current control cycle and the first-order difference ΔEd between the current error and the error of the previous cycle, which represents the error change trend.

[0232] Output: Pass Ed and ΔEd directly to the next level.

[0233] Layer 2: Blur layer;

[0234] Function: Converts clear input values ​​into fuzzy linguistic values, which is the starting point for fuzzy inference.

[0235] Key component: Membership function: Define multiple fuzzy subsets for each input variable (five in this embodiment: negative large NB, negative small NS, zero ZO, positive small PS, positive large PB). Each subset corresponds to a Gaussian membership function.

[0236] Operation: For an input value x (E_d or ΔE_d), calculate its membership degree μ(x) to each fuzzy subset, with a value between [0, 1]. For example, μ_{NS}(E_d) = 0.8 indicates that the current error belongs to the "negative small" category with a degree of 0.8.

[0237] Output: Each input variable yields 5 membership values, and the two input variables Ed_i or ΔEd_i yield a total of 10 values, which are then passed to the rule layer.

[0238] Layer 3: Rules Layer;

[0239] Function: Performs fuzzy "If-Then" rule reasoning.

[0240] Rule base: This embodiment uses full combination rules, totaling 5 × 5 = 25 rules. Each rule is in the form of: "IfEd_i is A_m and ΔEd_i is B_n, then output is f_mn", where A_m and B_n are fuzzy antecedents, f_mn is the consequent, m=1,2……,5, n=1,2……,5.

[0241] Operation (AND): Each rule node receives two corresponding membership values ​​from layer 2. For example, rule 1 receives μ_{NB} of E_d and μ_{NB} of ΔE_d. The firing strength ω_k of the rule is calculated using either the product method or the minimum method. Here, ω_k represents the degree to which the current input state matches the rule.

[0242] Output: 25 rule trigger strengths ω_1 ... ω_25, passed to the normalization layer.

[0243] Layer 4: Normalization layer and consequent calculation layer (core of neural network features);

[0244] Functionality: This is the core feature that distinguishes ANFIS from traditional fuzzy controllers, introducing adjustable parameters (p_k, q_k, r_k) and adaptive capabilities.

[0245] First, normalize the trigger strength of each rule: bar{ω}_k = ω_k / Σ(ω_t). This represents the relative weight of the rule in the final output, where t=1,2……,25, and k represents one of the 25.

[0246] Adaptive consequent: In traditional fuzzy controllers, the consequent of a rule is a fixed fuzzy set or a single point. In ANFIS, the consequent f_k of each rule is a linear function of the input variables: f_k = p_k×E_d + q_k ×ΔE_d + r_k. Here, p_k, q_k, and r_k are the connection weights (i.e., adjustable parameters) of the neural network.

[0247] Weighted calculation: The output of this layer is the product of the normalized weight of each rule and the consequent function: bar{ω}_k×f_k.

[0248] Layer 5: Output layer;

[0249] Function: Sum the weighted outputs of all rules to obtain a clear, unambiguous control quantity.

[0250] Fuzzy calculation formula: U = Σ (bar{ω}_k×f_k) = Σ (bar{ω}_k×(p_k×E_d + q_k×ΔE_d + r_k)).

[0251] Output: Finally, when the Ed_i and its rate of change ΔEd_i of each segment are input, the control command Ui representing the torque or current of each segment is obtained according to the fuzzy calculation formula and sent to the corresponding actuator; as follows:

[0252] The unwinding section braking current command U1 (mA) is output to the current controller to indirectly control the magnetic powder brake; the traction section torque difference compensation amount U2 is output to the third intermediate traction servo motor SM-2A and the fourth intermediate traction servo motor SM-2B of the traction section; the winding section torque command U3 (N·m) is output to the winding foil servo motor 28.

[0253] Adaptive learning mechanism (not directly shown in the diagram, but explained below)

[0254] Parameter training: The learning space is composed of the network's adjustable parameter set {p_k, q_k, r_k} and the parameters of the membership function (the center and width of the Gaussian function).

[0255] Hybrid learning algorithm: This method combines least squares (LSE) and backpropagation (BP). Typically, with the antecedent parameters (membership functions) fixed, LSE is used to optimize all consequent parameters {p_k, q_k, r_k} in one go, which is extremely efficient. Then, the BP algorithm is used to fine-tune the antecedent parameters to further reduce the error.

[0256] Online and offline learning: The system can be trained offline using historical high-quality production data to obtain an initial optimized parameter set. During operation, it can be fine-tuned online at an extremely low learning rate based on real-time performance metrics to adapt to slow changes in operating conditions.

