A closed-loop control method and system for improving the compaction density of LFP electrodes.

Abstract

This invention relates to a closed-loop control method and system for improving the compaction density of LFP electrodes. The method involves real-time acquisition of images of the micro-particle packing of the coating slurry, synchronized with the rolling mill operation data; extraction of micro-pore distribution features from the images, establishing a mapping relationship between these features and the material's fractional-order constitutive parameters; combining real-time rolling pressure and roll gap data, obtaining the equivalent fractional-order derivative order and rheological viscosity of the electrode material through rapid inversion calculation; inputting the viscoelastic properties into a physically constrained recurrent neural network to predict the full-cycle thickness evolution curve of the electrode, including elastic rebound, creep, and relaxation processes, and calculating the predicted compaction density based on the steady-state value of the curve; constructing a planning model with the goal of minimizing the residual between the predicted density and the set target, and jointly optimizing the roll gap position compensation and hot roller heating power online to achieve closed-loop control of the compaction density.

Description

Technical Field

[0001] This application belongs to the field of control, and in particular relates to a closed-loop control method and system for the preparation process of improving the compaction density of LFP electrodes. Background Technology

[0002] In the manufacturing process of lithium-ion batteries, electrode rolling uses mechanical compression to compact the coated, loose electrode sheets to a predetermined thickness, thereby increasing the compaction density. The LFP electrode rolling control system employs an "online thickness detection + roll gap feedback adjustment" mode. This involves real-time monitoring of the electrode thickness using a laser or X-ray thickness gauge installed at the mill exit. If the monitored value deviates from the set target, a servo motor is driven by a PID algorithm or simple logic control to adjust the roll gap. However, in the precise control of LFP material properties, LFP materials have high hardness and particle morphology, and the electrodes exhibit viscoelastic characteristics. After rolling, the electrodes not only experience instantaneous elastic rebound but also delayed creep rebound based on viscous flow during resting. The control method only provides feedback based on the online thickness immediately after exiting the roll, failing to obtain information on thickness changes, resulting in poor consistency in compaction density. The rolling process is a multivariate strongly correlated process involving pressure, roll gap, speed, temperature and material constitutive properties. Linear feedback makes it difficult to achieve coordinated optimization of heating power and mechanical compensation, and cannot predict and compensate for springback caused by microscopic differences in materials. Therefore, it is difficult to ensure the stability of electrode quality under extreme compaction processes. Summary of the Invention

[0003] This invention proposes a closed-loop control method for the preparation process of LFP electrode sheets to improve compaction density, comprising:

[0004] Real-time acquisition of particle accumulation texture images on the coated surface of LFP electrode sheets, simultaneous acquisition of real-time rolling pressure, roll gap and running speed data of the roller press, and spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance;

[0005] The micropore distribution features are obtained by extracting features from the particle packing texture image. The mapping relationship between the micropore distribution features and the initial modulus parameters in the fractional viscoelastic constitutive equation is constructed. Combined with real-time rolling pressure and roll gap data, the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material are calculated by rapid numerical inversion.

[0006] The equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained from the inversion are input into a recurrent neural network model embedded with viscoelastic physical constraints. Time-series deduction is performed to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound, and stress relaxation processes. Based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated.

[0007] A planning model is constructed with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. The position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll are jointly optimized and solved. The solution is then sent to the control unit to perform closed-loop operation.

[0008] In another aspect, the present invention also proposes a closed-loop control system for the preparation process of improving the compaction density of LFP electrodes, comprising the following modules:

[0009] The acquisition module is used to acquire real-time images of particle accumulation texture on the coated surface of LFP electrode sheets, and simultaneously acquire real-time rolling pressure, roll gap and running speed data of the roller press, and perform spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance.

[0010] The first calculation module is used to extract features from the particle accumulation texture image to obtain micropore distribution features, construct the mapping relationship between micropore distribution features and the initial modulus parameter in the fractional viscoelastic constitutive equation, and combine real-time rolling pressure and roll gap data to calculate the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material through rapid numerical inversion.

[0011] The second calculation module is used to input the equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained by inversion into a recurrent neural network model embedded with viscoelastic physical constraints, and perform time-series deduction to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound and stress relaxation processes; based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated.

[0012] The solution module is used to construct a planning model with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. It performs joint optimization on the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll, and sends the solution results to the control unit to perform closed-loop operation.

