Control method of tension adjusting device on composite copper foil continuous production line

Abstract

The application provides a control method of a tension adjusting device on a composite copper foil continuous production line, comprising the following steps: obtaining a tension set value and a tension actual value of adjacent workstations on the composite copper foil continuous production line, and calculating a difference value between the tension set value and the tension actual value to obtain tension deviation data; in the process of adjusting the tension between the adjacent workstations, it is judged whether the copper foil appears to snake, if the copper foil appears to snake, the copper foil snake deviation data is fed back to a feedforward compensation function and a feedback controller to cooperatively adjust the control of the tension; on the basis of balancing the tension between the adjacent workstations, the running parameters of each workstation of the composite copper foil continuous production line are collected in real time, a production line running mathematical description function is established; and the control parameters of the production line running are optimized in combination with preset rules in a preset knowledge base.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a control method for tension adjustment equipment on a continuous production line for composite copper foil. Background Technology

[0002] In continuous production lines for composite copper foil, tension differences exist between adjacent stations due to the physical distance between stations and the elastic-plastic deformation characteristics of the copper foil as it passes through multiple stages such as coating, drying, and cooling. Sudden tension changes during the copper foil's journey can easily lead to foil serpentine behavior, resulting in decreased product quality. Specifically, when the copper foil moves from the coating station to the drying station, the sudden temperature increase causes thermal expansion and contraction, leading to localized stress changes and subsequent tension abrupt changes. Similarly, the sudden temperature drop when the copper foil moves from the drying station to the cooling station also causes similar tension fluctuations. Furthermore, inconsistent tension applied to the copper foil at different stations also contributes to tension differences between stations. For example, the coating station typically requires a higher tension to ensure coating uniformity, while the drying station needs to appropriately reduce tension to prevent deformation. The cooling station has different tension requirements. These complex process requirements and the inherent physical properties of copper foil present numerous challenges to tension control in composite copper foil production lines. How to ensure stable copper foil tension while taking into account the specific needs of each workstation, coordinating the tension balance between adjacent workstations, and avoiding copper foil serpentination caused by sudden tension changes is a key technical problem that urgently needs to be solved. Summary of the Invention

[0003] This invention provides a control method for tension adjustment equipment on a continuous production line of composite copper foil, mainly including:

[0004] Obtain the tension setpoint and actual tension value of adjacent stations on the continuous production line of composite copper foil, and calculate the difference between the tension setpoint and the actual tension value to obtain tension deviation data;

[0005] By combining the tension fluctuation data and the tension deviation data in historical production data, the possible tension abrupt changes under the current production conditions are predicted, and a feedforward compensation signal is generated based on the prediction results to form a preliminary tension gradient correction signal.

[0006] Simultaneously, by combining tension deviation data and real-time operating status data of copper foil, the tension control parameters are dynamically adjusted to generate a feedback compensation signal, which corrects the initial tension gradient correction signal and forms a corrected tension gradient correction signal.

[0007] The corrected tension gradient correction signal is used to determine whether the tension between adjacent workstations is balanced. If they are not balanced, the corrected tension gradient correction signal is sent to the tension adjustment device of the adjacent workstation. The tension between the workstations is dynamically adjusted by the adjustment device until the tension between the adjacent workstations reaches a balanced state.

[0008] During the process of adjusting the tension between adjacent workstations, it is determined whether the copper foil is serpentine. If serpentine occurs, the copper foil serpentine offset data is fed back to the feedforward compensation function and the feedback controller to coordinate the adjustment of tension control.

[0009] Based on the tension balance between adjacent workstations, the operating parameters of each workstation in the continuous composite copper foil production line are collected in real time, and a mathematical description function of the production line operation is established.

[0010] The control parameters for production line operation are optimized by combining preset rules from the preset knowledge base.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses a control method for tension adjustment equipment on a continuous production line for composite copper foil. The method involves setting tension coordination control units between adjacent workstations to acquire the setpoint and actual tension values, and calculating the deviation data. This deviation data is input into a feedforward and feedback control model, and combined with historical data and real-time operating status, a tension gradient correction signal is generated. The tension between workstations is dynamically adjusted based on the correction signal to achieve balance. Simultaneously, the method monitors for copper foil serpentine behavior, triggering alarms and feeding back to the control model to adjust the strategy. This invention also collects operating parameters from each workstation in real time, establishes a mathematical model through big data analysis, and adaptively optimizes the control model parameters. This method, through a combination of feedforward prediction and feedback compensation, achieves precise and coordinated tension control in the composite copper foil production line. Furthermore, the introduction of big data analysis and adaptive optimization continuously improves control accuracy and stability. This invention effectively solves problems such as unstable tension control and serpentine behavior in the production of composite copper foil, significantly improving product quality consistency and production efficiency. Attached Figure Description

[0013] Figure 1 This is a flowchart of the control method for the tension adjustment equipment on the continuous production line of composite copper foil according to the present invention. Detailed Implementation

[0014] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0015] like Figure 1 The control method for tension adjustment equipment on a continuous composite copper foil production line in this embodiment may specifically include:

[0016] Step S101: Obtain the tension set value and actual tension value of adjacent stations on the continuous production line of composite copper foil, and calculate the difference between the tension set value and the actual tension value to obtain tension deviation data.