[0257] Advantages: In this step, the ANFIS structure perfectly combines expert experience (initialized in the form of fuzzy rules) with data-driven learning (optimized through neural network parameters). It not only possesses the fuzzy reasoning ability to handle nonlinear and uncertain problems, but also, through the adaptive learning capability of the neural network, enables the controller to continuously self-optimize and surpass initial experience. This is the fundamental reason why this invention can achieve ultra-high control accuracy of ±1% and significantly reduce reliance on debugging experience. This specific structure and its learning method are among the core technical features claimed in this patent.

[0258] Step S4: Collaborative execution;

[0259] U1, U2, and U3 are synchronously sent to each actuator via EtherCAT bus: foil feeding servo motor 2, two servo motors SM-2A and SM-2B of the traction section, and foil take-up servo motor 28;

[0260] Step S41, Unwinding Segment Control:

[0261] The core objective of the unwinding section is to maintain constant foil tension during the unwinding process, prevent sudden tension changes caused by factors such as changes in roll diameter and inertial fluctuations, and avoid foil stretching or breakage.

[0262] The unwinding section employs a closed-loop control scheme combining floating roller feedback and a magnetic powder brake. The magnetic powder brake outputs braking torque, which is connected to the unwinding shaft of the foil unwinding servo motor. By adjusting the excitation current, the braking torque is controlled, thereby regulating the foil unwinding tension. Torque-tension relationship:

[0263] The braking torque T_brake and the excitation current I_ex satisfy: T_brake = K_b·I_ex;

[0264] The foil tension T and the braking torque T_brake satisfy the following condition: T = T_brake / r;

[0265] Where K_b is the brake coefficient (N·m / A) and r is the unwinding radius (m).

[0266] Control output:

[0267] U1 is sent to the current controller via EtherCAT;

[0268] The current controller outputs the magnetizing current I_ex = K_c·U1;

[0269] The magnetic powder brake outputs braking torque T_brake = K_b·I_ex, which controls the speed of the unwinding section of the foil.

[0270] Step S42, Traction Section Control:

[0271] The torque difference compensation amount U2 is superimposed on the third intermediate traction servo motor SM-2A and the fourth intermediate traction servo motor SM-2B of the traction section to obtain torque commands U2A and U2B respectively; where U2A = U_base -U2 / 2; U2B = U_base +U2 / 2; U_base is the given base torque command to maintain linear velocity.

[0272] Step S43, Rewind Section Control:

[0273] The winding zone employs a variable diameter constant tension control strategy, which ensures constant winding tension by detecting the roll diameter in real time and dynamically adjusting the torque command.

[0274] The torque command U3 (N·m) output by ANFIS is corrected by performing feedforward compensation based on the measured roll diameter, resulting in the compensated torque command U_ff = (T_sp(D)·D) / (2·η).

[0275] The final output is: U3_cmd = U3 + U_ff. U3_cmd is sent to the winding servo motor driver via EtherCAT. The driver operates in torque control mode and drives the winding foil servo motor 28 to perform winding.

[0276] Step S5: Intelligent Integrated Monitoring;

[0277] like Figure 5 As shown, the data fusion and monitoring unit of this invention constitutes a parallel intelligent processing hub, and its logical flow is as follows:

[0278] Step S51, Background Data Fusion: The direct tension value T_fb2_1 of the unwinding section is fused with the indirect tension feedback value T_fb1_1 to obtain the optimal tension estimate T_fused_1 of the unwinding section; the direct tension value T_fb2_2 of the winding section is fused with the indirect tension feedback value T_fb1_2 to obtain the optimal tension estimate T_fused_2 of the winding section.

[0279] Step S52: The four intelligent functions are processed in parallel;

[0280] Step S521, Online calibration: When |T_fb2_i - T_fb1_i| > the threshold of the unwinding or rewinding section, automatically correct the floating roller calibration parameters Ki and Ci of the current section; i=1 represents the unwinding section, i=2 represents the rewinding section;

[0281] Step S522, Process Monitoring: Real-time calculation of process capability index CPK, drawing of control charts, and spectrum analysis;

[0282] Step S523, Predictive maintenance analysis;

[0283] Step S524, Quality Correlation and Optimization;

[0284] Step S53, Human-computer interaction.

[0285] The further specific procedures are as follows:

[0286] Step S51, First Stage: Multi-source data fusion;

[0287] Input: Synchronously receive four types of asynchronous data: direct measurement value T_fb2_i from the online tension meter (high-precision reference), floating roller conversion value T_fb1_i (continuous feedback), and other process parameters (speed V, current I, etc.).

