Method and system for paint film thickness fluctuation suppression regulation

Abstract

The present application relates to the technical field of coating process control, in particular to a paint film thickness fluctuation suppression regulation method and system. By constructing a virtual cell three-layer memory stack based on position coding pulse space-time alignment, using short-term, medium-term and long-term cyclic memory to generate directional thickness drift expected value, and introducing a temporary regulation alliance negotiation mechanism based on vector direction similarity and interference estimation matrix, combined with the technical scheme of lag compensation and verification write iteration of advance offset timestamp, the present application realizes the advanced prediction, non-conflict collaborative suppression and precise offset of the coating width full-field thickness fluctuation, and significantly improves the uniform consistency of the paint film thickness in dynamic continuous production.

Description

Technical Field

[0001] This invention relates to the field of coating process control technology, specifically to a method and system for suppressing and regulating film thickness fluctuations. Background Technology

[0002] In industrial coating production, the uniformity of paint film thickness is one of the core indicators for evaluating coating quality. Current coating processes typically rely on the relative movement between the die and the substrate, transferring the coating to the substrate surface via a glue pump or extruder. However, due to inherent errors in the mechanical transmission system, pressure pulsations in the feed pipeline, and variations in the rheological properties of the coating itself, the actual paint film thickness formed within the coating area often exhibits a non-uniform spatial distribution.

[0003] Existing paint thickness control technologies face three major drawbacks: First, point-based control using a single sensor ignores the lateral expansion of the fluid and stress transmission, easily leading to a seesaw effect and exacerbating overall thickness fluctuations. Second, the paint coating process involves significant nonlinear time delays, making traditional PID or model predictive controllers prone to overshoot or oscillation. Furthermore, misaligned signal timestamps in high-speed scenarios can cause thickness field mapping errors, exacerbating defects such as paint film streaks. Third, static grid block control ignores strong coupling conflicts between adjacent areas and lacks an interference prediction mechanism, resulting in control commands canceling each other out or causing abnormal accumulation. Summary of the Invention

[0004] To overcome the problems of delayed control response, interference between control actions in different regions leading to increased thickness fluctuations across the entire field, and lack of adaptive memory capability for substrate conditions in existing technologies, this invention provides a method and system for suppressing and controlling paint film thickness fluctuations. This method and system achieves advanced prediction of thickness fluctuations across the entire coating area, non-conflict collaborative suppression, and precise cancellation of actuator lag, significantly improving the uniformity of paint film thickness in dynamic continuous production.

[0005] The technical solution of this application specifically includes: According to one aspect of this application, a method for suppressing and controlling film thickness fluctuations is provided, comprising: During the coating process, the surface data of the coating film is continuously acquired; and the data is spatiotemporally aligned with the pulse signal of the position encoder of the coating machine to generate an initial thickness field snapshot. Based on the initial thickness field snapshot, the coating area is discretized into virtual cells. Each virtual cell compares its own thickness difference with that of its neighboring cells through a short-term circular memory stack to determine whether it is a bulge, depression or steady state and generate a basic thickness evolution sequence. An anti-phase pre-compensation component opposite to the basic thickness evolution sequence is generated through a medium-term circular memory stack. Then, the drift vector synthesized from the basic thickness evolution sequence and the anti-phase pre-compensation component is scaled proportionally through a long-term circular memory stack combined with the current substrate roughness, thereby outputting the expected thickness drift value. Based on the vector direction similarity and spatial adjacency of the expected thickness drift values, virtual cells are aggregated into temporary control alliances. Alliance-level risk index and interference prediction matrix are calculated. Each temporary control alliance applies for a control token from the central scheduler based on the alliance-level risk index and interference prediction matrix. After a limited number of rounds of negotiation and concession based on the prediction matrix, the alliance that obtains the control token will fit and map the expected internal drift values ​​to the target displacement command of the mold head fine-tuning screw or slit throttle. Alliances that do not obtain the control token will enter a forced cooling period. Generate an advance offset timestamp based on the average historical command response delay, perform lag compensation on the target displacement command to offset transmission and mechanical lag, and execute the command according to the advance offset timestamp. After the instruction is executed and a preset paint film stress relaxation time has elapsed, the actual thickness is scanned and compared with the expected drift value. If it is not up to standard, a single micro-compensation is initiated. If it is still out of tolerance, the expected drift value is doubled or halved depending on the deviation. If it is up to standard, the actual thickness data is written back to the three-layer loop memory stack for updating, and the actual thickness field of this round is used as the initial thickness field snapshot input for the next cycle to drive the loop.

[0006] As a further option of the method of the present invention, the method for generating the initial thickness field snapshot includes: An array-type optical interferometric thickness gauge is arranged behind the coating machine mold head along the width direction. An incremental rotary encoder is connected to the end of the coating machine conveyor roller shaft. A synchronous latching signal is sent to the thickness gauge data acquisition card and the encoder counting card with a unified clock source as a reference to obtain the one-dimensional linear array thickness value vector, the cumulative pulse count value of the encoder, and the corresponding high-precision timestamp output by the thickness gauge along the width direction. Based on the pre-calibrated physical installation distance from the die head lip exit plane to the thickness gauge measuring spot plane and the substrate travel speed calculated in real time by encoder pulses, the physical lag time experienced by the thickness data from the die head exit to the measuring point is calculated, and the original thickness value is assigned to the spatial grid point where the die head exit is located before the physical lag time. The scattered data obtained from the mapping is gridded and regularized. For the grid gaps where no data points fall, a linear interpolation algorithm based on Delaunay triangulation is used to fill them. For the regions where data points overlap, an inverse distance weighted average method is used to fuse them, generating an undistorted initial thickness field snapshot matrix.

[0007] As a further option of the method of the present invention, the method for generating the basic thickness evolution sequence of the short-term circular memory stack includes: Traverse all virtual cells, obtain the thickness value of the eight neighboring cells, and calculate the thickness difference between the current virtual cell thickness and the thickness of each neighboring cell. Set a bulge diffusion threshold and a depression convergence threshold. If there is a thickness difference in the neighborhood that exceeds a preset proportion and is positive and the amplitude exceeds the bulge diffusion threshold, the virtual cell is determined to be in a bulge diffusion state. If there is a thickness difference that exceeds a preset proportion and is negative and the absolute value of the amplitude exceeds the depression convergence threshold, the cell is determined to be in a depression convergence state. All other cases are determined to be in a steady state. For the bulging diffusion state, the thickness change rate is proportional to the Laplacian operator of the local thickness gradient; for the concave convergence state, the thickness change rate is proportional to the local thickness curvature; the short-term circular memory stack iteratively updates the thickness at each internal time step, generating a basic thickness evolution sequence that predicts the future set number of steps.

