Intelligent optimization method and system for coating process based on multi-modal data fusion

Abstract

The application discloses a coating process intelligent optimization method and system based on multi-modal data fusion, and relates to the technical field of intelligent control. The method comprises the following steps: collecting a multi-modal physical field tensor in a wet film stage; deploying a memristor array; receiving the multi-modal physical field tensor and converting it into a parallel voltage vector according to a front-end analog circuit of the memristor array, loading the parallel voltage vector into the memristor array, and determining a dry film quality prediction result; when the dry film quality prediction result does not meet an expected quality standard, triggering an optimization decision unit to perform iterative optimization, determining process characteristic values and converting the process characteristic values into analog control signals, and responding to an end effector of a coating production line to perform multi-thread regulation and control. The technical problems that the coating process lacks an effective feedforward regulation and control mechanism based on process states in the prior art, and coating quality fluctuations are difficult to suppress are solved, the dry film quality in the wet film stage is predicted in advance, and the coating process parameters are optimized and regulated in real time, thereby achieving the technical effect of improving the coating quality stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent optimization method and system for coating processes based on multimodal data fusion. Background Technology

[0002] Coating processes are widely used in industrial production processes such as lithium battery electrode manufacturing, optical thin film manufacturing, and functional coating material preparation. Coating quality directly affects the thickness uniformity, surface quality, and final product performance. During the coating process, the coating typically undergoes several stages, including wet film formation, leveling, solvent evaporation, and drying and curing. Dry film quality is a crucial indicator for evaluating the stability of the coating process and product quality. In existing technologies, coating process control usually relies on empirical parameter settings, open-loop control, or feedback adjustment based on dry film detection results. This involves obtaining dry film quality information through thickness and defect detection after coating is completed and dried, and then adjusting subsequent production parameters. This type of control is a post-processing feedback adjustment, which suffers from adjustment lag and long adjustment cycles. When coating conditions fluctuate, it is difficult to correct process parameters in a timely manner, easily causing fluctuations in coating quality and affecting product consistency and yield. Meanwhile, during the wet film stage, there are multiple physical fields such as thickness field, temperature field, solvent concentration field, and tension field that are coupled together. There is a complex nonlinear relationship between the wet film state and the final dry film quality. Existing technologies cannot accurately predict the dry film quality during the wet film stage, and therefore it is difficult to establish a feedforward control mechanism based on the wet film state during the coating process. As a result, the coating process mainly relies on post-process feedback adjustment, and the stability of coating quality still needs to be improved. Summary of the Invention

[0003] This application provides a method and system for intelligent optimization of coating process based on multimodal data fusion, which solves the technical problem that the coating process lacks an effective feedforward control mechanism based on process state in the prior art, resulting in difficulty in suppressing coating quality fluctuations.

[0004] The first aspect of this application provides an intelligent optimization method for coating processes based on multimodal data fusion, the method comprising: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor of the wet film stage. For the nonlinear mapping between the wet film and dry film stages, a memristor array is deployed at the edge of the coating production line. The memristor units at each intersection are deployed using a weight matrix based on wet film characteristics and dry film quality, employing convergent encoding based on the target conductance value. The front-end analog circuit of the memristor array receives the multimodal physical field tensor, converts it into a parallel voltage vector, loads it onto the memristor array, and determines the dry film quality prediction result. When the dry film quality prediction result does not meet the expected quality standard, an optimization decision unit is triggered to perform iterative optimization, determine process characteristic values, and convert them into analog control signals, responding to the end effector of the coating production line for multi-threaded control.

[0005] A second aspect of this application provides an intelligent optimization system for coating processes based on multimodal data fusion, the system comprising: Data Acquisition Module: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to acquire the multimodal physical field tensor of the wet film stage; Memristor Deployment Module: For the nonlinear mapping between the wet film and dry film stages, a memristor array is deployed at the edge of the coating production line. The memristor units at each intersection are deployed using a weight matrix based on wet film characteristics and dry film quality, employing convergent encoding based on the target conductance value; Quality Prediction Module: According to the front-end analog circuit of the memristor array, the multimodal physical field tensor is received and converted into a parallel voltage vector, which is then loaded onto the memristor array to determine the dry film quality prediction result; Control Module: When the dry film quality prediction result does not meet the expected quality standard, the optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values, and convert them into analog control signals, responding to the end effector of the coating production line for multi-threaded control.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, for the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor of the wet film stage. Next, to address the nonlinear mapping between the wet and dry film stages, a memristor array is deployed at the edge of the coating production line. A weight matrix based on wet film characteristics and dry film quality is used to perform convergent encoding based on the target conductance value for the memristor units at each intersection point. Then, according to the front-end analog circuit of the memristor array, the multimodal physical field tensor is received and converted into a parallel voltage vector, which is then loaded onto the memristor array to determine the predicted dry film quality. Finally, when the predicted dry film quality does not meet the expected quality standard, an optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values, and convert them into analog control signals, responding to the end effector of the coating production line for multi-threaded control. This solves the technical problem in existing technologies where the coating process lacks an effective feedforward control mechanism based on process state, leading to difficulty in suppressing coating quality fluctuations. It achieves the technical effect of predicting the dry film quality in advance during the wet film stage and driving real-time optimization and control of coating process parameters, thereby improving the stability of coating quality. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic diagram of the intelligent optimization method for coating process based on multimodal data fusion provided in this application embodiment; Figure 2 This is a schematic diagram of the intelligent optimization system structure for coating process based on multimodal data fusion provided in an embodiment of this application.

[0009] Figure labeling: Data acquisition module 11, memristor deployment module 12, quality prediction module 13, control module 14. Detailed Implementation

[0010] 1. To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose of the invention, the following detailed description of the specific implementation methods, structures, features and effects of the present invention is provided in conjunction with the accompanying drawings and preferred embodiments.

[0011] Example 1, as Figure 1 As shown, this application provides an intelligent optimization method for coating processes based on multimodal data fusion, wherein the method includes: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor during the wet film stage.

