Automatic coating method and system

Abstract

The invention discloses an automatic coating method and system, and relates to the technical field of industrial control, and the method comprises the steps: collecting coating distribution characteristics and gap distribution characteristics of each coating section of a target substrate based on a multi-section coating control model, and obtaining a characteristic array with a set regulation length through characteristic similarity dimension reduction processing; after the pre-trained coating retrieval features and gap retrieval features are combined, a cohesion mechanism is adopted to extract polymerization features of fusion core elements; respectively extracting gating fusion features of the polymerization features and the coating and gap feature arrays through a double-feature gating mechanism; and generating a multi-section intermittent coating control instruction based on the fused features. According to the method, the data processing efficiency is optimized through dynamic feature dimensionality reduction, accurate association of core process elements is achieved through a cohesion mechanism and double-feature gating fusion, the coating uniformity and the gap precision are effectively improved, the method is suitable for a high-precision multi-section coating scene of base materials such as the base coating foil, the multi-section cooperative control capacity is enhanced, and the product yield is increased.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial control, in particular to an automatic coating method and system. BACKGROUND

[0002] Multi-section intermittent coating is a key process in the field of precision coating of lithium battery pole pieces, flexible electronic devices, etc., and the core is to realize the coordinated control of coating section thickness uniformity and gap section precision. In the prior art, the traditional coating control method relies on manual experience or simple parameter closed-loop adjustment, and it is difficult to dynamically associate the coating distribution characteristics and gap distribution characteristics in the multi-section coating process. At the same time, the original collected characteristic data has high dimension and much redundant information, and direct use for control will lead to a sharp increase in model complexity and response delay, which cannot meet the demand of high-precision coating. SUMMARY

[0003] The purpose of the present application is to provide an automatic coating method and system.

[0004] In a first aspect, an embodiment of the present application provides an automatic coating method, comprising: Based on a multi-section coating control model, data acquisition is performed on each coating section in the target substrate to obtain corresponding coating distribution characteristics and gap distribution characteristics, and based on the similarity of the characteristics, dimension reduction processing is performed on the obtained coating distribution characteristic set and gap distribution characteristic set respectively to obtain coating distribution characteristic arrays and gap distribution characteristic arrays of a set control length; After merging the pre-trained coating retrieval features and gap retrieval features, the corresponding aggregation characteristics are extracted using a cohesion mechanism; the coating retrieval features and the gap retrieval features are obtained by training the corresponding adjustable retrieval features based on typical coating samples; the aggregation characteristics fuse the core elements in the coating retrieval features and the gap retrieval features; Using a double-feature gating mechanism, a first gating fusion feature between the aggregation characteristics and each coating distribution feature in the coating distribution characteristic array is extracted, and a second gating fusion feature between the aggregation characteristics and each gap distribution feature in the gap distribution characteristic array is extracted; Based on the obtained first gating fusion features and second gating fusion features, the multi-section intermittent coating control of the target substrate is controlled.

[0005] In a possible implementation, the multi-section coating control model includes a gating feature encoder; then the double-feature gating mechanism is used to extract the first gating fusion feature between the aggregation characteristics and each coating distribution feature in the coating distribution characteristic array, and the second gating fusion feature between the aggregation characteristics and each gap distribution feature in the gap distribution characteristic array, comprising: splitting the polymerization characteristics into first polymerization characteristics corresponding to the coating distribution characteristics and second polymerization characteristics corresponding to the gap distribution characteristics; extracting, by a first gating unit in the gating feature encoder, the first polymerization characteristics based on a double-feature gating mechanism, first gating fusion features between each of the coating distribution characteristics in the coating distribution characteristic array; extracting, by a second gating unit in the gating feature encoder, the second polymerization characteristics based on a double-feature gating mechanism, second gating fusion features between each of the gap distribution characteristics in the gap distribution characteristic array.

[0006] In a possible implementation, the gating feature encoder includes sequentially connected gating feature processors; for the first gating feature processor, the current processing features include the coating distribution characteristic array and the gap distribution characteristic array; for other gating feature processors, the current processing features include the first gating fusion features and the second gating fusion features output by the previous gating feature processor; controlling the multi-section intermittent coating control of the target substrate based on the obtained first gating fusion features and second gating fusion features, including: controlling the multi-section intermittent coating control of the target substrate based on the first gating fusion features and the second gating fusion features output by the last gating feature processor.

[0007] In a possible implementation, the multi-section coating control model includes a coating parameter generation model; controlling the multi-section intermittent coating control of the target substrate based on the obtained first gating fusion features and second gating fusion features, including: combining the first gating fusion features and the second gating fusion features, and converting the combined features into coating data features in the same feature domain as the coating process parameter dimension of the target substrate; inputting the coating data features, coating process description features, and coating control parameters of the target substrate as coating parameter guide information into the coating parameter generation model to generate multi-section intermittent coating control instructions for the target substrate.

[0008] In a possible implementation, each gap distribution characteristic corresponds to a gap sub-section in a coating section; the dimensionality reduction processing is performed on the obtained coating distribution feature set and gap distribution feature set based on feature similarity, to obtain a coating distribution characteristic array and a gap distribution characteristic array with a set regulation length, including: after each coating section is collected, the coating distribution characteristics of the coating section are written into a coating global storage unit, and the gap distribution characteristics of the coating section are written into a gap local storage unit; When the number of coating distribution features stored in the coating global storage unit reaches the set control length, the coating distribution features of at least two coating sections in the coating global storage unit are aggregated based on feature similarity, and the coating distribution features to be written are written into the coating global storage unit. And for each gap sub-section, when the number of corresponding gap distribution features stored in the gap local storage unit reaches the set control length, the corresponding gap distribution features of at least two coating sections in the gap local storage unit are aggregated based on feature similarity, and the gap distribution features to be written are written into the gap local storage unit.

[0009] In a possible implementation, before the first gating fusion features between the aggregation characteristics and each coating distribution feature in the coating distribution feature array are extracted respectively, and the second gating fusion features between the aggregation characteristics and each gap distribution feature in the gap distribution feature array are extracted respectively, the method further includes: When a set control node is reached, the coating distribution feature array stored in the coating global storage unit is read, and the gap distribution feature array stored in the gap local storage unit is read.

[0010] In a possible implementation, the features to be aggregated of at least two coating sections are aggregated by any of the following methods: The features to be aggregated of the continuous coating sections closest in parameters in the storage unit are calculated by mean value based on feature similarity; The features to be aggregated to be written are grouped and classified with the features to be aggregated written in the storage unit based on feature similarity, and the features to be aggregated of each coating section that has been grouped and classified are calculated by mean value; The features to be aggregated are coating distribution features or gap distribution features, and correspondingly, the storage unit is a coating global storage unit or a gap local storage unit.

[0011] In a possible implementation, if there are multiple groups of continuous coating sections closest in parameters, the calculation of the features to be aggregated of the continuous coating sections closest in parameters in the storage unit by mean value includes: The features to be aggregated of the continuous coating sections closest in parameters and written first in the storage unit are calculated by mean value; or The features to be aggregated of each group of continuous coating sections closest in parameters in the storage unit are calculated by mean value respectively.