[0288] Time stamp alignment: Timestamp synchronization is performed on all input data to ensure that the analysis is based on the same operating conditions at the same time.

[0289] Optimal Fusion Estimation: The core algorithm employs a Kalman filter or Bayesian estimator. This algorithm dynamically weights and fuses the direct tension value T_fb2_i (with low observation noise but potentially discontinuous) and the indirect tension feedback value T_fb1_i (continuous but potentially subject to model errors and drift) from the unwinding or rewinding segments. i=1 represents the unwinding segment, and i=2 represents the rewinding segment. Its output is the "fused tension estimate T_fused" and its estimated variance σ. 2 The dynamic weighted fusion formula is T_fused_i = a × T_fb1_i + b × T_fb2_i, where a + b = 1. T_fused_i is more reliable and accurate than any single signal, with a variance σ. 2 This reflects the current confidence level of the estimate. This is the data foundation for all subsequent advanced monitoring and analysis.

[0290] Step S52, Second Stage: Parallel processing of the four intelligent functions;

[0291] Step S521, Function 1: Online calibration and verification (closed-loop self-correction);

[0292] Trigger judgment: Continuously compare the deviation between T_fb2_i (reference) and T_fb1_i (object to be calibrated) of the unwinding or rewinding segment, and |T_fb2_i - T_fb1_i|> threshold.

[0293] Calibration execution: If the deviation continues to exceed the set threshold, it is determined that the "angle-tension" conversion model (K[i], C[i]) of the floating roller has drifted, and the parameter fine-tuning program (such as recursive least squares method) is automatically triggered to update K[i] and C[i] online.

[0294] Effect: Ensures the long-term accuracy of the main feedback channel and enables the system to self-calibrate.

[0295] Detailed steps of the standard method:

[0296] The purpose of this calibration method is to establish a precise mapping relationship between the rotation angle of the floating roller (Δθ) and the actual tension (T) acting on the foil, i.e., a calibration curve, and to ensure that this relationship can be verified and corrected periodically and automatically in production.

[0297] Calibration system components:

[0298] Calibration benchmark: A high-precision, traceable wireless handheld tension meter (or a fixed online tension meter installed at a specific calibration station on the production line) with an accuracy better than ±0.5%.

[0299] Calibration triggering and data acquisition: This is coordinated and completed by the data fusion and monitoring unit of the control system. This unit can synchronously read:

[0300] The angle value (θ[i]) of the absolute magnetic encoder of the floating roller.

[0301] The instantaneous measurement value (T_ref) of the calibrated tension meter.

[0302] The current output control quantity of the system (Ui, used to calculate the theoretical tension).

[0303] Static calibration procedure (after initial calibration or overhaul)

[0304] This process is used to establish a complete "angle-tension" characteristic curve.

[0305] Preparation: Thread the electrode foil along the normal path, and put the system in "calibration mode". Near the floating roller, gently press the wireless handheld tension meter onto the foil manually.

[0306] Tension step application: A series of discrete target tension values ​​T_set_j (j=1,2,...,m) covering the normal operating range (e.g., 50N to 150N) are manually set or automatically generated by the program through the HMI of the control system.

[0307] Data point acquisition: For each set T_set_j:

[0308] The system stabilizes the tension around this value using PID or other simple control methods.

[0309] Once the tension meter reading T_ref stabilizes (for example, fluctuations less than ±0.5N for 3 seconds), the data fusion unit synchronously captures and records a pair of data: [angle value θ_j, reference tension value T_ref_j].

[0310] Repeat the sampling 3-5 times and take the average value as the calibration data (θ_j, T_ref_j) for that tension point.

[0311] Curve Fitting and Model Establishment: After obtaining m sets of data points, polynomial fitting is performed using the least squares method within the control system. Typically, due to system rigidity and the lever principle, this relationship is highly linear within the effective working range. Therefore, the following univariate linear regression model can be established as the calibration curve: T_calc[i] = K[i]×θ[i] + C[i] where i=1 represents the unwinding section, i=2 represents the rewinding section; T_calc[i] is the tension value calculated based on the angle. θ[i] is the absolute angle measured by the absolute magnetic encoder (which needs to be converted to the offset Δθ relative to the "zero tension" equilibrium point). K[i] is the "angle-tension" conversion coefficient (unit: N / degree), representing the mechanical gain of the system. C[i] is the "zero position" tension offset constant (unit: N).

[0312] After fitting is complete, the system stores the coefficients K[i] and the constant C[i] in non-volatile memory as the core parameters of the main feedback channel.