[0017] A station sensor array is used to collect tension monitoring data on the composite copper foil conveyor line. The signal is processed by the tension data acquisition unit to obtain the station tension value. The station tension value is compared with a preset tension threshold range, and the position information is obtained from the optical encoder and converted into a digital quantity. If the station tension value exceeds the preset tension threshold range, the tension deviation value is obtained from the tension data acquisition unit and sent to the tension compensation unit.

[0018] For example, a workstation sensor array is used to deploy multiple sets of tension monitoring probes between adjacent workstations on the composite copper foil conveyor line and calculate real-time tension monitoring data. The collected data is processed by a tension data acquisition unit to obtain the workstation tension value. Optical encoders are installed at the drive shaft ends of adjacent workstations to collect rotational speed information. The workstation tension value is compared with a preset tension threshold range, and position information is obtained from the optical encoder and converted into a digital quantity. If the workstation tension value exceeds the preset tension threshold range, the tension deviation value is obtained from the tension data acquisition unit and input to the tension compensation unit. The compensation unit calculates the compensation coefficient using a recursive least squares algorithm. A state-space equation set including the tension deviation value, compensation coefficient, and workstation rotational speed is established based on the compensation coefficient output by the tension compensation unit, and a compensation control signal is output. The rotational speed of the drive motor of the drive shaft is adjusted according to the compensation control signal, and the adjusted rotational speed signal is fed back to the tension detection unit. The tension detection unit collects the workstation tension value after speed adjustment, compares it with the tension setting reference value, and outputs new tension deviation data to the tension compensation unit. Tension control in a composite copper foil production line involves dynamic coordination between multiple workstations. Taking a specific working condition as an example, when the copper foil is transferred from the first workstation to the second workstation, a tension monitoring probe collects real-time tension values. The probe uses a strain gauge sensor with a sensitivity of up to 0.1N and a sampling frequency of 200Hz. After signal conditioning, it outputs a 4-20mA standard signal. The tension data acquisition unit converts the analog signal into a digital quantity through a 16-bit AD converter. The workstation tension setting baseline is set to 50N, and the preset tension fluctuation threshold range is ±5N. An optical encoder is installed at the end of the drive shaft with a resolution of 2500 lines / revolution, which can accurately detect speed changes. When the detected tension exceeds the threshold range, compensation control is required. The compensation control uses a recursive least squares algorithm. By constructing a state vector [tension deviation, speed, compensation coefficient], a state space equation X(k+1)=AX(k)+Bu(k) is established, where A is the state transition matrix, B is the control matrix, u(k) is the compensation input, and X(k) is the state vector. Assuming a detected tension deviation of 6N and a rotational speed of 1200rpm, with an initial compensation coefficient of 0.8, a new compensation coefficient of 0.92 is calculated by substituting these values ​​into the equation. Closed-loop control of the drive motor speed is then implemented based on this compensation coefficient. The drive motor is a permanent magnet synchronous motor with a rated speed of 1500rpm, and speed regulation is achieved via a frequency converter with a response time of less than 50ms. After speed adjustment, the tension detection unit re-acquires the tension value at the workstation. If a deviation still exists, compensation control continues until the tension returns to the set range. The state-space equations are solved using a discrete Kalman filter algorithm, which effectively suppresses the influence of measurement noise. The system noise covariance matrix Q and the measurement noise covariance matrix R are respectively set as...

[0019] Q = diag(0.01, 0.01, 0.01), R = 0.1. The prediction step calculates the estimated state value for the next moment using the state equation, and the correction step corrects the estimated value by combining the measured value, finally obtaining the optimal state estimate. During the tension compensation control process, the update cycle of the compensation coefficient is 100ms to avoid system oscillation caused by frequent adjustments. The speed regulation of the drive motor adopts a ramp function to achieve a smooth transition, with the slope set to 200rpm / s to ensure the continuity of tension changes. When the distance between workstations is 2000mm and the copper foil running speed is 60m / min, the coupling effect of tension between adjacent workstations needs to be eliminated through feedforward compensation.

[0020] Step S102: Combining the tension fluctuation data and the tension deviation data in the historical production data, predict the possible tension abrupt changes under the current production conditions, and generate a feedforward compensation signal based on the prediction results to form a preliminary tension gradient correction signal.

[0021] A sliding time window is used to sample tension deviation data, and the sampled data is filtered through a median filter to obtain a filtered tension deviation sequence. Based on the tension deviation sequence and tension fluctuation feature values ​​in the historical production database, a long short-term memory network is used to train and obtain the tension fluctuation pattern. For the tension fluctuation pattern, the least squares method is used to fit the tension change curve to obtain a tension sudden change warning threshold. A feedforward compensation function is constructed based on the tension sudden change warning threshold, and an initial compensation signal is obtained by smoothing the feedforward compensation function through an adaptive filter. The initial compensation signal is then optimized and calculated using the gradient descent method to obtain a tension gradient correction signal.