[0012] In a possible implementation, the target substrate is a foil substrate with a primer; the data acquisition of each coating section in the target substrate is performed respectively to obtain the corresponding coating distribution characteristics and gap distribution characteristics, including: The primer data and the white space data of the current coating section are acquired by detecting the three different color regions of the foil substrate, i.e., the empty foil region, the primer region and the coating region, through the high-precision color scale color sensor. Based on the primer data and the white space data, the coating length characteristics of the coating section and the white space length characteristics of the gap sub-section are obtained, and the coating length characteristics are taken as the coating distribution characteristics, and the white space length characteristics are taken as the gap distribution characteristics.

[0013] In a second aspect, an embodiment of the present application provides a server system, comprising a server configured to execute the method of the first aspect.

[0014] Compared with the prior art, the present application provides the following beneficial effects: by using the automatic coating method and system disclosed in the present application, the coating distribution characteristics and the gap distribution characteristics of each coating section of the target substrate are acquired based on a multi-section coating control model, and the feature array of the set control length is obtained through feature similarity dimension reduction processing; after the pre-trained coating retrieval features and the gap retrieval features are combined, the cohesive mechanism is used to extract the aggregation characteristics of the fusion core elements; the gating fusion features of the aggregation characteristics and the coating and gap feature arrays are extracted through the double-feature gating mechanism; and the multi-section intermittent coating control instructions are generated based on the fusion features. The present application optimizes the data processing efficiency through dynamic feature dimension reduction, realizes the precise association of the core process elements through the cohesive mechanism and the double-feature gating fusion, effectively improves the coating uniformity and the gap accuracy, is suitable for the high-precision multi-section coating scene of the substrate such as the foil with primer, enhances the multi-section collaborative control capability, and improves the product yield. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0016] Figure 1 The step flowchart of the automatic coating method provided by the embodiment of the present application is shown in the figure. Figure 2 The structural schematic block diagram of the computer device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0018] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0019] To solve the technical problems in the foregoing background art, Figure 1 The flowchart of the automatic coating method provided by the embodiments of the present application is shown below, and the automatic coating method will be described in detail.

[0020] In step S201, based on the multi-section coating control model, data acquisition is performed on each coating section in the target substrate respectively, the corresponding coating distribution characteristics and gap distribution characteristics are obtained, and based on the characteristic similarity, the obtained coating distribution characteristic set and gap distribution characteristic set are processed for dimension reduction respectively, to obtain the coating distribution characteristic array and gap distribution characteristic array of the set control length; In step S202, after the coating search characteristics and gap search characteristics obtained by pre-training are combined, the corresponding aggregation characteristics are extracted by using the cohesion mechanism; the coating search characteristics and the gap search characteristics are obtained by training the corresponding adjustable search characteristics based on typical coating samples; the aggregation characteristics fuse the core elements in the coating search characteristics and the gap search characteristics; In step S203, by using the double-feature gating mechanism, the first gating fusion features between the aggregation characteristics and each coating distribution characteristic in the coating distribution characteristic array are extracted respectively, and the second gating fusion features between the aggregation characteristics and each gap distribution characteristic in the gap distribution characteristic array are extracted respectively; In step S204, based on the obtained each first gating fusion feature and each second gating fusion feature, the multi-section intermittent coating control of the target substrate is controlled.

[0021] In the embodiment of the present application, an example is provided. The embodiment takes a server as the execution subject and details the implementation process of the automatic coating method for the multi-section intermittent coating scenario of the foil substrate with primer coating (aluminum foil for lithium-ion battery pole piece). The target substrate is an aluminum foil with a thickness of 12 μm, the surface is pre-coated with a carbon black primer (black, RGB value 30, 30, 30), the coating material is lithium iron phosphate active material (dark gray, RGB value 100, 100, 100), and the empty foil area is the original color of the aluminum foil (silver white, RGB value 255, 255, 240). The color difference between the three is significant, which facilitates data acquisition through a high-precision color marker color sensor (resolution 0.01 mm, sampling frequency 1 kHz). The server is preloaded with a multi-section coating control model, which includes a data acquisition module, a feature storage unit (global coating layer storage unit, local gap storage unit), a feature dimensionality reduction processing module, a gated feature encoder (including a first / second gating unit and three sequentially connected gated feature processors GP1-GP3), and a coating parameter generation model. The control length is set to 5 (i.e., the final generated feature array length is 5).

[0022] The server controls the coating machine to transport the aluminum foil substrate at a speed of 0.5 m / s, and synchronously starts the color marker color sensor to perform real-time scanning on the current coating section. The sensor identifies the boundaries of the coating section and the gap sub-section through color jump: when the color jumps from the primer coating (30, 30, 30) to the coating layer (100, 100, 100) is detected, the start point of the coating section is recorded; when the color jumps from the coating layer to the primer coating, the end point of the coating section is recorded, and the distance between the two points is the "coating length feature" (coating layer distribution feature). For example, the length of the first coating section is 150 mm, the length of the second coating section is 152 mm, the length of the third coating section is 149 mm, the length of the fourth coating section is 151 mm, and the length of the fifth coating section is 153 mm. For the gap distribution feature, the white space area (gap sub-section) following each coating section is identified by the color jump from the end of the coating layer to the start of the next coating section, and the distance is the "white space length feature". For example, the first gap sub-section is 20 mm, the second gap sub-section is 19.8 mm, the third gap sub-section is 20.2 mm, the fourth gap sub-section is 19.9 mm, and the fifth gap sub-section is 20.1 mm.

[0023] The server writes the coating distribution features into the coating global storage unit in real time, and writes the gap distribution features into the corresponding gap local storage unit according to the gap sub-sections (for example, the first gap sub-section features are written into storage unit 1, the second gap sub-section features are written into storage unit 2, and so on). When the number of features in the coating global storage unit reaches the set control length 5, the server reduces the dimensionality of the feature set based on the feature similarity (Euclidean distance): the distance between the second coating section (152 mm) and the fourth coating section (151 mm) is the smallest (1.0), the mean of the two is aggregated (151.5 mm), the original features are deleted, and the aggregation result is written, at this time the storage unit features are [150, 151.5, 149, 153]; after writing the sixth coating section feature 150.5 mm, the storage unit restores five features, and generates the coating distribution feature array [150, 151.5, 149, 153, 150.5]. The gap local storage unit is processed in the same way, for example, when the number of features in the first gap sub-section storage unit reaches 5, the first (20 mm) and fifth (20.1 mm) features are aggregated (mean 20.05 mm), and after writing the sixth feature 19.7 mm, the gap distribution feature array [20.05, 19.8, 20.2, 19.9, 19.7] is generated.

[0024] The server loads the coating retrieval features (512 dimensions, including thickness deviation coefficient, uniformity index, etc. core dimensions) and gap retrieval features (512 dimensions, including width standard deviation, die cutting adaptation coefficient, etc. core dimensions) trained based on 1000 typical coating samples, and merges them into a 1024-dimensional joint feature vector. Through the cohesion mechanism, the information entropy of each element in the vector is calculated, and the top 256 core elements (such as the coating retrieval feature "thickness deviation coefficient ≤0.5 μm", the gap retrieval feature "width standard deviation ≤0.1 mm") are selected according to the entropy value, and a 256-dimensional aggregated feature is formed, which simultaneously carries the key constraints of coating uniformity and gap accuracy. The feature mapping layer compresses the 640-dimensional joint feature to 128-dimensional through a 128x640-dimensional weight matrix. The cohesion mechanism calculates the information entropy H = -∑p(x)logp(x) of each element in the joint feature vector, selects 256 core elements with entropy values below the threshold (such as H < 0.1), and retains the key control dimensions of the coating and gap features.