[0013] This invention integrates electrode surface texture images with equipment operating parameters, establishing a correspondence between multi-source data through spatiotemporal alignment, and constructing an intrinsic logical connection between microscopic morphology and macroscopic mechanical response. By utilizing microscopic pore distribution characteristics combined with fractional-order viscoelastic constitutive equations, the rheological properties of the material can be analyzed, and the equivalent fractional derivative order and viscosity characteristics can be inverted. Using a recurrent neural network model embedded with physical constraints, the full-cycle thickness evolution process of the electrode, including instantaneous elastic rebound, delayed creep, and stress relaxation, is simulated, obtaining highly reliable steady-state convergent thickness and compaction density predictions. Through a planning model, the roll gap compensation and hot roller heating power are collaboratively optimized and closed-loop adjusted, eliminating compaction density deviations caused by the rebound characteristics of lithium iron phosphate electrodes, improving the thickness consistency and compaction density compliance rate of electrode preparation, and achieving production quality control. Attached Figure Description

[0014] Figure 1 A flowchart of the first embodiment;

[0015] Figure 2 This is a schematic diagram illustrating the time-series evolution prediction of electrode thickness.

[0016] Figure 3 This is a schematic diagram of multivariate joint optimization. Detailed Implementation

[0017] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0018] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0019] Figure 1 For a flowchart of the first embodiment, see [link / reference]. Figure 1 The closed-loop control method shown for the preparation process of improving the compaction density of LFP electrodes includes:

[0020] S1, real-time acquisition of particle accumulation texture images on the coated surface of LFP electrode sheets, simultaneous acquisition of real-time rolling pressure, roll gap and running speed data of the roller press, and spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance;

[0021] A high-resolution camera is installed at the inlet of the roller press, using a high-frequency LED line light source to continuously acquire grayscale images of the electrode surface at a set frame rate. Simultaneously, a hydraulic pressure sensor mounted on the roller press frame reads the real-time pressure value, a laser displacement sensor reads the real-time roll gap value of the upper and lower rollers, and a rotary encoder reads the real-time linear velocity of the roller press. The real-time surface temperature of the hot roller is also acquired. Two first-in-first-out (FIFO) data queues are established in the central processing unit to store image data and roller press sensor data, respectively. The distance between the camera's center line and the roller press center line is measured, and the time delay is calculated by dividing the distance by the real-time linear velocity. The timestamps in the image data queue are corrected based on this time delay. Data frames with corrected timestamps that match the timestamps of the roller press sensor data are matched and merged to obtain a time-synchronized dataset containing texture images, pressure, gap, and velocity.

[0022] In an optional embodiment, the real-time acquisition of the particle accumulation texture image of the LFP electrode coating surface includes:

[0023] The linear CCD camera located at the entrance of the roller press is synchronized with the encoder of the roller press spindle. Under the condition that the LFP electrode sheet moves at a speed of 30-80m / min, a high-brightness line-scanning LED light source is used to illuminate the surface of the electrode sheet with a fixed light intensity to acquire a grayscale grain texture image with a resolution of 4096 pixels in width.

[0024] A rotary encoder with a resolution of at least 2000 pulses per revolution is rigidly connected to the main shaft of the roller press. The A and B phase pulse signals of the encoder are connected to the external trigger input of the line scan camera. Based on the electrode's moving speed range of 30m / min to 80m / min, the camera's line frequency is set to match the movement speed. For example, at a speed of 50m / min, the line frequency is set to approximately 15kHz to ensure consistent vertical and horizontal resolution and avoid image stretching or compression. For the light source, a high-brightness white LED line light source with a color temperature of 6000K to 6500K is used, with the installation angle set at a 30° to 45° angle to the normal to enhance the contrast of texture and shadows on the particle surface. During image acquisition, a 4K or 8K resolution line scan CCD camera is preferred, with line-by-line exposure triggered by encoder pulses. Image data is transmitted to the industrial control computer's video memory via a CameraLink or CoaXPress high-speed interface. The continuously acquired line data is stitched into frames, with each frame image size set to 4096 pixels × 1024 pixels or 4096 pixels × 2048 pixels. To eliminate ambient light and sensor dark current noise, black and white field correction is required before formal data acquisition: acquire a dark background image with the light source off as the black frame, and acquire an image with approximately 80% saturation under a standard white board as the white frame. Perform flat field correction calculations on the acquired LFP electrode grayscale image in real time to output a high signal-to-noise ratio grain accumulation texture image.

[0025] S2, the micropore distribution features are obtained by extracting features from the particle accumulation texture image, and the mapping relationship between the micropore distribution features and the initial modulus parameters in the fractional viscoelastic constitutive equation is constructed. Combined with real-time rolling pressure and roll gap data, the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material are calculated by rapid numerical inversion.

[0026] The total resistance generated when the electrode is rolled is determined by the elastic rebound force of its solid skeleton and the viscous friction force of the internal material flow. The gap between the rollers represents the degree of compression of the electrode, while the distribution of micropores determines the initial hardness of the electrode skeleton. According to the equilibrium relationship of the fractional viscoelastic constitutive equation, the degree of deformation and initial hardness of the electrode have been determined, and the actual total resistance generated has been measured in real time by the sensor. The mechanical response in the total resistance that deviates from the pure elastic calculation result is contributed by the proportion of viscoelastic properties and the flow friction resistance inside the electrode. Therefore, by back-calculating with the actual measured pressure as the target, the unknown physical parameters representing these two characteristics can be solved from the constitutive equation, namely the equivalent fractional derivative order and the equivalent rheological viscosity characteristics.