[0012] In this embodiment of the application, for a coating production line, a multimodal sensor array is deployed along the wet film transport path between the coating head outlet and the drying inlet to perform online detection of the short-term evolution state of the wet film after formation. The multimodal sensor array is sequentially arranged along the substrate running direction within a preset detection interval from the coating head outlet. This preset detection interval preferably covers the initial spreading area, leveling transition area, and stable transport area before drying. The multimodal sensor array includes at least a terahertz detection unit for detecting the wet film thickness and internal medium distribution, an infrared thermal imaging unit for detecting the surface thermal state, and a laser speckle imaging unit for detecting the surface flow morphology and rheological stability. Each detection unit is connected via a unified clock. The trigger module performs synchronous sampling and establishes a time-position correspondence with the substrate displacement encoder through the edge acquisition controller, ensuring that each detection data point is matched one-to-one with the spatial position of the same wet film area. During the specific acquisition process, when the substrate carrying the wet film passes through the detection zone, the edge acquisition controller triggers each detection unit to work synchronously according to a preset sampling cycle. Specifically, the terahertz detection unit outputs thickness amplitude information and solvent dielectric response information at different locations on the wet film, the infrared thermal imaging unit outputs two-dimensional temperature distribution information on the wet film surface, and the laser speckle imaging unit outputs a dynamic image sequence of speckle on the wet film surface. Subsequently, denoising and drift correction processing is performed on the acquired raw multi-source data to obtain the corresponding raw feature data. Further... Next, spatial registration processing is performed on the raw feature data output by each detection unit. A unified two-dimensional spatial coordinate system is established with the substrate running direction as the X-axis and the coating width direction as the Y-axis. Based on the rotation matrix and translation vector obtained by each detection unit through standard calibration parts during the installation stage, the extrinsic parameter matrix of each detection unit relative to the unified two-dimensional spatial coordinate system is established. The terahertz detection coordinates, infrared thermal imaging coordinates, and laser speckle imaging coordinates are uniformly mapped to the two-dimensional spatial coordinate system. A unified spatial grid is established according to a preset grid size and used as a common reference coordinate. For data with a spatial resolution lower than the unified spatial grid, bilinear interpolation is used for spatial resampling. Similarly, for data with a spatial resolution higher than the unified spatial grid, a regional averaging method is used for downsampling, so that the data of each detection channel corresponds to the same spatial grid node. After spatial registration is completed, a unified sampling time series is established using the unified timestamp output by the edge acquisition controller as the time reference. For detection channels with a sampling frequency lower than the unified sampling frequency, linear interpolation is used to supplement missing sampling points. For detection channels with a sampling frequency higher than the unified sampling frequency, a sliding window averaging process is performed according to the unified sampling period, so that each detection channel has a corresponding data value at the same sampling time, thereby realizing the time synchronization of multi-channel data. The unified sampling frequency is determined according to the maximum common sampling period of the sampling frequencies of each detection unit.After time synchronization is completed, the corresponding historical calibration ranges of the thickness channel, solvent content channel, surface temperature channel, evaporation rate channel, speckle contrast channel and flow stability channel are read respectively, and the scale is uniformly processed by the minimum-maximum normalization method. The calculation formula is: ,in, This indicates the data value of the current detection channel. This indicates the historical minimum calibration value of the detection channel. This indicates the historical maximum calibration value of the detection channel. The normalized dimensionless eigenvalues ​​represent the data from each detection channel after normalization, which converts the data to a uniform numerical range to eliminate dimensional differences between different physical quantities. After completing spatial registration, time synchronization, and scale unification, the data is constructed using the position index corresponding to the unified spatial grid as the first dimension, the unified sampling time series as the second dimension, and the normalized thickness features, solvent content features, surface temperature features, evaporation rate features, speckle contrast features, and flow stability features as the third dimension. Tensors are constructed for each dimension of data according to the organization method of spatial position × sampling time × feature channel, forming a multimodal physical field tensor corresponding to the wet film stage. The multimodal physical field tensor is then output to the front-end analog circuit of the memristor array as the data input for subsequent dry film quality prediction.

[0013] Furthermore, the acquisition of multimodal physical field tensors during the wet film stage includes: The thickness distribution and solvent content distribution of the wet film are determined by emitting broadband terahertz pulses onto the wet film surface using a spectral probe and receiving the reflected echoes. The temperature field of the wet film surface is acquired by an infrared thermal imager, and the solvent evaporation rate and droplet adhesion rate change rate are extracted by differential processing. The dynamic speckle pattern of the wet film surface is captured by a laser speckle imaging device, and the rheological state and flow stability of the wet film surface are determined by speckle contrast and time correlation analysis.

[0014] Preferably, a broadband terahertz pulse signal is emitted to the wet film surface via a broadband terahertz probe in the spectral detection unit, and echo signals formed by reflections from the wet film surface and internal medium are received. Time-domain analysis is performed on the echo signals, and the wet film thickness distribution is determined based on the time delay between the emitted pulse and the interface reflection echo. Specifically, the wet film thickness at corresponding locations is calculated using the time interval between the air-wet film interface reflection signal and the wet film-substrate interface reflection signal. ,in, Indicates spatial location The thickness of the wet film at that location; This indicates the propagation speed of terahertz waves in a vacuum; This indicates the time difference between two interface reflections at the corresponding location; This represents the refractive index of the wet film medium. Further, based on the amplitude attenuation and phase delay characteristics of terahertz echo signals at different frequencies, the spectral response parameters of the wet film medium are extracted, and based on a pre-established dielectric property calibration relationship, these spectral response parameters are converted into solvent content distribution. Specifically, the absorption coefficient of the terahertz signal at different frequencies is calculated: ,in, Indicates spatial location frequency The corresponding terahertz absorption coefficient; Indicates the amplitude of the reflected echo from the wet film; This represents the reference amplitude under conditions without a wet film. This indicates the wet film thickness at the corresponding location. Based on the calibration relationship between the absorption coefficient and solvent content: ,in, Indicates spatial location Solvent content; This represents the spectral response mapping function obtained based on experimental calibration.

[0015] Preferably, an infrared thermal imager positioned above the wet film transmission area continuously acquires infrared radiation from the wet film surface to obtain a surface temperature matrix corresponding to different sampling times. For each sampling time, the infrared radiation signal output by the infrared thermal imaging unit is converted into temperature data according to a preset temperature calibration relationship to establish a two-dimensional temperature field on the wet film surface. ,in, Indicates the spatial position of the wet film at time t. Surface temperature value at; This indicates the intensity of infrared radiation collected by the infrared thermal imaging unit; This represents the calibration conversion function between infrared radiation intensity and temperature. Furthermore, time-difference calculations are performed on the temperature field corresponding to continuous sampling times to obtain the temperature change rate at various locations on the wet film surface: ,in, Indicates spatial location The rate of temperature change on the surface of the wet film; This indicates the temperature value at the next sampling time; This indicates the temperature value at the current sampling time; This indicates the time interval between adjacent sampling.