[0008] As a further option of the method of the present invention, the method for generating the inverse pre-compensation component of the intermediate circular memory stack includes: Fast Fourier Transform is performed on the historical thickness evolution sequence stored in the intermediate circular memory stack. Peaks with significantly higher amplitudes than the noise floor and frequencies matching the known mechanical excitation frequencies are identified in the spectrum. The dominant frequency component and its corresponding amplitude and phase are extracted. For each extracted main frequency component, an anti-phase compensation signal with the same frequency, proportional amplitude, and 180° phase difference is generated. All anti-phase compensation signals are superimposed in the time domain to generate an anti-phase pre-compensation component sequence. Based on the actual thickness fluctuation suppression effect after the mid-term pre-compensation component is executed in advance according to recent historical data, the confidence weighting coefficient of the mid-term pre-compensation component is dynamically adjusted. If the recent pre-compensation success rate is higher than the preset first threshold, the confidence weighting coefficient is increased; if the pre-compensation success rate is lower than the preset second threshold, the confidence weighting coefficient is decreased.

[0009] As a further option of the method of the present invention, the method for scaling the long-term circular memory stack includes: Obtain the batch information of the substrate currently being coated and retrieve the corresponding surface roughness parameters from the process database; According to the formula Calculate the scaling factor, where This is the scaling factor. For the surface roughness of the substrate, denoted as the material's adsorption characteristic constant.

[0010] As a further option of the method of the present invention, the method for obtaining the expected thickness drift value includes: Substituting the evolution sequence of the basic thickness of the short-term circular stack output, the inverse pre-compensation component and its confidence weighting coefficient, and the scaling factor of the long-term circular stack output into the formula... The expected thickness drift vector of the virtual cell is calculated, where This is the expected vector of thickness drift. Based on the thickness evolution sequence components For the inverting pre-compensation component, These are confidence-weighted coefficients. The scaling factor is the direction of the expected thickness drift vector, which points in the direction of the expected thickness change. The modulus is... This represents the degree of drastic change expected.

[0011] As a further option of the method of the present invention, the aggregation method of the temporary regulatory alliance includes: Using virtual cells as nodes, if two virtual cells satisfy an eight-neighbor adjacency relationship in space, then an edge is created between the corresponding nodes in the graph structure. Calculate the angle between the expected values ​​of the virtual cell thickness drift at both ends of the connecting edge. If the angle is less than the preset angle similarity threshold, the weight of the edge is set to the first value. If the angle is greater than or equal to the angle similarity threshold, the weight of the edge is set to the second value. Starting from any unvisited edge with a weight of the first value, the virtual cell nodes connected by edges with the first weight are aggregated into an initial cluster using either breadth-first search or depth-first search. After traversing all edges, the entire field is divided into several temporary control alliances.

[0012] As a further option of the method of the present invention, the calculation method of the alliance-level risk index and the intervention prediction matrix includes: The arithmetic mean of the expected thickness drift values ​​of all virtual cells within the temporary control alliance is obtained. The current slope of the basic thickness evolution sequence output from the short-term cyclic memory stack of each virtual cell within the temporary control alliance is then obtained. The root mean square value of the slope sequence is calculated to obtain the dispersion. Combining the deviation weight and the dispersion weight, the alliance-level risk index is calculated using the following formula: ;in, It is an alliance-level risk index. This is the arithmetic mean of the expected drift values ​​of all cells within the temporary control alliance. This represents the root mean square value of the slope sequence within the alliance. and These are the deviation weight and the dispersion weight, respectively. The composite drift vector is obtained by summing and normalizing the expected thickness drift values ​​of all virtual cells within the temporary control alliance. For each pair of adjacent temporary control alliances with adjacent virtual cells on the spatial boundary, the formula is used... Calculate the elements of the interferometric prediction matrix, where For quantitative temporary control alliance Temporary regulatory alliance Matrix elements representing the expected level of interference. Drift vector With drift vector The included angle, Let Euclidean distance be the centroid of the boundary between the two alliances. It is the lateral flow attenuation index.

[0013] As a further option of the method of the present invention, the method of applying for and mapping the control token to the target displacement command includes: Each temporary control alliance submits an application to the central scheduler based on the alliance-level risk index as the initial application weight. The central scheduler checks whether there are strong interference conflicts between the applying alliances based on the interference prediction matrix. For alliance pairs with strong interference conflicts, the alliance with the lower risk index is forced to reduce its application weight. At the end of each round of negotiation, the central scheduler will grant control tokens to the top few temporary control alliances that are not in the mandatory cooling period and have the highest application weight. The negotiation cycle will terminate when the total risk index of all alliances decreases by less than the preset threshold or when the maximum number of negotiation rounds is reached. For a temporary control alliance that has obtained a control token, the thickness drift expectation value of the internal virtual cell is weighted and accumulated in the direction of travel to obtain a comprehensive drift demand curve. The comprehensive drift demand curve is fitted with a cubic spline function using least squares to generate a smooth actuator displacement curve and discretely sampled to obtain the target displacement command of the mold head fine-tuning screw or slit throttle.

[0014] As a further option of the method of the present invention, the method for generating the advance offset timestamp includes: Maintain a fixed-length first-in-first-out delay history queue to record the time interval from the moment the controller generates the digital value of the target displacement command to the moment the displacement sensor sends back a confirmation signal that the actuator has actually reached the command position. According to the formula Calculate the lag-compensated average delay, where To compensate for the average delay, To maintain a fixed length for the delayed history queue, This represents the average latency in the historical latency queue. The standard deviation of the delayed data in the delayed history queue. For index value; The original planned execution time of the target displacement command is received. The average delay due to lag compensation is subtracted from the original planned execution time to generate an advance offset command timestamp. The target displacement command with the additional advance offset command timestamp is then loaded into the hardware timer trigger queue for execution.

[0015] Another aspect of this application provides a coating film thickness fluctuation suppression and control system, the system comprising: The spatiotemporal alignment module is used to continuously acquire paint film surface data during the coating process; it performs spatiotemporal alignment with the pulse signal of the position encoder of the coating machine to generate an initial thickness field snapshot; The drift expectation value output module is used to discretize the coating area into virtual cells based on the initial thickness field snapshot. Each virtual cell compares its own thickness difference with that of its neighboring cells through a short-term circular memory stack to determine whether it is a bulge, depression or steady state and generate a basic thickness evolution sequence. An anti-phase pre-compensation component opposite to the basic thickness evolution sequence is generated through a medium-term circular memory stack. Then, the drift vector synthesized by the basic thickness evolution sequence and the anti-phase pre-compensation component is scaled proportionally through a long-term circular memory stack combined with the current substrate roughness, thereby outputting the thickness drift expectation value. The alliance negotiation and command mapping module is used to aggregate virtual cells into temporary control alliances based on the vector direction similarity and spatial adjacency of the expected thickness drift value. It calculates the alliance-level risk index and interference prediction matrix. Each temporary control alliance applies for a control token from the central scheduler based on the alliance-level risk index and interference prediction matrix. After a limited number of rounds of negotiation and concession based on the prediction matrix, the alliance that obtains the control token will fit and map the expected internal drift value to the target displacement command of the mold head fine-tuning screw or slit throttle. The alliance that does not obtain the control token will enter a forced cooling period. The lag compensation execution module is used to generate an advance offset timestamp based on the average historical command response delay, perform lag compensation on the target displacement command to offset transmission and mechanical lag, and execute the command according to the advance offset timestamp. The verification and write-back driver module is used to scan and obtain the actual thickness after the preset paint film stress relaxation time following instruction execution, and compare it with the expected drift value. If it is not qualified, a single micro-compensation is initiated. If it is still out of tolerance, the expected drift value is doubled or halved according to the positive or negative deviation. If it is qualified, the actual thickness data is written back to the three-layer loop memory stack for update, and the actual thickness field of this round is used as the initial thickness field snapshot of the next cycle input to the driver loop.