[0022] For example, a sliding time window is used to sample the tension deviation data sequence. A data preprocessing unit performs median filtering on the sampled data to remove outliers, and outputs the filtered tension deviation sequence from the data preprocessing unit. Based on the tension deviation sequence from the data preprocessing unit, tension fluctuation feature values ​​are read from a historical production database. A long short-term memory (LSTM) network, including the time step and the number of hidden layer nodes, is constructed using a time series prediction unit. For the tension fluctuation pattern obtained from the LSM network training, the least squares method is used to fit the tension change curve, and a tension abrupt change warning threshold is output from the curve fitting unit. A feedforward compensation function is constructed based on the tension abrupt change warning threshold, and the compensation function is input to the feedforward compensation unit, which calculates the compensation amount sequence. An adaptive filter is used to smooth the compensation amount sequence, outputting an initial compensation signal containing the compensation signal amplitude. The initial compensation signal is optimized using the gradient descent method, and a tension gradient correction signal is output from the feedforward compensation unit. Tension control in a composite copper foil production line involves several key aspects. Taking tension fluctuation control during production as an example, tension deviation data is collected during production with a sampling period of 200ms. A sliding time window with a length of 1000 data points is set. The data is preprocessed using a median filtering algorithm, with the filtering window containing 5 data points to remove sudden interference during measurement. Temporal series analysis is used to extract tension fluctuation features. Tension fluctuation data from the past 30 days are read from a historical database. Each data record contains three fields: timestamp, tension value, and production speed. Temporal prediction is performed using a Long Short-Term Memory (LSTM) network. The network structure contains two LSTM layers with 64 hidden nodes per layer, a time step of 20, and an input feature dimension of 3. Cubic spline interpolation is used to fit the tension change curve. The polynomial coefficients between adjacent nodes are solved using the least squares method. A tension mutation warning threshold is set to three times the standard deviation; that is, a warning signal is triggered when the detected tension fluctuation amplitude exceeds 3σ. The feedforward compensation function u(t) adopts a PI controller form, u(t) = Kpe(t) + Ki∫e(t)dt, where Kp is the proportional coefficient with a value of 0.8, Ki is the integral coefficient with a value of 0.05, and e(t) is the error signal, representing the difference between the expected output and the actual output. The smoothing of the compensation signal uses an adaptive Kalman filter algorithm, with the state equation being x(k+1) = x(k) + w(k) and the observation equation being y(k) = x(k) + v(k), where w(k) and v(k) are the process noise and observation noise, respectively, with covariance matrices Q = 0.01 and R = 0.1. The filter gain matrix K is dynamically adjusted according to changes in the measured values ​​to ensure the smoothness of the filtering result. During the gradient descent optimization process, the learning rate α = 0.01, the number of iterations is 100, and the loss function adopts the mean squared error form L = Σ(y-y')2 / n, where y is the target tension value, y' is the actual tension value, and n is the number of data points.The loss function is converged to its minimum through iterative calculation, ultimately outputting an optimized tension gradient correction signal. This multi-level linkage control scheme exhibits excellent dynamic characteristics in actual production. When the production line speed increases from 50m / min to 80m / min, the tension fluctuation amplitude is controlled within ±2N, achieving a 50% improvement in stability compared to traditional PID control. The control process forms a closed-loop feedback mechanism, achieving precise control of tension fluctuations through a combination of feedforward compensation and adaptive optimization.

[0023] In step S103, the tension control parameters are dynamically adjusted by combining the tension deviation data and the real-time operating status data of the copper foil, and a feedback compensation signal is generated to correct the preliminary tension gradient correction signal, thus forming a corrected tension gradient correction signal.

[0024] Based on the real-time operating status data of the copper foil, the recursive least squares algorithm is used to identify the operating parameters of the copper foil online. The online identification results are used to construct a feedback compensator. The feedback compensator outputs a tension compensation signal, which is used to dynamically correct the tension gradient correction value. A neural network is used to adaptively adjust the dynamically corrected compensation value, and the optimized correction parameters are obtained from the neural network. The correction parameters are used to adjust the tension control parameters.

[0025] For example, a sensor array is used to monitor the operating status of the copper foil production line in real time. Multiple sensors collect real-time data, including operating speed and tension fluctuations, to obtain a complete operating status dataset. For the collected operating status dataset, a recursive least squares algorithm is used to identify the copper foil operating parameters online. Based on the identification results, a feedback compensator containing proportional and integral terms is constructed, and a compensation signal is output from the feedback compensator. The tension gradient correction value is dynamically corrected based on the compensation signal. A neural network is used to adaptively adjust the corrected compensation value, and the optimized correction parameters are output from the neural network. The optimized correction parameters are used to dynamically adjust the tension control parameters, establishing a control function that includes feedback compensation. The corrected tension gradient correction signal is output through the feedback compensator. The feedback control process of the copper foil production line involves the real-time monitoring and dynamic adjustment of multiple key parameters. Taking a certain workstation on the production line as an example, a sensor array is arranged with strain gauge tension sensors, photoelectric encoders, and displacement sensors. The sampling frequency is set to 100Hz, and the data resolution is 16 bits, to collect operating parameters in real time. The tension sensor has a range of 0-500N and a sensitivity of 0.2mV / N; the photoelectric encoder has a resolution of 2500 lines / revolution and is used for speed detection; the displacement sensor has a range of 0-50mm and is used for position feedback. Online identification of operating parameters uses a recursive least squares method to construct the state equation x(k+1)=Ax(k)+Bu(k) and the observation equation y(k)=Cx(k)+Du(k), where A is the state transition matrix, B is the control matrix, C is the output rectangle, and D is the direct transfer matrix. The state vector x(k) contains two components: velocity and tension, and the control input u(k) is the torque command for the drive motor. When a sudden change in speed from 50m / min to 80m / min is detected, a transient fluctuation in tension occurs. The recursive algorithm calculates the velocity response time constant to be 0.5s and the tension response time constant to be 0.3s. The feedback compensator adopts a PI control structure, with the proportional coefficient Kp determined to be 0.85 using the Ziegler-Nichols tuning method, and the integral time Ti set to 2 seconds. The compensator's u(t) = Kp[e(t) + 1 / Ti∫e(t)dt], where e(t) is the tension deviation. For a tension setpoint of 50N, when the measured tension is 45N, the calculated proportional term output is 4.25N, the integral term output is 3.75N, and the total compensation is 8N. The adaptive neural network adopts a three-layer structure: two nodes in the input layer corresponding to speed and tension deviation, eight nodes in the hidden layer, and one node in the output layer outputting the compensation correction. The activation function of the network is the hyperbolic tangent function, and the learning rate is set to 0.01. When the input speed is 80m / min and the tension deviation is -5N, the output compensation correction coefficient after neural network mapping is 1.2, correcting the original compensation amount.In closed-loop feedback control, the compensation signal and gradient correction signal are combined using a weighted summation method, with weighting coefficients of 0.7 and 0.3, respectively. Under actual operating conditions, when the production line starts and accelerates, the feedback compensation loop can quickly respond to tension fluctuations caused by speed changes. The compensated tension fluctuation amplitude is reduced from ±8N to ±2N, and the dynamic response time is shortened by 40%.