[0016] Since the solvent evaporation process in the wet film is accompanied by the absorption of latent heat of vaporization, a temperature change corresponding to the amount of solvent loss occurs in the wet film region. Therefore, a calibration relationship between the temperature change rate and the solvent evaporation rate is established before formal testing. Specifically, while maintaining consistency in the coating material system, substrate type, ambient temperature and humidity, and drying conditions, multiple wet film evaporation stages are selected as calibration samples. Infrared temperature change data are collected synchronously at each calibration time, and the amount of solvent loss in the wet film within the corresponding time period is obtained by weighing. The actual solvent evaporation rate is calculated based on the change in solvent mass per unit time. ,in, This indicates the actual solvent evaporation rate measured during the calibration phase; Indicates time Wet film sample quality; Indicates time Wet film sample quality; Indicates the sampling time interval.

[0017] Furthermore, the temperature change rate data obtained during the calibration phase are correlated with the corresponding actual solvent evaporation rate data. A mapping relationship between the temperature change rate and the solvent evaporation rate is established based on multiple calibration samples. The evaporation rate conversion coefficient is determined through linear fitting, resulting in the solvent evaporation rate calculation function. ,in, Indicates spatial location The solvent evaporation rate corresponding to the wet film; This indicates the rate of temperature change on the wet film surface at the corresponding location; This represents the conversion coefficient of the rate of temperature change obtained by fitting the calibration sample; This indicates the amount of compensation correction caused by factors such as the thermal response of the substrate and environmental heat dissipation.

[0018] In the actual online detection process, the wet film temperature field acquired in real time by the infrared thermal imaging unit is input into the solvent evaporation rate calculation function. The solvent evaporation rate at each spatial location is calculated based on the temperature change rate at each location, thereby obtaining the solvent evaporation rate distribution within the wet film area.

[0019] Furthermore, based on the changes in local temperature anomaly regions within the wet film surface temperature field, the rate of change in droplet adhesion state is determined. Specifically, temperature gradients are calculated on continuous temperature images, and regions where the temperature gradient exceeds a preset threshold and persists are identified as candidate droplet adhesion regions. The area ratio of droplet anomaly regions at each sampling time is then statistically analyzed. ,in, Indicates time The percentage of the area of ​​the abnormal region of the droplet; This represents the number of pixels in the abnormal temperature region detected at the current moment; N represents the total number of pixels in the wet film detection area.

[0020] The rate of change of droplet adhesion rate is calculated based on the change in droplet region area at consecutive sampling times: ,in, This indicates the rate of change of droplet adhesion rate; This indicates the percentage of the droplet's abnormal region area at the next sampling time; This indicates the percentage of the droplet's abnormal region area at the current sampling time.

[0021] Preferably, a laser speckle imaging unit positioned above the wet film transmission area projects a coherent laser beam onto the wet film surface, causing the laser to scatter in the microstructure region of the wet film surface. An image acquisition module then continuously acquires a sequence of speckle images of the wet film surface at a preset sampling frequency. Since particle migration, liquid flow, and surface morphology changes within the wet film cause dynamic changes in the speckle intensity distribution, the acquired continuous speckle images are converted to grayscale to obtain the speckle intensity matrix corresponding to each sampling time. ,in, Indicates time Spatiotemporal location The speckle light intensity value at that location; This represents the raw image signal acquired by the laser speckle imaging device; This represents the image grayscale conversion function.

[0022] Furthermore, based on the degree of dispersion of light intensity distribution in a single-frame speckle image, speckle contrast parameters are calculated to characterize the degree of change in the microstructure of the wet film surface: ,in, Indicates time Corresponding speckle contrast; This represents the standard deviation of the light intensity values ​​in the current speckle image; This represents the average light intensity in the current speckle image. The correlation coefficient is obtained by calculating the correlation between speckle images from adjacent sampling times. ,in, This represents the speckle time correlation coefficient between adjacent time frames; This represents the speckle intensity matrix at the current moment; This represents the speckle intensity matrix at the next sampling time; This represents the average light intensity of the speckle pattern at the current moment; This indicates the average speckle intensity at the next sampling time; This indicates the time interval between adjacent sampling.

[0023] Furthermore, the degree of flow change on the wet film surface is calculated based on the changes in speckle contrast and the trend of the time correlation coefficient. When the flow inside the wet film intensifies, the rate of change in the liquid microstructure increases, leading to more severe changes in the speckle pattern, manifested as an increase in the amplitude of speckle contrast change and a decrease in the time correlation coefficient. When the wet film gradually levels out and stabilizes, the changes in the speckle pattern weaken, manifested as an increase in the time correlation coefficient and a decrease in speckle contrast fluctuation. Specifically, the dynamic change index of the speckle is calculated: ,in, Indicators representing dynamic changes on the wet film surface; This indicates the change in speckle contrast between adjacent sampling times; This represents the correlation coefficient of speckle time; This represents the corresponding feature weight coefficient. The rheological state parameters of the wet film surface are determined based on the speckle dynamic change index. Specifically, when the speckle dynamic change index increases, it indicates enhanced flow on the wet film surface, signifying that the wet film is in the spreading or leveling stage; when the speckle dynamic change index decreases and remains stable, it indicates weakened flow on the wet film surface, signifying that the wet film has entered a stable transport stage.

[0024] Furthermore, based on the degree of fluctuation of the speckle dynamic change index within a continuous time window, the wet film flow stability parameters are calculated: ,in, Indicates parameters related to the stability of wet film flow; The standard deviation of the speckle dynamic change index within the time window; This represents the preset normalized maximum dynamic change value. The stability of the wet film flow state is determined based on the flow stability parameter. When the flow stability parameter increases, it indicates that the flow on the wet film surface tends to be stable; when the flow stability parameter decreases, it indicates that there are still strong flow changes or local instability phenomena on the wet film surface.