[0313] Dynamic online verification and fine-tuning process (periodic calibration in production)

[0314] This process utilizes the existing online non-contact tension testing instrument on the production line to periodically verify and fine-tune the static calibration curve without interrupting production.

[0315] Triggering conditions: Automatically triggered when any of the following conditions are met:

[0316] Planned trigger: after every 8 hours of continuous production or after every change in product specifications.

[0317] Event-triggered: When the deviation between the reading T_online of the online non-contact ultrasonic tension meter and T_calc calculated based on the floating roller angle continuously exceeds a preset threshold (e.g., ±3%) for a certain period of time.

[0318] Data Synchronous Acquisition: During periods of stable production line operation (small tension fluctuations), the data fusion unit synchronously acquires θ(t) and T_online(t) at a high sampling rate (e.g., 100Hz), continuously collecting data sequences for a period of time (e.g., 30 seconds).

[0319] Parameter comparison and correction decision: Substitute the collected θ(t) sequence into the current calibration model T_calc(t) = K_current×θ(t) + C_current to obtain the calculated tension sequence.

[0320] Calculate the average deviation and correlation coefficient of the two sequences T_online(t) and T_calc(t).

[0321] If the average deviation is within the allowable range (e.g., <±2%) and the correlation coefficient is high (>0.95), the calibration curve is considered to be accurate and no adjustment is needed; only this verification log should be recorded.

[0322] If the average deviation exceeds the limit, a fine-tuning algorithm is initiated. Using algorithms such as recursive least squares, the current coefficients K and C are updated slightly and smoothly using newly acquired data pairs [θ(t), T_online(t)], resulting in K_new and C_new. The update formula needs to include a weighting factor to prevent sudden changes caused by a single abnormal data point.

[0323] Update and record: Update the fine-tuned new parameters K_new and C_new to the running model T_calc[i] = K[i]×θ[i] + C[i], and record the calibration time, old parameters, new parameters and deviation data in the system log to form a complete calibration traceability chain.

[0324] Method for determining the "zero point" (θ0):

[0325] "Zero position" corresponds to the equilibrium position of the floating roller when there is theoretical zero tension. Due to the weight of the floating roller and the spring force, the zero position needs to be determined indirectly through a two-point calibration method:

[0326] In static calibration, two tension points with a large difference, T_ref_low and T_ref_high, and their corresponding angles θ_low and θ_high, are selected.

[0327] Calculate the conversion factor K = (T_ref_high - T_ref_low) / (θ_high - θ_low).

[0328] The zero-position angle θ0 is then calculated as θ_low - T_ref_low / K. This θ0 will serve as the reference for converting all angle measurements θ into the offset Δθ = θ - θ0.

[0329] Advantages: This step, through the calibration method described above, achieves the following:

[0330] High-precision benchmark: Initial calibration is performed using a high-precision external benchmark, establishing a reliable source of accuracy for the entire feedback system.

[0331] Closed-loop self-calibration: Utilizing online testing instruments to achieve automatic verification and parameter drift compensation during the production process ensures the long-term and continuous accuracy of the feedback system, overcoming the accuracy degradation caused by mechanical wear, temperature drift, and other factors.

[0332] Data traceability: Complete calibration logs provide key equipment status data for product quality analysis, enabling the correlation between "equipment parameters" and "process quality".

[0333] Step S522, Function 2: Process Progress Monitoring (SPC - Statistical Process Control);

[0334] Real-time calculation: Based on the estimated fusion tension T_fused_i sequence of the unwinding or rewinding section, its mean, standard deviation, and process capability index CP / CPK are calculated in real time.

[0335] Monitoring and Early Warning: Compare the CPK value with the preset quality standard and plot the Xbar-R control chart. If the CPK remains below the requirement or shows a trend of increasing / decreasing for 7 consecutive points, an "Early Warning on Process Capability" will be generated immediately.

[0336] Effect: It elevates tension control from "maintaining stability" to the level of "progress control" through "quantitative evaluation and prediction".

[0337] Step S523, Function 3: Predictive Maintenance Analysis (State-Based Maintenance);

[0338] Signal analysis: Perform Fast Fourier Transform or Wavelet Analysis on the fused tension estimate T_fused_i signal of the unwinding or winding section to extract its frequency domain features.

[0339] Fault diagnosis: Match abnormal peaks in the spectrum (such as harmonics that are consistent with the rotation frequency of a certain guide roller) with a pre-established mechanical fault feature library.

