Feed-forward intelligent coating thickness closed-loop control method and system
By collecting multi-dimensional data during the coating process and building a neural network model, predicting the coating trend and adjusting parameters in advance, the problems of poor uniformity and low automation production efficiency in coating thickness control are solved, and more efficient coating is achieved Improved thickness uniformity and product quality.
Patent Information
- Application Number
- CN202410925957.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-05
AI Technical Summary
In the prior art, coating thickness control has problems such as poor uniformity and low automation production efficiency, which is mainly due to large manual adjustment errors and measurement feedback adjustment hysteresis, which affects the consistency and safety of the battery pole sheet.
Feedforward intelligent coating thickness closed-loop control method is adopted, and neural network models are constructed and trained by collecting multi-dimensional data, predicting the coating thickness trend and adjusting the coating parameters in advance, reducing the hysteresis of closed-loop control and improving the uniform coating thickness sex.
By predicting and adjusting the coating parameters in advance, the uniformity of the coating thickness and product quality are significantly improved, manual adjustment errors are reduced, and automated production needs are adapted.
Smart Images

Figure CN118892969A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of coating thickness control, and in particular to a feedforward intelligent coating thickness closed-loop control method and system. Background Art
[0002] The electrode plate is the core of lithium-ion batteries, and coating is an important process in the production of battery plates. During the coating process, the coating amount, that is, the coating thickness control directly affects the consistency and safety of the battery. The current mainstream production method is for operators to observe and detect, and manually adjust the coating amount, which will affect the timeliness of the coating amount adjustment and human misoperation, resulting in poor consistency of the electrode coating amount and material loss beyond the process.
[0003] During the coating process, due to the influence of factors such as slurry temperature or viscosity, the thickness of the coating is uneven, and the coating thickness needs to be adjusted. In the prior art, the manual adjustment structure is mainly performed by manually observing whether the data fluctuation detected by the coating thickness gauge deviates from the target value. The main adjustments include the slurry pump adjustment mechanism, the distance between the coating head lip and the substrate, and the size of the coating lip slit to perform manual overall or local adjustments. However, manual adjustment is affected by human factors and the adjustment error is large. It is easy to fail to adjust or adjust too much, resulting in uneven coating thickness and poor coating quality. In addition, manual operation adjustment leads to low adjustment efficiency, which is not conducive to automated production. Summary of the invention
[0004] The main purpose of the present invention is to solve the technical problems of uniform coating thickness and poor yield in the prior art. A feedforward intelligent coating thickness closed-loop control method comprises the following steps:
[0005] Collect data to train the model, including coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount;
[0006] Construct data time series features and multi-dimensional correlation features between data;
[0007] Build and train a neural network, input the above time series features and multi-dimensional correlation features between the data, output a neural network model for the classification of coating thickness trend adjustment direction and the adjustment amount of each actuator, and obtain a trained neural network model;
[0008] If the coating thickness does not meet expectations, calculate △P61, the calculation formula of △P61 is:
[0009]
[0010] Where, σ represents the surface tension of the slurry, H represents the coating height, θ is the static contact angle between the slurry and the substrate, φ is the static contact angle between the slurry and the coating head mold, l 1 is the upstream length of the coating bead, l 2 is the downstream length of the coating bead, Uc is the coating speed, W is the slit width of the coating head, h m is the predicted coating thickness, h represents the stable coating thickness, b represents the coating width, and μ represents the density of the slurry;
[0011] According to the calculation result of △P61, the coating speed, coating gap and lip flow rate are adjusted.
[0012] The second aspect of the present invention provides a feedforward intelligent coating thickness closed-loop control system, comprising:
[0013] A data acquisition unit is used to collect data for training the model, wherein the data includes coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount;
[0014] Feature construction unit, used to construct data time series features and multi-dimensional correlation features between data;
[0015] The model building and training unit is used to build and train the neural network, input the above time series features and the multi-dimensional correlation features between the data, and output the neural network model of the classification of the coating thickness trend adjustment direction and the adjustment amount of each actuator to obtain a trained neural network model;
[0016] The control execution unit is used to perform closed-loop control of coating thickness according to the trained neural network model.