[0025] To address the nonlinear mapping between the wet film stage and the dry film stage, a memristor array is deployed at the edge of the coating line. The memristor cells at each intersection are deployed by performing convergence encoding based on the target conductivity value using a weight matrix based on wet film characteristics and dry film quality.

[0026] Specifically, during the coating process, the multimodal physical field tensor of the wet film stage is acquired by a multimodal sensing array. This tensor is then expanded according to spatial location, sampling time, and physical feature channels to extract characteristic parameters representing the wet film state. These parameters include at least wet film thickness, solvent content, surface temperature, solvent evaporation rate, droplet adhesion rate, speckle contrast, and flow stability parameters. The characteristic parameters of different dimensions are then normalized to obtain the wet film feature vector. ,in, Represents the feature vector of the wet film; This represents the m-th wet film state feature; m represents the number of wet film features.

[0027] Further, dry film quality inspection data corresponding to the wet film feature vector is obtained, wherein the dry film quality inspection data is obtained by inspecting the coated film layer after drying, and includes at least the dry film thickness uniformity index, residual solvent content index, and surface defect rate index; the wet film feature vectors collected in the same coating area during the wet film stage are spatially matched and temporally correlated with the dry film quality inspection data after drying to establish a wet film-dry film correlation sample set: Where D represents the wet-dry film associated sample set; This represents the feature vector of the k-th wet film group; This represents the dry film quality index vector corresponding to the k-th group; N represents the number of samples.

[0028] Based on the aforementioned wet-film-dry-film associated sample set, a nonlinear mapping relationship is established from wet film features to dry film quality indicators. Specifically, the wet film feature vector is used as input, and the coupling relationship between different wet film physical features is calculated through a nonlinear feature mapping layer to obtain the implicit features of the wet film. ,in, This represents the implicit characteristics of the k-th group of wet films; This represents the wet film feature mapping weight matrix; Indicates the bias parameter; This represents a nonlinear mapping function.

[0029] Furthermore, the implicit features of the wet film are input into the dry film quality mapping layer to calculate the corresponding predicted dry film quality value: ,in, This indicates the predicted dry film quality indicators; The weight matrix represents the mapping from the implicit characteristics of the wet film to the quality indicators of the dry film. This represents the quality output bias parameter.

[0030] In the process of establishing the nonlinear mapping relationship, the difference between the predicted dry film quality index and the actual detected dry film quality index is used as the optimization basis to iteratively update the wet film feature mapping weight matrix and the quality output mapping weight matrix: Where L represents the mapping error loss value; This represents the actual dry film quality index of the k-th group; This represents the predicted dry film quality index for the k-th group.

[0031] Once the mapping error loss value reaches the preset convergence condition, the weight update stops, and the wet film feature-dry film quality weight matrix is ​​calculated based on the converged wet film feature mapping weight matrix and dry film quality mapping weight matrix: ,in, Represents the wet film characteristics-dry film quality weight matrix; This represents the wet film feature mapping weight matrix; This represents the quality output mapping weight matrix.

[0032] Furthermore, deploying memristor arrays at the edge of the coating production line includes: A memristor array with a cross-shaped structure is deployed, wherein M row lines and N column lines intersect perpendicularly, and each intersection point integrates a memristor unit; a weight matrix based on wet film characteristics and dry film quality is determined, the memristor array is initially encoded, and deployed at the edge of the coating production line to establish data interaction with the multimodal sensing end.

[0033] Furthermore, the initial encoding of the memristor array includes: Using wet-dry film paired data as sample data, multimodal wet film features and dry film quality indicators are extracted and a nonlinear mapping is established to determine an M×N dimensional weight matrix. For the weight matrix, the target conductance value of each intersection point of the memristor array is converted through the mapping transformation rule. The target conductance value is written into the memristor array to complete the initial encoding.

[0034] A cross-shaped memristor array is deployed at the edge of the coating line. The memristor array includes M row lines and N column lines, which intersect to form multiple intersection points. Each intersection point integrates a memristor unit. The conductivity state of each memristor unit is used to characterize a weight parameter in the wet film characteristic-dry film quality weight matrix. Based on the correspondence between the memristor array size and the dimensions of the wet film characteristic-dry film quality weight matrix, each element in the weight matrix is ​​mapped to the corresponding intersection point of the memristor array to obtain the target conductivity value of each memristor unit. ,in, Indicates the first The target conductance of the memristor cell at the crossover point; The first element in the wet film characteristic-dry film quality weight matrix represents the... Each weight value; and These represent the minimum and maximum weight values ​​in the weight matrix, respectively. and These represent the minimum and maximum adjustable conductance values ​​of the memristor unit, respectively.

[0035] Convergence coding is performed on each crossover memristor cell based on the target conductance value. Specifically, for any memristor cell at a target crossover point, the current conductance value of that memristor cell is first read. Calculate the conductance deviation between the current conductance value and the target conductance value: ,in, Indicates the first Conductivity deviation of the position memristor unit; Indicates the target conductivity value; This indicates the current conductance value. The programming direction is determined based on the conductance deviation. When the conductance deviation is greater than zero, the current conductance value is determined to be lower than the target conductance value. A conductance-boosting SET voltage pulse is applied to the corresponding memristor unit, causing the memristor's resistance state to change towards lower resistance, thus increasing the conductance value. When the conductance deviation is less than zero, the current conductance value is determined to be higher than the target conductance value. The memristor conductance value is reduced by adjusting the reset pulse. The pulse amplitude, pulse width, and number of pulse applications are dynamically adjusted based on the difference between the current and target conductance values. The conductance detection and pulse adjustment process is repeated until a preset convergence condition is met. ,in, This represents the preset conductance convergence error threshold. After all memristor units at all intersection points in the memristor array have converged to the target conductance value, the conductance state of each unit in the memristor array is made to correspond to the wet film characteristic-dry film quality weight matrix. The nonlinear mapping relationship between the multimodal physical state of the wet film stage and the dry film quality index is solidified into the memristor array, thereby forming a wet film-dry film quality prediction calculation unit deployed at the edge of the coating production line.