[0025] The server calls the gated feature encoder to split the 256-dimensional aggregated features into 128-dimensional first aggregated features (coating related) and 128-dimensional second aggregated features (gap related). The first gating unit receives the first aggregated features and the coating distribution feature array, calculates the fusion weight of each coating feature and the first aggregated features through a double-feature gating mechanism (G1 = sigma (W1 · [P1; Ci] + b1), sigma is the sigmoid function), and outputs five 128-dimensional first gated fusion features; the second gating unit outputs five 128-dimensional second gated fusion features in the same way. These features are input into the sequential gated feature processor: GP1 receives the original feature array and outputs the first round of fusion features; GP2 receives the output of GP1 and optimizes the feature interaction information (such as the correlation between coating and gap features); GP3, as the last processor, outputs five 64-dimensional first gated fusion features (G1 final5 ) and five 64-dimensional second gated fusion features (G2 final1 ~G2 final5 ).

[0026] The server maps the fusion features (5x64x2=640 dimensions) output by GP3 through linear transformation into 128-dimensional coating data features, splices them with target substrate process description features (“lithium iron phosphate positive electrode coating, target thickness 80 pm”, 128 dimensions) and current control parameters (pressure 0.3 MPa, flow 50 ml / min, speed 0.5 m / s, 128 dimensions) into a 256-dimensional input vector, and inputs it into the coating parameter generation model (Transformer sequence generation model). The model captures feature correlations through self-attention mechanisms and generates multiple control instructions: for example, based on G1 final1 (corresponding to a 150 mm coating section), adjust the pressure to 0.32 MPa to compensate for the thickness; based on G2 final1 (corresponding to a 20.05 mm gap), adjust the speed to 0.49 m / s to control the margin deviation; and optimize the pressure and flow parameters for subsequent coating sections. The server sends instructions to the coating machine PLC through the Profinet bus to achieve closed-loop control, with a final coating thickness standard deviation of ≤0.8 pm and a gap deviation of ≤0.1 mm, meeting the precision requirements of the pole piece production.

[0027] In the embodiment of the present application, the multi-section coating control model includes a gated feature encoder; then the double-feature gating mechanism is used to extract the first gated fusion features between the aggregated features and each coating distribution feature in the coating distribution feature array, and the second gated fusion features between the aggregated features and each gap distribution feature in the gap distribution feature array, which can be implemented through the following examples.

[0028] splitting the polymerization characteristics into first polymerization characteristics corresponding to the coating distribution characteristics, and second polymerization characteristics corresponding to the gap distribution characteristics; extracting, by a first gating unit in the gating feature encoder, based on a double-feature gating mechanism, first gating fusion features between each of the coating distribution characteristics in the coating distribution characteristic array, respectively, for the first polymerization characteristics; extracting, by a second gating unit in the gating feature encoder, based on a double-feature gating mechanism, second gating fusion features between each of the gap distribution characteristics in the gap distribution characteristic array, respectively, for the second polymerization characteristics.

[0029] In an embodiment of the present application, in an exemplary multi-section coating control model loaded by the server, the gating feature encoder includes a first gating unit (coating feature processing) and a second gating unit (gap feature processing) that are independent of each other and run in parallel to improve processing efficiency. The target substrate is a bottom-coated aluminum foil, which has generated 256-dimensional polymerization characteristics (fusion of coating and gap core control elements), a 5-dimensional coating distribution characteristic array ([150, 151.5, 149, 153, 150.5] mm, coating length characteristics), and a 5-dimensional gap distribution characteristic array ([20.05, 19.8, 20.2, 19.9, 19.7] mm, white space length characteristics) after pre-processing. The server executes the following steps based on the above data: The server splits the 256-dimensional polymerization characteristics into 128-dimensional first polymerization characteristics (related to coating control) and 128-dimensional second polymerization characteristics (related to gap control) by a feature domain division algorithm. The first polymerization characteristics include coating thickness deviation coefficients (dimensions 1-32, representing the correlation between coating length and thickness, such as 1 mm increase in length corresponding to thickness compensation), coating uniformity indicators (dimensions 33-64, such as thickness standard deviation threshold at different lengths), slurry fluidity adaptation parameters (dimensions 65-128, such as matching coefficients of coating length and slurry flow); the second polymerization characteristics include gap width standard deviation (dimensions 1-32, controlling the fluctuation range of white space length), die cutting adaptation coefficients (dimensions 33-64, such as die cutting knife position parameters corresponding to a 20 mm standard gap), substrate tension correlation parameters (dimensions 65-128, adjustment relationship between white space length and substrate tension). After splitting, the two types of polymerization characteristics are respectively input into the first and second gating units.

[0030] The first gating unit loads pre-trained gating parameters (weight matrix W1 is 128x129-dimensional, bias vector b1 is 128-dimensional), receives 128-dimensional first polymerization characteristics (denoted as P1) and 5-dimensional coating distribution characteristic array (denoted as C=[C1, C2, C3, C4, C5]=[150, 151.5, 149, 153, 150.5] mm), and performs double-feature gating fusion for each coating distribution characteristic Ci: Feature concatenation: concatenate P1 (128 dimensions) and the current coating distribution feature Ci (1 dimension, normalized to 0-1 interval value, such as C1=150mm normalized to 0.32) into a 129-dimensional vector [P1;Ci]; Weight calculation: calculate the fusion weight Gi=σ(W1·[P1;Ci]+b1) by the sigmoid activation function, to obtain a 128-dimensional weight vector (element value 0-1, representing the fusion ratio of P1 and Ci); Feature fusion: the first gating fusion feature Fi=Gi⊙P1+(1-Gi)⊙(Ci mapping vector), where "⊙" is element-wise multiplication, and the Ci mapping vector is converted from the normalized Ci to a 128-dimensional feature by a 1x128-dimensional linear layer.

[0031] For example, C1=150mm (normalized to 0.32): After the concatenation vector [P1;0.32] is calculated by W1·[P1;0.32]+b1, the sigmoid output Gi=[0.85,0.82,...,0.79] (128 dimensions, mean 0.81) is obtained; The Ci mapping vector is [0.32,0.32,...,0.32] (128 dimensions); The fusion feature F1=0.85×P11+0.15×0.32,...,0.79×P1 128 +0.21×0.32, to obtain a 128-dimensional first gating fusion feature F1.

[0032] Repeat the above process to process C2-C5 in turn, and finally output five 128-dimensional first gating fusion features {F1,F2,F3,F4,F5}.

[0033] The second gating unit is symmetrical to the first gating unit in structure, loads gating parameters (weight matrix W2 is 128x129 dimensions, bias vector b2 is 128 dimensions), receives a 128-dimensional second aggregation feature (denoted as P2) and a 5-dimensional gap distribution feature array (denoted as G=[G1,G2,G3,G4,G5]=[20.05,19.8,20.2,19.9,19.7]mm), and performs double feature gating fusion on each gap distribution feature Gj: Feature concatenation: concatenate P2 (128 dimensions) and the current gap distribution feature Gj (1 dimension, normalized, such as G1=20.05mm normalized to 0.45) into a 129-dimensional vector [P2;Gj]; Weight calculation: calculate the fusion weight Hj=σ(W2·[P2;Gj]+b2) by the sigmoid function, to obtain a 128-dimensional weight vector; Feature fusion: the second gating fusion feature Hj=Hj⊙P2+(1-Hj)⊙(Gj mapping vector), and the Gj mapping vector is converted into a 128-dimensional feature through a 1*128-dimensional linear layer.