[0016] The beneficial effects of this application are as follows: This invention overcomes the challenges of traditional methods by constructing three major mechanisms: spatiotemporally aligned virtual cell multilayer memory prediction, distributed collaborative interference prediction and token negotiation decoupling, and hysteresis adaptive advance compensation. These mechanisms address issues such as the lateral seesaw effect caused by single-point feedback, phase misalignment caused by nonlinear time delay, and strong coupling conflicts between static blocks. In actual coating dynamics, this method can accurately predict fluid widening and stress transmission to the sub-millimeter level spatial grid. Through dynamic aggregation of temporary control alliances driven by directional similarity and bounded negotiation of interference matrices, it reduces inter-region control conflicts by more than 80%. Simultaneously, by using advance timestamp command execution based on historical delay statistics, it cancels the hysteresis of signal transmission and mechanical response to millisecond-level synchronization accuracy. Ultimately, it can reduce the standard deviation of the full-area wet film thickness by 45% to 60%, significantly eliminate stripes and edge build-up defects, and improve the one-time coating yield of continuous production to over 99%. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method for suppressing and controlling coating film thickness fluctuations. Figure 2 Flowchart of the S100 method for suppressing and controlling coating film thickness fluctuations; Figure 3 Flowchart of the S200 method for suppressing and controlling coating film thickness fluctuations; Figure 4 Flowchart of the S300 method for suppressing and controlling coating thickness fluctuations; Figure 5 Flowchart of the S400 method for suppressing and controlling coating thickness fluctuations; Figure 6 The flowchart of the S500 method for suppressing and controlling coating thickness fluctuations. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The theoretical foundation of this invention is built upon three pillars: the theory of virtual cell multilayer spatiotemporal memory, the theory of distributed collaborative interference prediction and token negotiation, and the theory of hysteresis adaptive advance compensation. By constructing a virtual cell grid spatiotemporally aligned with the position-encoded pulse, and maintaining a three-layer circular memory stack within each cell—short-term state evolution, mid-term anti-phase pre-compensation, and long-term roughness scaling—to generate directional thickness drift prediction values, a temporary control alliance dynamic aggregation and finite-round negotiation token allocation mechanism based on vector direction similarity and interference prediction matrix is ​​implemented to resolve inter-region control conflicts. Furthermore, a mechanism for advance offset timestamp instruction execution and post-relaxation micro-compensation verification write-back based on historical instruction response delay is designed. Ultimately, this achieves advance prediction of the entire coating thickness fluctuation, non-conflict collaborative suppression, and precise cancellation of actuator hysteresis, significantly improving the uniformity of coating thickness in dynamic continuous production.

[0020] The definitions of the core variables and the derivation of the formulas are as follows: 1. Spatiotemporal alignment mapping relationship: at time... Wet film thickness data collected by an optical thickness gauge Pulse count with the position encoder of the paint coating machine Alignment is performed. The physical installation distance from the die exit to the thickness gauge's measuring spot is defined as... The substrate travel speed is The time when the data was generated. With sampling time Satisfying Relationships By interpolating the encoder pulses, the sampled data is traced back to the spatial coordinates of the moment the material leaves the die lip. Generate the initial thickness field snapshot matrix .

[0021] 2. Formula for synthesizing the expected value of the thickness drift of the three-layer memory stack of a virtual cell: For the first... Each virtual cell defines the basic thickness evolution sequence component generated by the short-term memory stack as... The inverse pre-compensation component generated by the intermediate memory stack is Long-term memory stacks are based on the roughness of the substrate. The calculated scaling factor is The expected value of cell thickness drift. Represented as: ;in The confidence weighting coefficient for the mid-term pre-compensation component, with a value range of [value range missing]. The scaling factor is dynamically adjusted based on a sliding window statistical analysis of historical pre-compensation success rates. , denoted as the material's adsorption characteristic constant.

[0022] 3. Cell state determination function: Defines the virtual cell. With adjacent cells The thickness difference is Set the bulge diffusion threshold. Convergence threshold of the depression If a neighborhood exists Make and If so, it is determined to be a raised and diffused state; if and If the condition is as described above, it is determined to be a concave convergence state; otherwise, it is a steady-state state. It is a symbolic function.

[0023] 4. Definition of Interference Prediction Matrix Elements: Assume a temporary control coalition. Alliance with neighboring countries The composite drift vectors are respectively and Define the interferometric prediction matrix. China Quantitative Alliance Against the alliance The element of the expected interference level is : ;in The angle between the two resultant drift vectors. Let Euclidean distance be the distance between the centroids of the two alliance boundaries in space. The lateral flow attenuation index is related to the current viscosity of the coating. Positive correlation.

[0024] 5. Formula for calculating the advance offset timestamp of lag compensation: Define the historical instruction response delay record sequence as follows: Then the lag compensation average delay for: ;in The standard deviation of the delay sequence is used as a dynamic margin compensation term. The original planned execution time of the target displacement command is... The advanced offset command timestamp after lag compensation Determined as .

[0025] 6. Deviation Comparison and Drift Expected Value Correction Function: Defines the deviation between the measured thickness and the expected drift value after execution as... If two consecutive verifications fail, the corresponding temporary regulatory alliance will be... The expected thickness drift vector of all cells within the cell Make corrections. The correction function is: ;in To correct the step size coefficient, when the deviation sign is positive, take... To achieve doubling correction, take the sign of the deviation when it is negative. Achieve halving correction.

[0026] 7. Formula for calculating the alliance-level risk index: Temporary regulatory alliance Risk Index It is determined by a combination of internal thickness deviation and evolutionary dispersion: ;in For the number of cells within the alliance, The expected drift amplitude, The current slope of the basic thickness evolution sequence, The average slope within the league. and These are the deviation weight and the dispersion weight, respectively.

[0027] The specific embodiments of the present invention will be described in detail below.