[0036] Preferably, a cross-shaped memristor array is deployed within the edge computing node of the coating production line. The memristor array includes M row lines arranged along a first direction and N column lines arranged along a second direction. The M row lines and N column lines intersect perpendicularly, and a memristor unit is integrated at each intersection point. The memristor unit stores weight parameters corresponding to the wet film characteristic-dry film quality mapping relationship. Based on the determined wet film characteristic-dry film quality weight matrix, initial encoding is performed on the memristor units corresponding to each intersection point in the memristor array, converting the weight parameters in the weight matrix into the conductance state of the corresponding memristor unit. This forms a computing structure corresponding to the nonlinear evolution relationship from the wet film stage to the dry film stage. The memristor array, after initial encoding, is deployed at the edge of the coating production line. A data interaction connection is established with a multimodal sensing end through an edge data interface, allowing the wet film stage physical field data acquired by the multimodal sensing end to be input into the memristor array. Based on the pre-encoded weight relationship, the mapping calculation from wet film state to dry film quality is performed.

[0037] Furthermore, the initial encoding of the memristor array includes: Using wet-dry film paired data as sample data, multimodal wet film features and dry film quality indicators are extracted and a nonlinear mapping is established to determine an M×N dimensional weight matrix. For the weight matrix, the target conductance value of each intersection point of the memristor array is converted through the mapping transformation rule. The target conductance value is written into the memristor array to complete the initial encoding.

[0038] Preferably, paired sample data corresponding to the wet film stage and the dry film stage are obtained, wherein the paired sample data includes wet film feature data acquired by the multimodal sensing array and dry film quality indicators detected after drying; based on the paired sample data, the correlation between wet film feature parameters and dry film quality indicators is extracted to determine a wet film feature-dry film quality weight matrix, wherein the row direction of the weight matrix corresponds to the wet film feature dimension of the input memristor array, and the column direction corresponds to the dry film quality indicator dimension of the output memristor array, forming an M×N dimensional weight matrix: ,in, Represents the wet film characteristics-dry film quality weight matrix; Indicates the first The wet film characteristics and the first Mapping weights between dry film quality indicators; Indicates the number of wet film features; This indicates the quantity of dry film quality indicators.

[0039] Furthermore, based on the correspondence between each weight parameter in the weight matrix and the position of the memristor array crossover point, a mapping conversion rule between the weight value and the memristor conductance state is established, and each weight parameter in the weight matrix is ​​converted into the target conductance value of the corresponding memristor unit, so that each weight parameter corresponds to a memristor crossover point unit. The target conductance value is used to characterize the influence weight of the corresponding wet film characteristics on the dry film quality index.

[0040] Furthermore, the target conductivity value is loaded into the memristor array programming control terminal, and the conductivity state of the memristor cells at each intersection position is adjusted according to the target conductivity value so that the actual conductivity state of each memristor cell matches the corresponding target conductivity value. After all memristor cells have completed the conductivity state writing, the initial encoding of the memristor array is completed, so that the memristor array stores the mapping relationship between wet film characteristics and dry film quality indicators, and is used for dry film quality prediction calculation under subsequent wet film stage data input.

[0041] Furthermore, the target conductance value is written into the memristor array to complete the initial encoding, including: For the first target crossover point, the current conductance value of the first memristor cell is identified, wherein the first target crossover point is any crossover point in the memristor array; based on the difference between the current conductance value and the first target conductance value, the first SET voltage pulse is determined by measuring the pulse duration; by applying the first SET voltage pulse to the first target crossover point, the conductance value of the first memristor cell converges to the first target conductance value.

[0042] Preferably, a first target crossover point is determined for any crossover point in the memristor array, and the first memristor cell corresponding to the first target crossover point is activated by row and column gating, and the current conductance value of the first memristor cell is read; the current conductance value is compared with the predetermined first target conductance value, and the current conductance deviation is calculated. ,in, This indicates the conductance deviation of the first memristor unit; Indicates the first target conductivity value; This indicates the current conductance value of the first memristor unit.

[0043] The parameters of the first SET voltage pulse are determined based on the conductance deviation. When the conductance deviation is greater than a preset gain threshold, it is determined that the current conductance value of the first memristor unit is lower than the first target conductance value, and the conductance state of the memristor unit needs to be increased. Based on the correspondence between the conductance deviation and the conductance change characteristics of the memristor unit, the required conductance increment is calculated, and the pulse amplitude and duration of the first SET voltage pulse are further determined. The pulse duration is determined based on the target conductance increment and the conductance change per unit time. ,in, Indicates the duration of the SET voltage pulse; This represents the difference in conductivity between the current conductivity value and the target conductivity value. This indicates the rate of change of conductance of the memristor cell under a set pulse condition.

[0044] Furthermore, based on the parameters of the first SET voltage pulse, a first SET voltage pulse is applied to the first target crossover point through the memristor array driving circuit, causing a change in the internal resistance state of the first memristor unit, thereby adjusting the conductance value; after the pulse application is completed, the current conductance value of the first memristor unit is reread, and it is determined whether the current conductance value meets the preset convergence condition. ,in, This indicates the preset conductance convergence error threshold.

[0045] When the current conductance value of the first memristor cell does not meet the convergence condition, the parameters of the next SET voltage pulse are recalculated based on the updated conductance deviation, and the process of "conductance reading - deviation calculation - pulse adjustment - convergence judgment" is executed cyclically until the conductance value of the first memristor cell converges to the first target conductance value. Then, conductance writing is performed on the memristor cells corresponding to other intersections in the memristor array in the same way, so that all memristor cells complete the target conductance state configuration and complete the initial encoding of the memristor array.

[0046] The multimodal physical field tensor is received and converted into a parallel voltage vector by the front-end analog circuit of the memristor array, and then loaded onto the memristor array to determine the dry film quality prediction result.

[0047] Furthermore, determining the predicted dry film quality includes: The front-end analog circuit of the memristor array receives the multimodal physical field tensor and converts it into an analog voltage amplitude proportional to the physical quantity; the analog voltage amplitude is sampled in a preset order to determine a parallel voltage vector; the parallel voltage vector is loaded onto M rows of the memristor array to output N columns of current; and the N columns of current are converted into dry film quality prediction results.

[0048] The front-end analog circuit of the memristor array receives the multimodal physical field tensor, performs analog quantity mapping on the channel features of the multimodal physical field tensor, and converts the feature values ​​corresponding to different physical quantities into analog voltage signals proportional to their amplitudes; the analog voltage signals are synchronously sampled according to a preset feature arrangement order to form a parallel voltage vector containing multiple input voltage components: Where V represents the parallel voltage vector; The represents the analog voltage value corresponding to the i-th input feature; M represents the number of input features.

