A functional fabric antibacterial coating uniform spraying control method and system

Abstract

The application discloses a functional fabric antibacterial coating uniform spraying control method and system, and relates to the technical field of program automatic control.The steps of the method comprise the following steps: obtaining the spraying operation parameters, the environmental structure parameters, the antibacterial coating fabric to be sprayed and the coating distribution characteristics with the same time label under the distributed collection node; processing and analyzing the spraying operation parameters and the environmental structure parameters, generating a basic fluctuation mode through task combination; processing and analyzing the spraying operation parameters and the coating distribution characteristics, constructing a dynamic behavior tree, screening out the maximum effective combination mode and the minimum effective combination mode, generating a composite control matrix, converting the composite control matrix into a supplementary coating control instruction and issuing the supplementary coating control instruction to the corresponding distributed execution node to execute the spraying operation represented by the supplementary coating control instruction; and the application realizes dynamic compensation control of the antibacterial coating fabric to be sprayed while guaranteeing the coating quality.

Description

Technical Field

[0001] This invention relates to the field of automated control technology, specifically to a method and system for controlling the uniform spraying of antibacterial coatings on functional fabrics. Background Technology

[0002] Fabric spraying is a highly efficient and precise coating technology that is widely used in clothing, furniture, and automobiles. It can enhance the functionality and aesthetics of fabrics. In particular, in the clothing industry, the introduction of spraying technology can achieve multiple functional finishing effects on textile fabrics, such as antibacterial, stain-resistant, waterproof, and flame-retardant properties, thereby increasing the added value of textiles and meeting market demands.

[0003] Existing spraying technologies typically employ program-defined DCS (Distributed Control System) automatic control, integrating continuous, sequential, and batch control through distributed program control. It can also easily incorporate specific control algorithms for intelligent control of fabric spraying processes. However, this approach has limitations: Firstly, nonlinear coupling and fluctuations in program control parameters make it difficult for spraying control algorithms or models to cover complex and diverse working conditions or processes. Secondly, the control of antibacterial coating quality often relies on independent end-layer coating detection feedback or adjustments to single discrete process parameters. For example, analysis of fabric spraying quality may only consider ambient temperature and humidity, while spraying operation parameters such as spraying speed and spray width control pressure, along with environmental structural parameters such as material viscosity, fabric tension, and temperature, are not considered. In high-frequency dynamic changes, if the analysis is not conducted in conjunction with environmental and fabric substrate data, it can easily lead to coating process drift and make it impossible to analyze the core disturbance conditions. On the other hand, on modern high-precision coating lines, when film-forming defects are found on high-speed production lines, adjustment commands are often issued directly in situ. However, by this time, the fabric area where the error occurred has already slipped past the coating area, making it impossible to accurately locate the source of the defect. This causes the multi-nozzle distributed control system to experience control overshoot, lag, or misalignment of the recoating position due to data misalignment, which greatly limits the stable improvement of the film-forming quality of the full-width coating. At the same time, the control algorithm used may try to avoid this tiny defect in order to calculate quickly and minimize the global impact, which can easily lead to uneven antibacterial coating on the fabric and a significantly increased probability of lag and failure to identify the defect, making it impossible to respond to recoating. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for controlling the uniform spraying of antibacterial coatings on functional fabrics. The method acquires spraying operation parameters, environmental structural parameters, the fabric to be coated with the antibacterial coating, and coating distribution characteristics from distributed acquisition nodes with the same time label. It then processes and analyzes the spraying operation parameters and environmental structural parameters, generating basic fluctuation modes through task combinations. Further processing and analysis of the spraying operation parameters and coating distribution characteristics constructs a dynamic behavior tree, selecting the maximum effective combination mode and the minimum effective combination mode, generating a composite control matrix. This matrix is ​​then converted into a recoating control command and issued to the corresponding distributed execution nodes to execute the spraying operation represented by the recoating control command, thus solving the problems mentioned in the background technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides a method for controlling the uniform spraying of antibacterial coatings on functional fabrics. The method includes: collecting spraying operation parameters for each working condition through distributed acquisition nodes, and obtaining environmental structural parameters, the fabric to be sprayed with antibacterial coating, and coating distribution characteristics that have the same time sequence label as the spraying operation parameters. The spraying operation parameters and environmental structure parameters are processed and analyzed. Based on the first and second abnormal time points, the fluctuation behavior of the antibacterial coating thickness is analyzed. Within the preset dynamic sliding window, the basic fluctuation mode is generated by task combination. The spraying operation parameters and coating distribution characteristics are processed and analyzed to construct a dynamic behavior tree. Based on the dynamic behavior tree, non-standard branch links are obtained through tree comparison. By introducing contribution scores and combining with a CNN model, the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are selected. Based on the effective pairing of the maximum effective combination mode and the minimum effective combination mode, a composite control matrix is ​​generated; the composite control matrix is ​​converted into a recoating control command, and the recoating control command is sent to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command.

[0006] Furthermore, the spraying operation parameters for each working condition are collected, including: In response to the spraying control command, the register status data of the programmable logic controller is read to obtain the spraying operation parameters that characterize the opening and closing sequence of the nozzle valve and the reciprocating motion under different working conditions, including at least the spraying moving speed, spray width control pressure, coating thickness and spraying execution indicator; The system synchronously receives data from the sensor array, including the antibacterial coating fabric to be sprayed, the viscosity of the coating solution, the fabric tension, the temperature, and the humidity. The system then splices the data from the coating solution viscosity, the fabric tension, the temperature, and the humidity to generate environmental structural parameters. Simultaneously, the ultraviolet excitation light source is controlled to irradiate the fabric to be coated with the antibacterial coating. The coating image is captured by an industrial camera and divided into a number of image grid data. The pixel grayscale mean of each image grid data is extracted, and the pixel mean of all image grids is combined to obtain the coating distribution characteristics.

[0007] Furthermore, fundamental wave modes are generated through task combination, including: Based on the environmental structure parameters, extract the first abnormal time point where the environmental structure parameters do not conform to the standard parameter range; The time-domain and frequency-domain features of the spraying operation parameters and environmental structure parameters are extracted by Fourier transform and labeled as feature evaluation parameters. A dynamic sliding window is preset, and the feature evaluation parameters within the dynamic sliding window are collected into a feature evaluation parameter set G. n feature evaluation parameters are selected from G using the multiplication counting principle to construct the spraying unevenness task: A(m,n), where the spraying unevenness task is the fluctuation behavior of the antibacterial coating thickness under a certain working condition; where A(·) represents permutation and combination calculation, and m represents the feature evaluation parameter at any time under a certain working condition. Based on the spraying operation parameters, the second abnormal time point where the coating thickness does not meet the standard thickness range was extracted; After the first abnormal time point is counted, the number of times the second abnormal time point occurs within the preset maximum lag time window is counted. The uneven spraying task is then carried out, and several mutually orthogonal fused feature vectors are decoupled. The CNN model is used to analyze the several fused feature vectors and output the basic fluctuation mode.

[0008] Furthermore, a preset dynamic sliding window includes: Collect the spraying operation parameters of all first abnormal time points and the environmental structure parameters of all second abnormal time points within the historical time period; calculate the correlation coefficient using the cross-correlation function and construct the time-delay response curve; identify the time-delay point with the largest absolute value of the correlation coefficient in the time-delay response curve, and extract a pre-set information interval range on both sides of the time-delay point as the center.