[0027] The gray-level co-occurrence matrix algorithm is used to calculate four statistical quantities of the texture image: energy, contrast, correlation, and inverse difference moment, which serve as feature vectors representing the micropore distribution. Beforehand, offline tensile and compression experiments are conducted to determine the stress-strain curves of electrode samples with different porosities at a reference temperature. The initial elastic modulus corresponding to different feature vectors is obtained by fitting the curves using the least squares method, and a multinomial regression model is established as the mapping relationship. Preferably, sklearn.linear_model.Ridge or sklearn.neural_network.MLPRegressor is used to establish the mapping relationship. During the online computation phase, the real-time extracted feature vectors are input into the regression model to obtain the initial modulus parameters at the reference temperature.

[0028] In one embodiment, real-time surface temperature data of the hot roller is also acquired simultaneously, and the WLF empirical equation is introduced based on the time-temperature equivalence principle of polymer viscoelasticity to calculate the relaxation time displacement factor of the current surface temperature relative to the reference temperature. The displacement factor is used to perform nonlinear temperature correction on the initial modulus parameter at the reference temperature to obtain the actual equivalent initial modulus of the electrode at the current hot roller temperature.

[0029] A fractional Zener model was selected as the viscoelastic constitutive equation. The actual compressive strain was calculated using the initial thickness of the electrode before entering the roll and the real-time roll gap as the strain input. The corrected actual equivalent initial modulus was substituted into the constitutive equation to calculate the theoretical stress. An optimization problem was constructed with the goal of minimizing the mean square error between the theoretical stress calculated by the model and the actual measured roll pressure. The Levenberg-Marquardt algorithm was used for iterative solution until the error converged. The fractional derivative order and the sticky pot coefficient at this point were output as the current equivalent rheological viscosity characteristics. The equivalent fractional derivative order is mainly used to measure the viscoelasticity of LFP electrode material during the rolling process. The value of the fractional derivative order is usually between 0 and 1. If the calculated order is close to 0, it is biased towards a pure elastic body, indicating that the electrode material is like a spring, and its thickness will rebound rapidly and significantly after being flattened by the roller and exiting the roller. If the order is close to 1, it is biased towards a pure viscous fluid, indicating that the electrode material is like clay, and the deformation after being rolled will be permanently retained, and almost no thickness rebound will occur after exiting the roller. The equivalent rheological viscosity characteristic refers to the magnitude of the internal frictional resistance exhibited by the particles and binder inside the LFP electrode when they slip relative to each other under stress. It determines the speed at which the electrode undergoes delayed creep rebound and stress relaxation after exiting the roll. If the equivalent rheological viscosity characteristic value of the electrode is large, the internal deformation resistance of the material is large, and the delayed rebound action of the electrode after exiting the roll will be very slow, which may require 24 hours or even longer to reach the final stable thickness. Conversely, if the viscosity characteristic value is small, the internal stress is released quickly, and the electrode can quickly complete the creep process and be fixed at the final compacted thickness in a very short time after exiting the roll.

[0030] In an optional embodiment, the step of extracting features from the particle-stacking texture image to obtain micropore distribution features includes:

[0031] For each frame of grayscale grain texture image acquired, calculate the grayscale co-occurrence matrix in four directions: 0°, 45°, 90°, and 135°.

[0032] Texture parameters of four dimensions—contrast, correlation, energy, and inverse difference moment—are extracted from the gray-level co-occurrence matrices of the four directions, respectively.

[0033] Principal component analysis was used to reduce the dimensionality of the texture parameters in the four dimensions mentioned above, and the principal component vectors with a cumulative contribution rate of more than 95% were selected as the micropore distribution features.

[0034] Gray-level compression is performed on the preprocessed grayscale image, representing the original 256 gray levels as 64 or 32 levels, reducing computational load and suppressing noise. A pixel distance d=1 is defined as adjacent pixels. The frequency of gray-level pairs is statistically analyzed in four directions: 0° horizontal, 45° right diagonal, 90° vertical, and 135° left diagonal, constructing four N×N gray-level co-occurrence matrices, where N is the number of gray levels after representation. For each matrix, according to the Haralick texture feature definition, the contrast reflecting the depth of texture grooves, the correlation reflecting the linear relationship of local gray levels, the energy reflecting texture uniformity, and the inverse difference moment reflecting the regularity of texture are calculated, resulting in a total of 16-dimensional original feature vectors. In the data dimensionality reduction stage, the 16 feature parameters are Z-score standardized. A covariance matrix is ​​constructed, and eigenvalues ​​and eigenvectors are calculated, sorted in descending order of eigenvalues. The cumulative contribution rate of the top k eigenvalues ​​is calculated, and when the cumulative contribution rate... The first k eigenvectors are extracted to form a transformation matrix. The original 16-dimensional eigenvectors are projected onto this transformation matrix to obtain principal component vectors, typically 2 to 4 dimensions. These principal component vectors encapsulate the porosity distribution information on the LFP electrode surface.