[0026] Step S104: Determine whether the tension between adjacent workstations is balanced based on the corrected tension gradient correction signal. If not, send the corrected tension gradient correction signal to the tension adjustment device of the adjacent workstation respectively, and dynamically adjust the tension between the workstations through the adjustment device until the tension between the adjacent workstations reaches a balanced state.

[0027] A sensor array is used to collect real-time tension values ​​of adjacent workstations; based on the real-time tension values, the tension difference between workstations is calculated to obtain tension deviation data; a correction signal is obtained through a tension gradient correction unit, which is generated when the tension deviation data exceeds a preset tension balance threshold; based on the correction signal, a workstation speed adjustment command is output from a signal distributor, which is calculated by the signal distributor according to the speed adjustment ratio of the transmission shaft of each workstation; a feedback controller is used to process the workstation speed adjustment command, and a speed control signal is output from the feedback controller, which is obtained by the feedback controller calculating the speed compensation amount based on the workstation operating parameters.

[0028] For example, a sensor array is used to collect real-time tension values ​​of adjacent workstations. A tension comparison unit calculates the tension difference between workstations and outputs tension deviation data from the tension comparison unit. The tension deviation data is compared with a preset tension balance threshold. If the tension deviation exceeds the balance threshold range, a correction signal is obtained from the tension gradient correction unit. Based on the correction signal, a signal distributor calculates the speed adjustment ratio of the drive shafts at each workstation and outputs a workstation speed adjustment command. A feedback controller processes the workstation speed adjustment command, calculates the speed compensation amount based on the workstation operating parameters, and outputs a speed control signal. The speed control signal adjusts the speed of the drive shafts at adjacent workstations, and a speed feedback signal is collected from the drive shaft encoder. The workstation tension value is calculated based on the speed feedback signal. If the tension deviation still exceeds the balance threshold range, the process returns to step three until the tension deviation falls within the balance threshold range. The application of tension balance control in copper foil production lines involves the coordinated operation of multiple processes. Taking two adjacent workstations as an example, strain gauge tension sensors are used for real-time monitoring with a sampling frequency of 200Hz, a signal resolution of 16 bits, and a measurement range of 0-500N. The sensor outputs a 4-20mA standard signal, which is then conditioned and input to the tension comparison unit. The tension balance control between workstations adopts a two-stage structure. The first stage is for tension difference judgment, with a preset balance threshold of ±3N. When the tension value at workstation 1 is detected to be 52N and the tension value at workstation 2 is detected to be 47N, the calculated tension deviation is 5N, exceeding the balance threshold range, triggering a correction signal output. The amplitude of the correction signal is obtained through PID calculation, with a proportional coefficient Kp = 0.8, an integral time Ti = 2s, and a derivative time Td = 0.1s. Signal allocation employs a weighted allocation method. Based on the station characteristics, an allocation matrix of [0.6, 0.4; 0.4, 0.6] is established, indicating that 60% of the speed control command for station 1 applies to this station, and 40% applies to adjacent stations, while the opposite applies to station 2. For a tension deviation of 5N, the allocated speed control command for station 1 is -1.8rpm, and for station 2 it is 1.2rpm. The feedback controller uses a model predictive control structure, with a prediction time domain of 10 sampling periods and a control time domain of 3 sampling periods. The state equation is x(k+1) = 0.95x(k) + 0.2u(k), where x is the tension value and u is the speed control command. Substituting the initial state of 52N for station 1, the future tension change trend in the time domain is predicted, and the optimal speed control sequence is calculated through rolling optimization. The transmission shaft speed is adjusted using a frequency converter; a rated frequency of 50Hz corresponds to a speed of 1500rpm, with a speed control resolution of 0.01Hz. The encoder's feedback signal is 2500 lines / revolution, which, after being quadrupled, provides 10000 pulses / revolution for position feedback. In closed-loop control, the speed signal is filtered by a first-order low-pass filter with a cutoff frequency of 5Hz to remove high-frequency interference components.During the iterative process of tension balance control, the tension value of the workstation is re-acquired and compared after each adjustment. When the detected tension deviation is 2.8N, it is within the balance threshold range, and the adjustment process is completed. Compared with the traditional single-workstation control method, multi-workstation collaborative control can more effectively suppress the transmission and amplification of tension fluctuations.