[0028] Example 1: Please see Figure 1 , Figure 1 This diagram illustrates the overall flowchart of a method for suppressing and controlling film thickness fluctuations according to an embodiment of the present invention. The method includes: S100: Spatiotemporal alignment of coating data generates an initial thickness field snapshot; S200: Three-layer memory stack operation outputs the expected drift value of each cell; S300: Alliance negotiation decoupling mapping module target displacement command; S400: Lead timestamp performs displacement compensation for mechanical lag in transmission; S500: Deviation micro-compensation verification write-back memory drives new cycle.

[0029] The specific plan is as follows: Please refer to Figure 2 , Figure 2 A detailed flowchart of stage S100 in an exemplary embodiment of this application is shown, including stages S110 to S130.

[0030] In a method for suppressing and controlling coating film thickness fluctuations, S100 continuously acquires coating film surface data during the coating process and aligns it with the pulse signal of the coating machine position encoder in time and space to generate an initial thickness field snapshot containing the mapping relationship between spatial coordinates and thickness values, providing high-precision reference data for subsequent cell discretization.

[0031] S110: Synchronously collect multi-source data during the coating process; In the implementation steps, an array-type optical interferometric thickness gauge or a line laser profile scanner is arranged behind the coating machine die head along the width direction, and a high-resolution incremental rotary encoder is connected to the end of the coating machine conveyor roller shaft or the encoder interface of the drive motor.

[0032] Specifically, using a unified clock source as a reference, a synchronization latch signal is sent to the thickness gauge data acquisition card and the encoder counting card. Within each sampling period, raw data in three dimensions is acquired: the first dimension is the one-dimensional linear array thickness value vector output by the thickness gauge along the width direction. ,in For the first The width coordinates corresponding to each thickness measurement channel For the first The first sampling time; the second dimension is the encoder's cumulative pulse count value. The third dimension is the high-precision timestamp corresponding to the pulse count value. All raw data, along with a unified timestamp label, is stored in a circular buffer, awaiting spatiotemporal alignment processing.

[0033] S120: Perform spatiotemporal backtracking alignment based on physical installation distance and travel speed; In the implementation steps, for each frame of thickness linear array data acquired in S110 The physical installation distance from the pre-calibrated exit plane of the die lip to the plane of the thickness gauge's measuring spot is determined. And the substrate travel speed calculated in real time by encoder pulses. To perform spatiotemporal inversion calculations on the location where the data was generated.

[0034] In one possible implementation of this embodiment, the spatiotemporal backtracking alignment execution step includes: Based on the time interval between adjacent encoder pulses Physical travel corresponding to a single pulse Real-time calculation of the substrate at the sampling time instantaneous speed of travel To suppress encoder quantization noise, a cutoff frequency of is applied to the velocity sequence. Hz low-pass filtering processing.

[0035] Calculate the physical lag time experienced by the thickness data from the die exit to the measurement point. .

[0036] Thickness sampling time The corresponding physical spatial location should be the mold head outlet at The current position at any given time. Retrieve data based on encoder pulse records. The encoder count value corresponding to the time is used to calculate the coating travel direction. precise coordinates of the axis Finally, the original thickness value Assign to spatial grid points After accumulating and interpolating multiple consecutive frames of data, an initial thickness field snapshot matrix covering the entire coating area is generated. .

[0037] S130: Generate a distortion-free initial thickness field snapshot; In the implementation steps, the scattered data obtained by S120 mapping is subjected to gridding and normalization processing to form a standard two-dimensional matrix format.

[0038] In one possible implementation of this embodiment, the grid normalization process includes: Define a regular grid directional spacing and directional spacing The spacing is set to be consistent with the size of the virtual cell. For mesh gaps where no data points fall, a linear interpolation algorithm based on Delaunay triangulation is used for filling; for overlapping data points, an inverse distance weighted average method is used for fusion. The final generated initial thickness field snapshot matrix... As input for stage S200.

[0039] Please refer to Figure 3 , Figure 3 A detailed flowchart of stage S200 in an exemplary embodiment of this application is shown, including stages S210 to S240.

[0040] In a method for suppressing and controlling coating thickness fluctuations, S200 discretizes the coating area into a virtual cell grid based on the initial thickness field snapshot generated by S100, and constructs an intelligent agent structure containing a short-term, medium-term, and long-term three-layer circular memory stack for each cell. Through the collaborative operation of the three layers of memory, a thickness drift expectation value representing the future thickness evolution trend is generated.

[0041] S210: Discretization of the coated area and initialization of virtual cells; In the implementation steps, the geometric dimensions of the virtual cell are set according to the maximum coating width of the coating machine and the process control accuracy requirements.

[0042] Each virtual cell During initialization, bind the following data structure: cell center space coordinates. ; Snapshot from the initial thickness field Extracted current thickness value A list of pointers to four- or eight-neighbor cells; and three circular memory stack objects for short-term, medium-term, and long-term memory that are about to be initialized.

[0043] S220: Short-term circular stack operations and generation of the basic thickness evolution sequence; In the implementation steps, the short-lived circular memory stack primarily simulates the leveling and diffusion behavior of coatings under the combined effects of surface tension and gravity within a very short time. The short-lived stack maintains a length of... The thickness history sequence is used to capture the current local dynamic trend.

[0044] In one possible implementation of this embodiment, the execution of the short-lived circular stack includes: Traverse all virtual cells Get the thickness value of the eight neighboring cells. Calculate the thickness of the current cell. With each neighboring thickness The difference If there are more than a preset proportion in the neighborhood. It is positive and the amplitude exceeds the bulge diffusion threshold. Then determine the cell unit Currently, the coating is in a raised and spreading state, exhibiting a tendency to extend outwards; if the amount exceeds the preset ratio... It is negative and the absolute value of the amplitude exceeds the concave convergence threshold. If the coating is in a concave convergence state, the surrounding coating tends to converge and fill towards the center; otherwise, it is considered to be in a steady state.

[0045] For the bulging diffusion state, the rate of thickness change is defined as proportional to the Laplacian operator of the local thickness gradient; for the concave convergence state, the rate of thickness change is defined as proportional to the local thickness curvature. The short-term stack at each internal time step... The thickness is iteratively updated to generate a prediction of the future. The basic thickness evolution sequence of the step .sequence It describes the natural evolutionary trajectory of cell thickness over a short period of time under the current local state and neighborhood influence, without external regulatory intervention.

[0046] S230: Perform intermediate loop memory stack operations and generate inverted pre-compensation components; In the implementation steps, the intermediate-cycle memory stack is designed to counteract regular thickness ripples caused by periodic mechanical excitation sources such as glue pump pulsation and conveyor roller eccentricity. The intermediate-cycle stack maintains a length of... The historical thickness evolution sequence library stores the actual thickness change trajectory that occurred in several past control cycles.