[0035] In an optional embodiment, the step of calculating the current equivalent fractional derivative order and equivalent rheological viscosity characteristics of the electrode material through rapid numerical inversion includes:

[0036] An objective function based on a fractional Zener mechanical model is constructed, wherein the objective function is the mean square error between the theoretical stress and the real-time collected roller pressure.

[0037] The modulus parameter obtained by mapping the micropore distribution characteristics of the image is set as the initial value of the iteration. The fractional derivative order and rheological viscosity parameter are solved by fast gradient descent within a preset sliding time window using the Levenberg-Marquardt algorithm or the trust region reflection algorithm.

[0038] When the gradient magnitude of the objective function is less than the preset convergence threshold or the maximum number of iterations is reached, the current parameter solution is output as the current equivalent fractional derivative order and equivalent rheological viscosity characteristics.

[0039] objective function ,in For pressure sensor data, The theoretical stress is calculated for the fractional-order model, where N is the number of sampling points within the sliding time window. For example, if the window length is set to 100 ms and the sampling frequency is 1 kHz, then N = 100. The initial elastic modulus in the model... Instead of being treated as an unknown variable, it is predicted through a pre-trained regression model from image texture features to modulus, such as support vector regression, and serves as prior knowledge for inversion. Rheological viscosity is taken into account. Sensitivity to temperature; the inversion process utilizes real-time roll surface temperature. As a correction factor or state label, ensure that the inversion results The values ​​reflect the material properties under the current thermodynamic state. The Levenberg-Marquardt least-squares iteration is used in the solution process. The fractional order of the parameters to be inverted is set. The search range is 0 to 1, rheological viscosity The search scope is to Initial damping factor Set it to 0.01. If a certain iteration causes the objective function to decrease, then decrease it. Approaching Newton's method; conversely increasing Approximate gradient descent. The convergence condition is set as follows: the change in the objective function value is less than... or gradient magnitude less than The process can be repeated up to a maximum of 50 iterations. This process is handled in parallel using multi-threading in the control system, ensuring that the material's rheological parameters are updated every certain time interval, such as 200ms.

[0040] S3, the equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained by inversion are input into a recurrent neural network model embedded with viscoelastic physical constraints, and time-series deduction is performed to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound and stress relaxation processes; based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated.

[0041] Specifically, a Long Short-Term Memory (LSTM) network is constructed as the basic architecture. The network's input layer receives the equivalent fractional derivative order, equivalent rheological viscosity characteristics, initial roll gap and pressure data, and real-time roll surface temperature at the current moment. A physical constraint term is added to the LSTM's loss function, which is the residual between the thickness change rate output by the network and the theoretical change rate calculated by the fractional constitutive equation. During model training and derivation, the fractional derivative is numerically discretized using the Grünwald-Letnikov definition. The inverted material parameters are input into the trained model, and the electrode thickness values ​​at a series of future time points are recursively output according to a set time step. The thickness values, connected in chronological order, constitute a full-cycle thickness evolution prediction curve representing the electrode's elastic recovery at the moment of roll exit, its long-term creep, and stress relaxation to stability. Figure 2 .

[0042] Numerical differentiation is performed on the generated full-cycle thickness evolution prediction curve to calculate the thickness change rate of adjacent time steps. When the absolute value of the thickness change rate of multiple consecutive time steps is less than the preset small threshold, it is determined that the electrode thickness has reached a steady state. The thickness value corresponding to this moment is extracted as the steady-state convergence thickness. The preset areal density parameter of the electrode is obtained, and the areal density is divided by the steady-state convergence thickness to calculate the predicted compaction density of the electrode after undergoing a complete springback process.

[0043] In an optional embodiment, the recurrent neural network model embedded with viscoelastic physical constraints includes:

[0044] It adopts a Long Short-Term Memory (LSTM) network architecture, which includes an input layer, two stacked LSTM hidden layers, and a fully connected output layer;

[0045] A physical regularization term is added to the loss function of the Long Short-Term Memory network. The physical regularization term is composed of the square of the residual value of the constitutive differential equation of the fractional Zener rheological model at the current time, and is used to penalize the network prediction output that does not conform to the physical laws of viscoelasticity.