[0009] Furthermore, the spraying operation parameters and coating distribution characteristics are processed and analyzed to construct a dynamic behavior tree, including: Extract each spraying execution identifier from the spraying operation parameters, map each spraying execution identifier to a spraying execution node, and mark it as the parent node of the dynamic behavior tree; at the same time, obtain the timestamps of adjacent spraying execution identifiers in the spraying operation parameters, and retrieve the coating distribution characteristics within the time difference range as the child nodes of the dynamic behavior tree; For each spray execution node, the spray gun height and jet direction are determined. Spatial decomposition is performed based on the jet direction to determine the spray tilt angle. The spray gun height and spray tilt angle are combined to generate spray evaluation indicators, which serve as edge weights of the dynamic behavior tree.

[0010] Furthermore, the maximal and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are screened, including: The parent-child node relationships of the dynamic behavior tree and the standard behavior tree are compared level by level. If an unknown child node not included in the standard behavior tree is detected in the dynamic behavior tree, the rule engine is activated. Using the parent node of the unknown child node as the center, a circle is drawn to determine the local tree branch starting from the parent node containing the unknown child node, and it is marked as a non-standard branch link. The absolute value set of all edge weights in the non-standard link is obtained, the integral sum of the absolute value set is calculated, and it is marked as the contribution score. Using the spraying execution identifier as an index, the basic fluctuation mode of the CNN model is projected onto the dynamic behavior tree after N rounds of debugging and optimization. By identifying the mapping gain, the maximum lag time window is introduced, and the Dice loss function is called to perform N rounds of iterative training. The connection edges of the non-standard branch link are adaptively updated to generate the modal response matrix. The modal response matrix is ​​analyzed, and the contribution score of each basic wave mode update is extracted along the non-standard branch link. The contribution score change curve is plotted. For basic wave mode combinations whose curve change slope is greater than or equal to the standard change slope, the maximum effective combination mode is obtained; for basic wave mode combinations whose curve change rate is less than the standard change slope, the minimum effective combination mode is obtained.

[0011] Furthermore, several fused feature vectors are analyzed using a CNN model to output the fundamental wave modes, including: CNN is used as the basic architecture of the model; Multiple fused feature vectors are obtained, spatiotemporal alignment and stacking are performed to construct a two-dimensional spatiotemporal feature matrix; a centrosymmetric mapping layer is set at the end of the CNN, the CNN model is iteratively trained using historical datasets, the gradient descent algorithm is selected, and through backpropagation, the centrosymmetric mapping layer outputs multiple sets of mutually orthogonal feature vectors, and outputs the basic wave mode. Using a pre-trained CNN model, the new fused feature vector is trained to output the corresponding basic wave mode.

[0012] Furthermore, the mapping gain, which represents the correlation between the fused feature vector and the CNN model, is achieved through at least one of the following methods: Calculate the Pearson correlation coefficient or Spearman rank correlation coefficient between the fused feature vector and the spraying unevenness task constructed by permutation and combination using the multiplication counting principle.

[0013] Furthermore, after generating the touch-up control command, it also includes: The repainting control command and the associated timestamp data are encapsulated into a synchronization control data packet; Synchronous control data packets are broadcast to multiple distributed execution nodes via fieldbus; Each distributed execution node parses the synchronization control data packet and synchronously executes the dynamic adjustment spraying operation within the clock period specified by the timestamp data.

[0014] Secondly, this application provides a functional fabric antibacterial coating uniform spraying control system, the system including: a parameter acquisition module, which collects spraying operation parameters for each working condition through distributed acquisition nodes, and acquires environmental structure parameters, antibacterial coating fabric to be sprayed, and coating distribution characteristics that have the same time sequence label as the spraying operation parameters; The modal analysis module processes and analyzes the spraying operation parameters and environmental structural parameters. Based on the first and second abnormal time points, it analyzes the fluctuation behavior of the antibacterial coating thickness. Within a preset dynamic sliding window, it generates basic fluctuation modes through task combination. The behavior tree construction module processes and analyzes the spraying operation parameters and coating distribution characteristics to construct a dynamic behavior tree. Based on the dynamic behavior tree, non-standard branch links are obtained through tree comparison. By introducing contribution scores and combining with a CNN model, the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are selected. The control module is adjusted to generate a composite control matrix based on the effective pairing of the maximum and minimum effective combination modes. The composite control matrix is ​​then converted into a recoating control command, which is sent to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command.

[0015] (III) Beneficial Effects This invention provides a method and system for controlling the uniform spraying of antibacterial coatings on functional fabrics, which has the following beneficial effects: This invention avoids the problem of single-data-driven processing caused by a distributed system by setting up a distributed system to acquire spraying operation parameters, environmental structure parameters, and coating distribution characteristics with the same time sequence through distributed acquisition nodes. By introducing a preset sliding window to analyze the fluctuation behavior of antibacterial coating thickness under complex working conditions, and based on the first and second abnormal time points in the processing and analysis of spraying operation parameters and environmental structure parameters, the invention extracts the basic fluctuation modes with mutually orthogonal properties using a CNN model, providing interpretable and highly reliable basic data for subsequent analysis. This invention analyzes spraying operation parameters and coating distribution characteristics to dynamically construct a dynamic behavior tree under a distributed architecture. It provides tree comparison, activates a rule engine, and performs a circle-drawing operation, focusing on local tree branches from complex ones, greatly simplifying the strong coupling of multiple spray heads in industrial settings. By generating contribution branches based on edge weights and combining them with a CNN model to capture the correlation between the fused feature vector and the CNN model, it obtains mapping gain and adaptively updates each connection edge of non-standard branch links, generating a modal response matrix. Through curve analysis, it amplifies the characteristics of positional abrupt changes, more comprehensively summarizing the maximum and minimum effective combination modes of the uneven spraying task, significantly improving the system's robustness. Furthermore, through effective pairing of the maximum and minimum effective combination modes, it transforms the recoating control instructions for distributed execution nodes, achieving dynamic compensation control of the antibacterial coating fabric to be sprayed. This achieves high-precision uniform spraying, effectively reducing the defect rate while ensuring coating quality, and realizing fully automated operation for coating uniformity. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] The core of this invention lies in acquiring spraying operation parameters, environmental structure parameters, and coating distribution characteristics with the same time label. In terms of data integration, the spraying operation parameters and environmental structure parameters are first processed and analyzed. Through multi-dimensional task permutations and combinations, a fundamental wave mode characterizing multi-physics interference is abstractly constructed. Simultaneously, the spraying operation parameters and coating distribution characteristics are processed and analyzed to construct a dynamic behavior tree. The dominant maximum effective combination mode and minimum effective combination mode are adaptively selected through the edge weights of the tree topology to determine the composite control matrix. This matrix is ​​then directly converted into recoating control commands to precisely target and counteract distributed execution nodes. In this way, the response blind spot caused by the spatiotemporal misalignment of physical signals and control models in traditional methods is completely overcome, significantly improving the accuracy of distributed high-concurrency recoating of antibacterial coatings.