[0029] In step S105, during the process of adjusting the tension between adjacent workstations, it is determined whether the copper foil exhibits serpentine behavior. If serpentine behavior occurs, the copper foil serpentine offset data is fed back to the feedforward compensation function and the feedback controller to coordinate the adjustment of tension control.

[0030] A photoelectric sensor array is used to acquire the position information of the copper foil edge. The position detection unit calculates the serpentine offset data of the copper foil centerline relative to the baseline. Based on the serpentine offset data, feature parameters are extracted by a deep neural network, and serpentine correction parameters are obtained from a historical database as feedforward compensation. For the serpentine offset data, a dynamic compensation signal is calculated by the feedback controller as feedback compensation. The feedforward compensation and the feedback compensation are weighted by a fusion unit to obtain a station tension compensation command, which is then used by a tension control device to adjust the tension distribution of adjacent stations.

[0031] For example, a photoelectric sensor array is used to detect the edge position of the copper foil in real time. The offset distance between the center line of the copper foil and the baseline is calculated by the position detection unit, and the serpentine offset data is output from the position detection unit. The serpentine offset data is compared with a preset serpentine warning threshold. If the serpentine offset exceeds the warning threshold range, the serpentine alarm device is triggered and an alarm signal is output. Features are extracted from the serpentine offset data using a deep neural network, and serpentine correction parameters under similar working conditions are obtained from a historical database to output a feedforward compensation amount. A feedback controller processes the real-time serpentine offset data, calculates the feedback compensation amount based on the offset change trend, and outputs a dynamic compensation signal from the feedback controller. The feedforward compensation amount and the feedback compensation amount are weighted and fused, and a comprehensive compensation parameter is calculated by the fusion unit to output a station tension compensation command. The station tension compensation command is used to adjust the tension distribution of adjacent stations, and the adjusted serpentine offset is collected by the position detection unit. It is determined whether the adjusted serpentine offset still exceeds the warning threshold. If it does, the feedforward compensation operation is returned until the serpentine offset is within the warning threshold range. The serpentine control of the copper foil production line involves a multi-level, closed-loop adjustment process. Taking actual operating conditions as an example, through-beam photoelectric sensors are arranged on both sides of the production line. The sensors have a detection range of ±50mm, a resolution of 0.1mm, and a sampling frequency of 100Hz. Five sensor arrays are used to detect the edge position of the copper foil. The least squares method is used to fit the center line of the copper foil, and its offset distance from the baseline is calculated. The serpentine warning threshold is set in stages. When the detected copper foil offset exceeds ±10mm, a first-level warning is triggered, and an alarm signal is output; when the offset exceeds ±20mm, a second-level warning is triggered, and a deep neural network is activated for feature extraction. The deep neural network adopts a five-layer structure, including three convolutional layers and two fully connected layers. The input features include parameters such as offset sequence, running speed, and tension distribution. The historical database stores nearly six months of operating data, with each record containing fields such as timestamp, offset, speed, and tension. Similar operating conditions are retrieved using a feature matching algorithm, and correction parameters are extracted as feedforward compensation. Feedforward compensation employs proportional-derivative control with a proportional coefficient Kp = 0.8 and a derivative coefficient Kd = 0.1. For a 20mm offset, the calculated feedforward compensation is 16N. The feedback controller uses a fuzzy control structure, with input variables being the offset e and the rate of change of offset ec, and the output variable being the compensation amount u. The fuzzy rules are in the form of a two-dimensional lookup table, containing 49 rules. When the detected offset is 15mm and the offset rate is 2mm / s, the fuzzy output obtained from the lookup table is 0.6, which, after defuzzification, yields a feedback compensation amount of 12N. The weighted fusion of the compensation amounts uses an adaptive weighting method, with the weight coefficients dynamically adjusted using a Kalman filter algorithm. The state equation is w(k+1) = w(k) + v(k), where w(k) is the weight vector and v(k) is the process noise.The measurement equation is y(k) = Hw(k) + e(k), where H is the observation matrix and e(k) is the measurement noise. When the feedforward compensation effect is good, its weighting coefficient is increased to 0.7, and the feedback compensation weighting is reduced to 0.3. The station tension adjustment uses frequency converter control, and closed-loop speed regulation is achieved through a PI controller. The control cycle is 50ms, the proportional coefficient Ki = 0.5, and the integral time Ti = 1s. In actual operation, when the copper foil serpentine offset gradually decreases from 20mm to 5mm, the dynamic adjustment process of the station tension exhibits good tracking characteristics and stability.

[0032] Step S106: Based on the tension balance between adjacent workstations, collect the operating parameters of each workstation of the composite copper foil continuous production line in real time, and establish the mathematical description function of the production line operation.