[0046] The recurrent neural network model consists of an input layer, hidden layers, and an output layer. The hidden layers use double-layered stacked long short-term memory network units. The first layer preferably has 128 neurons, and the second layer preferably has 64 neurons. A dropout operation with a utilization ratio of 0.2 between layers prevents overfitting. The output layer is a fully connected layer. The total loss function... Mean square error term With physical regularization term The weighted sum, where the physical regularization term is derived from the constitutive differential equation of the fractional Zener rheological model. The model is constructed by squared residuals at the current time. The input to the model is the aforementioned multidimensional feature sequence within a time window, and the output is a predicted curve of electrode thickness variation over a future preset time period. This allows the neural network to be constrained within a solution space that conforms to the laws of polymer rheology when generating the full-cycle thickness evolution curve of the electrode, thus improving the reliability of the thickness rebound prediction over long periods, such as 24 hours, after the electrode exits the roll.

[0047] In an optional embodiment, calculating the predicted compaction density of the electrode after it has been left to stand after exiting the roll, based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, includes:

[0048] The thickness value corresponding to 24 hours after exiting the roll is extracted from the full-cycle thickness evolution prediction curve of the electrode sheet and used as the steady-state convergence value.

[0049] The mass of the dressing per unit area of ​​the LFP electrode is measured, and the mass of the dressing is divided by the steady-state convergence value to obtain the predicted compaction density.

[0050] Because LFP electrodes are viscoelastic materials, their thickness rebound exhibits creep characteristics, typically stabilizing within 24 hours after rolling. Therefore, on the time axis of the model output, the location is... The time index h is used to read the predicted thickness value at that moment. The unit is μm. To eliminate the influence of the current collector thickness, actual calculations should use... ,in The nominal thickness of the aluminum foil current collector is, for example, 13μm or 15μm. For quality data acquisition, online areal density detection data uploaded during the coating process or offline stamping weighing data is used to obtain the coating mass per unit area of ​​the current roll of LFP electrodes. The unit is or Predicting compaction density The unit is The calculation formula is: The coefficient is used for unit conversion. This calculation process is performed in real time in the control system. Whenever a new steady-state thickness value is predicted, the predicted compaction density is updated once in combination with the corresponding areal density data, thereby reflecting the potential deviation in compaction density caused by uneven electrode coating thickness or fluctuations in material properties.

[0051] S4. Construct a planning model with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. Perform joint optimization on the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll, and send the solution results to the control unit to perform closed-loop operation.

[0052] The objective function is the square of the difference between the predicted compaction density and the standard compaction density required by the process. Constraints are set, including the maximum mechanical stroke limit of the roll gap adjustment mechanism, the response speed limit of the servo motor, and the maximum allowable temperature and power limit of the electromagnetic heating roll. A sequential quadratic programming algorithm (SQP) is used to optimize the objective function, with the roll gap position compensation and roll surface temperature as optimization variables. Within the constraints, the optimal combination of solutions that minimizes the objective function is searched, such as... Figure 3 Specifically, scipy.optimize.minimize is used, with method='SLSQP' specified for optimization. The calculated optimal roll gap position compensation is converted into a pulse control signal for the servo motor and sent to the hydraulic servo system to adjust the roll gap; the optimal roll surface temperature is converted into a pulse width modulation (PWM) signal and sent to the electromagnetic induction heating controller to adjust the roll surface temperature; the control unit executes the above steps cyclically with a control cycle of hundreds of milliseconds to correct deviations in real time.

[0053] In an optional embodiment, the joint optimization solution for the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll includes:

[0054] An optimization equation was established with the roll gap position compensation and heating power adjustment as decision variables. The roll gap adjustment range was constrained to ±50μm, and the heating roll surface temperature change rate was constrained to ±5℃ / min.

[0055] The optimization equation is solved iteratively using a sequential quadratic programming algorithm.

[0056] The optimal control solution is the combination of decision variables that minimizes the absolute value of the difference between the predicted compaction density and the standard compaction density under the given constraints.

[0057] The objective function of the optimization problem is The decision variable vector These represent the roll gap compensation and temperature adjustment, respectively. Constraints include hard physical constraints and process safety constraints: roll gap adjustment. Prevent mechanical impact or crushing; temperature regulation. Thermal inertia constraints must be met, meaning the adjustment amount is converted into a rate-of-change constraint on the heating power P, ensuring the roll surface temperature change rate does not exceed ±5℃ / min to prevent temperature overshoot from causing electrode sticking or strip breakage. A sequential quadratic programming algorithm is used for the solution because it has convergence in handling constrained optimization problems. Within each control cycle (e.g., 5 seconds), a second-order Taylor expansion of the objective function is performed using the current operating point, approximating the programming subproblem as a quadratic programming problem. The optimal step size is iteratively searched using a quasi-Newton method with the Hessian matrix. After typically 5 to 10 iterations, the optimal roll gap compensation value is obtained. Such as +2.5μm and optimal roller surface temperature adjustment value For example, +0.5℃, this combination will ensure that the electrode compaction density is closest to the process setting value after 24 hours. For example, 2.4 The best control strategy.