[0017] The third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned feedforward intelligent coating thickness closed-loop control method.
[0018] A fourth aspect of the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned feedforward intelligent coating thickness closed-loop control method.
[0019] The present invention has the following beneficial effects:
[0020] The present invention predicts the coating thickness trend direction and the adjustment amount of each control mechanism, and inputs them to the corresponding adjustment mechanisms in advance, thereby reducing the hysteresis of closed-loop control and improving the uniformity of coating thickness.
[0021] In order to solve the hysteresis of measurement feedback adjustment, the present invention constructs a self-learning multi-feature fusion network model in advance according to the internal factors affecting the coating thickness and the external trend characteristics, predicts the subsequent trend changes of the coating weight thickness in advance, and then controls the slurry adjustment structure in advance, further improves the uniformity of the coating thickness, and improves product quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 A flow chart of a feedforward intelligent coating thickness closed-loop control method provided by an embodiment of the present invention;
[0023] Figure 2 This is an architecture diagram of the interactive feature extraction module of the present invention.
[0024] Figure 3 Schematic diagram of the initial flow field of slit coating.
[0025] Figure 4 The morphology of the upstream side of the coating beads in four critical cases. DETAILED DESCRIPTION
[0026] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0027] The coating machine is composed of multiple fully automatic die heads. In order to make the coating machine run better, the fully automatic die heads need to be adjusted according to the parameters of the coating machine. In existing methods, the coating thickness is usually measured at the wet film or dry film coating location based on laser, ultrasound, x-ray, beta ray, etc. This lagged measurement value is used as the feedback amount, and then the coating thickness is automatically adjusted according to the die slurry control mechanism to achieve closed-loop regulation. Although this method basically achieves automatic adjustment, it is based on readjustment after measurement feedback, and has a large hysteresis, which in turn affects the uniformity of the coating thickness and the improvement of the yield.
[0028] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , the first embodiment of the feedforward intelligent coating thickness closed-loop control method in the embodiment of the present invention includes:
[0029] The collected data is used to train the model, and the data includes coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount.
[0030] Specifically include:
[0031] Continuous data collection: Use sensors and measuring equipment to collect various parameters of the coating process in real time, including coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, mold cavity pressure, feeding pressure, slurry temperature, viscosity, etc.
[0032] Adjustment mechanism data: The adjustment label and adjustment amount of the adjustment mechanism are recorded at the same time. These data are crucial for the subsequent analysis of the relationship between the adjustment process and the coating thickness.
[0033] Construct data time series features and multi-dimensional correlation features between data;
[0034] The prediction of surface density thickness using the time series data of the surface density thickness gauge as input is based on the premise of "the fluctuation law implied by the time series data". Therefore, the main task of the model is to learn this fluctuation law. Manually constructing combined features is a time-consuming and laborious task. The present invention uses a neural network model based on a factor decomposition machine to enhance the degree of mining of input data and automatically learn the influence of the combined effect between different time series features on the trend of surface density thickness. Specifically including:
[0035] Time series characteristics:
[0036] Timestamp: Record the timestamp of each data point for subsequent analysis of data trends over time.
[0037] Time interval: Calculates the time interval between adjacent data points to analyze the frequency and rate of data change.
[0038] Time series statistics: Calculate statistical data such as average, maximum, minimum, standard deviation, etc. in each time period (such as every minute, every hour) to capture the dynamic changes of the coating process.
[0039] Multi-dimensional correlation features:
[0040] Correlation analysis: Calculate the correlation coefficients between different parameters to understand the linear relationship between them.
[0041] Lag and lead features: Considering the time delay effects between different parameters, lagging and leading features are constructed to capture the temporal dependencies between them.
[0042] Feature combination: Based on process knowledge and experience, different parameters are combined to create new features to capture the nonlinear relationship between them.
[0043] Build and train a neural network, input the above time series features and multi-dimensional correlation features between the data, output a neural network model for the classification of coating thickness trend adjustment direction and the adjustment amount of each actuator, and obtain a trained neural network model;
[0044] Model selection: Choose a suitable neural network structure according to the characteristics of the problem, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), or a combination of a convolutional neural network (CNN) and a fully connected layer (DNN) to process time series data and capture multi-dimensional correlation features.