[0047] In one possible implementation of this embodiment, the execution of the intermediate circular memory stack includes: A Fast Fourier Transform (FFT) is performed on the historical thickness evolution sequence stored in the intermediate stack to transform the time-domain sequence into the frequency domain. Peaks with amplitudes significantly higher than the noise floor and frequencies matching the known mechanical excitation frequency are identified in the spectrogram. The dominant frequency component is then extracted. and its corresponding amplitude With phase .

[0048] For each extracted main frequency component, an inverse compensation signal with the same frequency, proportional amplitude, and 180° phase difference is generated. All inverse compensation signals are then superimposed in the time domain to generate an inverse pre-compensation component sequence. .sequence This represents the amount of reverse thickness adjustment that needs to be applied in advance to counteract upcoming cyclical fluctuations.

[0049] Based on recent historical data, the confidence weighting coefficient of the mid-term pre-compensation component is dynamically adjusted according to the effectiveness of suppressing actual thickness fluctuations after the mid-term stack pre-compensation component is executed in advance. If the recent pre-compensation success rate is higher than ,but near To increase the level of compensation; if the success rate is lower than ,but Reduce to To prevent erroneous compensation from exacerbating volatility.

[0050] S240: Long-term circular memory stack operations and thickness drift expected value output; In the implementation steps, a long-term circular memory stack stores and maintains physical property parameters related to the currently coated substrate, particularly the substrate surface roughness. Roughness directly affects the boundary slip velocity and effective diffusion coefficient of coatings on the substrate surface.

[0051] In one possible implementation of this embodiment, the execution of the long-lived circular memory stack includes: By communicating with the host computer production execution system, information on the current substrate batch being coated is obtained, and the corresponding surface roughness parameters are retrieved from the process database. .

[0052] According to the formula in the definition of core variables Calculate the scaling factor. This includes the material adsorption characteristic constant. Roughness was pre-determined and stored in a long-term stack through offline rheological experiments on specific coating and substrate combinations. The larger, The smaller the value, the greater the resistance to diffusion of the coating on a rough surface, and the expected drift thickness amplitude should be reduced under the same conditions.

[0053] The evolution sequence of the base thickness of the short-term stack output Inverted pre-compensation component of intermediate stack output and its confidence coefficient Scaling factor of long-term stack output Substitute into the synthesis formula The final thickness drift expected value vector of the virtual cell is calculated. .vector The direction points in the direction of the expected thickness change, and the mold length This represents the degree of drastic change expected.

[0054] The expected thickness drift value of all virtual cells in the entire field. The encapsulated output is used as the input for the S300 stage.

[0055] Please refer to Figure 4 , Figure 4 A detailed flowchart of stage S300 in an exemplary embodiment of this application is shown, including stages S310 to S360.

[0056] In a method for suppressing and controlling film thickness fluctuations, S300 dynamically aggregates virtual cells across the entire field into several temporary control alliances based on the vector direction similarity and spatial adjacency relationship of the expected thickness drift values ​​of each cell generated by S200. Each alliance calculates the alliance-level risk index and the interference prediction matrix, and performs a finite round negotiation through the central scheduler to decouple spatial interference conflicts. Finally, the alliance that obtains the control token maps the expected drift values ​​of its internal cells to the target displacement command of the head actuator through least squares fitting.

[0057] S310: Dynamic alliance aggregation based on vector direction similarity and spatial adjacency; In the implementation step, the thickness drift expectation vector of all virtual cells output by S200 is received. Each vector includes a direction angle. and amplitude .

[0058] In one possible implementation of this embodiment, the dynamic alliance aggregation step includes: Using virtual cells as nodes, if two cells satisfy an eight-neighbor adjacency relationship in space, then an edge is created between the corresponding nodes in the graph.

[0059] For each edge in the graph, the two connected cells and Calculate the angle between the expected thickness drift vectors of the two. If the included angle is less than the preset angle similarity threshold, the weight of that side is set to 1, indicating that the two cells have the same regulatory direction and can belong to the same alliance; otherwise, the weight is set to 0, indicating that there is a conflict of regulatory intentions.

[0060] Starting with any unvisited edge with a weight of 1, a breadth-first or depth-first search is used to aggregate all cell nodes connected by edges with a weight of 1 into an initial cluster. After traversing all edges, the entire field is divided into several initial clusters. For isolated nodes or very small clusters, they are merged into the nearest larger cluster in the same direction based on their vector direction. After this step, the entire field is formed. A temporary regulatory alliance Each alliance represents a spatially contiguous region with the same regulatory tendencies.

[0061] S320: Calculate the alliance-level risk index for each temporary regulatory alliance; In the implementation steps, for each temporary control alliance generated in S310 Calculate an alliance-level risk index that represents the urgency and risk level of internal thickness control. .

[0062] One possible implementation in this embodiment In the implementation method, the risk index The calculations include: Against the alliance The arithmetic mean of the expected drift amplitudes of all cells is calculated, i.e. The larger the average amplitude, the more severe the deviation from the expected thickness.

[0063] Obtain the current slope of the basic thickness evolution sequence output by each cell's short-term memory stack within the alliance. Calculate the root mean square value of the slope sequence within the alliance, i.e. The greater the dispersion, the more inconsistent the thickness evolution behavior within the alliance, and the higher the difficulty and uncertainty of regulation.

[0064] Based on the formula in the definition of core variables Calculate the risk index. Risk Index The higher the value, the more the alliance needs to prioritize obtaining control tokens.

[0065] S330: Calculate the interference prediction matrix between each temporary control alliance and its neighboring alliances; In the implementation steps, when adjacent alliances perform their respective control actions, the lateral flowability of the coating will inevitably cause interference between their respective areas. To quantify the degree of interference, an interference prediction matrix is ​​calculated. .

[0066] In one possible implementation of this embodiment, the calculation of the interferometry prediction matrix includes: Iterate through all alliances, if alliance With Alliance If there are adjacent cells on the spatial boundary, then the two are defined as an adjacent alliance.

[0067] For the alliance The expected drift values ​​of all cells within the consortium are summed and normalized to obtain the composite drift vector of the consortium. Vector This represents the dominant regulatory intention of the alliance as a whole.

[0068] For each pair of adjacent alliances According to the formula in the definition of core variables Computing Consortium Against the alliance Expected level of interference .in for and The angle between them, the closer the angle is to The larger the numerator, the greater the degree. Let Euclidean distance be the centroids of the two alliance boundaries. The closer the distance, the smaller the denominator and the stronger the interference. The lateral flow attenuation index is related to the current viscosity of the coating. Positively correlated, determined in real-time by process parameters. Matrix elements. The larger the size, the more likely it is to form an alliance. Implementing regulation will affect the alliance This will cause greater negative interference.

[0069] S340: Initiate a limited round negotiation cycle to allocate control tokens; In the implementation steps, each alliance bases its actions on its own risk index. and interferometric prediction matrix A weighted control token request is sent to the central scheduler, which then organizes at most [number missing] control tokens. Rounds of negotiations to resolve the conflict.