[0058] In an optional embodiment, the step of sending the solution result to the control unit to perform the closed-loop operation includes:

[0059] The position compensation value obtained by solving is sent to the servo hydraulic control system via the EtherCAT industrial real-time Ethernet bus, driving the hydraulic cylinder to adjust the roll gap with an adjustment accuracy of 1μm.

[0060] The heating power adjustment obtained from the solution is simultaneously sent to the electromagnetic heating controller to adjust the duty cycle of the induction coil and change the working temperature of the roller surface.

[0061] The main control unit uses a PC-based controller such as Beckhoff TwinCAT or Keyence, communicating with each actuator via an EtherCAT bus with a communication cycle set to 1ms. For position control, the calculated optimal roll gap target value is mapped to the PDO (Process Data Object) of the servo hydraulic system. After receiving the command, the servo valve controls the high-frequency response hydraulic cylinder to move, and combines this with feedback from the grating ruler to achieve fully closed-loop position control, ensuring that the actual roll gap adjustment accuracy reaches ±1μm and the response time is less than 50ms. For temperature control, the roll surface temperature adjustment is converted into the duty cycle of a PWM (Pulse Width Modulation) signal or a 4-20mA analog signal, and sent to the electromagnetic induction heating controller. The controller adjusts the high-frequency current output of the IGBT inverter accordingly, changing the magnetic field strength of the induction coil, thereby regulating the eddy current heating power on the hot roll surface. Simultaneously, feedback data from the roll surface infrared thermometer is monitored, and the output power is fine-tuned through a PID algorithm to achieve a smooth approach to the target temperature. The dual-variable collaborative control strategy utilizes the rapidity of hydraulic adjustment to compensate for high-frequency thickness fluctuations, and the characteristics of temperature regulation to change the material's rheological properties to compensate for low-frequency rebound trends, thus achieving closed-loop control of compaction density.

[0062] In the second embodiment, the present invention also proposes a closed-loop control system for the preparation process of improving the compaction density of LFP electrodes, comprising the following modules:

[0063] The acquisition module is used to acquire real-time images of particle accumulation texture on the coated surface of LFP electrode sheets, and simultaneously acquire real-time rolling pressure, roll gap and running speed data of the roller press, and perform spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance.

[0064] The first calculation module is used to extract features from the particle accumulation texture image to obtain micropore distribution features, construct the mapping relationship between micropore distribution features and the initial modulus parameter in the fractional viscoelastic constitutive equation, and combine real-time rolling pressure and roll gap data to calculate the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material through rapid numerical inversion.

[0065] The second calculation module is used to input the equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained by inversion into a recurrent neural network model embedded with viscoelastic physical constraints, and perform time-series deduction to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound and stress relaxation processes; based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated.

[0066] The solution module is used to construct a planning model with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. It performs joint optimization on the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll, and sends the solution results to the control unit to perform closed-loop operation.

[0067] In an optional embodiment, the real-time acquisition of the particle accumulation texture image of the LFP electrode coating surface includes:

[0068] The linear CCD camera located at the entrance of the roller press is synchronized with the encoder of the roller press spindle. Under the condition that the LFP electrode sheet moves at a speed of 30-80m / min, a high-brightness line-scanning LED light source is used to illuminate the surface of the electrode sheet with a fixed light intensity to acquire a grayscale grain texture image with a resolution of 4096 pixels in width.

[0069] In an optional embodiment, the step of extracting features from the particle-stacking texture image to obtain micropore distribution features includes:

[0070] For each frame of grayscale grain texture image acquired, calculate the grayscale co-occurrence matrix in four directions: 0°, 45°, 90°, and 135°.

[0071] Texture parameters of four dimensions—contrast, correlation, energy, and inverse difference moment—are extracted from the gray-level co-occurrence matrices of the four directions, respectively.

[0072] Principal component analysis was used to reduce the dimensionality of the texture parameters in the four dimensions mentioned above, and the principal component vectors with a cumulative contribution rate of more than 95% were selected as the micropore distribution features.

[0073] In an optional embodiment, the step of calculating the current equivalent fractional derivative order and equivalent rheological viscosity characteristics of the electrode material through rapid numerical inversion includes:

[0074] An objective function based on a fractional Zener mechanical model is constructed, wherein the objective function is the mean square error between the theoretical stress and the real-time collected roller pressure.

[0075] The modulus parameter obtained by mapping the micropore distribution characteristics of the image is set as the initial value of the iteration. The fractional derivative order and rheological viscosity parameter are solved by fast gradient descent within a preset sliding time window using the Levenberg-Marquardt algorithm or the trust region reflection algorithm.