[0019] Example 1: This embodiment of the invention provides a method for controlling the uniform spraying of an antibacterial coating on functional fabrics; Figure 1 This is a schematic diagram of the steps of the present invention; please refer to it. Figure 1 The method includes the following steps; S1: Collect the spraying operation parameters for each working condition through distributed acquisition nodes, and obtain the environmental structure parameters, the antibacterial coating fabric to be sprayed, and the coating distribution characteristics that have the same time series label as the spraying operation parameters. By acquiring the spraying operation parameters for each working condition sent by the distributed acquisition nodes, the environmental structure parameters, the antibacterial coating material to be sprayed, and the coating distribution characteristics with the same time sequence label as the spraying operation parameters are obtained. This includes: responding to the spraying control command, periodically reading the register status data of the programmable logic controller through the fieldbus at a sampling frequency of ≥200Hz to obtain the spraying operation parameters characterizing the opening and closing sequence of the nozzle valve and the execution of reciprocating motion under different working conditions, including but not limited to spraying movement speed, spray width control pressure, coating thickness, and spraying execution identifier: the spraying movement speed is calculated by feedback from the servo drive encoder, with a control range of 0.5m / s to 2.5m / s; the spray width control pressure is read through the analog input unit corresponding to the proportional regulating valve, with a reference range of 0.15MPa to 0.45MPa; the coating thickness is acquired in real time at high speed by an infrared thickness gauge; the spraying execution identifier is the array-type nozzle physical hardware code pre-written in the EEPROM, and each independent nozzle or control loop is a unique hardware physical layer code or software index label in the bus network, used to subsequently lock the data source on the dynamic behavior tree; The system synchronously receives data from the sensor array, including the antibacterial coating fabric to be sprayed, liquid viscosity, fabric tension, temperature, and humidity. After initialization and establishing a full-link communication handshake, it reads the fabric spraying process parameters and directs the PLC's data register pointer to the corresponding buffer. The sensor array includes a tension sensor, an ultrasonic viscometer, and a temperature and humidity transmitter mounted on the fabric guide roller. The tension sensor collects fabric tension, the ultrasonic viscometer collects liquid viscosity, and the temperature and humidity transmitter collects temperature and humidity. The system then performs scaling and data normalization on the received heterogeneous data streams of liquid viscosity, fabric tension, temperature, and humidity at a unified gateway layer. Through row vector stitching technology, it generates environmental structure parameters in the form of a multi-dimensional structured matrix. It should be noted that, to ensure data accuracy, the system sets standard parameter ranges for the environmental structure parameters. Simultaneously, ultraviolet light is used to irradiate the fabric to be coated with the antibacterial coating. The fluorescent tracer in the fabric is used to excite the fluorescence properties of the fabric to be coated with the antibacterial coating. The spraying image is captured by an industrial camera and divided into a number of image grid data. The average gray value of each image grid data is extracted. Since the magnitude of the average gray value is monotonically positively correlated with the spatial adhesion density of the antibacterial coating, the average pixel values ​​of all image grids are matrix-recombined and vector-concatenated to finally obtain the coating distribution characteristics that can accurately characterize the film formation quality in the width and evolution directions.

[0020] By setting up distributed acquisition nodes to collect spraying operation parameters, environmental structure parameters, and coating distribution characteristics, the acquired data sources have time-series labels and spraying execution identifiers. This enables precise addressing in the subsequent generation of recoating compensation instructions, ensuring accurate response to the corresponding distributed execution nodes and dynamic adjustment of the corresponding intensity within a specific clock cycle. This truly realizes a control closed loop from multi-source heterogeneous input to distributed collaborative output.

[0021] S2: Process and analyze the spraying operation parameters and environmental structure parameters. Based on the first and second abnormal time points, analyze the fluctuation behavior of the antibacterial coating thickness. Within the preset dynamic sliding window, generate the basic fluctuation mode through task combination. The basic fluctuation mode is generated through task combination, including: based on environmental structural parameters, retrieving preset standard parameter ranges, and extracting the first abnormal time point where the environmental structural parameters do not conform to the standard parameter ranges. Specifically, under a certain working condition, for each environmental structural parameter, it is compared with its corresponding standard parameter range. The standard parameter ranges include standard liquid concentration, standard tension range, standard temperature range, and standard humidity range: comparing the liquid viscosity with the standard viscous range, the fabric tension with the standard tension range, the temperature with the standard temperature range, and the humidity with the standard humidity. If the comparison result shows that any environmental structural parameter is greater than the maximum value of the standard parameter range or any environmental structural parameter is greater than the standard parameter range, the error is considered. If the value in the range is less than the minimum value of the standard parameter range, it is determined that the environmental structural parameters do not conform to the standard parameter range, and an anomaly is marked. It should be noted that among the environmental structural parameters, temperature, humidity, liquid viscosity, and fabric tension have mutual influences. For example, the influence of environmental humidity on fabric spraying: environmental humidity affects the evaporation rate of liquid viscosity, which in turn affects the leveling and sagging performance of the coating. Spraying under high environmental humidity will cause the liquid viscosity to evaporate, resulting in the surface temperature of the fabric film layer being lower than the dew point temperature, producing water vapor condensation on its surface, causing the paint film to turn white, thus affecting the fabric spraying. Therefore, if the comparison result of any environmental structural parameter shows an anomaly, the current moment is locked and marked as the first time anomaly point. The time-domain and frequency-domain features of the spraying operation parameters and environmental structure parameters are extracted by Fourier transform and labeled as feature evaluation parameters, including the first time-domain feature and the first frequency-domain feature corresponding to the spraying operation parameters, and the second time-domain feature and the second frequency-domain feature corresponding to the environmental structure parameters. Among them, the time-domain features include, but are not limited to, mean, variance, peak value, valley value, peak-to-peak value, kurtosis, skewness, trend slope and rate of change, and the frequency-domain features include, but are not limited to, principal vibration frequency, power spectral density principal peak value and spectral centroid. By collecting spraying operation parameters and environmental structure parameters over historical time periods, including collecting spraying operation parameters at all first abnormal time points and environmental structure parameters at all second abnormal time points within the historical time period; using a cross-correlation function to calculate the correlation coefficient between the two, and constructing a time-delay response curve with time as the horizontal axis and the correlation coefficient as the vertical axis; It should be noted that the magnitude of the correlation coefficient determines the confidence level of the time delay point. If the absolute value of the correlation coefficient is larger and closer to 1, the time delay response curve will show a steep and clear single peak, indicating that an anomaly has occurred at this moment, and this moment is marked as the time delay point. In this case, it means that the relationship between the spraying operation parameters and the environmental structure parameters is highly focused. Narrowing the preset confidence interval on both sides of the time delay point results in a narrower and more precise time window, requiring only a very small time range to be calculated, reducing computing power consumption and improving the controller's response speed. If the absolute value of the correlation coefficient is larger and closer to 0, the time delay response curve tends to be flat overall, and multiple peaks or spurious peaks may appear, making it difficult to distinguish whether an anomaly has occurred. In this case, it means that the relationship between the spraying operation parameters and the environmental structure parameters is divergent. Expanding the preset confidence interval on both sides of the time delay point results in a wider time window, ensuring that potential interference signals are not missed, which increases the computational load on the backend.