[0070] In one possible implementation of this embodiment, the execution of the negotiation loop includes: Each alliance Initial application weight In other words, initially, alliances with higher risks are more likely to apply. All alliances submit applications to the central scheduler.

[0071] The central scheduler checks all requesting coalitions for strong interference conflicts based on the interference prediction matrix. If a coalition... and Alliance All applications were submitted, and matrix elements or Exceeding the preset interference threshold If the two are in conflict, then a conflict is determined. For conflicting alliance pairs, the alliance with the lower risk index is forced to adjust its application weights according to the intensity of the conflict. Alliances with lower risk indices made concessions by reducing their weighting.

[0072] At the end of each round of negotiation, the central scheduler will select all applicants who are not in a mandatory cooling-off period and have the highest application weight. Each alliance is granted a control token. Alliances that receive a token are marked as executing. Alliances that do not receive a token will proceed to the next round of applications if their application weight is not zero, or abandon their application in this round if their weight has decayed to near zero.

[0073] After each round of negotiations, the total league-level risk index is calculated. If, after one round of token allocation, the total no longer decreases or decreases by less than a preset threshold, or the maximum number of negotiation rounds has been reached... If this happens, the negotiation cycle terminates.

[0074] S350: Maps the expected drift value of the obtained token alliance to the target displacement command; In the implementation steps, for the alliance that ultimately obtains the control token The expected thickness drift values ​​of all virtual cells within the alliance need to be converted into continuous target displacement commands for the corresponding physical actuators.

[0075] In one possible implementation of this embodiment, the mapping process includes: Based on the die head mechanical design drawings, the coating width range affected by each fine-tuning screw is pre-defined, and actuator numbering is established. With virtual cells A mapping table of coordinate indices.

[0076] For each actuator number ,exist Along the axis, there is a corresponding column of cells with the same coordinates along the width direction. The expected drift values ​​of all cells in this column that have obtained the token alliance are... By weighting and accumulating along the direction, a comprehensive drift demand curve is obtained. .

[0077] To avoid stress concentration or transverse stripes in the coating at the die lip due to a step adjustment of the fine-tuning screw, a cubic spline function is used to represent the discrete composite drift demand curve. Perform least-squares fitting to generate a smooth actuator displacement curve. .Will Discrete sampling is performed in space with the screw spacing as the interval to obtain the first... Each fine-tuning screw is in the corresponding The target displacement command value at the coordinate position.

[0078] S360: Alliance without tokens enters mandatory cooling-off period; In the implementation steps, for temporary control alliances that have not obtained a control token after a negotiation cycle, in order to avoid system oscillation caused by an immediate reaction after neighboring alliances perform control, a forced cooling-off period of a preset duration is imposed.

[0079] In one possible implementation of this embodiment, the duration of the forced cooling period The time required for the coating to complete one stress relaxation cycle on the substrate is set. to The system will ignore any token request requests initiated during the mandatory cooling-off period, even if the alliance's risk index continues to rise. After the mandatory cooling-off period ends, the alliance's state will be reset, and it will be able to participate in the next major cycle of token requests. This mechanism ensures the temporal separation and stable convergence of system control actions.

[0080] Please refer to Figure 5 , Figure 5 A detailed flowchart of stage S400 in an exemplary embodiment of this application is shown, including stages S410 to S430.

[0081] In a method for suppressing and controlling film thickness fluctuations, S400 performs hysteresis compensation processing on the target displacement command generated by S300. By introducing an advance offset command timestamp generated based on the average historical command response delay, the hysteresis effect of signal transmission and mechanical transmission links is offset, ensuring that the actuator responds to the thickness deviation in the correct phase.

[0082] S410: Historical command response delay recording and statistics; In the implementation steps, after each successful execution of a displacement command, a complete command response delay time is recorded. Delay time Defined as the moment when the digital quantity of the target displacement command is generated from the controller. From the moment the high-precision displacement sensor mounted on the die head sends back a confirmation signal indicating that the actuator has actually reached the commanded position. The time interval between stopping.

[0083] In one possible implementation of this embodiment, a fixed length is maintained. A first-in, first-out (FIFO) delayed history queue. Whenever new delayed data is generated, it is pushed to the tail of the queue, and the oldest data at the head of the queue is popped. This is based on the formula in the core variable definition. The average delay time required for lag compensation is calculated in real time. The standard deviation of the delayed data in the queue is used as a dynamic margin to cover the instantaneous jitter of the delay.

[0084] S420: Generate advance offset instruction timestamp; In the implementation steps, for each target displacement command output by S350, the command itself carries an original planned execution time calculated based on the spatial position and the current travel speed. The original plan's execution time was calculated based on the corresponding module exit space. Coordinates and current substrate travel speed This allows us to estimate the flight time required for materials to reach their destination.

[0085] In one possible implementation of this embodiment, the lag compensation module receives the original planned execution time. Average delay calculated with S410 The advance offset instruction timestamp is generated by subtraction. .

[0086] S430: Execute the instruction based on the advance offset instruction timestamp; In the implementation steps, an advance offset timestamp is added. The target displacement command is loaded into the hardware timer trigger queue. When the system's high-precision real-time clock reaches... At any given moment, the hardware immediately sends the displacement command value to the corresponding servo driver or piezoelectric ceramic drive power supply via a high-speed fieldbus.

[0087] In one possible implementation of this embodiment, the actuator begins to move after receiving the instruction. Due to the lead time... This precisely offsets the system's average response lag, and the actuator will... The displacement action is completed and stabilized near the specified time. At this time, the coating fluid just flows through the die head area, thus achieving precise synchronization between the control action and the material passage time.

[0088] Please refer to Figure 6 , Figure 6 A detailed flowchart of stage S500 in an exemplary embodiment of this application is shown, including stages S510 to S560.

[0089] In a method for suppressing and controlling film thickness fluctuations, after the target displacement command is executed and a preset film stress relaxation time has elapsed, the S500 quickly scans and obtains the actual thickness after execution, compares it with the expected drift value output by each virtual cell, performs micro-compensation or memory correction based on the comparison result, and writes the finally verified thickness data back to the three-layer memory stack to drive the next cycle.

[0090] S510: Waiting time for the paint film stress to relax; In the implementation steps, after S430 executes the target displacement command, the effect verification is not performed immediately. This is because after the mold head moves, the wet paint film needs to undergo a brief leveling process, the shear stress inside the paint needs to be released, and the surface tension needs to drive the coating to reach a new equilibrium state.

[0091] In one possible implementation of this embodiment, the preset paint film stress relaxation time The relaxation time spectrum measured by the coating rheometer and the current wet film thickness are combined to determine the optimal thickness.

[0092] S520: Quickly scan to obtain the actual thickness field after execution; In the implementation steps, after the waiting time has elapsed, the thickness gauge is immediately controlled to perform a rapid scan of the painted area where the adjustment action was just performed. The scanning mode employs a local high-density scan only near the adjusted area to save time.