[0076] When the gradient magnitude of the objective function is less than the preset convergence threshold or the maximum number of iterations is reached, the current parameter solution is output as the current equivalent fractional derivative order and equivalent rheological viscosity characteristics.

[0077] In an optional embodiment, the recurrent neural network model embedded with viscoelastic physical constraints includes:

[0078] It adopts a Long Short-Term Memory (LSTM) network architecture, which includes an input layer, two stacked LSTM hidden layers, and a fully connected output layer;

[0079] A physical regularization term is added to the loss function of the Long Short-Term Memory network. The physical regularization term is composed of the square of the residual value of the constitutive differential equation of the fractional Zener rheological model at the current time, and is used to penalize the network prediction output that does not conform to the physical laws of viscoelasticity.

[0080] In an optional embodiment, calculating the predicted compaction density of the electrode after it has been left to stand after exiting the roll, based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, includes:

[0081] The thickness value corresponding to 24 hours after exiting the roll is extracted from the full-cycle thickness evolution prediction curve of the electrode sheet and used as the steady-state convergence value.

[0082] The mass of the dressing per unit area of ​​the LFP electrode is measured, and the mass of the dressing is divided by the steady-state convergence value to obtain the predicted compaction density.

[0083] In an optional embodiment, the joint optimization solution for the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll includes:

[0084] An optimization equation was established with the roll gap position compensation and heating power adjustment as decision variables. The roll gap adjustment range was constrained to ±50μm, and the heating roll surface temperature change rate was constrained to ±5℃ / min.

[0085] The optimization equation is solved iteratively using a sequential quadratic programming algorithm.

[0086] The optimal control solution is the combination of decision variables that minimizes the absolute value of the difference between the predicted compaction density and the standard compaction density under the given constraints.

[0087] In an optional embodiment, the step of sending the solution result to the control unit to perform the closed-loop operation includes:

[0088] The position compensation value obtained by solving is sent to the servo hydraulic control system via the EtherCAT industrial real-time Ethernet bus, driving the hydraulic cylinder to adjust the roll gap with an adjustment accuracy of 1μm.

[0089] The heating power adjustment obtained from the solution is simultaneously sent to the electromagnetic heating controller to adjust the duty cycle of the induction coil and change the working temperature of the roller surface.

[0090] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A closed-loop control method for the preparation process of LFP electrode sheets to improve compaction density, characterized in that, include: Real-time acquisition of particle accumulation texture images on the coated surface of LFP electrode sheets, simultaneous acquisition of real-time rolling pressure, roll gap and running speed data of the roller press, and spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance; The micropore distribution features are obtained by extracting features from the particle packing texture image. The mapping relationship between the micropore distribution features and the initial modulus parameters in the fractional viscoelastic constitutive equation is constructed. Combined with real-time rolling pressure and roll gap data, the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material are calculated by rapid numerical inversion. The current equivalent fractional derivative order and equivalent rheological viscosity characteristics of the electrode material are calculated by rapid numerical inversion, including: constructing an objective function based on a fractional Zener mechanical model, wherein the objective function is the mean square error between the calculated theoretical stress and the real-time collected rolling pressure; setting the modulus parameter obtained by mapping the microscopic pore distribution characteristics of the image as the initial value of the iteration, and using the Levenberg-Marquardt algorithm or the trust region reflection algorithm to solve the fractional derivative order and rheological viscosity parameters by rapid gradient descent within a preset sliding time window; when the gradient magnitude of the objective function is less than the preset convergence threshold or the maximum number of iterations is reached, the current parameter solution is output as the current equivalent fractional derivative order and equivalent rheological viscosity characteristics; The equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained from the inversion are input into a recurrent neural network model embedded with viscoelastic physical constraints. Time-series deduction is performed to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound, and stress relaxation processes. Based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated. A planning model is constructed with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. The position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll are jointly optimized and solved. The solution is then sent to the control unit to perform closed-loop operation.

2. The method according to claim 1, characterized in that, The real-time acquisition of particle accumulation texture images on the LFP electrode coating surface includes: The linear CCD camera located at the entrance of the roller press is synchronized with the encoder of the roller press spindle. Under the condition that the LFP electrode sheet moves at a speed of 30-80m / min, a high-brightness line-scanning LED light source is used to illuminate the surface of the electrode sheet with a fixed light intensity to acquire a grayscale grain texture image with a resolution of 4096 pixels in width.

3. The method according to claim 1, characterized in that, The step of extracting features from the particle accumulation texture image to obtain micropore distribution features includes: For each frame of grayscale grain texture image acquired, calculate the grayscale co-occurrence matrix in four directions: 0°, 45°, 90°, and 135°. Texture parameters of four dimensions—contrast, correlation, energy, and inverse difference moment—are extracted from the gray-level co-occurrence matrices of the four directions, respectively. Principal component analysis was used to reduce the dimensionality of the texture parameters in the four dimensions mentioned above, and the principal component vectors with a cumulative contribution rate of more than 95% were selected as the micropore distribution features.