[0034] For example, G1=20.05mm (normalized 0.45): After the splicing vector [P2; 0.45] is calculated through W2・[P2; 0.45]+b2, the sigmoid output Hj=[0.78, 0.80,..., 0.83] (128-dimensional, average 0.80); The Gj mapping vector is [0.45, 0.45,..., 0.45] (128-dimensional); The fusion feature H1=0.78×P21+0.22×0.45,...,0.83×P2 128 +0.17×0.45, to obtain 128-dimensional second gating fusion feature H1.

[0035] After the processing is completed, the second gating unit outputs five 128-dimensional second gating fusion features {H1, H2, H3, H4, H5}, which are used as inputs of a subsequent gating feature processor together with {F1-F5} output by the first gating unit, to further optimize the coating control instruction.

[0036] In the embodiment of the application, the gating feature encoder comprises sequentially connected gating feature processors; for the first gating feature processor, the current processing feature comprises the coating distribution feature array and the gap distribution feature array; for other gating feature processors, the current processing feature comprises each first gating fusion feature and each second gating fusion feature output by the previous gating feature processor; Based on the acquired each first gating fusion feature and each second gating fusion feature, the multi-section intermittent coating control of the target substrate can be executed by the following examples.

[0037] Based on the each first gating fusion feature and the each second gating fusion feature output by the last gating feature processor, the multi-section intermittent coating control of the target substrate is controlled.

[0038] In the embodiment of the present application, the server configured gating feature encoder contains 3 sequentially connected gating feature processors (GP1, GP2, GP3), each processor is built-in feature interaction module and dimension optimization layer, and sequentially performs progressive enhancement processing on the gating fusion features, and finally realizes coating control based on the output of the last processor. The target substrate is an aluminum foil with a primer coating, which has generated a 5-dimensional coating distribution feature array (C=[150, 151.5, 149, 153, 150.5]mm), a 5-dimensional gap distribution feature array (G=[20.05, 19.8, 20.2, 19.9, 19.7]mm), and 5 128-dimensional first gating fusion features (F1-F5) and 5 128-dimensional second gating fusion features (H1-H5) preliminarily fused by a double-feature gating mechanism.

[0039] The first gating feature processor (GP1): the current processing features are the original coating distribution feature array C and the gap distribution feature array G. GP1 loads the feature interaction matrix (128×5 dimensions, pre-trained to capture the position correlation of coating and gap features), aligns the 5 F1-F5 (coating related) and 5 H1-H5 (gap related) by position (the ith F corresponds to the ith H), calculates the interaction weight of F and H in the same position (such as the correlation degree of F1 and H1, representing the cooperative adjustment relationship between the length of the 1st coating section and the length of the 1st gap) through a multi-head attention mechanism (2 heads, each head is 64 dimensions), and outputs 5 96-dimensional first gating fusion features (F1'=96 dimensions, corresponding to C1=150mm) and 5 96-dimensional second gating fusion features (H1'=96 dimensions, corresponding to G1=20.05mm) through the dimension optimization layer (128→96 dimensions, delete redundant features).

[0040] The second gating feature processor (GP2): the current processing features are F1'-F5' (96 dimensions) and H1'-H5' (96 dimensions) output by GP1. GP2 focuses on strengthening the cross-position feature dependence (such as the influence of the length of the 1st coating section and the 3rd coating section on the uniformity of the overall coating), respectively models the sequences of F1'-F5' and H1'-H5' through a bidirectional long short-term memory network (Bi-LSTM, hidden layer dimension 96), extracts time sequence features (such as the thickness compensation cumulative effect corresponding to the increasing trend of the lengths of the first three coating sections), and then outputs 5 64-dimensional first gating fusion features (F1''=64 dimensions) and 5 64-dimensional second gating fusion features (H1''=64 dimensions) through the dimension optimization layer (96→64 dimensions).

[0041] The third gate feature processor (GP3, the last one): the current processing features are F1''-F5'' (64 dimensions) and H1''-H5'' (64 dimensions) output by GP2. GP3 focuses on the core control features (such as the first 32 dimensions in F1'' directly related to the thickness deviation, and the 20th-45th dimensions in H1'' related to the gap die-cutting adaptation) through a self-attention mechanism (4 heads, each head 16 dimensions), and fuses the original features output by GP2 through a residual connection, finally outputs 5 first gate fusion features (F1 final =64 dimensions, containing a thickness compensation coefficient of 0.02 MPa / mm for a 150 mm coating section) and 5 second gate fusion features (H1 final =64 dimensions, containing a speed adjustment coefficient of-0.01 m / s / mm for a 20.05 mm gap).

[0042] The server merges F1 final -F5 final (5x64 dimensions) and H1 final -H5 final (5x64 dimensions) output by GP3 into a 640-dimensional feature vector, and inputs the coating parameter generation model. The model calculates the pressure adjustment amount for the 150 mm coating section according to the thickness compensation coefficient (0.02 MPa / mm) in F1 final : 150 mm x 0.02 MPa / mm = 3 MPa, superimposes the basic pressure 0.3 MPa, and generates the coating head pressure instruction 3.3 MPa; according to the speed adjustment coefficient (-0.01 m / s / mm) in H1 final : (20.05-20) mm x (-0.01 m / s / mm) =-0.0005 m / s, superimposes the basic speed 0.5 m / s, and generates the substrate speed instruction 0.4995 m / s ≈ 0.5 m / s (fine-tuned to 0.49 m / s to leave a margin). Similar calculations are performed on F2 final -F5 final and H2 final -H5 final , respectively, to generate 5 groups of coating section pressure / flow parameters and 5 groups of gap section speed / tension parameters, which are sent to the coating machine PLC through the industrial bus to realize precise closed-loop control of multi-section intermittent coating.

[0043] In the embodiment of the application, the multi-section coating control model includes a coating parameter generation model; and the multi-section intermittent coating control of the target substrate based on the obtained first gate fusion features and second gate fusion features can be implemented through the following examples.

[0044] After merging the first gated fusion features and the second gated fusion features, they are converted into coating data features that are located in the same feature domain as the coating process parameters of the target substrate. The coating process description features and coating control parameters of the target substrate are used as coating parameter guidance information, and the coating data features are combined with the coating parameter generation model to generate multi-segment intermittent coating control instructions for the target substrate.

[0045] In an embodiment of the invention, for example, the multi-segment coating control model configured on the server includes a coating parameter generation model (a sequence generation network based on the Transformer architecture) used to convert gating fusion features into actual coating control commands. The target substrate is aluminum foil with a base coating (used for the positive electrode sheet of a lithium-ion battery). The last processor (GP3) of the gating feature encoder outputs five 64-dimensional first gating fusion features (F1). final -F5 final (corresponding to 5 coating segments) and 5 64-dimensional second-gated fusion features (H1) final -H5 final (corresponding to 5 gap segments), the server performs the following steps based on these characteristics: The server first sends 5 F1 final -F5 final (Total dimensions 5 × 64 = 320) and 5 H1s final -H5 final (Total dimensions 5 × 64 = 320) Segmented by position (F1) final With H1 final Corresponding to the first coating segment and the gap sub-segment, and so on, a 640-dimensional joint feature vector is formed. This vector is transformed into a 128-dimensional coating data feature through a pre-trained feature mapping layer (linear transformation matrix W = 128 × 640 dimensions, bias vector b = 128 dimensions). This feature is located in the same feature domain as the coating process parameters of the target substrate (coating head pressure, slurry flow rate, substrate velocity, etc.). For example, dimensions 1-32 of the coating data feature correspond to "coating segment thickness compensation" (such as F1). final The extracted 150mm coating section requires a pressure compensation coefficient of 0.02MPa / mm. Dimensions 33-64 correspond to "gap section speed adjustment" (e.g., H1). final The 20.05mm gap requires a speed adjustment coefficient of -0.01m / s / mm. Dimensions 65-128 correspond to "cross-segment coordination parameters" (such as the flow accumulation adjustment factor corresponding to the increasing trend of the first 3 coating segment lengths).