[0045] Input layer design: The preprocessed time series features and correlation features are used as the input of the neural network to ensure that the format and dimension of the input data match the model structure.
[0046] Hidden layer design: Design the appropriate number of hidden layers and neurons, as well as activation functions and regularization strategies based on the complexity of the problem and the characteristics of the data.
[0047] Output layer design: The output layer should include neurons for classifying the coating thickness trend adjustment direction (such as using the softmax function for multi-classification) and neurons for adjusting the amount of each actuator (such as using a linear output layer for regression).
[0048] Model training: Use the collected data set to train the neural network, optimize the model parameters through the back propagation algorithm, and minimize the prediction error.
[0049] Model evaluation: Use the validation set or test set to evaluate the trained model to verify its generalization ability and predictive performance.
[0050] According to the trained neural network model, closed-loop control of coating thickness is performed.
[0051] Real-time data input: The real-time collected data is preprocessed and then input into the trained neural network model.
[0052] Prediction output: The model outputs the classification of coating thickness trend adjustment direction and the predicted value of the adjustment amount of each actuator based on the input data.
[0053] Actuator control: Based on the predicted output of the model, the actuator of the coating equipment is controlled to make corresponding adjustments, such as adjusting the pump speed, lip opening, etc., to achieve closed-loop control of coating thickness.
[0054] Feedback and adjustment: Continuously collect new data and monitor the actual value of coating thickness, compare the actual value with the predicted value, and adjust the model parameters or control strategy according to the error to optimize the control effect of coating thickness.
[0055] If the coating thickness does not meet expectations, calculate △P61, the calculation formula of △P61 is:
[0056]
[0057] Where, σ represents the surface tension of the slurry, H represents the coating height, θ is the static contact angle between the slurry and the substrate, φ is the static contact angle between the slurry and the coating head mold, l 1 is the upstream length of the coating bead, l 2 is the downstream length of the coating bead, Uc is the coating speed, W is the slit width of the coating head, h m is the predicted coating thickness, h represents the stable coating thickness, b represents the coating width, and μ represents the density of the slurry;
[0058] According to the calculation result of △P61, the coating speed, coating gap and lip flow rate are adjusted.
[0059] For example, the negative electrode slurry of the power lithium battery is transported to the mixing chamber through the inlet channel of the die head through the feeding system. After filling the mixing chamber, the slurry flows out evenly from the slit channel, generates relative flow between the outside of the die head and the substrate, and gradually stabilizes to form an outer flow field. When the slurry contacts the substrate, due to the relative speed between the coating head and the substrate, the slurry will stabilize after a short accumulation at the lip of the coating head, forming a half-moon shape. The half-moon-shaped liquid in the slit forms a liquid bridge due to the relative movement of the coating head and the substrate. The liquid part contained between the lip and the substrate is called a coating bead. The flow field diagram is shown as follows: Figure 3 After the coating is stable, the shape of the coated beads remains constant, and its flow process can be regarded as the steady flow of viscous incompressible fluid. The coated beads are analyzed and the calculation formula of △P61 is obtained.
[0060] like Figure 4 During the slit coating process, there are four types of morphology at the upstream of the coating bead. Each type corresponds to a critical pressure drop value. The morphology of each type is as follows: Figure 4-1 As shown. According to the comparison between the value of △P61 and the critical pressure drop value of the four cases, the shape and location of the upstream of the coating bead can be determined, so as to predict the stability of the coating in this case. The specific coating results corresponding to each case are as follows:
[0061] According to the calculation result of △P61, the coating speed, coating gap and lip flow rate are adjusted as follows:
[0062] When the value of △P61 is greater than 200,000:
[0063] At this time, adjust the coating speed first, then adjust the coating gap, and do not adjust the lip flow;
[0064] ②When the value of △P61 is 150,000-200,000:
[0065] At this time, adjust the coating speed first, then adjust the coating gap, if it does not meet expectations, fine-tune the lip flow;
[0066] ③When the value of △P61 is 100,000-150,000:
[0067] At this time, the coating gap, coating speed, and lip flow rate have equal weights and can be adjusted simultaneously without regard to order.