[0022] Identify the time delay point with the largest absolute value of the correlation coefficient in the time delay response curve, and use this time delay point as the center to extract a preset confidence interval range on both sides to obtain a preset dynamic sliding window. For example, if the identified maximum absolute value of the correlation coefficient is 0.88, the corresponding time delay point is 1.4 seconds. Using 1.4 as the midline, symmetrically extract a preset confidence interval range on both sides of the time axis, setting a span of ±wc%, where wc is a positive integer. It should be noted that a 95% confidence level is usually selected, corresponding to wc=20, resulting in a dynamic sliding window with a span of [1.2, 1.6]. The feature evaluation parameters within the preset dynamic sliding window are collected into a feature evaluation parameter set G. Using the multiplication counting principle, n feature evaluation parameters are selected from G to construct a spraying unevenness task: A(m, n), where the spraying unevenness task represents the fluctuation behavior of the antibacterial coating thickness under a certain working condition. Here, A(·) represents permutation and combination calculation, and m represents the feature evaluation parameter at any moment under a certain working condition. Simultaneously, based on the preset dynamic sliding window, wc=30 is set, and the obtained dynamic sliding window is defined as the preset maximum lag time window and stored in the system for easy future retrieval. Example: To obtain the fabric spraying process under a certain working condition, within a dynamic sliding window, acquire feature evaluation parameters. The total number of parameter types is 4, including the first time-domain feature, the first frequency-domain feature, the second time-domain feature, and the second frequency-domain feature. Set the single spraying unevenness task to 3. Using the multiplication counting principle, A(4,3)=4×3×2×1=24 ordered single-item spraying unevenness tasks, specifically including: the task chain generated by the first time-domain feature as the driving source is as follows: Task 1: {first time-domain feature, first frequency-domain feature, second time-domain feature}, Task 2: {first time-domain feature, first frequency-domain feature, second frequency-domain feature}, Task 3: {first time-domain feature, second time-domain feature, first frequency-domain feature}, Task 4: {first time-domain feature, second time-domain feature, first frequency-domain feature}, Task 5: {first time-domain feature, second frequency-domain feature, first frequency-domain feature}, Task 6: {first time-domain feature, second frequency-domain feature, first frequency-domain feature}. The task chain generated using the first frequency domain feature as the driving source is as follows: Task 7: {first frequency domain feature, first time domain feature, second time domain feature}, Task 8: {first frequency domain feature, first time domain feature, second frequency domain feature}, Task 9: {first frequency domain feature, second time domain feature, first time domain feature}, Task 10: {first frequency domain feature, second time domain feature, second frequency domain feature}, Task 11: {first frequency domain feature, second frequency domain feature, first time domain feature}, Task 12: {first frequency domain feature, second frequency domain feature, second time domain feature}; The task chain generated by the second time-domain feature as the driving source is as follows: Task 13: {second time-domain feature, first time-domain feature, first frequency-domain feature}, Task 14: {second time-domain feature, first time-domain feature, second frequency-domain feature}, Task 15: {second time-domain feature, first frequency-domain feature, first time-domain feature}, Task 16: {second time-domain feature, first frequency-domain feature, second frequency-domain feature}, Task 17: {second time-domain feature, second frequency-domain feature, first time-domain feature}, Task 18: {second time-domain feature, second frequency-domain feature, first frequency-domain feature}; The task chain generated using the second frequency domain feature as the driving source is as follows: Task 19: {second frequency domain feature, first time domain feature, first frequency domain feature}, Task 20: {second frequency domain feature, first time domain feature, second time domain feature}, Task 21: {second frequency domain feature, first frequency domain feature, first time domain feature}, Task 22: {second frequency domain feature, first frequency domain feature, second time domain feature}, Task 23: {second frequency domain feature, second time domain feature, first time domain feature}, Task 24: {second frequency domain feature, second time domain feature, first frequency domain feature}; Based on the principle of permutation and combination, the above-mentioned set of 24 non-uniform tasks containing third-order ordered chains is decomposed into 24 task chains with clear dynamic causal transmission directions. Each ordered item represents a transmission path under a specific working condition, providing interpretable basic data for the subsequent generation of fundamental wave modes. For some non-uniform spraying tasks, for example, the task chain is: {first time domain feature, first frequency domain feature, second frequency domain feature}, which characterizes the movement speed jitter of the servo beam in the time domain and serves as the excitation source. First, high-frequency pressure pulsation in the frequency domain is induced in the nozzle atomized fluid, which then leads to high-frequency standing wave variation in the tension of the high-speed conveying fabric through gas-liquid impact; for example, the task chain is: {second time domain feature, first frequency domain feature, first time domain feature}, which characterizes the viscosity trend drift of the antibacterial liquid in the time domain caused by temperature change, which changes the fluid resistance of the pipeline and induces periodic pressure pulsation in the frequency domain of the liquid system. This fluid reaction force is ultimately fed back to the mechanical transmission shaft, causing the crossbeam movement speed to produce slip variance in the time domain; Based on the spraying operation parameters, the second abnormal time point where the coating thickness does not conform to the standard thickness range is extracted. Specifically, the coating thickness is retrieved and compared with the standard thickness range. If the coating thickness is less than the minimum value of the standard thickness range or the standard thickness is greater than the maximum value of the standard thickness range, it is determined that the coating thickness does not conform to the standard thickness range, and the corresponding time is marked as the second abnormal time point. After the occurrence of the first abnormal time point, within the preset maximum lag time window, the number of occurrences of the second abnormal time point is counted. The spraying unevenness task is expanded, and several mutually orthogonal fused feature vectors are decoupled. The above 24 ordered task items are standardized individually, and features are extracted again to calculate the covariance. The variance matrix is ​​decomposed into eigenvalues ​​of the covariance matrix. For example, principal component analysis yields multiple eigenvectors. The eigenvectors with the highest eigenvalues ​​are selected in descending order. These eigenvectors are naturally orthogonal to each other, forming the basis vectors for the fused eigenvector. The two-dimensional eigenma matrix is ​​projected onto these orthogonal basis vectors to obtain the fused eigenvector. For example, for Task 2, the eigenvectors are further extracted, including feature magnitude, phase difference, and the similarity between feature magnitude and phase difference calculated based on cosine similarity. These are then combined into an initial eigenvector. The initial eigenvector is then decomposed into singular values, and the first k principal component eigenvectors are extracted and recombined to generate the fused eigenvector. Here, k is a positive integer. This paper analyzes several fused feature vectors using a CNN model to output the basic fluctuation mode. Specifically, it includes: using CNN as the basic architecture of the model; obtaining multiple fused feature vectors; introducing a sliding time window and performing spatiotemporal alignment stacking: arranging the fused feature vectors from multiple consecutive time points vertically along the time axis according to the time label to generate a two-dimensional spatiotemporal feature matrix; the CNN model includes, but is not limited to, an input layer, a first convolutional layer, a batch normalization layer, a second convolutional layer, and a pooling layer. A center-symmetric mapping layer is set at the end of the CNN, after the pooling layer, which is essentially a fully connected layer. During the online training phase of the CNN model, historical datasets were used for iterative training. Gradient descent was selected, and backpropagation was used to enable the output of the central symmetric mapping layer, which outputs the basic wave mode. The input layer consists of a two-dimensional spatiotemporal feature matrix; the first convolutional layer has 32 3×3 kernels with a stride of (1,1), zero-padding, and ReLU activation to obtain the first feature map; the batch normalization layer performs mean scaling on the first feature map to smooth the control variable drift; and the second convolutional layer has 64 3×3 kernels with a stride of (2,2) to perform spatial dimensionality reduction extraction. The second feature map is output; the number of channels in the second feature map output by the second convolutional layer is determined by the number of convolutional kernels in that layer, i.e., 64. Pooling layer: compresses the second feature map to generate a one-dimensional feature vector; specifically, the pooling layer adopts global average pooling, and by receiving the second feature map with 64 channels output by the second convolutional layer, it performs mean dimensionality reduction compression on the two-dimensional spatial dimension of each channel, compressing the two-dimensional feature matrix of each channel into a scalar feature value to eliminate translation redundancy in spatial position, and concatenating the obtained 64 scalar feature values ​​according to the channel order, thereby directly generating a one-dimensional feature vector with a length of 64. Centrosymmetric mapping layer: Multiple sets of one-dimensional feature vectors are input, and the constraint W(i,j) = W(d_in+1-i, d_out+1-j) is applied. In this formula, W(i,j) represents any element of the weight matrix configured for this layer, d_in represents the length of the one-dimensional feature vector output by the pooling layer, and d_out represents the number of basic fluctuation modes to be extracted; here, d_out is set to 4. This step is a standard matrix operation and will not be elaborated upon here. During the online training iteration phase of the model, the system introduces an orthogonal penalty term into the total loss function. In the formula, L orth W represents the loss due to the orthogonal penalty term. T Let W be the transpose of W, λ be the preset weight coefficients, and I be a 4×4 identity matrix. The expression represents the norm calculation; the optimizer performs backpropagation gradient descent to force the column vectors of the centrosymmetric weight matrix to converge to the orthogonal basis space. The forward logic structure of its centrosymmetric layer is V_mobe=xW+b; where V_mobe is the modal activation vector output by the matrix-vector multiplication operation, which has a dimension of 1×4, x represents the input one-dimensional feature vector with a length of 64, and b is the bias vector with a dimension of 1×4; the centrosymmetric layer is used to force the matrix to be centrosymmetric during initialization and iteration, so that multiple sets of one-dimensional feature vectors are mutually orthogonal, and outputs the basic wave mode; It should be noted that the basic fluctuation mode characterizes the typical fluctuation pattern of the functional fabric antibacterial coating during the spraying process. Specifically, the mode is represented as a time-series vector, including the coating thickness fluctuation value and a timestamp. The mode types include sinusoidal fluctuation, exponentially decaying fluctuation, step fluctuation, and random fluctuation. The above basic fluctuation modes are automatically extracted from the fused feature vector using a CNN model. Specifically, the sinusoidal fluctuation type characterizes the periodic high and low fluctuations in the thickness of the functional fabric antibacterial coating, possibly due to unstable spraying speed; the exponentially decaying type characterizes sudden jumps in the thickness of the functional fabric antibacterial coating, possibly due to nozzle valve opening and closing delays; the step fluctuation type characterizes a gradual decrease in the amplitude of the functional fabric antibacterial coating thickness fluctuation, possibly due to changes in the viscosity of the coating solution; and the random fluctuation type characterizes irregular fluctuations in the functional fabric antibacterial coating, possibly due to environmental temperature and humidity. Once training converges, the system saves the optimal weight parameters and switches the CNN model to online real-time inference mode. Using the trained CNN model, the new fused feature vector is trained. At this time, the CNN model closes the backpropagation and gradient update channels, and the data performs unidirectional forward propagation inference. After passing through the central symmetric mapping layer, the corresponding basic fluctuation mode is output, realizing online fully automatic and accurate recognition.