[0340] Warning generation: If the matching is successful (e.g., abnormal vibration at the same frequency as the rotation speed of guide roller #2 in the current section is detected), a "predictive maintenance warning" is generated, indicating that "guide roller #2 in the current section may have dynamic balance failure or premature bearing wear".

[0341] Effect: Shifting from "repair after failure" to "early warning before failure", greatly reducing unplanned downtime.

[0342] Step S524, Function 4: Quality Correlation and Optimization (Data-Driven Optimization);

[0343] Big data correlation: Long-term collection and correlation of historical tension stability data (such as standard deviation, CPK) with final product test data (such as average capacity and distribution, leakage current pass rate).

[0344] Modeling: Machine learning regression algorithms (such as support vector regression (SVR) and random forest) are used to establish a quantitative relationship model between "tension control quality indicators" and "final product performance indicators".

[0345] Target feedback: The optimal control target parameter range that the tension system should pursue to achieve the best product quality is derived by reverse derivation of the model, and this is used as the optimization target feedback to the learning engine of the ANFIS controller.

[0346] Effect: It achieves a leap from "controlling tension" to "actively ensuring and optimizing the quality of the final product through tension control", forming a manufacturing closed loop.

[0347] Step S53, Third Stage: Output and Execution;

[0348] Human-computer interaction: All warnings, reports, and real-time charts are centrally displayed in the HMI.

[0349] Data accumulation: All data throughout the process (raw data, integrated data, events, and early warnings) is written into the database to form a traceable "digital twin" record.

[0350] Optimization Feedback: The optimization targets derived from quality correlation analysis are fed back to the parameter learning engine of the ANFIS controller, driving the controller to continuously and adaptively evolve in the direction of improving the quality of the final product.

[0351] This flowchart fully reveals how this invention transcends the simple logic control of traditional PLCs and constructs an advanced intelligent system integrating "sensor fusion, self-calibration, process monitoring, predictive maintenance, and quality optimization".

[0352] like Figure 6 As shown, the execution flow of the core algorithm in a single control cycle (1ms) is described in the following steps:

[0353] Step 1: Synchronous acquisition and intelligent preprocessing of multi-source signals;

[0354] Synchronous reading: angle signal of absolute magnetic encoder in each tension zone, direct tension value of non-contact ultrasonic tension meter, lateral deviation of visual correction instrument, pressure value of pressure sensor, and real-time roll diameter of roll diameter sensor.

[0355] Intelligent preprocessing: The Kalman filter is used to perform optimal estimation of the original signal, filtering out mechanical vibration and electrical noise;

[0356] Output: Clean indirect tension feedback value T_fb1_i, direct tension reference value T_fb2.

[0357] Step S2: Multivariable dynamic feedforward decoupling calculation;

[0358] Calculate the original error: E_i = T_sp_i - T_fb1_i (i=1,2,3 correspond to the three tension zones);

[0359] Feedforward decoupling based on the coupled model G(s): Ed_i = E_i - Σ(δE_ij) - δE_epc;

[0360] Where δE_ij is the estimated coupling effect of segment j on segment i;

[0361] δE_epc is the estimated disturbance for the correction action;

[0362] Output: Independent error Ed_i after eliminating coupling interference;

[0363] Step S3: Adaptive neural fuzzy reasoning system control;

[0364] Each tension zone is equipped with an independent adaptive neurofuzzy inference system controller;

[0365] Input: Ed_i and its rate of change ΔEd_i;

[0366] Five-layer feedforward network inference: Input layer → Fuzzification layer → Rule layer → Normalization layer → Output layer;

[0367] Output: High-precision control quantity Ui (torque or current command);

[0368] Step S4: Collaborative execution and intelligent integrated monitoring;

[0369] The UI is synchronously sent to each actuator via the EtherCAT bus;

[0370] Background data fusion: Fuse T_fb2 and T_fb1_i to obtain the optimal tension estimate;

[0371] Online calibration: When |T_fb2 - T_fb1_i| > the threshold, the floating roller calibration parameters K and C are automatically corrected;

[0372] Process monitoring: Real-time calculation of process capability index CPK, drawing of control charts, and spectrum analysis.