[0033] The system acquires output signals from tension sensors, speed sensors, and temperature sensors, and outputs raw parameter data from the data acquisition unit. Based on the raw parameter data, a data preprocessing unit performs normalization processing to obtain a standardized parameter dataset. For the standardized parameter dataset, principal component analysis is used to reduce the dimensionality of tension, copper foil speed, and temperature parameters, and a parameter correlation coefficient matrix is ​​obtained from the correlation analysis unit. A neural network is used to perform nonlinear mapping on the operating condition characteristic parameters, and a state prediction value is obtained from the prediction unit. The state prediction value is then output from the prediction unit. A mathematical description function for production line operation is established based on the state prediction value.

[0034] For example, a sensor array is used to collect operating parameters at each workstation of the production line. Multiple data acquisition units acquire output signals from tension sensors, speed sensors, and temperature sensors, outputting raw parameter data. A data preprocessing unit normalizes the raw parameter data and removes outliers, generating a standardized parameter dataset. Based on the standardized parameter dataset, principal component analysis is used to reduce the dimensionality of tension, copper foil speed, and temperature parameters, yielding a parameter correlation coefficient matrix from the correlation analysis unit. For the parameter correlation coefficient matrix, eigenvalue decomposition is used to extract key influencing factors, outputting operating condition characteristic parameters from the feature extraction unit. A neural network is used to perform nonlinear mapping on the operating condition characteristic parameters, constructing a state predictor with hidden nodes, outputting predicted state values ​​from the prediction unit. A multivariate state-space equation system is established based on the predicted state values, and the equation parameters are identified online using recursive least squares. A data communication unit uploads the identified equation parameters to the central control unit, where they are stored to form the mathematical description function of the production line operation. The parameter modeling process for the copper foil production line involves data processing and analysis across multiple stages. Multiple sensor arrays are deployed on the production line, including tension sensors, speed sensors, and temperature sensors. The tension sensor uses a strain gauge structure with a range of 0-500N and a sensitivity of 0.2mV / N; the speed sensor uses a photoelectric encoder with a resolution of 2500 lines / revolution; and the temperature sensor uses a PT100 with a measurement range of 0-200℃ and an accuracy of 0.1℃. The sampling frequency is set to 100Hz, and the data resolution is 16 bits. The preprocessing of the raw data employs a standardization method. For each measured value x, a standard score is calculated using the formula z = (x - μ) / σ, where μ is the mean and σ is the standard deviation. Outliers are identified using the 3σ criterion; a measured value is considered an outlier if its standard score exceeds ±3. For example, at a certain workstation, the mean tension is 50N, and the standard deviation is 2N. Measured values ​​greater than 56N are discarded. During principal component analysis, the correlation coefficient matrix R between the parameters is calculated first. Taking three parameters as an example, the correlation coefficient matrix R = [1.0, 0.8, 0.3; 0.8, 1.0, 0.4; 0.3, 0.4, 1.0] is obtained, where the correlation coefficient between tension and velocity is 0.8, showing a strong correlation. Eigenvalue decomposition yields eigenvalues ​​λ1 = 2.1, λ2 = 0.6, and λ3 = 0.3, with corresponding eigenvectors as follows:

[0035] The first principal component, v1 = [0.707, 0.707, 0.000], v2 = [-0.408, 0.408, 0.816], and v3 = [0.577, -0.577, 0.577], contributes 70%, reflecting that tension-velocity coupling is the main influencing factor. The neural network adopts a three-layer structure: the three nodes in the input layer correspond to the standardized parameters, the eight nodes in the hidden layer use the hyperbolic tangent activation function, and the three nodes in the output layer correspond to the predicted state values. Training uses the Levenberg-Marquardt algorithm with a learning rate of 0.01. Training stops when the validation set error does not decrease for five consecutive iterations. The state-space equation x(k+1) is in discrete form: x(k+1) = Ax(k) + Bu(k), where the state vector x(k) contains three components: tension, velocity, and temperature. The control input u(k) is the torque command of the drive motor, and B is the control matrix, describing how the control input u(k) affects the system's state variables. The forgetting factor for the recursive least squares method is set to 0.95, and the initial covariance matrix P0 = 100I. After 1000 iterations, the state transition matrix A = [0.95, 0.04, 0.01; 0.03, 0.96, 0.01; 0.01, 0.01, 0.98] is obtained, reflecting the dynamic coupling relationship between parameters. Data communication adopts the OPCUA protocol. The data frame contains three fields: timestamp, parameter value, and quality stamp. The communication period is 50ms, and the data is uploaded to the central control unit via Ethernet.

[0036] Step S107: Optimize the control parameters of the production line operation by combining the preset rules in the preset knowledge base.

[0037] The preset rules in the preset knowledge base are parsed by the rule matcher to extract the parameter tuning rules; the parameter tuning rules are quantified by the fuzzy inferencer and the parameter adjustment direction is obtained by calculating the membership degree; based on the parameter adjustment direction, the action space is constructed by the deep reinforcement learning method, and the tension deviation sequence and the control parameter combination are set as the state space. The optimized control parameter set is output from the reinforcement learning unit.