[0023] Based on the first and second abnormal time points in the analysis of spraying operation parameters and environmental structural parameters, a preset sliding window is introduced to analyze the fluctuation behavior of antibacterial coating thickness under complex working conditions. Through a trained CNN model and orthogonal matrix projection of the central symmetric mapping layer, the corresponding basic fluctuation mode is directly output online in real time. This provides a zero-delay, high-purity data instruction benchmark for the back-end control system to adaptively generate composite control matrices and drive each distributed nozzle to perform precise targeted collaborative recoating.

[0024] S3: Process and analyze the spraying operation parameters and coating distribution characteristics to construct a dynamic behavior tree; based on the dynamic behavior tree, obtain non-standard branch links through tree comparison; by introducing contribution scores and combining with a CNN model, screen the maximum and minimum effective combination modes of the basic fluctuation mode relative to the dynamic behavior tree; The process involves analyzing and processing spraying operation parameters and coating distribution characteristics to construct a dynamic behavior tree. This includes: extracting each spraying execution identifier from the spraying operation parameters, mapping each spraying execution identifier to a spraying execution node, and marking it as the parent node; simultaneously, obtaining the timestamps of adjacent spraying execution identifiers in the spraying operation parameters, retrieving coating distribution characteristics (i.e., the average pixel grayscale value of the image grid) within the time difference range as child nodes; and establishing a directed topological relationship from the parent node to the corresponding child node through the connection of parent and child nodes to construct the dynamic behavior tree. For each spraying execution node, the spray gun height and jet direction, which characterize the node position, are obtained through a laser displacement sensor and a vision sensor, respectively: Laser displacement sensor: vertically mounted on the spray gun action beam, it measures and obtains in real time the absolute vertical distance between the current nozzle end face and the surface of the antibacterial coating fabric to be sprayed, and marks it as the spray gun height; the spray gun height determines the flight time of the atomized droplets in the air. The farther the distance, the greater the probability of the droplets being slowed down by air resistance, evaporating, and deviating from the trajectory; Vision sensor: through vision sensor capture and edge line segment detection algorithm, the central axis of the fluid atomization cone surface sprayed from the nozzle is extracted and marked as the jet direction. The more perpendicular the jet direction is to the antibacterial coating fabric to be sprayed, the higher the energy flux density; Spatial decomposition of the jet direction is performed, specifically: using the vertical normal axis of the horizontal plane where the antibacterial coating fabric to be sprayed is located as the baseline, the angle formed by the jet direction of the spraying execution node and the normal axis is marked as the spray tilt angle; the spray gun height and spray tilt angle are combined: the spray gun height and spray tilt angle are normalized to eliminate the influence of dimensions; during the spray gun normalization process, the minimum safe distance H_min and the maximum safe distance H_max allowed by the process are retrieved, then the normalized spray gun height = (spray gun height - H_min) / (H_max - spray gun height); first use Subtract the normalized spray gun height from the value 1, take the cosine of the normalized spray angle, multiply the two results, and then multiply by the preset correction coefficient to obtain the spray evaluation index, that is, spray evaluation index = (1 - normalized spray gun height) × cos(normalized spray angle) × preset correction coefficient, and use the spray evaluation index as the edge weight of the dynamic behavior tree; where the preset correction coefficient is a custom setting, the value range is 0 to 1, and it is used to adaptively translate and scale the edge weight to the numerical range most suitable for the model processing, which will not be elaborated here; It should be noted that if the normalized value of the spray gun height is closer to 0, while still within the minimum safe distance allowed by the process, it indicates that the spray gun position is almost completely perpendicular to the fabric to be coated with the antibacterial coating. When the spray gun emits viscous liquid, the mass transfer efficiency of the droplets reaching the fabric surface is higher. When the spray gun position is changed, such as when the spray gun is raised higher, the transfer efficiency gets closer to 0 by subtracting the spray gun height from 1. If the normalized value of the spray angle is closer to 0, the cosine value is closer to 1, representing fluid kinetic energy, indicating no spatial slip dissipation. Dividing the two values ​​reflects the effect of the spray gun position being raised or tilted. This results in the relative energy density of the liquid material that finally reaches the surface of the fabric to be coated with the antibacterial coating. The higher the value, the more perfect the coating quality and the more focused the energy; the lower the value, the more serious the process deterioration and the further away from the standard. Through this calculation method, the spray gun height and spray angle are condensed into a single scalar, namely the edge weight, through cosine transformation. When the dynamic behavior tree is compared with the standard behavior tree in subsequent steps, it is not necessary to compare high-dimensional images and complex motion trajectories. Only a simple scalar value subtraction is needed to compare the difference in edge weights, which greatly improves the real-time performance of the online control of the system. Screening the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree, including: The parent-child node relationships of the dynamic behavior tree and the standard behavior tree are compared level by level. If an unknown child node in the dynamic behavior tree that is not included in the standard behavior tree is detected, the rule engine is activated. Using the parent node of the unknown child node as the center and R as the radius, a circle is drawn to obtain the connected nodes and edges falling into the circle. This is identified as a local tree branch starting from the parent node containing the unknown child node and marked as a non-standard branch link. Here, R is greater than 0. The absolute value set of all edge weights in the non-standard link is obtained, and the integral sum of the absolute value set is calculated and marked as the contribution score. It should be noted that by extracting the integral sum and marking it as the contribution score of the non-standard branch link, its value ranges from 0 to 1. This provides a decision-making basis for subsequent generation of composite control matrix and implementation of two-level precise collaborative control of large-scale reconstruction and local fine-tuning. Using the spraying execution identifier as an index, the basic fluctuation mode of the CNN model is projected onto the dynamic behavior tree after N rounds of debugging and optimization, establishing a connection interface between the CNN model and the dynamic behavior tree structure. By identifying the mapping gain, a maximum lag time window is introduced, automatically tracing back along the time axis of the dynamic behavior tree to lock the corresponding non-standard branch links. The Dice loss function is called as the core gradient optimizer, and N rounds of iterative training are performed. Based on the parent node of the dynamic behavior tree, the intersection overlap multiplication operation is performed between the retrieved real-time edge weights and the edge weights analyzed by the CNN model. At the same time, the retrieved real-time edge weights are recorded as true edge weights and assigned true labels. In addition, the maximum lag time window here is set based on the preset dynamic sliding window in the above process, and can be retrieved here. It should be noted that the specific structure of the dynamic behavior tree includes a parent node layer and a child node layer. As mentioned above, the parent node is directly mapped to the spraying execution identifier on the pipeline; the child node is directly mapped to the coating distribution characteristics under the corresponding timestamp of the nozzle. The weight of the directed topological edge connecting the parent and child nodes is defined as the spraying evaluation index. Therefore, each edge of this dynamic behavior tree represents the execution control link between a specific physical nozzle and its corresponding controlled fabric image grid. When