[0093] In one possible implementation of this embodiment, the latest thickness data acquired by scanning is used to generate a snapshot matrix of the actual thickness field after execution, using the same spatiotemporal alignment algorithm in S100. .

[0094] S530: Perform deviation comparison and verification judgment; In the implementation steps, the actual thickness field after execution will be... The expected thickness drift value output by each virtual cell in S240 The expected thickness field it points to Perform a comparison. For each cell... Calculate the deviation value .

[0095] In one possible implementation of this embodiment, a tolerance range is set. Iterate through the temporary control alliances that have obtained control tokens. If all cells in the cell exceed Cellular satisfaction If the verification is successful, the current regulation is deemed acceptable; otherwise, the verification is deemed unacceptable.

[0096] S540: Memory update and loop drive after successful verification; In the implementation steps, if S530 determines that the verification is qualified, it indicates that the prediction model and control execution in this round are accurate and effective.

[0097] Actual thickness data after execution As the new true value, it is pushed into the short-term and medium-term circulating memory stacks of each corresponding cell, and the oldest data in the stacks is popped. The medium-term stack adjusts the confidence coefficients of subsequent pre-compensation components based on the success of this adjustment. Long-term stacks can adjust the adsorption characteristic constants based on the slight difference between the actual and expected diffusion effects. Perform minor online corrections.

[0098] A snapshot of the actual thickness field obtained after this round of execution. It is directly used as the initial thickness field snapshot input in the next control cycle S100 step, driving the entire S100 to S500 process into a new cycle, realizing rolling time domain optimization control.

[0099] S550: Single micro-compensation cycle when verification fails; In the implementation steps, if S530 determines that the verification is unqualified, it does not immediately perform a large-scale model correction, but first attempts to remedy the situation through a single micro-compensation cycle.

[0100] In one possible implementation of this embodiment, the process of a single micro-compensation cycle is as follows: Analyze the deviation distribution that failed verification, and for the local area with the largest deviation, generate a displacement command with an amplitude equal to the original target displacement. to The same direction micro-compensation instruction.

[0101] By bypassing the complex calculations of S200 to S300, the micro-compensation instructions are directly sent to S400 for lag compensation and execution.

[0102] Waiting again after micro-compensation is executed The time is then used for rapid scanning and verification. If the deviation falls back to the tolerance range after micro-compensation, the remediation is considered successful, and the process jumps to S540 to complete the memory update; if the deviation still exceeds the tolerance range after micro-compensation, a significant model mismatch is determined, and the process proceeds to S560.

[0103] S560: Forced correction of drift expectation value when model mismatch occurs; In the implementation steps, if a single micro-compensation cycle of S550 still fails to achieve the required thickness, it indicates that the current virtual cell model has a significant deviation in its description of the physical processes of the region.

[0104] In one possible implementation of this embodiment, the forced correction step is as follows: Temporary control alliance for verification failure All virtual cells within the core variable are modified according to the correction function defined in the core variable definition. The expected thickness drift vector is corrected. Specifically, if the measured thickness continues to be too high, then... This doubles the magnitude of the expected drift vector, thus enhancing the thinning effect of the next round of adjustments; if the measured thickness remains too thin, then... This halves the magnitude of the expected drift vector, thereby reducing the intensity of the next round of regulation and preventing overshoot.

[0105] The revised drift expectation will forcibly overwrite the original value, and at the end of this cycle, data including the experience gained from this failure will be written back to the medium- and long-term memory stacks for adjusting model parameters. At the end of this adjustment cycle, the system will enter the next S100 cycle starting with the current actual thickness field.

[0106] Example 2: This embodiment underwent a 6-month on-site deployment and comparative verification on a gantry-type automated electrostatic spraying line in a large automotive painting workshop. The spraying target was the outer body panel of a passenger car, with a target paint film thickness of 120 micrometers and a process tolerance requirement of [missing information]. Within micrometers. The sprayed coating is a high-solids acrylic amino topcoat, with an application viscosity of 22 seconds as measured by a Ford No. 4 cup. The substrate is a pre-treated and electrophoretically primed steel body panel with a surface roughness of... The roughness was measured to be between 0.8 micrometers and 1.2 micrometers using a stylus roughness tester. The production line cycle time is 45 seconds per car body. The painting machine is a 6-axis robot equipped with a rotary cup atomizer. Within the painting area, there are 5 sets of independently adjustable forming air rings arranged along the direction of travel, along with corresponding high-voltage electrostatic regulating units.

[0107] Monitoring system hardware configuration: Two sets of line laser contour scanners are arranged at the exit of the painting booth along the vehicle's travel direction. Each set covers a lateral width of 600 mm, with a sampling frequency of 2000 Hz and a lateral resolution of 0.5 mm. Absolute encoders are installed at each joint and the travel rail of the robot, synchronized with the main control PLC via an EtherCAT bus. The central computing unit is an industrial computer equipped with an Intel Core i7 processor, 32 GB of memory, and running a real-time Linux operating system. The actuators consist of five sets of high-voltage electrostatic controllers and forming air pressure proportional valves, each controlling an area of ​​approximately 120 mm in width.

[0108] The specific parameter configurations used during implementation are as follows: Virtual cell side length The length of the short-term memory stack in the S200 is set to 5 mm. Set to 80, internal iteration step size The duration is 3 milliseconds; the medium-term memory stack stores 15 historical periods of data; the long-term memory stack contains material adsorption characteristic constants. The surface calibration value is 0.15 based on the electrophoretic primer. The angular similarity threshold in S300 is set to 12 degrees, with a maximum negotiation round. Three rounds, interference threshold The length of the delayed history queue in S400 is 0.3. The stress relaxation time in S500 is 120. Based on the leveling characteristics of the paint, a tolerance of 25 milliseconds was set. Set as Micrometer.

[0109] I. Detailed records of typical regulatory cases; Case Study: Coordinated Suppression and Conflict Resolution of Large-Area Thickness Deviations in the Hood Area of ​​a Vehicle Body; During a certain batch of production, the system detected a large area of ​​thinner paint film on the leading edge of the hood, with an average thickness of 98 micrometers and a deviation of -22 micrometers. Meanwhile, the adjacent fender transition area showed a slight increase in thickness, with an average thickness of 128 micrometers and a deviation of +8 micrometers. Under traditional control strategies, if thickening commands for the hood area and thinning commands for the fender area are sent simultaneously, the spatial distance between the two areas is only about 80 millimeters, and electrostatic spraying exhibits an edge-enhancing effect. This easily leads to the thickened paint migrating to the fender area, creating a control dilemma where one area cannot compensate for the other.