[0038] For example, an optimization calculation unit analyzes the mathematical description function of the production line operation, uses a genetic algorithm to encode the control parameters into binary form, constructs a fitness function based on the tension stability index, and outputs the optimization range of the control parameters from the optimization calculation unit. Based on the optimization range of the control parameters, a rule matcher analyzes the empirical rules in the preset knowledge base, extracting expert experience data, including parameter tuning rules, from the rule base. For the expert experience data, a fuzzy inference engine quantifies the parameter tuning rules, calculating the parameter adjustment direction and adjustment step size through membership degree calculation. Based on the parameter adjustment direction, deep reinforcement learning constructs the action space, setting the tension deviation sequence and control parameter combination as the state space, and outputs the optimized control parameter set from the reinforcement learning unit. A proportional-integral controller runs the optimized control parameters online, calculating the tension output deviation using a closed-loop control method. For the tension output deviation, an evaluation index function including overshoot and settling time is established, and the parameter evaluation result is output from the evaluation calculation unit. The parameter evaluation result is compared with the preset optimization target. If the evaluation index does not meet the optimization target, the process returns to constructing the action space until the evaluation index meets the preset optimization target. Parameter optimization in composite copper foil production lines involves iterative calculations across multiple stages. Taking control parameter optimization as an example, the parameters are represented using 16-bit binary encoding, and the proportional coefficient Kp ranges from [value missing].

[0039] The integral time Ti ranges from [0.1, 2.0] to [0.5s, 5.0s]. The fitness function f is in weighted form: f = 0.6 / Mp + 0.4 / ts, where Mp is the overshoot and ts is the settling time. The rule base stores parameter tuning rules for typical operating conditions, described in If-Then form. For example, when the tension deviation is positive and the rate of change of the deviation is positive, the corresponding rule is Ife>0 and de / dt>0, then ΔKp = -0.1. Fuzzy inference uses the Mamdani method, with the input variables using a triangular membership function, dividing the [-1, 1] interval into 5 equal parts, and the output using a single-value membership function. The state space of deep reinforcement learning contains 8 dimensions, including the tension deviation sequence [e(k), e(k-1), e(k-2)], control parameters [Kp, Ti], and operating condition parameters [v, T]. The action space is a two-dimensional continuous space, corresponding to the parameter increments [ΔKp, ΔTi]. The reward function *r* is designed as *r = -|e| - 0.1|de / dt| - 0.01|u|, where *e* is the tension deviation and *u* is the control output. The controller uses a cascaded PI structure, with the inner loop for speed control and the outer loop for tension control. The control period for the speed loop is 10 ms, and the control period for the tension loop is 50 ms. When the tension setpoint is detected to jump from 50 N to 60 N, the unoptimized parameters Kp = 0.8 and Ti = 2.0 s are used for control, resulting in an overshoot Mp = 15% and a settling time ts = 3.5 s. After 100 generations of genetic optimization iterations, with a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.05, the optimized parameters Kp = 1.2 and Ti = 1.5 s are obtained. Deep reinforcement learning uses an Actor-Critical structure, with both the Actor and Critical networks employing a three-layer structure and 32 hidden nodes. After 10,000 training steps, the parameters converged to Kp = 1.25 and Ti = 1.4s. The evaluation index function is J = w1Mp + w2ts + w3*σe, where σe is the steady-state error standard deviation, and the weighting coefficients [w1, w2, w3] = [0.4, 0.4, 0.2]. Step response tests were conducted using the optimized parameters, and the overshoot was reduced to 8%, the settling time was shortened to 2.2s, and the steady-state error standard deviation was reduced from 0.8N to 0.3N. The optimization process formed a closed-loop optimization chain of parameter encoding, rule matching, reinforcement learning, and evaluation feedback.

[0040] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A control method for tension adjustment equipment on a continuous production line of composite copper foil, characterized in that, The method includes: Obtain the tension setpoint and actual tension value of adjacent stations on the continuous production line of composite copper foil, and calculate the difference between the tension setpoint and the actual tension value to obtain tension deviation data; By combining the tension fluctuation data and the tension deviation data in historical production data, the possible tension abrupt changes under the current production conditions are predicted, and a feedforward compensation signal is generated based on the prediction results to form a preliminary tension gradient correction signal. Simultaneously, by combining tension deviation data and real-time operating status data of copper foil, the tension control parameters are dynamically adjusted to generate a feedback compensation signal, which corrects the initial tension gradient correction signal and forms a corrected tension gradient correction signal. The corrected tension gradient correction signal is used to determine whether the tension between adjacent workstations is balanced. If they are not balanced, the corrected tension gradient correction signal is sent to the tension adjustment device of the adjacent workstation. The tension between the workstations is dynamically adjusted by the adjustment device until the tension between the adjacent workstations reaches a balanced state. During the process of adjusting the tension between adjacent workstations, it is determined whether the copper foil exhibits serpentine behavior. If serpentine behavior occurs, the copper foil serpentine offset data is fed back to the feedforward compensation function and the feedback controller to coordinate the adjustment of tension control.

2. The method according to claim 1, characterized in that, The process of obtaining the tension setpoint and actual tension value of adjacent stations on the continuous composite copper foil production line, and calculating the difference between the tension setpoint and the actual tension value to obtain tension deviation data, includes: A workstation sensor array is used to collect tension monitoring data on the composite copper foil conveyor line, and the workstation tension value is obtained by signal processing through a tension data acquisition unit. The position information is obtained from the optical encoder and converted into a digital quantity by comparing the station tension value with the preset tension threshold range. If the tension value at the workstation exceeds the preset tension threshold range, the tension deviation value is obtained from the tension data acquisition unit and sent to the tension compensation unit.