non-standard branch links, i.e., local control chains with defective grids, are filtered out through tree comparison, the edges contained in this link are definite and specific. The edge weights analyzed by the CNN model are directly and only assigned to the directed edge of the parent node with the specific spraying execution identifier, pointing to the child node of the coating distribution characteristics. This ensures that the multi-layer tree structure can be directly traversed and mapped, and the interface is extremely accurate. For example: the first layer consists of 8 spray execution identifiers, corresponding to 8 distributed nozzles as parent nodes; the second layer consists of 4 coating distribution features, corresponding to 4 image grids (top left, top right, bottom left, bottom right) as child nodes. A total of 32 weighted edges are formed through the directed topological relationships from parent nodes to child nodes. For a given weighted edge, the current spray gun height time series and spray tilt angle time series of nozzle number 3, as well as the pixel grayscale history vector of the top left grid, are collected and concatenated into a two-dimensional spatiotemporal feature matrix, which is then input into the CNN model and processed through a 32-core first convolutional layer... The batch normalization layer and the 64-core second convolutional layer decouple the spatial and temporal features, and then normalize them into a one-dimensional feature vector of length 64 through the global average pooling layer. This vector is then input to the centrosymmetric mapping layer and outputs a 1×4-dimensional temporal vector, which corresponds to the trigger probability of the four basic wave modes at the current time. The vector is then expanded backward along the prediction time axis and inferred continuously for 10 forward calculations. These 10 sets of vectors are then concatenated horizontally to form a 4×10 two-dimensional matrix, which is marked as the edge weights analyzed by the CNN model. Assuming the set of edges for all non-standard branch links in the dynamic behavior tree is E, for one edge, two sets are constructed: Set B0: the edge weight of this edge analyzed by the CNN model. The interface is then activated, and precise spatial and logical addressing is performed through the directed topological relationships already constructed in the dynamic behavior tree. The spraying execution identifier of the parent node in the currently identified non-standard branch link is automatically extracted and used as a retrieval index to call the corresponding CNN model. The CNN model outputs 1×4-dimensional time vectors corresponding to the four basic wave modes, forming a true feature vector, which is precisely and unidirectionally assigned to the parent node. On the directed topological connection edge of its corresponding child node; by using the physical encoding of the parent node as the dictionary key, the neural network tensor achieves accurate local positioning of multi-layer local edges of the behavior tree; set B1: retrieves the real edge weight of the edge according to the current working condition; by performing the intersection overlap multiplication operation on set B0 and set B1: binarize set B0 and set B1 respectively, calculate the intersection of the two binarized sets, calculate the Dice coefficient, the Dice loss function is calculated as the value 1 minus the Dice coefficient, and used as the loss function for iterative training, and adaptively update the edge weight of each connection edge of the dynamic behavior tree through gradient descent backpropagation; When using the Dice loss function to train the model, in order to ensure the convergence of the CNN model when updating the edge weights of the behavior tree connection edges, the specific training architecture includes setting the label data format, namely the true labels of set B1. Specifically, for the edge of a specific non-standard branch link, the true edge weights are discretized into a 4×Tn-dimensional binary state activation vector after being processed within a continuous Tn control cycle. This vector is used to represent the true triggering state of the branch edge in time, including: 1 for activation and 0 for inactivation; where Tn represents the number of continuous control cycles set within the preset dynamic sliding window. The specific steps for acquiring the label data include: based on the current uneven spraying conditions, by setting up an industrial camera or a precision thickness gauge, the defect image of the finished antibacterial coating is detected in real time. Based on this image, the coating distribution characteristics, i.e., the gray-scale mean of the image grid, are extracted. When the gray-scale mean is detected to deviate from the standard uniformity threshold, it is determined that uneven spraying may occur. At the same time, the dynamic behavior tree under the current uneven spraying conditions is retrieved, each child node is identified, and the control engine introduces the maximum lag time window to trace backward on the time axis, accurately locating the non-standard branch link edge that caused the uneven spraying at that time. The edge weight value is read from the memory and converted into a 4×Tn binary state activation vector containing four basic fluctuation modes. This vector is then assembled to form set B1 and used as the real label. The direct goal of the CNN network is to minimize the Dice loss function value in the total loss function, that is, to maximize the overlap similarity between the set B0 and the real label set B1 in the triggering sequence. By maximizing the distribution overlap between the two in the 4-dimensional orthogonal space, the internal network weights of the CNN model are adaptively updated and corrected through the gradient descent backpropagation algorithm, so that the activation sequence of the edge weights inferred in the forward inference is completely aligned with the actual uneven spraying triggering sequence on site. For example, in the first round of iterative training, due to inaccurate initial model values ​​and the sparsity of non-standard link samples, the Dice loss value was initially recorded as 0.7. When the number of iterations reaches N, the system stops training. At this point, the edge weights updated by all spraying execution nodes in the non-standard link are extracted, arranged horizontally according to the spraying execution identifier, and vertically according to the decoupled basic fluctuation modes. Finally, a structured modal response matrix is ​​generated by cascading. Each element in this matrix precisely quantifies the mapping gain that a nozzle needs to perform when encountering a certain basic fluctuation mode disturbance, which is directly used to guide the digital recoating hedging of subsequent distributed nozzles. The mapping gain is the correlation between the real-time fused feature vector and the CNN model, and is achieved in at least one of the following ways: by dimensional alignment, the Pearson correlation coefficient or Spearman rank correlation coefficient between the fused feature vector and the one-dimensional feature vector input to the central mapping layer in the CNN model is calculated, and the calculated Pearson correlation coefficient matrix or Spearman nonlinear rank correlation scalar is defined as the corresponding mapping gain. It should be noted that the mapping gain directly maps the excitation intensity of the orthogonal features of the uneven spraying task in each batch to a specific permutation and combination of task chains, providing direct basic data for the CNN model to allocate attention weights among the feature maps. The modal response matrix is ​​analyzed, and the contribution score of each basic wave mode update is extracted along the non-standard branch link. The contribution score change curve of each basic wave mode is plotted with the iteration number as the horizontal axis and the contribution score as the vertical axis. The combination of basic wave modes with a curve slope greater than or equal to the standard slope is obtained as the maximum effective combination mode; the combination of basic wave modes with a curve slope less than the standard slope is obtained as the minimum effective combination mode. By sorting the contribution score curves of multiple non-standard branch links by slope and thresholding them, the basic fluctuation mode corresponding to the link with the largest contribution score and the steepest slope is selected and marked as the maximum effective combination mode; the basic fluctuation mode corresponding to the link with the smallest contribution score, although exceeding the safety boundary, is selected and marked as the minimum effective combination mode. These are effectively paired to generate a composite control matrix. This approach ensures both rapid hedging against macroscopic large gradient defects and smooth suppression of microscopic edge local thickness valleys, effectively preventing over-adjustment or under-adjustment during the control process.