[0373] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A multi-segment collaborative high-precision constant tension control method for an electrode foil production line, characterized in that, Includes the following steps: Step S1: Collect multimodal sensing data from each process section of the electrode foil production line; Step S2: Calculate the original tension error E_i of the foil in each process section based on the pre-processed multimodal sensing data, and convert the original tension error E_i into a set of decoupled errors Ed_i for independent closed-loop control of the foil tension. Step S3: Construct an independent adaptive neural fuzzy inference network for each tension segment, taking Ed_i and its time rate of change ΔEd_i of each segment as input, and outputting a command Ui to characterize torque or current: Step S4: Synchronously send the Ui to each actuator: the magnetic powder brake of the unwinding section, the two servo motors SM-2A and SM-2B of the traction section, and the foil take-up servo motor (28) to perform multi-segment collaborative high-precision constant tension control; In step S5, during the multi-segment collaborative high-precision constant tension control process, multimodal sensing data from each process segment are fused, calibrated online, and subjected to process monitoring and predictive maintenance analysis. The results are then visualized and presented to the user through human-computer interaction.

2. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, The multimodal sensing data includes: The angle signal θ[1] of the first absolute magnetic encoder in the unwinding section and the angle signal θ[3] of the second absolute magnetic encoder in the winding section; The direct tension value T_fb2_1 of the first non-contact ultrasonic tension meter (5) in the unwinding section and the direct tension value T_fb2_2 of the second non-contact ultrasonic tension meter (31) in the winding section; The pressure value F_A of the third pressure sensor and the pressure value F_B of the fourth pressure sensor in the traction section; The real-time roll diameter D(t) of the roll diameter sensor in the take-up section.

3. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, The data preprocessing includes: Step S1201, Filtering: Use a Kalman filter to filter the multimodal sensing data to remove mechanical vibration and electrical noise; Step S1202: Calculate the indirect tension feedback value T_fb1_1 of the unwinding section and the indirect tension feedback value T_fb1_3 of the winding section, using the following formula: T_calc[i] = K[i]×θ[i] + C[i]; Where i=1 represents the unwinding section and i=3 represents the winding section; K[i] and C[i] represent the floating roller calibration parameters of the unwinding or winding area, which are obtained after pre-calibration; Step S1203: Calculation of tension and error in the traction section; Calculate the inlet tension T_in = F_A / (2·sinα); Calculate the outlet tension T_out = F_B / (2·sinα); Calculate the tension difference ΔT in the traction section: ΔT = T_out - T_in; The output includes the indirect tension feedback value T_fb1_1 of the unwinding section, the indirect tension feedback value T_fb1_3 of the winding section, the tension difference ΔT within the traction section, the direct tension reference value T_fb2 of the three sections, and the real-time winding diameter D(t) of the winding diameter sensor in the winding section.

4. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 3, characterized in that, The original error of the foil in each process section is E_i = [E_1, E_2, E_3]^T, where i=1,2,3 correspond to the three tension sections, and ^T represents transposition. It is calculated and obtained as follows: The tension error of the unwinding section is E_1 = T_sp_1 - T_fb1_1; where T_sp_1 is the target tension of the unwinding section, which is set by the process. The tension difference error of the traction section is E_2 = E_diff = ΔT - ΔT_sp; where ΔT_sp is the target tension difference of the traction section, which is set by the process. The take-up section employs a tapered control strategy with variable roll diameter and constant tension. This strategy involves real-time detection of the roll diameter and dynamic adjustment of torque commands to ensure constant take-up tension. Based on the current real-time roll diameter D(t) and the taper formula T_sp(D)_3 = T0 · (D0 / D)^k, calculate the target tension T_sp(D)_3 of the take-up section; calculate the tension error E_3 of the take-up section = T_sp(D)_3 - T_fb1_3; where T0 represents the initial tension setting value, D0 represents the initial roll diameter, ^k represents the exponentiation operation, k represents the taper coefficient, and 0≤k≤1.

5. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, In step S2, converting the original tension error E_i into a set of decoupled errors Ed_i for independent closed-loop control of foil tension specifically involves: obtaining a process coupling mathematical model G(s) in advance through system identification. This model is used as input to convert the original tension error E_i, which has a strong coupling relationship, into a set of decoupled errors E_di for independent closed-loop control of foil tension. Specifically, the following steps are included: Ei is the original error, ΔUi is the control input; G(s) is a 3×3 coupling matrix; The feedforward decoupling compensator D(s) adopts the diagonal matrix decoupling method and calculates the coupling compensation amount δE_ij= G(s) ×ΔUi in real time; The output error of the feedforward decoupler after decoupling is Ed_i = E_i - Σ(δE_ij); After decoupling, the independent error vector Ed_i = [Ed_1, Ed_2, Ed_3]^T; Where δE_ij is the estimated coupling effect of segment j on segment i; the output is the independent error Ed_i after eliminating coupling interference.

6. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, The adaptive neural fuzzy inference network is a five-layer feedforward network, including an input layer, a fuzzification layer, a rule layer, a normalization layer, a consequent calculation layer, and an output layer. The input layer receives two precise quantities that characterize the error change trend of tension or tension difference in each segment: the decoupled tension error Ed of the current control cycle and the first-order difference ΔEd between the current error and the error of the previous cycle, and transmits them to the fuzzification layer. The fuzzification layer defines five fuzzy subsets for each input variable: negative large NB, negative small NS, zero ZO, positive small PS, and positive large PB; each subset corresponds to a Gaussian membership function; for input variable E_d or ΔE_d, its membership degree μ(x) belonging to each fuzzy subset is calculated, and its value is between [0, 1]; this is then passed to the rule layer; The rule layer, based on the five membership values ​​of each of the two input variables Ed_i and ΔEd_i, generates 25 rules using a full combination approach. Each rule takes the form: if Ed is A_m and ΔEd is B_n, then output f_mn; where A_m and B_n are the fuzzy antecedents (i.e., membership function values), and f_mn is the rule consequent function value, m=1,2……,5, n=1,2……,5. The trigger strength ω_k of this rule is calculated using a product method or a minimum method, where ω_k represents the degree of matching of the current input state to this rule. The 25 rule trigger strengths ω_1 ... ω_25 are output and passed to the normalization layer. The normalization layer and consequent calculation layer normalize the trigger strength of each rule as the normalized weight parameter bar{ω}_k = ω_k / Σ(ω_t); t=1,2……,25, k represents one of 25; for each rule, a linear function with respect to the input variable is defined, namely the adaptive consequent function f_k = p_k×E_d + q_k ×ΔE_d +r_k, where p_k, q_k, r_k are the connection weights of the neural network; for each rule, the product of the normalized weight and the consequent function is calculated by weighting: bar{ω}_k×f_k; The output layer sums the weighted outputs of all rules to obtain a clear, unambiguous control quantity: U = Σ (bar{ω}_k×f_k) = Σ (bar{ω}_k×(p_k×E_d + q_k×ΔE_d + r_k)); Finally, when the input of E_di and its rate of change ΔE_di for each section is obtained, the control command Ui representing the torque or current of each section is obtained according to the fuzzy calculation formula, including: unwinding section braking current command U1, output traction section torque difference compensation amount U2, and winding section torque command U3.

7. A multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 6, characterized in that, The connection weights {p_k, q_k, r_k} of the neural network are obtained through training using the following adaptive learning mechanism: The learning space is constructed using the network's adjustable parameter set {p_k, q_k, r_k} and the parameters of the membership function; the parameters of the membership function include the center and width of the Gaussian function. By combining least squares (LSE) with backpropagation (BP), under the condition of a fixed antecedent membership function, LSE is used to optimize all consequent parameters {p_k, q_k, r_k} in one go; then, the BP algorithm is used to fine-tune the antecedent membership function. Offline training is performed using historical production data to obtain an initial optimized set of adjustable parameters {p_k, q_k, r_k}, and then online fine-tuning is performed at an extremely low learning rate based on real-time performance metrics during operation to adapt to slow changes in operating conditions.

8. The multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, The unwinding section adopts a closed-loop control method of floating roller feedback + magnetic powder brake. U1 is sent to the current controller, the current controller outputs excitation current I_ex = K_c·U1, and the magnetic powder brake outputs braking torque T_brake = K_b·I_ex to control the foil speed in the unwinding section. In the traction section, the torque difference compensation amount U2 is superimposed on the third intermediate traction servo motor SM-2A and the fourth intermediate traction servo motor SM-2B of the traction section to obtain torque commands U2A and U2B respectively; Where, U2A = U_base -U2 / 2; U2B = U_base +U2 / 2; U_base is the given base torque command to maintain linear velocity; In the winding section, the torque command U3 is corrected to obtain the compensated torque command U_ff = (T_sp(D)·D) / (2·η); the final output U3_cmd = U3 + U_ff is sent to the winding servo motor driver. The driver operates in torque control mode and drives the winding foil servo motor (28) to perform winding.