[0068] ④When the value of △P61 is 50,000-100,000:
[0069] At this time, adjust the lip flow rate first, then adjust the coating gap and coating speed;
[0070] ⑤When the value of △P61 is less than 50,000:
[0071] At this time, adjust the lip flow rate first, then adjust the coating gap, and finally adjust the coating speed.
[0072] As a preferred implementation, the construction of data time series features and multi-dimensional correlation features between data includes the following steps:
[0073] Based on dynamic time warping algorithm:
[0074] Input: Two columns of thickness gauge detection value time series X = {x 0 ,x 1 ,…x m-1} and Y={y 0 ,y 1 ,…y n-1}, their lengths are m and n respectively, where m may not be equal to n,
[0075] Output: The best matching path between time series X and Y is W = {w 0 ,w 1 …w k}, and the cumulative distance DTW(X,Y) of path W;
[0076] Specifically include:
[0077] Step 1: Construct an mxn cost matrix, where the matrix element D mn The matrix d i,j Represents x i and i ; Euclidean distance between two points Path W in w k represents a position in a matrix and satisfies the following three constraints:
[0078] (1) Boundary conditions, w 1 =D(0,0) and w k =D(m-1,n-1), the two sequences can be of different lengths, as long as the starting and ending positions are aligned;
[0079] (2) Continuity, if the path element w has been determined k-1 =D(a', b'), then for the next path point w k =D(a,b) needs to satisfy (a-a')≤1 and (b-b')≤1, so as to ensure that the two time series are continuous when matching, without skipping or missing any point;
[0080] (3) Monotonicity: If the path element w is determined, k-1 =D(a', b'), then for the next path point w k =D(a,b) needs to satisfy (a-a')≥0 and (b-b')≥0 to ensure that the points on the path are monotonically increasing on the time axis, that is, only the right, top and top-right of the current point can be selected, ensuring that the path will not return;
[0081] Step 2: Initialize the cumulative distance: γ(0,0) = d(x 0 ,y 0 )
[0082] Step 3: Calculate the minimum cumulative distance γ(m-1,n-1) according to the recursive formula:
[0083] γ(i,j)=d(x i ,y j )+min{γ(i-1,j-1),γ(i-1,j),γ(i,j-1)}
[0084] Step 4: Backtrack to find all the matching point pairs in path W: Based on the best local path of each step, backtrack from the matching point pair (m-1, n-1) to the point (0,0).
[0085] As a preferred implementation, the building and training of a neural network includes:
[0086] The factor decomposition model is used to combine the low-order features of multi-source time series data to obtain the feature vector F M ;
[0087] Use deep neural network to combine high-order features to get feature vector F DNN ,
[0088] The feature vector F M and F DNN Splice them together to get the spliced feature vector;
[0089] The concatenated feature vector is passed as input to the input layer of the neural network;
[0090] The cross entropy loss function is selected as the loss function, and the gradient descent optimization algorithm is used for model training.
[0091] Specifically: Figure 2 The present invention adopts an interactive feature extraction module. The present invention first uses a factor decomposition model to combine low-order (first-order and second-order) features of multi-source time series data to obtain F M , and then use the deep neural network to combine high-order features to get F DNN Finally, add the two parts of the results and activate them to get As the extraction result of the combined feature. As shown in the formula:
[0092]
[0093] The factor decomposition model is used to combine the low-order features of multi-source time series data to obtain the feature vector F M ,include:
[0094] F M The formula represents the calculation process of the factorization machine:
[0095]
[0096] In the formula, f i is the state vector T in the input data t or I t The i-th dimension feature, f i ,w i ∈R, Part I is the first-order combination of features, w i is the feature weight to be learned; Indicates V j1 and F j1 Multiply the elements at corresponding positions in The feature vector formed by this is called embedding.