[0046] The server combines the coating process description characteristics of the target substrate and the current coating control parameters into coating parameter guidance information. The process description characteristics are "lithium iron phosphate positive electrode coating, target thickness 80 pm ± 0.5 pm, substrate width 300 mm, primer thickness 2 pm, standard gap 20 mm ± 0.2 mm", which are processed by text embedding (Word2Vec) and normalization into 128-dimensional vectors (including the value 0.65 corresponding to the target thickness 80 pm and the value 0.42 corresponding to the standard gap 20 mm, etc.); the current coating control parameters are real-time collected equipment operation data: coating head pressure 0.3 MPa (normalized 0.3), slurry flow 50 ml / min (normalized 0.5), substrate transmission speed 0.5 m / s (normalized 0.5), substrate tension 30 N (normalized 0.3), which are encoded into 128-dimensional vectors. The two are spliced into 256-dimensional coating parameter guidance information as the "control target anchor point" of the model.

[0047] The server splices the 128-dimensional coating data characteristics and the 256-dimensional coating parameter guidance information into a 384-dimensional input vector, which is input into the coating parameter generation model. The model captures feature associations through a self-attention mechanism (8 heads, each 48-dimensional): for example, the matching relationship between "0.02 MPa / mm pressure compensation coefficient" in the coating data characteristics and "target thickness 80 pm" in the guidance information, the pressure adjustment amount of the 1st coating section (150 mm) = 150 mm x 0.02 MPa / mm = 3 MPa (base pressure 0.3 MPa + adjustment amount 0.02 MPa = 0.32 MPa); through the association of "-0.01 m / s / mm speed adjustment coefficient" and "standard gap 20 mm", the speed adjustment amount of the 1st gap sub-section (20.05 mm) = (20.05-20) mm x (-0.01 m / s / mm) = -0.0005 m / s (current speed 0.5 m / s + adjustment amount ≈ 0.49 m / s). The model calculates the parameters for the 5 coating sections and 5 gap sub-sections in turn to generate control instructions: the 1st coating section "pressure 0.32 MPa, flow 50.5 ml / min", the 1st gap sub-section "speed 0.49 m / s, tension 30.5 N", and the subsequent section parameters are optimized according to this logic. The server issues the instructions to the coating machine PLC through the Profinet bus to achieve precise control, and finally the coating thickness deviation is ≤0.4 pm and the gap length deviation is ≤0.08 mm, meeting the requirements of the pole piece production.

[0048] In the embodiments of the present application, each gap distribution characteristic corresponds to one gap sub-section in the coating section; the obtained coating distribution characteristic set and gap distribution characteristic set are respectively processed by dimension reduction based on feature similarity to obtain coating distribution characteristic arrays and gap distribution characteristic arrays of a set control length, which can be implemented by the following examples.

[0049] For each coating section collected, the coating distribution characteristics of the coating section are written into the coating global storage unit, and the gap distribution characteristics of the coating section are written into the gap local storage unit; When the number of coating distribution characteristics stored in the coating global storage unit reaches the set control length, the coating distribution characteristics of at least two coating sections in the coating global storage unit are aggregated based on feature similarity, and the coating distribution characteristics to be written are written into the coating global storage unit; And, for each gap sub-section, when the number of corresponding gap distribution characteristics stored in the gap local storage unit reaches the set control length, the corresponding gap distribution characteristics of at least two coating sections in the gap local storage unit are aggregated based on feature similarity, and the gap distribution characteristics to be written are written into the gap local storage unit.

[0050] In the embodiment of the present application, for example, the server implements dynamic management of characteristics for the multi-section intermittent coating scenario of the aluminum foil substrate with primer (targeting lithium ion battery pole piece) through the coating global storage unit (storing the coating distribution characteristics of all coating sections) and the gap local storage unit (storing the corresponding gap distribution characteristics independently according to the gap sub-section), reduces the dimension of the feature set based on feature similarity, and ensures that the number of stored characteristics always maintains the set control length of 5. The coating section of the target substrate is continuously generated at a speed of 0.5 m / s, each coating section corresponds to a gap sub-section, and the server collects the "coating length characteristics" (coating distribution characteristics) of the coating section and the "blank length characteristics" (gap distribution characteristics) of the gap sub-section through a high-precision color sensor. The specific processing procedure is as follows: Real-time writing process: the server writes the coating distribution characteristics (coating length) of each coating section collected into the coating global storage unit (initially empty) immediately. For example: The coating length of the first coating section is 150 mm, and after writing, the feature set S of the storage unit is coat =

[150] (number=1<5, no dimension reduction triggered); The coating length of the second coating section is 152 mm, and after writing, S is coat =[150, 152] (number=2<5); The coating length of the third coating section is 149 mm, and after writing, S is coat =[150, 152, 149] (number=3<5); The coating length of the fourth coating section is 151 mm, and after writing, S is coat =[150, 152, 149, 151] (number=4<5); The coating length of the fifth coating section is 153 mm, and after writing, S is coat =[150, 152, 149, 151, 153] (number=5=the set control length, triggering dimension reduction).

[0051] Feature Similarity Calculation and Aggregation: The server calculates the similarity (distance) of all feature pairs in S coat by Euclidean distance (the smaller the distance, the higher the similarity): 150 and 152 distance = 2; 152 and 149 distance = 3; 149 and 151 distance = 2; 151 and 153 distance = 2; 150 and 149 distance = 1 (smallest), determine that 150 mm (1st coating section) and 149 mm (3rd coating section) are the two features with the highest similarity.

[0052] Perform mean aggregation on the two: (150 + 149) / 2 = 149.5 mm, delete the original 150 and 149, and update the unit to S coat = [149.5, 152, 151, 153] (number = 4 < 5).

[0053] Write new feature: collect the coating length of the 6th coating section as 150.5 mm, and write S coat = [149.5, 152, 151, 153, 150.5] (number = 5 = set control length), complete the dimension reduction of the coating distribution feature array, and the array is [149.5, 152, 151, 153, 150.5]. Repeat the "write - full 5 - aggregation - write new feature" process for each new coating section, and always maintain 5 features.