[0049] The parallel voltage vector is applied to M row lines of the memristor array, and the input voltage of each row line is simulated by multiplying and adding the corresponding memristor cell's conductance state to obtain N column line output currents: ,in, This indicates the output current of the j-th column; This represents the conductance value of the memristor cell in the i-th row and j-th column; Let M represent the i-th input voltage; M represent the number of rows in the memristor array; and N represent the number of columns in the memristor array. Current sampling and analog-to-digital conversion are performed on the N columns of output current to obtain a digital prediction feature vector. Based on a pre-calibrated mapping relationship between current response and dry film quality indicators, quality parameters are calculated from the digital prediction feature vector to obtain the dry film quality prediction result. This dry film quality prediction result includes at least the dry film thickness uniformity index, the residual solvent content index, and the defect occurrence probability index.

[0050] Furthermore, converting the N columns of current into dry film quality prediction results includes: Define a preset conversion method, wherein the preset conversion method is a weighted sum of a set of weights in the memristor array corresponding to each column of current and the input voltage vector, the amplitude represents a prediction index of dry film quality, and the input voltage vector corresponds to the wet film characteristics; according to the preset conversion method, the N columns of current are converted and solved to the dry film quality prediction result.

[0051] The preset conversion method includes converting the output current of each column into a corresponding dry film quality prediction index based on the multiplicative mapping relationship between the conductance weight matrix of the memristor array and the input voltage vector; wherein, the input voltage vector is obtained by converting the multimodal physical field features collected during the wet film stage, the conductance state of each memristor unit in the memristor array represents the mapping weight between wet film characteristics and dry film quality indicators, and each column of output current corresponds to a dry film quality prediction index: ,in, This indicates the output current of the j-th column; This represents the conductance weight of the memristor cell in the i-th row and j-th column of the memristor array; The input voltage corresponding to the i-th wet film feature is represented by ; M represents the number of input features. According to the preset conversion method, the N columns of output currents are sampled, normalized, and mapped to indicators. The amplitude of each column of output current is converted into corresponding dry film quality prediction parameters to obtain the dry film quality prediction result. The dry film quality prediction result includes predicted dry film thickness, predicted residual solvent content, and predicted defect risk. When the dry film quality prediction result does not meet the expected quality standard, the optimization decision unit is triggered to perform iterative optimization, determine the process feature values, and convert them into analog control signals, responding to the end effector of the coating line for multi-threaded control.

[0052] Furthermore, it is determined whether the predicted dry film quality meets the expected quality standard. If not, an optimization decision unit is triggered. An optimization objective is defined, which includes maximizing thickness uniformity, minimizing solvent residue, and minimizing the probability of surface defects. Decision variable parameters are defined, which include coating gap, drying temperature zoning, and airflow distribution. The difference between the predicted dry film quality and the expected quality standard is measured. Based on the optimization objective and the decision variable parameters, the optimal parameter combination is determined through multiple rounds of iterative optimization and used as the process characteristic value.

[0053] Specifically, the predicted dry film quality is compared with a preset quality standard to calculate the predicted quality deviation: ,in, This indicates a deviation in the dry film quality prediction. This indicates the target quality indicator corresponding to the preset quality standard. This represents the predicted dry film quality output from the memristor array.

[0054] When the predicted quality deviation exceeds a preset allowable range, an optimization decision unit is triggered. Based on the predicted quality deviation, a set of process parameters to be optimized is determined. This set includes parameters such as the coating head die gap, drying zone temperature, and drying airflow distribution. A parameter adjustment search space is established based on the current wet film characteristics, dry film quality deviation, and historical process adjustment results. The optimization objectives are to improve dry film thickness uniformity, reduce residual solvent content, and decrease the probability of surface defects. The set of process parameters to be optimized is then iteratively optimized. Specifically, a process optimization objective function is constructed: ,in, This represents the objective function for process optimization. Indicators representing the deviation in dry film thickness uniformity; Indicates the residual solvent content; This represents a surface defect probability index. , , These represent the weight coefficients corresponding to the optimization objectives.

[0055] During the iterative optimization process, the current set of process parameters is used as the initial solution. The process parameters are perturbed and updated according to the preset parameter adjustment step size to obtain candidate set of process parameters. The quality improvement degree corresponding to each candidate set of process parameters is evaluated using the dry film quality prediction results. ,in, Indicates the first The change in the objective function before and after the iteration; Indicates the first The objective function value corresponding to the next iteration; Indicates the first The objective function value corresponding to each iteration. When the objective function value reaches the preset convergence condition or meets the expected quality standard, the iteration stops, and the corresponding optimal combination of process parameters is output as the process characteristic value.

[0056] Furthermore, the signals are converted into analog control signals, which are then used to perform multi-threaded control of the end effector in the coating production line, including: Based on the process characteristic values, analog control signals are determined through digital-to-analog conversion. The end effector driven by the coating process assigns the analog control signals to multiple threads to perform coating optimization control. The first control thread drives the piezoelectric ceramic fine-tuning array embedded in the coating head die to independently fine-tune the coating gap of each segment in the width direction, dynamically compensating for slurry rheological fluctuations and substrate deformation. The second control thread drives the opening degree of the air valves and the heating power of each zone of the drying oven to perform spatial distribution optimization of the temperature field and air field.

[0057] Preferably, based on the coating gap adjustment, drying temperature adjustment, and airflow adjustment included in the process characteristic values, the dimensions of each parameter are unified and normalized to obtain the corresponding digital control quantities; the digital control quantities are converted into continuous analog control signals through a digital-to-analog conversion module, wherein the output voltage or current amplitude of the analog control signal is mapped to the target adjustment quantity of the corresponding end effector. ,in, This represents the i-th analog control signal; This represents the scaling factor for the i-th digital-to-analog conversion. This indicates the corresponding digital control quantity; This represents the initial bias control value of the actuator.