[0097] The feature vector F is obtained by combining high-order features using a deep neural network. DNN ,include:
[0098] The embedding is used as the input of the DNN part, denoted as a 0 :
[0099] a 0 =[e 0 ,e 1 …,e m ]
[0100] The output of the lth hidden layer is denoted as a l+1 :
[0101] a l+1 =σ(W l a l +b l )
[0102] W l , b l is the parameter of the lth layer. Finally, the output of DNN is:
[0103] F DNN =σ(W H+1 a H +b H+1 )
[0104] Where H is the number of hidden layers.
[0105] The input layer of the neural network includes three layers, namely:
[0106] A. The first fully connected layer: Assume that the input feature vector is x∈R d , the weight matrix of the first layer is W (1) ∈R h×d , where h is the number of neurons in the first layer, and the output of the first layer is Z (1) =W (1) x+b (1) ;
[0107] Among them, b (1) is the bias vector, and then passes through a nonlinear activation function f (1) (·), and the output after activation is obtained as a (1) =f (1) (z (1) );
[0108] B. The second fully connected layer: Assume that the weight matrix of the second layer is W (2) ∈R h'×h , where h' is the number of neurons in the second layer, and the output of the second layer is Z (2) =W(2) a (1) +b (2) , and then through a nonlinear activation function f (2) (·) Get the output a after activation (2) =f (2) (z (2) );
[0109] C. The third fully connected layer: Assume that the weight matrix of the third layer is W (3) ∈R k×h' , where k is the number of output categories, and the output of the third layer is Z (3) =W (3) a (2) +b (3) .
[0110] The output layer of the neural network is:
[0111] Use the softmax function for multi-classification prediction.
[0112] For , the Softmax function can be expressed as:
[0113]
[0114] where p = (y = i | z (3) ), Z (3) is the input vector; y represents the category and i represents the predicted probability.
[0115] Loss function and optimization algorithm:
[0116] Select the cross entropy loss function as the loss function and use the gradient descent optimization algorithm for model training. Cross entropy loss function: Assume that the true label is y,
[0117] The predicted probability is p, and the cross entropy loss can be expressed as:
[0118]
[0119] During training, the data set is divided into training set, validation set and test set. The training set is used for model training, the validation set is used for model parameter adjustment, and finally the test set is used to evaluate the model performance.
[0120] The above describes the feedforward intelligent coating thickness closed-loop control method in the embodiment of the present invention. The following describes the feedforward intelligent coating thickness closed-loop control device in the embodiment of the present invention. The first embodiment of the feedforward intelligent coating thickness closed-loop control device in the embodiment of the present invention includes:
[0121] A data acquisition unit is used to collect data for training the model, wherein the data includes coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount;
[0122] Feature construction unit, used to construct data time series features and multi-dimensional correlation features between data;
[0123] The model building and training unit is used to build and train the neural network, input the above time series features and the multi-dimensional correlation features between the data, and output the neural network model of the classification of the coating thickness trend adjustment direction and the adjustment amount of each actuator to obtain a trained neural network model;
[0124] The control execution unit is used to perform closed-loop control of coating thickness according to the trained neural network model.
[0125] Specifically, the feedforward intelligent coating thickness closed-loop control device performs the following steps:
[0126] Step 1: Collect data to train the model. The data includes coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, adjustment amount, etc.
[0127] Step 2: Preparation before training, including data cleaning, preprocessing, building data time series features and multi-dimensional correlation features between data;
[0128] Step 3: Build a classification and regression network, input the above time series features and multi-dimensional correlation features between the data, and output a model for the classification of the coating thickness trend adjustment direction and the adjustment amount of each actuator;
[0129] Step 4: When deploying specifically, the coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity and other data shall be included, including the adjustment label of the adjustment mechanism and the adjustment amount data.
[0130] Step 5: Preprocess the data collected in step 4, construct features according to the processing method of the training data set in step 2, and input them into the model trained in step 4;
[0131] Step 6: Output the adjustment direction label and the adjustment amount of each actuator, and enter the actuator controller such as pump speed, lip opening, distance between coating head and back roller, etc.
[0132] Step 7: The actuator performs closed-loop online adjustment according to the adjustment direction and adjustment amount.
[0133] The structure of an electronic device provided by an embodiment of the present invention may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memories and storage media may be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the electronic device. Furthermore, the processor may be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.