[0054] Each coating section corresponds to a gap sub-section (e.g., the 1st coating section corresponds to the 1st gap sub-section, the 2nd coating section corresponds to the 2nd gap sub-section, etc.), and the server configures an independent gap local storage unit (denoted as LS1, LS2, …, LS n , n is the gap sub-section number) for each gap sub-section, which only stores the blank length features of the corresponding sub-section and is independently dimensionally processed. Take the 1st gap sub-section (LS1) as an example: Real-time writing process: the server immediately writes the blank length feature of the gap sub-section (the mth gap sub-section) corresponding to the mth coating section to LS m after completing the coating of the mth coating section. For LS1 (storing all blank lengths of the 1st gap sub-section): After the 1st coating, the blank length of the 1st gap sub-section is 20 mm, and LS1 is written as

[20] (number = 1 < 5); After the 2nd coating, the blank length of the 1st gap sub-section (repeated due to substrate circulation coating) is 19.8 mm, and LS1 is written as [20, 19.8] (number = 2 < 5); After the 3rd coating, the blank length of the 1st gap sub-section is 19.9 mm, and LS1 is written as [20, 19.8, 19.9] (number = 3 < 5); ​After the 4th coating, the 1st gap sub-segment is 20.1 mm, and the write-in LS1 is [20, 19.8, 19.9, 20.1] (number = 4 < 5); After the 5th coating, the 1st gap sub-segment is 20.2 mm, and the write-in LS1 is [20, 19.8, 19.9, 20.1, 20.2] (number = 5 = set control length, trigger dimension reduction).

[0055] Feature similarity calculation and aggregation: the server calculates the Euclidean distance of the features in LS1: The distance between 20 and 19.8 is 0.2, the distance between 19.8 and 19.9 is 0.1, the distance between 19.9 and 20.1 is 0.2, the distance between 20.1 and 20.2 is 0.1, and the distance between 20 and 20.1 is 0.1 (minimum), and it is determined that 20 mm (1st) and 20.1 mm (4th) are the two features with the highest similarity.

[0056] Perform mean aggregation on both: (20+20.1) / 2=20.05 mm, delete the original 20 and 20.1, and LS1 is updated to [20.05, 19.8, 19.9, 20.2] (number = 4 < 5).

[0057] Write in new features: after the 6th coating, the 1st gap sub-segment has a blank length of 19.7 mm, and the write-in LS1 is [20.05, 19.8, 19.9, 20.2, 19.7] (number = 5 = set control length), complete the dimension reduction of the gap distribution feature array of the 1st gap sub-segment, and the array is [20.05, 19.8, 19.9, 20.2, 19.7]. Other gap sub-segments (LS2, LS3…) are independently processed according to the same logic to ensure that the length of the gap distribution feature array of each sub-segment is always 5.

[0058] Through the above processing, the server dynamically maintains the coating distribution feature array and the length of the gap distribution feature array of each gap sub-segment to be the set control length of 5, which not only retains the key feature information, but also avoids the expansion of the number of features, thereby reducing the efficiency of subsequent gating fusion, and laying a data foundation for accurate generation of coating control instructions.

[0059] In the embodiments of the present application, before the first gating fusion feature between the aggregation characteristic and each coating distribution feature in the coating distribution feature array and the second gating fusion feature between the aggregation characteristic and each gap distribution feature in the gap distribution feature array are extracted respectively, the following implementation is provided.

[0060] Determine that the set control node is reached, read the coating distribution feature array stored in the coating global storage unit, and read the gap distribution feature array stored in the gap local storage unit.

[0061] In the embodiment of the present application, the server needs to trigger the feature array reading through the preset regulation node before executing the double feature gating mechanism, to ensure that the feature data of the input gating unit meets the set regulation length and is stable and effective. The target substrate is an aluminum foil with a primer, and the set regulation length is 5. The trigger condition of the regulation node is dynamically determined based on the feature state of the coating global storage unit and the gap local storage unit. The specific process is as follows: The regulation node determination rule of the server configuration is: “when the coating global storage unit completes the kth feature aggregation and writes new features, the stored coating distribution feature quantity is stable at the set regulation length of 5, and all enabled gap local storage units (corresponding to the appeared gap sub-sections) complete at least one feature aggregation and maintain 5 features stably, the regulation node is triggered”. When the number of gap sub-sections is ≥3 times, the server determines that it is a ‘persistent sub-section’ and enables the corresponding gap local storage unit, to ensure that the node trigger condition is quantifiable.

[0062] Taking the target substrate coating process as an example: after the server completes the 6th coating section coating, the coating global storage unit is “written with the 6th coating section feature (150.5mm) -> the storage unit feature quantity is restored to 5 ([149.5, 152, 151, 153, 150.5])”, and the local storage units (LS1 to LS5) of the first five gap sub-sections (1st to 5th gap sub-sections) have all completed the first feature aggregation (such as LS1 stores [20.05, 19.8, 19.9, 20.2, 19.7], LS2 stores [19.9, 20.1, 19.7, 20.0, 19.8], etc., all of which are 5 features). At this time, the server confirms through the state detection module that the coating global storage unit feature quantity = 5, all gap local storage unit feature quantities = 5, and there are no temporary features (such as redundant features to be aggregated) that have not completed dimension reduction, and it is determined that the set regulation node is reached.

[0063] After the server triggers the regulation node, the feature array of the coating global storage unit and each gap local storage unit is read synchronously through the feature reading interface: Read the coating distribution feature array: the coating global storage unit outputs the currently stored 5 coating distribution features to the input buffer of the gating feature encoder through the cache interface (transmission rate 1 GB / s), and the array is [149.5, 152, 151, 153, 150.5] mm (including the aggregated 1 / 3 coating section feature 149.5 mm, the original 2nd coating section 152 mm, etc.); Read the gap distribution characteristic array: each gap local storage unit (LS1 to LS5) outputs the characteristic array in turn according to the sub-section number, for example, LS1 outputs [20.05, 19.8, 19.9, 20.2, 19.7] mm for the first gap sub-section, LS2 outputs [19.9, 20.1, 19.7, 20.0, 19.8] mm for the second gap sub-section, and so on. After the verification (characteristic dimension 5x1, numerical range within 19.5-20.5 mm), all arrays are written into the buffer area.

[0064] After reading is completed, the server sends a "feature ready" signal to the gating feature encoder, triggering the start of the double-feature gating mechanism to ensure that subsequent feature fusion is based on the latest and stable feature array, providing a data basis for accurate generation of coating control instructions.

[0065] In the embodiments of the present application, the to-be-aggregated features of at least two coating sections are aggregated by any of the following methods: Based on the feature similarity, the to-be-aggregated features of the continuous coating sections with the closest parameters in the storage unit are calculated by the mean value; Based on the feature similarity, the to-be-aggregated features to be written are grouped and classified with the to-be-aggregated features already written in the storage unit, and the to-be-aggregated features of each coating section after grouping and classification are calculated by the mean value; Wherein, the to-be-aggregated features are coating distribution features or gap distribution features, and correspondingly, the storage unit is a coating global storage unit or a gap local storage unit.

[0066] In the embodiments of the present application, for example, when the server aggregates the to-be-aggregated features (coating distribution features or gap distribution features) in the coating global storage unit or the gap local storage unit, two typical methods are used based on the feature similarity to ensure that the aggregated features retain the core control information and the dimension conforms to the set control length. The following will be described in combination with specific scenarios: Application scenario: the coating global storage unit stores the coating distribution characteristics (coating length) of continuous coating segments, and the continuously aggregated characteristics with the closest parameters need to be aggregated. For example, the coating global storage unit stores 5 continuous coating segment characteristics after the 5th coating: [150, 152, 151, 153, 154] (unit: mm, all are the coating lengths of continuous coating segments). The server calculates the Euclidean distance (parameter proximity) of all continuous characteristic pairs: 150 and 152 distance = 2, 152 and 151 distance = 1 (minimum), 151 and 153 distance = 2, 153 and 154 distance = 1. The continuously coated segments with the closest parameters are the 2nd and 3rd coating segments (152 mm and 151 mm), and the server performs mean aggregation on the two: (152+151) / 2 = 151.5 mm, deletes the original 152 and 151, and the storage unit is updated to [150, 151.5, 153, 154]. Then, a new coating segment characteristic 150.5 mm is written, and the 5-feature array [150, 151.5, 153, 154, 150.5] is restored.