3. The method according to claim 1, characterized in that, The process involves combining tension fluctuation data from historical production data with tension deviation data to predict potential tension abrupt changes under current production conditions. Based on the prediction results, a feedforward compensation signal is generated to form a preliminary tension gradient correction signal, including: A sliding time window is used to sample the tension deviation data, and the sampled data is passed through a median filter to obtain the filtered tension deviation sequence. Based on the tension deviation sequence and the tension fluctuation characteristic values ​​in the historical production database, the tension fluctuation pattern is obtained by training a long short-term memory network. Based on the aforementioned tension fluctuation pattern, the least squares method is used to fit the tension change curve to obtain the tension sudden change warning threshold; A feedforward compensation function is constructed based on the tension sudden change warning threshold. An initial compensation signal is obtained by smoothing the feedforward compensation function through an adaptive filter. The initial compensation signal is then optimized and calculated using the gradient descent method to obtain the tension gradient correction signal.

4. The method according to claim 1, characterized in that, The process involves simultaneously combining tension deviation data and real-time operating status data of the copper foil to dynamically adjust tension control parameters, generate a feedback compensation signal, and correct the initial tension gradient correction signal to form a corrected tension gradient correction signal, including: Based on the real-time operating status data of the copper foil, the recursive least squares algorithm is used to identify the operating parameters of the copper foil online, and the online identification results are used to construct a feedback compensator. The feedback compensator outputs a tension compensation signal, which is used to dynamically correct the tension gradient correction value. A neural network is used to adaptively adjust the dynamically corrected compensation value, and optimized correction parameters are obtained from the neural network. These correction parameters are used to adjust the tension control parameters.

5. The method according to claim 1, characterized in that, The step of determining whether the tension between adjacent workstations is balanced based on the corrected tension gradient correction signal, and if not balanced, sending the corrected tension gradient correction signal to the tension adjustment devices at the adjacent workstations respectively, and dynamically adjusting the tension between the workstations until the tension between the adjacent workstations reaches a balanced state, includes: A sensor array is used to collect real-time tension values ​​from adjacent workstations; Based on the real-time tension value, the tension difference between workstations is calculated to obtain tension deviation data; A correction signal is obtained through a tension gradient correction unit, and the correction signal is generated when the tension deviation data exceeds a preset tension balance threshold. In response to the correction signal, a station speed adjustment command is output from the signal distributor. The station speed adjustment command is calculated by the signal distributor based on the speed adjustment ratio of the transmission shaft of each station. The speed control command of the workstation is processed by a feedback controller, and a speed control signal is output from the feedback controller. The speed control signal is obtained by the feedback controller calculating the speed compensation amount based on the workstation operating parameters.

6. The method according to claim 1, characterized in that, During the process of adjusting the tension between adjacent workstations, it is determined whether the copper foil exhibits serpentine movement. If serpentine movement occurs, the copper foil serpentine offset data is fed back to the feedforward compensation function and the feedback controller to coordinate the tension control adjustment, including: A photoelectric sensor array is used to acquire the position information of the copper foil edge, and the snake offset data of the copper foil centerline relative to the baseline is calculated by the position detection unit. Based on the snake offset data, feature parameters are extracted through a deep neural network, and snake correction parameters are obtained from the historical database as feedforward compensation. For the snake offset data, the feedback controller calculates a dynamic compensation signal as the feedback compensation amount; The feedforward compensation amount and the feedback compensation amount are weighted and calculated by the fusion unit to obtain the station tension compensation command, and the tension control device adjusts the tension distribution of adjacent stations.

7. The method according to claim 1, characterized in that, The method also includes: based on the tension balance between adjacent workstations, collecting the operating parameters of each workstation in the continuous composite copper foil production line in real time, and establishing a mathematical description function for the production line operation.

8. The method according to claim 7, characterized in that, Based on the tension balance between adjacent workstations, the operating parameters of each workstation in the continuous composite copper foil production line are collected in real time to establish a mathematical description function for the production line operation, including: Acquire the output signals from the tension sensor, speed sensor, and temperature sensor, and output the raw parameter data from the data acquisition unit; Based on the original parameter data, a data preprocessing unit is used to perform normalization processing to obtain a standardized parameter dataset from the preprocessing unit. For the standardized parameter dataset, principal component analysis was used to reduce the dimensionality of the tension, copper foil speed and temperature parameters, and the parameter correlation coefficient matrix was obtained from the correlation analysis unit. A neural network is used to perform nonlinear mapping on the operating condition characteristic parameters, and the state prediction value is obtained from the prediction unit and output from the prediction unit. Establish a mathematical description function for the production line operation based on the predicted state values.

9. The method according to claim 1, characterized in that, The method also includes optimizing the control parameters of the production line operation by combining preset rules in a preset knowledge base.

10. The method according to claim 9, characterized in that, The optimization of production line operation control parameters by combining preset rules from a preset knowledge base includes: The rule matcher parses the preset rules in the preset knowledge base and extracts parameters to adjust the rules. A fuzzy inference engine is used to quantify the parameter tuning rules, and the parameter adjustment direction is obtained by calculating the membership degree. Based on the parameter adjustment direction, a deep reinforcement learning method is used to construct the action space, and the tension deviation sequence and control parameters are combined to form the state space. The optimized control parameter set is output from the reinforcement learning unit.

Images