[0025] S4: Generate a composite control matrix based on the effective pairing of the maximum effective combination mode and the minimum effective combination mode; convert the composite control matrix into a recoating control command, and send the recoating control command to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command; Based on the effective pairing of the maximum and minimum effective combination modes, features of the maximum and minimum effective combination modes are extracted respectively. Cosine similarity is introduced to calculate the similarity of corresponding features of the maximum and minimum effective combination modes. A composite control matrix is ​​generated by cascading the similarity rows and columns. According to the real-time spraying execution node and nozzle valve opening and closing sequence, the row and column indices of the composite control matrix are retrieved, the element values ​​are locked, and each element value of the composite control matrix is ​​mapped to the write value of the data register. The composite control matrix is ​​converted into a repainting control command. After generating the repainting control command, the process also includes: encapsulating the repainting control command and associated timestamp data into a synchronous control data packet; broadcasting the synchronous control data packet to multiple distributed execution nodes through the fieldbus; each distributed execution node parses the synchronous control data packet and synchronously executes the dynamically adjusted spraying operation within the clock period specified by the timestamp data.

[0026] By broadcasting synchronous control data packets encapsulated with unified timestamp data to multiple distributed execution nodes via fieldbus, multiple sets of independent actuators in space can achieve high-precision concurrent response within the same absolute clock cycle. This effectively overcomes the inherent large hysteresis physical characteristics of high-speed production lines for functional fabrics, and achieves fine online uniform repair of the entire width of the coating without interrupting the fully automated operation of the main production line. This significantly reduces the defect rate of the fabric and improves the overall production efficiency and film quality of the line.

[0027] Example 2: This embodiment of the invention provides a uniform spraying control system for antibacterial coatings on functional fabrics; the system includes: a parameter acquisition module, a modal analysis module, a behavior tree construction module, and an adjustment control module, and the parameter acquisition module, modal analysis module, behavior tree construction module, and adjustment control module are communicatively connected; Parameter acquisition module: Collects spraying operation parameters for each working condition through distributed acquisition nodes, and obtains environmental structure parameters, antibacterial coating fabric to be sprayed, and coating distribution characteristics that have the same time series label as the spraying operation parameters; Modal analysis module: Processes and analyzes spraying operation parameters and environmental structural parameters. Based on the first and second abnormal time points, it analyzes the fluctuation behavior of antibacterial coating thickness. Within a preset dynamic sliding window, it generates basic fluctuation modes through task combination. Behavior tree construction module: Processes and analyzes spraying operation parameters and coating distribution characteristics to construct a dynamic behavior tree; Based on the dynamic behavior tree, non-standard branch links are obtained through tree comparison; By introducing contribution scores and combining with a CNN model, the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are selected. Adjust the control module: Generate a composite control matrix based on the effective pairing of the maximum effective combination mode and the minimum effective combination mode; convert the composite control matrix into a recoating control command, and send the recoating control command to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command.

[0028] In the application, the various formulas mentioned are all calculated by removing dimensions and taking their numerical values. The formulas are derived from the most recent real-world situation by collecting a large amount of data and simulating it with software. The formulas are set by those skilled in the art according to the actual situation.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for controlling the uniform spraying of an antibacterial coating on functional fabrics, characterized in that, The method includes: collecting spraying operation parameters for each working condition through distributed acquisition nodes, and obtaining environmental structure parameters, antibacterial coating fabric to be sprayed, and coating distribution characteristics that have the same time series label as the spraying operation parameters; The spraying operation parameters and environmental structure parameters are processed and analyzed. Based on the first and second abnormal time points, the fluctuation behavior of the antibacterial coating thickness is analyzed. Within the preset dynamic sliding window, the basic fluctuation mode is generated by task combination. The spraying operation parameters and coating distribution characteristics are processed and analyzed to construct a dynamic behavior tree. Based on the dynamic behavior tree, non-standard branch links are obtained through tree comparison. By introducing contribution scores and combining with a CNN model, the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are selected. Based on the effective pairing of the maximum effective combination mode and the minimum effective combination mode, a composite control matrix is ​​generated; the composite control matrix is ​​converted into a recoating control command, and the recoating control command is sent to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command.