[0067] Application scenario: the gap local storage unit (such as LSI of the 1st gap subsegment) stores the gap distribution characteristics (gutter length) to be written, which needs to be aggregated after grouping with the written characteristics. For example, LSI has stored 4 characteristics: [19.8, 20.0, 20.2, 19.9] (unit: mm, all are the gutter lengths of the 1st gap subsegment), and a new characteristic 20.1 mm is to be written. The server presets 3 characteristic grouping intervals: [19.7-19.9] (low interval), [19.9-20.1] (middle interval), and [20.1-20.3] (high interval). Among the written characteristics, 19.8 and 19.9 belong to the low interval (average 19.85), 20.0 belongs to the middle interval, and 20.2 belongs to the high interval; the to-be-written 20.1 mm belongs to the middle interval and is grouped with the written 20.0 mm, and mean aggregation is performed on the group: (20.0+20.1) / 2 = 20.05 mm, replacing the original 20.0 mm, the storage unit is updated to [19.85, 20.05, 20.2], and then a new characteristic 19.7 mm (low interval) is written, the low interval characteristics become [19.85, 19.7], and the average (19.85+19.7) / 2 = 19.775 mm. Finally, the LSI characteristic array is [19.775, 20.05, 20.2], and after supplementing the new characteristic, 5 characteristics are maintained, and the grouping and classification aggregation is completed.

[0068] In the embodiments of the present application, if there are multiple groups of continuously coated segments with the closest parameters, the mean calculation of the to-be-aggregated characteristics of the continuously coated segments with the closest parameters in the storage unit can be implemented through the following examples.

[0069] performing mean value calculation on the to-be-aggregated features of the continuous coating sections closest in parameters in the storage unit; respectively performing mean value calculation on the to-be-aggregated features of each group of continuous coating sections closest in parameters in the storage unit.

[0070] In the embodiments of the present application, when the server in the storage unit has multiple groups of to-be-aggregated features of continuous coating sections closest in parameters, the server performs mean value calculation through two strategies to ensure that the features after aggregation retain the time sequence correlation and meet the set control length. The following describes the scenario of the coating global storage unit: The coating global storage unit stores the coating distribution features (coating length) of five continuous coating sections: [150, 151, 152, 151, 150] mm (arranged in the order of writing, and the parameters are the first to the fifth coating sections in turn). The server calculates the Euclidean distance (parameter proximity) of all pairs of continuous features: (150, 151) = 1, (151, 152) = 1, (152, 151) = 1, (151, 150) = 1, a total of 4 groups of parameters closest (distance is 1). At this time, according to the "first written" principle, the first group of continuous coating sections (the first and second coating sections, 150 mm and 151 mm) is selected to perform mean value aggregation: (150+151) / 2=150.5 mm, the original 150 and 151 are deleted, and the storage unit is updated to [150.5, 152, 151, 150], and then the new feature 153 mm is written, and the five-feature array [150.5, 152, 151, 150, 153] is restored.

[0071] The same storage unit feature array [150, 151, 152, 151, 150] mm, the distance of 4 groups of continuous feature pairs is 1 (parameters closest). According to the "separate aggregation" principle, the server independently performs mean value calculation on each group of continuous coating sections: the first group (150, 151) has a mean value of 150.5 mm, the second group (151, 152) has a mean value of 151.5 mm, the third group (152, 151) has a mean value of 151.5 mm, and the fourth group (151, 150) has a mean value of 150.5 mm. The original five features are deleted, and the four aggregated features [150.5, 151.5, 151.5, 150.5] are retained, and after writing the new feature 153 mm, the storage unit feature array is [150.5, 151.5, 151.5, 150.5, 153] (length 5), and the multiple group aggregation is completed.

[0072] In the embodiments of the present application, the target substrate is a foil substrate with a primer; the data of each coating section in the target substrate is collected respectively to obtain the corresponding coating distribution features and gap distribution features, including: The high-precision color scale color sensor detects the three different color regions of the foil base material, i.e. the blank foil region, the primer region and the coating region, and collects the primer data and the blank data of the current coating section; Based on the primer data and the blank data, the coating length feature of the coating section and the blank length feature of the gap sub-section are obtained, and the coating length feature is taken as the coating distribution feature, and the blank length feature is taken as the gap distribution feature.

[0073] In the embodiment of the present application, for example, the server detects the boundaries of the three color regions of the empty foil, the primer and the coating layer of the foil substrate with primer (such as the aluminum foil for lithium ion battery pole piece, thickness 12 pm, primer is carbon black conductive layer) by high-precision color sensor, collects key data and extracts coating distribution characteristics and gap distribution characteristics, and the specific process is as follows: the server controls the coating machine to transmit the foil substrate at a speed of 0.5 m / s, and synchronously starts the high-precision color sensor (resolution 0.01 mm, sampling frequency 1 kHz, spectral response range 400-700 nm) deployed in the detection station. The sensor performs line scanning on the surface of the substrate, and distinguishes the three regions by RGB color value: the empty foil region (aluminum foil original color, silver white color) outputs RGB value (255, 255, 240), the primer region (carbon black layer, black color) outputs (30, 30, 30), and the coating layer region (active material, dark gray color) outputs (100, 100, 100). The sensor outputs the RGB value of one sampling point every 1 ms, and the server receives and stores it in real time through the data acquisition module. For example, when the substrate passes through the detection area of the sensor, the sampling sequence appears in turn: (30, 30, 30) (primer) → (100, 100, 100) (coating layer start point) → (100, 100, 100) (coating layer region) → (30, 30, 30) (coating layer end point / gap start point) → (30, 30, 30) (gap region) → (100, 100, 100) (next coating layer start point). The server identifies the boundaries of “primer-coating layer” and “coating layer-primer” based on the color jump points, which correspond to the start point and end point of the coating section respectively; the boundaries of “coating layer-primer” and “primer-coating layer” correspond to the start point and end point of the gap sub-section, and accordingly the primer data (the length of the primer region before the coating layer start point) and the white space data (the length of the primer region of the gap sub-section) of the current coating section are collected. The server calculates the length of the collected primer data and white space data to generate the coating distribution characteristics and gap distribution characteristics: coating length characteristics (coating distribution characteristics): the start point of the coating section is the color jump point of “primer-coating layer” (RGB jumps from (30, 30, 30) to (100, 100, 100)), and the end point is the jump point of “coating layer-primer” (RGB jumps from (100, 100, 100) to (30, 30, 30)). The server calculates the distance between the two points according to the sampling frequency and the substrate speed: for example, the start point sampling time t1=100 ms, the end point sampling time t2=400 ms, the sampling interval Δt=300 ms, the substrate speed v=0.5 m / s=500 mm / s, and the coating length L=v×Δt=500 mm / s×0.3 s=150 mm, which is the coating length characteristics (coating distribution characteristics) of the current coating section.The white space length feature (gap distribution feature): the gap sub-section starts at the end of the coating section (t2=400 ms), and ends at the start of the next coating section (t3=440 ms), the sampling interval Δt=40 ms, the white space length L=500 mm / s*0.04 s=20 mm, which is the white space length feature (gap distribution feature) of the current gap sub-section. For example, the "bottom coating→coating layer" jump point of the 2nd coating section is at t=500 ms, the "coating layer→bottom coating" jump point is at t=804 ms, Δt=304 ms, the coating length=500*0.304=152 mm; the corresponding gap sub-section starts at t=804 ms, and ends at t=843.6 ms, Δt=39.6 ms, the white space length=500*0.0396=19.8 mm. The server stores 152 mm as the coating distribution feature of the 2nd coating section, and 19.8 mm as the gap distribution feature of the corresponding gap sub-section to the corresponding storage unit.