[0134] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the electronic device structure does not constitute a limitation on the electronic device, and may include more or less components than the aforementioned, or combine certain components, or arrange the components differently.
[0135] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the feedforward intelligent coating thickness closed-loop control method.
[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.
[0137] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.
[0138] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A feedforward intelligent coating thickness closed-loop control method, characterized in that: The following steps are involved: Collect data to train the model, including coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount; Construct data time series features and multi-dimensional correlation features between data; Build and train a neural network, input the above time series features and multi-dimensional correlation features between the data, output a neural network model for the classification of coating thickness trend adjustment direction and the adjustment amount of each actuator, and obtain a trained neural network model; Predict coating thickness based on the trained neural network model.
2. A feedforward intelligent coating thickness closed-loop control method according to claim 1, characterized in that: If the predicted coating thickness does not meet expectations, ΔP61 is calculated. The calculation formula of ΔP61 is: Wherein, σ represents the surface tension of the slurry, H represents the coating height, θ is the static contact angle between the slurry and the substrate, φ is the static contact angle between the slurry and the coating head mold, l1 is the length of the upstream of the coating bead, l2 is the length of the downstream of the coating bead, Uc is the coating speed, W is the slit width of the coating head, and h m is the predicted coating thickness, h represents the stable coating thickness, b represents the coating width, and μ represents the density of the slurry; According to the calculation result of △P61, the coating speed, coating gap and lip flow rate are adjusted.
3. A feedforward intelligent coating thickness closed-loop control method according to claim 1, characterized in that: According to the calculation result of △P61, the coating speed, coating gap and lip flow rate are adjusted as follows: When the value of △P61 is greater than 200,000: At this time, adjust the coating speed first, then adjust the coating gap, and do not adjust the lip flow; ②When the value of △P61 is 150,000-200,000: At this time, adjust the coating speed first, then adjust the coating gap, if it does not meet expectations, fine-tune the lip flow rate; ③When the value of △P61 is 100,000-150,000: At this time, the coating gap, coating speed, and lip flow rate have equal weights and can be adjusted simultaneously without regard to order. ④When the value of △P61 is 50,000-100,000: At this time, adjust the lip flow rate first, then adjust the coating gap and coating speed; ⑤When the value of △P61 is less than 50,000: At this time, adjust the lip flow rate first, then adjust the coating gap, and finally adjust the coating speed.
4. A feedforward intelligent coating thickness closed-loop control method according to claim 1, characterized in that: The construction of data time series features and multi-dimensional correlation features between data includes the following steps: Based on dynamic time warping algorithm: Input: Two columns of thickness gauge detection value time series X = {x0, x1, ... x m-1 } and Y={y0,y1,…y n-1 }, their lengths are m and n respectively, where m may not be equal to n, Output: The best matching path between time series X and Y is W = {w0,w1…w k }, and the cumulative distance DTW(X,Y) of path W; Specifically include: Step 1: Construct an mxn cost matrix, where the matrix element D mn The matrix d i,j Represents x i and i ; Euclidean distance between two points Path W in w k represents a position in a matrix and satisfies the following three constraints: (1) Boundary conditions, w1 = D(0,0) and w k =D(m-1,n-1), the two sequences can be of different lengths, as long as the starting and ending positions are aligned; (2) Continuity, if the path element w has been determined k-1 =D(a', b'), then for the next path point w k =D(a,b) needs to satisfy (a-a')≤1 and (b-b')≤1, so as to ensure that the two time series are continuous when matching, without skipping or missing any point; (3) Monotonicity: If the path element w is determined, k-1 =D(a', b'), then for the next path point w k =D(a,b) needs to satisfy (a-a')≥0 and (b-b')≥0 to ensure that the points on the path are monotonically increasing on the time axis, that is, only the right, top and top-right of the current point can be selected, ensuring that the path will not return; Step 2: Initialize the cumulative distance: γ(0,0)=d(x0,y0) Step 3: Calculate the minimum cumulative distance γ(m-1,n-1) according to the recursive formula: γ(i,j)=d(x i ,y j )+min{γ(i-1,j-1),γ(i-1,j),γ(i,j-1)} Step 4: Backtrack to find all the matching point pairs in path W: Based on the best local path of each step, backtrack from the matching point pair (m-1, n-1) to the point (0,0).