9. A multi-segment collaborative high-precision constant tension control method for an electrode foil production line according to claim 1, characterized in that, The data fusion includes: performing time-stamp alignment processing on all collected data, and performing weighted fusion of the direct tension value T_fb2_i and the indirect tension feedback value T_fb1_i in the unwinding and rewinding sections to obtain the optimal tension estimate T_fused_i for the current section; The online calibration involves automatically correcting the floating roller calibration parameters K[i] and C[i] of the current section when |T_fb2_i - T_fb1_i| > the threshold in the unwinding or rewinding section, based on the floating roller's "angle-tension" conversion model T_calc[i] = K[i]×θ[i] + C[i]. i=1 represents the unwinding section, and i=2 represents the rewinding section. The process monitoring includes: calculating the mean, standard deviation, and process capability index CP / CPK in real time based on the estimated fusion tension value T_fused_i sequence of the unwinding or rewinding section; comparing the CPK value with the preset quality standard to draw an Xbar-R control chart; and generating a "process capability warning" if the CPK is continuously lower than the requirement or shows a trend of continuous increase / decrease of a preset number of points. The predictive maintenance analysis includes: performing Fast Fourier Transform or Wavelet Analysis on the fused tension estimate T_fused_i signal of the unwinding or winding section to extract its frequency domain features; matching abnormal peaks in the spectrum with a pre-established mechanical fault feature library to identify faults and provide alerts.

10. A multi-segment collaborative high-precision constant tension control system for an electrode foil production line, characterized in that, This includes multimodal sensors and data acquisition cards installed in the industrial site of the electrode foil production line, industrial computers and multi-segment ANFIS controllers, and multi-segment collaborative actuators; The multimodal sensor includes: a first absolute magnetic encoder for the unwinding section and a second absolute magnetic encoder for the winding section, used to collect the floating angle signals θ[1] and θ[3] of the unwinding and winding sections respectively; a first non-contact ultrasonic tension meter (5) and a second non-contact ultrasonic tension meter (31) for the unwinding and winding sections, used to collect the direct tension values ​​T_fb2_1 and T_fb2_2 of the unwinding and winding sections respectively; a third pressure sensor and a fourth pressure sensor for the traction section, used to collect the pressure values ​​F_A and F_B of the two rollers in the traction section respectively; and a roll diameter sensor for the winding section, used to collect the real-time roll diameter D(t); the above collected data constitute the multimodal sensing data of each process section of the electrode foil production line. The data acquisition card connects to the multimodal sensor and the industrial computer, and is used to acquire and upload data; The industrial computer and the multi-segment ANFIS controller are equipped with a real-time data interface layer, an intelligent control algorithm layer, and a fusion monitoring and optimization layer; the industrial computer is equipped with a control backend and a front-end human-machine interface. The real-time data interface layer is located in the industrial computer control backend, which collects multimodal sensing data from each process section of the electrode foil production line and performs data preprocessing. The intelligent control algorithm layer includes a multivariable dynamic decoupling unit located in the industrial computer control backend and an adaptive neural fuzzy inference network unit located in the multi-segment ANFIS controller. The multivariable dynamic decoupling unit calculates the original tension error E_i of the foil in each process segment based on the preprocessed data, establishes a multivariable dynamic feedforward decoupling calculation network, and converts the original tension error E_i with strong coupling relationship into a set of decoupled errors Ed_i for independent closed-loop control of the foil tension. The adaptive neural fuzzy inference network unit is used to construct an independent adaptive neural fuzzy inference network for each tension segment. Through fuzzy inference and self-learning engine, it takes Ed_i and its rate of change ΔEd_i of each segment as input and outputs torque or current commands Ui: unwinding segment torque command U1, traction segment torque difference compensation U2, and winding segment torque command U3 to the multi-segment cooperative actuator. The fusion monitoring and optimization layer is located in the industrial computer control backend. During the multi-segment collaborative high-precision constant tension control process, it performs data fusion, online calibration, process monitoring, and predictive maintenance analysis on the multimodal sensor data of each process segment. The front-end human-machine interface is used to visually display to users the data acquisition, process parameters and command calculation results of multi-segment collaborative high-precision constant tension control, as well as early warning information of the process control process; The multi-segment cooperative actuator includes the following actuators installed in the unwinding section, traction section, and winding section of the multi-segment cooperative high-precision constant tension system of the electrode foil production line: magnetic powder brake of the unwinding section, two servo motors SM-2A and SM-2B of the traction section, and foil winding servo motor (28), which are used to receive torque or current commands Ui and perform multi-segment cooperative high-precision constant tension control on the foil in each section; The multi-segment collaborative high-precision constant tension system of the electrode foil production line includes three parts: unwinding section, traction section, and rewinding section. The unwinding section is equipped with a first groove (001), the traction section is equipped with a second groove (002), and the rewinding section is equipped with a third groove (003). The first groove (001), the second groove (002), and the third groove (003) can be a formation groove or an etching groove according to actual needs. The electrode foil completes the etching or formation process through the unwinding section, the traction section, and the rewinding section.

Images