2. The method for controlling the uniform spraying of an antibacterial coating on functional fabrics according to claim 1, characterized in that, Collect spraying operation parameters for each working condition, including: In response to the spraying control command, the register status data of the programmable logic controller is read to obtain the spraying operation parameters that characterize the opening and closing sequence of the nozzle valve and the reciprocating motion under different working conditions, including at least the spraying moving speed, spray width control pressure, coating thickness and spraying execution indicator; The system synchronously receives data from the sensor array, including the antibacterial coating fabric to be sprayed, the viscosity of the coating solution, the fabric tension, the temperature, and the humidity. The system then splices the data from the coating solution viscosity, the fabric tension, the temperature, and the humidity to generate environmental structural parameters. Simultaneously, the ultraviolet excitation light source is controlled to irradiate the fabric to be coated with the antibacterial coating. The coating image is captured by an industrial camera and divided into a number of image grid data. The pixel grayscale mean of each image grid data is extracted, and the pixel mean of all image grids is combined to obtain the coating distribution characteristics.

3. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 2, characterized in that, The fundamental wave mode is generated through task combination, including: Based on the environmental structure parameters, extract the first abnormal time point where the environmental structure parameters do not conform to the standard parameter range; The time-domain and frequency-domain features of the spraying operation parameters and environmental structure parameters are extracted by Fourier transform and labeled as feature evaluation parameters. A dynamic sliding window is preset, and the feature evaluation parameters within the dynamic sliding window are collected into a feature evaluation parameter set G. n feature evaluation parameters are selected from G using the multiplication counting principle to construct the spraying unevenness task: A(m,n), where the spraying unevenness task is the fluctuation behavior of the antibacterial coating thickness under a certain working condition; where A(·) represents permutation and combination calculation, and m represents the feature evaluation parameter at any time under a certain working condition. Based on the spraying operation parameters, the second abnormal time point where the coating thickness does not meet the standard thickness range was extracted; After the first abnormal time point is counted, the number of times the second abnormal time point occurs within the preset maximum lag time window is counted. The uneven spraying task is then carried out, and several mutually orthogonal fused feature vectors are decoupled. The CNN model is used to analyze the several fused feature vectors and output the basic fluctuation mode.

4. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 3, characterized in that, Preset dynamic sliding windows include: Collect the spraying operation parameters of all first abnormal time points and the environmental structure parameters of all second abnormal time points within the historical time period; calculate the correlation coefficient using the cross-correlation function and construct the time-delay response curve; identify the time-delay point with the largest absolute value of the correlation coefficient in the time-delay response curve, and extract a pre-set information interval range on both sides of the time-delay point as the center.

5. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 1, characterized in that, The spraying operation parameters and coating distribution characteristics were processed and analyzed to construct a dynamic behavior tree, including: Extract each spraying execution identifier from the spraying operation parameters, map each spraying execution identifier to a spraying execution node, and mark it as the parent node of the dynamic behavior tree; at the same time, obtain the timestamps of adjacent spraying execution identifiers in the spraying operation parameters, and retrieve the coating distribution characteristics within the time difference range as the child nodes of the dynamic behavior tree; For each spray execution node, the spray gun height and jet direction are determined. Spatial decomposition is performed based on the jet direction to determine the spray tilt angle. The spray gun height and spray tilt angle are combined to generate spray evaluation indicators, which serve as edge weights of the dynamic behavior tree.

6. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 3, characterized in that, Screening the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree, including: The parent-child node relationships of the dynamic behavior tree and the standard behavior tree are compared level by level. If an unknown child node not included in the standard behavior tree is detected in the dynamic behavior tree, the rule engine is activated. Using the parent node of the unknown child node as the center, a circle is drawn to determine the local tree branch starting from the parent node containing the unknown child node, and it is marked as a non-standard branch link. The absolute value set of all edge weights in the non-standard link is obtained, the integral sum of the absolute value set is calculated, and it is marked as the contribution score. Using the spraying execution identifier as an index, the basic fluctuation mode of the CNN model is projected onto the dynamic behavior tree after N rounds of debugging and optimization. By identifying the mapping gain, the maximum lag time window is introduced, and the Dice loss function is called to perform N rounds of iterative training. The connection edges of the non-standard branch link are adaptively updated to generate the modal response matrix. The modal response matrix is ​​analyzed, and the contribution score of each basic wave mode update is extracted along the non-standard branch link. The contribution score change curve is plotted. For basic wave mode combinations whose curve change slope is greater than or equal to the standard change slope, the maximum effective combination mode is obtained; for basic wave mode combinations whose curve change rate is less than the standard change slope, the minimum effective combination mode is obtained.

7. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 3, characterized in that, The CNN model is used to analyze several fused feature vectors and output the basic wave modes, including: CNN is used as the basic architecture of the model; Multiple fused feature vectors are obtained, spatiotemporal alignment and stacking are performed to construct a two-dimensional spatiotemporal feature matrix; a centrosymmetric mapping layer is set at the end of the CNN, the CNN model is iteratively trained using historical datasets, the gradient descent algorithm is selected, and through backpropagation, the centrosymmetric mapping layer outputs multiple sets of mutually orthogonal feature vectors, and outputs the basic wave mode. Using a pre-trained CNN model, the new fused feature vector is trained to output the corresponding basic wave mode.

8. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 7, characterized in that, The mapping gain represents the correlation between the fused feature vector and the CNN model, and is achieved through at least one of the following methods: By dimensional alignment, the Pearson correlation coefficient or Spearman rank correlation coefficient is calculated between the fused feature vector and the one-dimensional feature vector input to the central mapping layer in the CNN model.

9. The method for controlling the uniform spraying of an antibacterial coating on a functional fabric according to claim 1, characterized in that, After generating the touch-up control command, it also includes: The repainting control command and the associated timestamp data are encapsulated into a synchronization control data packet; Synchronous control data packets are broadcast to multiple distributed execution nodes via fieldbus; Each distributed execution node parses the synchronization control data packet and synchronously executes the dynamic adjustment spraying operation within the clock period specified by the timestamp data.

10. A system for controlling the uniform spraying of an antibacterial coating on a functional fabric as described in any one of claims 1-9, characterized in that, The system includes: a parameter acquisition module, which collects spraying operation parameters for each working condition through distributed acquisition nodes, and acquires environmental structure parameters, antibacterial coating fabric to be sprayed, and coating distribution characteristics that have the same time series label as the spraying operation parameters; The modal analysis module processes and analyzes the spraying operation parameters and environmental structural parameters. Based on the first and second abnormal time points, it analyzes the fluctuation behavior of the antibacterial coating thickness. Within a preset dynamic sliding window, it generates basic fluctuation modes through task combination. The behavior tree construction module processes and analyzes the spraying operation parameters and coating distribution characteristics to construct a dynamic behavior tree. Based on the dynamic behavior tree, non-standard branch links are obtained through tree comparison. By introducing contribution scores and combining with a CNN model, the maximum and minimum effective combination modes of the basic fluctuation modes relative to the dynamic behavior tree are selected. The control module is adjusted to generate a composite control matrix based on the effective pairing of the maximum and minimum effective combination modes. The composite control matrix is ​​then converted into a recoating control command, which is sent to the corresponding distributed execution node to execute the spraying operation represented by the recoating control command.

Images