5. The feedforward intelligent coating thickness closed-loop control method according to claim 1, characterized in that: The said building and training of the neural network comprises: The factor decomposition model is used to combine the low-order features of multi-source time series data to obtain the feature vector F M ; Use deep neural network to combine high-order features to get feature vector F DNN , The feature vector F M and F DNN Splice them together to get the spliced feature vector; The concatenated feature vector is passed as input to the input layer of the neural network; The cross entropy loss function is selected as the loss function, and the gradient descent optimization algorithm is used for model training.
6. A feedforward intelligent coating thickness closed-loop control method according to claim 5, characterized in that: The factor decomposition model is used to combine low-order features of multi-source time series data features to obtain a feature vector F M ,include: F M The formula represents the calculation process of the factorization machine: In the formula, f i is the state vector T in the input data t or I t The i-th dimension feature, f i ,w i ∈R, Part I is the first-order combination of features, w i is the feature weight to be learned; Indicates V j1 and F j1 Multiply the elements at corresponding positions in The feature vector formed by the embedding is called embedding; The feature vector F is obtained by combining high-order features using a deep neural network. DNN ,include: The embedding is used as the input of the DNN part, denoted as a 0 : a 0 =[e0,e1…,e m ] The output of the lth hidden layer is denoted as a l+1 : a l+1 =σ(W l a l +b l ) W l , b l is the parameter of the lth layer. Finally, the output of DNN is: F DNN =σ(W H+1 a H +b H+1 ) Where H is the number of hidden layers.
7. A feedforward intelligent coating thickness closed-loop control method according to claim 5, characterized in that: The input layer of the neural network includes three layers, namely: A. The first fully connected layer: Assume that the input feature vector is x∈R d , the weight matrix of the first layer is W (1) ∈R h×d , where h is the number of neurons in the first layer, and the output of the first layer is Z (1) =W (1) x+b (1) ; Among them, b (1) is the bias vector, and then passes through a nonlinear activation function f (1) (·), and the output after activation is obtained as a (1) =f (1) (z (1) ); B. The second fully connected layer: Assume that the weight matrix of the second layer is W (2) ∈R h,×h , where h' is the number of neurons in the second layer, and the output of the second layer is Z (2) =W (2) a (1) +b (2) , and then through a nonlinear activation function f (2) (·) Get the output a after activation (2) =f (2) (z (2) ); C. The third fully connected layer: Assume that the weight matrix of the third layer is W (3) ∈R k×h, , where k is the number of output categories, and the output of the third layer is Z (3) =W (3) a (2) +b (3) .
8. A feedforward intelligent coating thickness closed-loop control system, characterized in that: The system comprises: A data acquisition unit is used to collect data for training the model, wherein the data includes coating thickness, pump speed, lip opening, distance between coating head and back roller, vehicle speed, cavity pressure, feeding pressure, slurry temperature, viscosity, adjustment mechanism adjustment label, and adjustment amount; Feature construction unit, used to construct data time series features and multi-dimensional correlation features between data; The model building and training unit is used to build and train the neural network, input the above time series features and the multi-dimensional correlation features between the data, and output the neural network model of the classification of the coating thickness trend adjustment direction and the adjustment amount of each actuator to obtain a trained neural network model; The control execution unit is used to perform closed-loop control of coating thickness according to the trained neural network model.
9. An electronic device, comprising a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes each step of the feedforward intelligent coating thickness closed-loop control method as described in any one of claims 1-7.
10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the feedforward intelligent coating thickness closed-loop control method as described in any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Pole piece coating closed-loop control method
CN120605844A
A kind of pole piece coating closed loop control method
CN120605844B
Prediction model training method and device, equipment control method and device, product and equipment
CN121278380A
Self-adaptive control method of cement mixing pile and pile forming quality real-time prediction method
CN121348774A
Coating method, device, equipment and storage medium
CN121565782A