[0074] The embodiment of the present application provides a computer device 100, which comprises a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the automatic coating method described above. As shown in the figure, Figure 2 Figure 2 The structure block diagram of the computer device 100 provided by the embodiment of the present application. The computer device 100 comprises a memory 111, a processor 112 and a communication unit 113. In order to realize the transmission or interaction of data, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected with each other. For example, the electrical connection between these elements can be realized by one or more communication buses or signal lines.

[0075] The foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teachings. The embodiments are chosen and described in order to best explain the principles of the disclosure and its practical application to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated.​

Claims

1. An automated coating method, characterized in that, include: Based on the multi-segment coating control model, data is collected from each coating segment in the target substrate to obtain the corresponding coating distribution features and gap distribution features. Based on feature similarity, the obtained coating distribution feature set and gap distribution feature set are dimensionality reduced to obtain coating distribution feature array and gap distribution feature array with set control length. After merging the pre-trained coating retrieval features and gap retrieval features, a cohesive mechanism is used to extract the corresponding aggregation characteristics. The coating retrieval features and the gap retrieval features are obtained by training the corresponding adjustable retrieval features based on typical coating samples. The aggregation characteristics integrate the core elements of the coating retrieval features and the gap retrieval features. A dual-feature gating mechanism is adopted to extract the first gating fusion feature between the polymerization characteristic and each coating distribution feature in the coating distribution feature array, and to extract the second gating fusion feature between the polymerization characteristic and each gap distribution feature in the gap distribution feature array. Based on the acquired first-gated fusion features and second-gated fusion features, the multi-segment intermittent coating control of the target substrate is controlled.

2. The method according to claim 1, characterized in that, The multi-segment coating control model includes a gated feature encoder; the adoption of a dual-feature gating mechanism, extracting the first gated fusion feature between the aggregation characteristic and each coating distribution feature in the coating distribution feature array, and extracting the second gated fusion feature between the aggregation characteristic and each gap distribution feature in the gap distribution feature array, includes: The polymerization characteristics are broken down into a first polymerization characteristic corresponding to the coating distribution characteristics and a second polymerization characteristic corresponding to the gap distribution characteristics; The first gating unit in the gated feature encoder extracts the first aggregated feature and the first gating fusion feature between each coating distribution feature in the coating distribution feature array based on the dual feature gating mechanism. The second gating unit in the gated feature encoder extracts the second aggregated feature and the second gating fusion feature between each gap distribution feature in the gap distribution feature array based on the dual feature gating mechanism.

3. The method according to claim 2, characterized in that, The gated feature encoder includes gated feature processors connected in sequence; for the first gated feature processor, the currently processed features include the coating distribution feature array and the gap distribution feature array; for other gated feature processors, the currently processed features include each first gated fusion feature and each second gated fusion feature output by the previous gated feature processor. The control of multi-segment intermittent coating of the target substrate based on the acquired first-gated fusion features and second-gated fusion features includes: Based on the first gated fusion features and the second gated fusion features output by the last gated feature processor, the multi-segment intermittent coating control of the target substrate is controlled.

4. The method according to claim 1, characterized in that, The multi-segment coating control model includes a coating parameter generation model; the multi-segment intermittent coating control of the target substrate based on the acquired first-gated fusion features and second-gated fusion features includes: After merging the first gated fusion features and the second gated fusion features, they are converted into coating data features that are located in the same feature domain as the coating process parameters of the target substrate. The coating process description features and coating control parameters of the target substrate are used as coating parameter guidance information, and the coating data features are combined with the coating parameter generation model to generate multi-segment intermittent coating control instructions for the target substrate.

5. The method according to any one of claims 1 to 4, characterized in that, Each gap distribution feature corresponds to a gap sub-segment in the coating segment; based on feature similarity, the obtained coating distribution feature set and gap distribution feature set are respectively subjected to dimensionality reduction processing to obtain a coating distribution feature array and a gap distribution feature array with a set adjustable length, including: For each coating segment collected, the coating distribution characteristics of the coating segment are written into the global coating storage unit, and the gap distribution characteristics of the coating segment are written into the gap local storage unit. When the number of coating distribution features stored in the global coating storage unit reaches the set control length, based on feature similarity, feature aggregation is performed on the coating distribution features of at least two coating segments in the global coating storage unit, and the coating distribution features to be written are written into the global coating storage unit. Furthermore, for each gap segment, when the number of corresponding gap distribution features stored in the gap local storage unit reaches the set control length, based on feature similarity, feature aggregation is performed on the corresponding gap distribution features of at least two coating segments in the gap local storage unit, and the gap distribution features to be written are written into the gap local storage unit.

6. The method according to claim 5, characterized in that, Before employing a dual-feature gating mechanism to extract the first gated fusion feature between the aggregation characteristic and each coating distribution feature in the coating distribution feature array, and before extracting the second gated fusion feature between the aggregation characteristic and each gap distribution feature in the gap distribution feature array, the method further includes: Once the set control node is determined to be reached, the coating distribution feature array stored in the global coating storage unit and the gap distribution feature array stored in the gap local storage unit are read.

7. The method according to claim 5, characterized in that, Feature aggregation is performed on the features to be aggregated from at least two coating segments using any of the following methods: Based on feature similarity, the average value of the features to be aggregated in the continuous coating segments with the closest parameters in the storage unit is calculated. Based on feature similarity, the features to be written and aggregated are grouped and classified with the features to be written and aggregated in the storage unit, and the average value of the features to be aggregated in each grouped and classified coating segment is calculated. Wherein, the feature to be aggregated is a coating distribution feature or a gap distribution feature, and correspondingly, the storage unit is a coating global storage unit or a gap local storage unit.

8. The method according to claim 7, characterized in that, If multiple sets of consecutive coated segments with the closest parameters exist, the step of averaging the features to be aggregated in the storage unit for the consecutive coated segments with the closest parameters includes: The average value of the features to be aggregated in the consecutive coated segments that were written first and have the closest parameters in the storage unit is calculated; or The average value of the features to be aggregated for each group of consecutive coated segments in the storage unit with the closest parameters is calculated.

9. The method according to claim 1, characterized in that, The target substrate is a foil substrate with a primer coating; the data acquisition of each coated segment in the target substrate to obtain the corresponding coating distribution characteristics and gap distribution characteristics includes: A high-precision color mark sensor is used to detect three different color areas on the foil substrate: the empty foil area, the base coating area, and the coating area, to collect the base coating data and blanking data of the current coating section. Based on the base coating data and the blanking data, the coating length characteristics of the coating segment and the blanking length characteristics of the gap segment are obtained, and the coating length characteristics are used as the coating distribution characteristics, and the blanking length characteristics are used as the gap distribution characteristics.

10. A server system, characterized in that, Includes a server, the server being used to perform the method according to any one of claims 1-9.

Images