4. The method according to claim 1, characterized in that, The recurrent neural network model embedded with viscoelastic physical constraints includes: It adopts a Long Short-Term Memory (LSTM) network architecture, which includes an input layer, two stacked LSTM hidden layers, and a fully connected output layer; A physical regularization term is added to the loss function of the Long Short-Term Memory network. The physical regularization term is composed of the square of the residual value of the constitutive differential equation of the fractional Zener rheological model at the current time, and is used to penalize the network prediction output that does not conform to the physical laws of viscoelasticity.

5. The method according to claim 1, characterized in that, The calculation of the predicted compaction density of the electrode after it has been left to stand after exiting the roll, based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, includes: The thickness value corresponding to 24 hours after exiting the roll is extracted from the full-cycle thickness evolution prediction curve of the electrode sheet and used as the steady-state convergence value. The mass of the dressing per unit area of ​​the LFP electrode is measured, and the mass of the dressing is divided by the steady-state convergence value to obtain the predicted compaction density.

6. The method according to claim 1, characterized in that, The joint optimization solution for the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll includes: An optimization equation was established with the roll gap position compensation and heating power adjustment as decision variables. The roll gap adjustment range was constrained to ±50μm, and the heating roll surface temperature change rate was constrained to ±5℃ / min. The optimization equation is solved iteratively using a sequential quadratic programming algorithm. The optimal control solution is the combination of decision variables that minimizes the absolute value of the difference between the predicted compaction density and the standard compaction density under the given constraints.

7. The method according to claim 1, characterized in that, The step of sending the solution results to the control unit to perform closed-loop operation includes: The position compensation value obtained by solving is sent to the servo hydraulic control system via the EtherCAT industrial real-time Ethernet bus, driving the hydraulic cylinder to adjust the roll gap with an adjustment accuracy of 1μm. The heating power adjustment obtained from the solution is simultaneously sent to the electromagnetic heating controller to adjust the duty cycle of the induction coil and change the working temperature of the roller surface.

8. A closed-loop control system for the preparation process of LFP electrode sheets to improve compaction density, characterized in that, Includes the following modules: The acquisition module is used to acquire real-time images of particle accumulation texture on the coated surface of LFP electrode sheets, and simultaneously acquire real-time rolling pressure, roll gap and running speed data of the roller press, and perform spatiotemporal alignment of image data and roller press data based on running speed and sensor installation distance. The first calculation module is used to extract features from the particle accumulation texture image to obtain micropore distribution features, construct the mapping relationship between micropore distribution features and the initial modulus parameter in the fractional viscoelastic constitutive equation, and combine real-time rolling pressure and roll gap data to calculate the current equivalent fractional derivative order and equivalent rheological viscosity features of the electrode material through rapid numerical inversion. The first calculation module is further configured to construct an objective function based on a fractional Zener mechanical model, wherein the objective function is the mean square error between the theoretical stress and the real-time collected roller pressure; set the modulus parameter obtained by mapping the microscopic pore distribution characteristics of the image as the initial value of the iteration, and use the Levenberg-Marquardt algorithm or the trust region reflection algorithm to solve the fractional derivative order and rheological viscosity parameters by fast gradient descent within a preset sliding time window; when the gradient magnitude of the objective function is less than the preset convergence threshold or the maximum number of iterations is reached, output the current parameter solution as the current equivalent fractional derivative order and equivalent rheological viscosity characteristics; The second calculation module is used to input the equivalent fractional derivative order and equivalent rheological viscosity characteristics obtained by inversion into a recurrent neural network model embedded with viscoelastic physical constraints, and perform time-series deduction to generate a prediction curve of the electrode's full-cycle thickness evolution that includes instantaneous elastic rebound, delayed creep rebound and stress relaxation processes; based on the steady-state convergence value of the electrode's full-cycle thickness evolution prediction curve in the time domain, the predicted compaction density of the electrode after it has been left to stand after exiting the roller is calculated. The solution module is used to construct a planning model with the goal of minimizing the residual between the predicted compaction density and the standard compaction density setting value. It performs joint optimization on the position compensation amount of the roll gap adjustment mechanism and the heating power of the electromagnetic induction heating of the hot roll, and sends the solution results to the control unit to perform closed-loop operation.

9. The system according to claim 8, characterized in that, The real-time acquisition of particle accumulation texture images on the LFP electrode coating surface includes: The linear CCD camera located at the entrance of the roller press is synchronized with the encoder of the roller press spindle. Under the condition that the LFP electrode sheet moves at a speed of 30-80m / min, a high-brightness line-scanning LED light source is used to illuminate the surface of the electrode sheet with a fixed light intensity to acquire a grayscale grain texture image with a resolution of 4096 pixels in width.

Images