[0058] A control channel allocation relationship is established based on the execution object corresponding to the analog control signal, and different types of analog control signals are allocated to the corresponding control threads. Among them, the first control thread sends the coating gap adjustment signal to the piezoelectric ceramic fine-tuning array embedded in the coating head die. By controlling the driving voltage of each piezoelectric ceramic unit, the die lip at different positions in the width direction is adjusted by micro-displacement, thereby independently correcting the coating gap of each area and compensating for thickness deviations caused by changes in slurry viscosity, rheological fluctuations, and local deformation of the substrate. The second control thread sends the drying adjustment signal to the air valve actuator and heating actuator corresponding to each temperature zone of the drying chamber. By adjusting the opening degree of the air valve and the heating power of each zone, the temperature field distribution and airflow field distribution inside the drying chamber are changed, so that the solvent evaporation process matches the changes in the wet film state.

[0059] During the multi-threaded control process, feedback displacement data of the piezoelectric ceramic fine-tuning array and feedback data of drying oven temperature and air volume are collected respectively. The feedback data is compared with the target control parameters in a closed loop, and the analog control signal is dynamically corrected according to the deviation until the coating gap, drying temperature distribution and air volume distribution reach the corresponding target range.

[0060] In summary, the embodiments of this application have at least the following technical effects: First, for the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor of the wet film stage. Next, to address the nonlinear mapping between the wet and dry film stages, a memristor array is deployed at the edge of the coating production line. A weight matrix based on wet film characteristics and dry film quality is used to perform convergent encoding based on the target conductance value for the memristor units at each intersection point. Then, according to the front-end analog circuit of the memristor array, the multimodal physical field tensor is received and converted into a parallel voltage vector, which is then loaded onto the memristor array to determine the predicted dry film quality. Finally, when the predicted dry film quality does not meet the expected quality standard, an optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values, and convert them into analog control signals, responding to the end effector of the coating production line for multi-threaded control. This solves the technical problem in existing technologies where the coating process lacks an effective feedforward control mechanism based on process state, leading to difficulty in suppressing coating quality fluctuations. It achieves the technical effect of predicting the dry film quality in advance during the wet film stage and driving real-time optimization and control of coating process parameters, thereby improving the stability of coating quality.

[0061] Example 2, based on the same inventive concept as the intelligent optimization method for coating process based on multimodal data fusion in the foregoing examples, such as... Figure 2 As shown, this application provides an intelligent optimization system for coating processes based on multimodal data fusion, wherein the system includes: Data acquisition module 11: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to acquire the multimodal physical field tensor of the wet film stage; Memristor deployment module 12: For the nonlinear mapping between the wet film stage and the dry film stage, a memristor array is deployed at the edge of the coating production line, wherein the memristor units at each intersection point are deployed by performing convergence encoding based on the target conductance value using a weight matrix based on wet film characteristics and dry film quality; Quality prediction module 13: According to the front-end analog circuit of the memristor array, the multimodal physical field tensor is received and converted into a parallel voltage vector, which is then loaded onto the memristor array to determine the dry film quality prediction result; Control module 14: When the dry film quality prediction result does not meet the expected quality standard, the optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values ​​and convert them into analog control signals, and respond to the end actuator of the coating production line for multi-threaded control.

[0062] Furthermore, the data acquisition module 11 is used to perform the following methods: The thickness distribution and solvent content distribution of the wet film are determined by emitting broadband terahertz pulses onto the wet film surface using a spectral probe and receiving the reflected echoes. The temperature field of the wet film surface is acquired by an infrared thermal imager, and the solvent evaporation rate and droplet adhesion rate change rate are extracted by differential processing. The dynamic speckle pattern of the wet film surface is captured by a laser speckle imaging device, and the rheological state and flow stability of the wet film surface are determined by speckle contrast and time correlation analysis.

[0063] Furthermore, the memristor deployment module 12 is used to perform the following method: A memristor array with a cross-shaped structure is deployed, wherein M row lines and N column lines intersect perpendicularly, and each intersection point integrates a memristor unit; a weight matrix based on wet film characteristics and dry film quality is determined, the memristor array is initially encoded, and deployed at the edge of the coating production line to establish data interaction with the multimodal sensing end.

[0064] Furthermore, the memristor deployment module 12 is used to perform the following method: Using wet-dry film paired data as sample data, multimodal wet film features and dry film quality indicators are extracted and a nonlinear mapping is established to determine an M×N dimensional weight matrix. For the weight matrix, the target conductance value of each intersection point of the memristor array is converted through the mapping transformation rule. The target conductance value is written into the memristor array to complete the initial encoding.

[0065] Furthermore, the memristor deployment module 12 is used to perform the following method: For the first target crossover point, the current conductance value of the first memristor cell is identified, wherein the first target crossover point is any crossover point in the memristor array; based on the difference between the current conductance value and the first target conductance value, the first SET voltage pulse is determined by measuring the pulse duration; by applying the first SET voltage pulse to the first target crossover point, the conductance value of the first memristor cell converges to the first target conductance value.

[0066] Furthermore, the quality prediction module 13 is used to perform the following method: The front-end analog circuit of the memristor array receives the multimodal physical field tensor and converts it into an analog voltage amplitude proportional to the physical quantity; the analog voltage amplitude is sampled in a preset order to determine a parallel voltage vector; the parallel voltage vector is loaded onto M rows of the memristor array to output N columns of current; and the N columns of current are converted into dry film quality prediction results.

[0067] Furthermore, the quality prediction module 13 is used to perform the following method: Define a preset conversion method, wherein the preset conversion method is a weighted sum of a set of weights in the memristor array corresponding to each column of current and the input voltage vector, the amplitude represents a prediction index of dry film quality, and the input voltage vector corresponds to the wet film characteristics; according to the preset conversion method, the N columns of current are converted and solved to the dry film quality prediction result.

[0068] Furthermore, the control module 14 is used to perform the following methods: The process involves determining whether the predicted dry film quality meets the expected quality standard. If not, an optimization decision unit is triggered. An optimization objective is defined, including maximizing thickness uniformity, minimizing solvent residue, and minimizing the probability of surface defects. Decision variable parameters are defined, including coating gap, drying temperature zones, and airflow distribution. The difference between the predicted dry film quality and the expected quality standard is measured. Based on the optimization objective and decision variable parameters, multiple rounds of iterative optimization are performed to determine the optimal parameter combination, which serves as the process characteristic value.

[0069] Furthermore, the control module 14 is used to perform the following methods: Based on the process characteristic values, analog control signals are determined through digital-to-analog conversion. The end effector driven by the coating process assigns the analog control signals to multiple threads to perform coating optimization control. The first control thread drives the piezoelectric ceramic fine-tuning array embedded in the coating head die to independently fine-tune the coating gap of each segment in the width direction, dynamically compensating for slurry rheological fluctuations and substrate deformation. The second control thread drives the opening degree of the air valves and the heating power of each zone of the drying oven to perform spatial distribution optimization of the temperature field and air field.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent optimization of coating process based on multimodal data fusion, characterized in that, The method includes: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor during the wet film stage; To address the nonlinear mapping between the wet film stage and the dry film stage, a memristor array is deployed at the edge of the coating line. The memristor cells at each intersection are deployed by performing convergence encoding based on the target conductivity value using a weight matrix based on wet film characteristics and dry film quality. According to the front-end analog circuit of the memristor array, the multimodal physical field tensor is received and converted into a parallel voltage vector, which is then loaded onto the memristor array to determine the dry film quality prediction result. When the predicted dry film quality does not meet the expected quality standard, the optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values ​​and convert them into analog control signals, and respond to the end effector of the coating production line to perform multi-threaded control.

2. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 1, characterized in that, The multimodal physical field tensors during the wet film phase are collected, including: The thickness distribution and solvent content distribution of the wet film were determined by emitting broadband terahertz pulses onto the surface of the wet film using a spectral probe and receiving the reflected echoes. The surface temperature field of the wet film was acquired by an infrared thermal imager, and the solvent evaporation rate and droplet adhesion rate were extracted by differential processing. Based on the dynamic speckle pattern captured by the laser speckle imaging device on the wet film surface, the rheological state and flow stability of the wet film surface are determined by speckle contrast and time correlation analysis.

3. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 1, characterized in that, Deploying a memristor array at the edge of the coating line includes: A memristor array with a cross-shaped structure is deployed, wherein M row lines and N column lines intersect perpendicularly, and each intersection point integrates a memristor unit; A weight matrix based on wet film characteristics and dry film quality is determined, the memristor array is initially encoded, deployed at the edge of the coating production line, and establishes data interaction with the multimodal sensing end.

4. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 3, characterized in that, The initial encoding of the memristor array includes: Using wet-dry film paired data as sample data, we extract multimodal wet film features and dry film quality indicators and establish a nonlinear mapping to determine an M×N dimensional weight matrix. The weight matrix is ​​converted into the target conductance values ​​at each intersection of the memristor array using a mapping transformation rule. The target conductivity value is written into the memristor array to complete the initial encoding.

5. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 4, characterized in that, The target conductance value is written into the memristor array to complete the initial encoding, including: For the first target crossover point, identify the current conductance value of the first memristor cell, wherein the first target crossover point is any crossover point in the memristor array; Based on the difference between the current conductance value and the first target conductance value, the first SET voltage pulse is determined by calculating the pulse duration; By applying the first SET voltage pulse to the first target intersection, the conductance of the first memristor cell converges to the first target conductance.

6. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 1, characterized in that, Determine the dry film quality prediction results, including: The front-end analog circuit of the memristor array receives the multimodal physical field tensor and converts it into an analog voltage amplitude proportional to the physical quantity; The simulated voltage amplitudes are sampled in a preset order to determine the parallel voltage vector; The parallel voltage vector is applied to M rows of the memristor array, resulting in N columns of current output. The N columns of current are converted into dry film quality prediction results.

7. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 6, characterized in that, Converting the N columns of current into dry film quality prediction results includes: Define a preset conversion method, wherein the preset conversion method is a weighted sum of a set of weights in the memristor array corresponding to each column of current and the input voltage vector, the amplitude represents a predictive index of dry film quality, and the input voltage vector corresponds to the wet film characteristics; According to the preset conversion method, the N columns of current are converted and calculated into the dry film quality prediction result.

8. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 1, characterized in that, Determine whether the predicted dry film quality meets the expected quality standard; if not, trigger the optimization decision unit. Define optimization objectives, wherein the optimization objectives include maximizing thickness uniformity, minimizing solvent residue, and minimizing the probability of surface defects; Define decision variable parameters, wherein the decision variable parameters include coating gap, drying temperature zone and air volume distribution; The difference between the predicted dry film quality and the expected quality standard is measured. Based on the optimization objective and decision variable parameters, the optimal parameter combination is determined through multiple rounds of iterative optimization and used as the process characteristic value.

9. The intelligent optimization method for coating process based on multimodal data fusion as described in claim 1, characterized in that, Converted into analog control signals, these signals are used to perform multi-threaded control of the end effectors in the coating production line, including: Based on the aforementioned process characteristic values, the analog control signal is determined through digital-to-analog conversion; An end effector driven by a coating process performs multi-threaded management of the analog control signals and executes coating optimization control. The first control thread drives the piezoelectric ceramic fine-tuning array embedded in the coating head die to independently fine-tune the coating gap of each segment in the width direction, dynamically compensating for slurry rheological fluctuations and substrate deformation. The second control thread drives the opening degree of the air valves and the heating power of each zone of the drying chamber, and performs spatial distribution optimization of the temperature field and air field.

10. A coating process intelligent optimization system based on multimodal data fusion, characterized in that, The system is used to implement the intelligent optimization method for coating processes based on multimodal data fusion as described in any one of claims 1-9, the system comprising: Data acquisition module: For the coating production line, a multimodal sensor array is deployed between the coating head outlet and the drying inlet to collect the multimodal physical field tensor during the wet film stage; Memristor deployment module: For the nonlinear mapping between the wet film stage and the dry film stage, a memristor array is deployed at the edge of the coating line. The memristor cells at each intersection are deployed by performing convergence encoding based on the target conductivity value using a weight matrix based on wet film characteristics and dry film quality. Quality prediction module: Based on the front-end analog circuit of the memristor array, it receives the multimodal physical field tensor and converts it into a parallel voltage vector, loads it onto the memristor array, and determines the dry film quality prediction result; Control module: When the dry film quality prediction result does not meet the expected quality standard, the optimization decision unit is triggered to perform iterative optimization, determine the process characteristic values ​​and convert them into analog control signals, and respond to the end effector of the coating production line to perform multi-threaded control.

Images