[0110] The execution process of this method is as follows: S100 Stage: The system acquires the first frame of valid thickness data 3.2 seconds after the vehicle body enters the paint booth. After processing with a spatiotemporal alignment algorithm, the 1.8-meter physical lag caused by the scanner's installation at the paint booth exit is precisely compensated, generating an initial thickness field snapshot that strictly corresponds to the vehicle body's spatial coordinates. The snapshot clearly shows a low-thickness area of ​​approximately 220 square centimeters, roughly trapezoidal in shape, at the leading edge of the hood.

[0111] S200 stage: Approximately 880 virtual cells were generated in this region. The short-term memory stack determined that the cells in the hood region were in a concave convergence state, based on the basic thickness evolution sequence. Without intervention, the thickness is predicted to decrease by approximately 5 micrometers within the next 0.5 seconds. Mid-term memory stack analysis of 15 historical cycles did not detect a main frequency component matching the known mechanical pulsation frequency; therefore, an inverted pre-compensation component was identified. The amplitude is close to 0. The long-term memory stack is based on the current substrate roughness. Micrometer scaling factor .

[0112] S300 stage: Expected drift vector of all cells in the thinner area of ​​the hood. With consistent orientation and a standard deviation of only 8.5 degrees, the system aggregates them into a temporary control alliance. The thicker areas of adjacent fenders converge into an alliance. .alliance Risk Index The calculation is 0.78, Alliance Risk Index The calculated value is 0.32. The interferometric prediction matrix calculation shows that... and Composite drift vector angle Degree, value of interference element Far exceeding the interference threshold This indicates that if the two alliances act simultaneously, a serious conflict will occur.

[0113] In the subsequent three rounds of negotiation: Round 1: Initial application weight , Initial weights Due to the existence of strong interference, Weight adjustment . Obtain the token. Those who do not obtain a token will enter a mandatory cooling-off period.

[0114] Round 2: After implementation, the risk index dropped to 0.41. The risk index naturally rose to 0.38 due to inaction during the forced cooling period. The interference intensity decreased to... , Weight adjustment It has not yet exceeded The weight corresponding to residual risk Continue to hold the tokens.

[0115] Round 3: The total risk index of the entire event decreased by only 3% compared to Round 2, which is below the convergence threshold of 5%, and the negotiation was terminated.

[0116] S400 Phase: Alliance The expected internal drift value is fitted using least squares to generate target displacement commands for the 2nd, 3rd, and 4th groups of the 5 actuators, specifically increasing the high-voltage electrostatic voltage by 2.8 kV, 3.5 kV, and 2.9 kV respectively. The hysteresis compensation module calculates the historical average delay... Milliseconds allow the command to be triggered ahead of time.

[0117] S500 Phase: After execution, wait 25 milliseconds for relaxation. A subsequent scan shows the average thickness in the hood area has rebounded to 117 micrometers. The fender area thickness, due to undisturbed operation during the forced cooling period, naturally leveled off to 124 micrometers. The overall thickness range has narrowed from 30 micrometers to 7 micrometers, with all deviations falling within the acceptable range. Micrometer tolerance range. Verification passed, data written back to memory stack.

[0118] II. Long-term operating performance comparison experiment; To objectively evaluate the technical effectiveness of the method of this invention, a 30-day A / B comparison experiment was designed during the stable operation phase of the production line. During the experiment, the same painting robot system was alternately switched between a traditional independent control mode based on single-point PID and the mode of this invention on odd and even workdays, with each mode running for 15 production days. The collected comparison data are as follows.

[0119] Table 1 Comparison of film thickness quality indicators under the two control modes Table 2 Comparison of response performance under different control scenarios Table 3. Average time consumption statistics for each major step in the present invention. Table 4. Statistics on the Conflict Resolution Effectiveness of the Consultation Mechanism III. Experimental Conclusions; The above experimental data fully demonstrates that the paint film thickness fluctuation suppression and control method and system provided by this invention exhibit excellent performance in actual automotive painting production environments. Compared with the traditional PID independent control mode, the method of this invention reduces the thickness range by 62.7%, increases the process capability index to 1.42, reduces the rework rate by 90.5%, and shortens the adaptation time to changing working conditions by approximately 70%. The effective cooperation of the three-layer memory stack and the alliance negotiation mechanism successfully resolves 98.4% of the inter-region control conflicts. A single complete cycle takes 26.0 milliseconds, fully meeting the real-time requirement of a 45-second cycle time for the production line. This invention effectively solves the problems of lagging control response, inter-region action interference, and lack of substrate condition adaptive capability in existing technologies, possessing significant technological advancements and industrial application value.

[0120] Example 3: A coating film thickness fluctuation suppression and control system includes six interfaces, including: 1. System Overview Interface: Displays a 3D thermal cloud map of the wet film thickness across the entire field, with real-time peak and valley values ​​marked.

[0121] 2. Spatiotemporal alignment interface: By backtracking the sampling data through point cloud, the true spatial coordinates of the material are restored by combining the spiral pulse axis.

[0122] 3. Virtual cell interface: The honeycomb columnar array represents discrete cells, and the thickness drift expectation of the three-layer memory stack is presented in three dimensions.

[0123] 4. Alliance Negotiation Interface: Simulates alliance aggregation using a network of glowing nodes, visualizing token allocation and interference risk index.

[0124] 5. Lag Compensation Interface: The dual waveform curves compare the original command and the compensation signal, demonstrating the synchronization effect of the advance offset timestamp.

[0125] 6. Deviation Verification Interface: Compare the expected surface with the measured surface from top to bottom, and mark the deviation value with lines to drive model correction.

[0126] The system includes: The spatiotemporal alignment module is used to continuously acquire paint film surface data during the coating process; it performs spatiotemporal alignment with the pulse signal of the position encoder of the coating machine to generate an initial thickness field snapshot; The drift expectation value output module is used to discretize the coating area into virtual cells based on the initial thickness field snapshot. Each virtual cell compares its own thickness difference with that of its neighboring cells through a short-term circular memory stack to determine whether it is a bulge, depression or steady state and generate a basic thickness evolution sequence. An anti-phase pre-compensation component opposite to the basic thickness evolution sequence is generated through a medium-term circular memory stack. Then, the drift vector synthesized by the basic thickness evolution sequence and the anti-phase pre-compensation component is scaled proportionally through a long-term circular memory stack combined with the current substrate roughness, thereby outputting the thickness drift expectation value. The alliance negotiation and command mapping module is used to aggregate virtual cells into temporary control alliances based on the vector direction similarity and spatial adjacency of the expected thickness drift value. It calculates the alliance-level risk index and interference prediction matrix. Each temporary control alliance applies for a control token from the central scheduler based on the alliance-level risk index and interference prediction matrix. After a limited number of rounds of negotiation and concession based on the prediction matrix, the alliance that obtains the control token will fit and map the expected internal drift value to the target displacement command of the mold head fine-tuning screw or slit throttle. The alliance that does not obtain the control token will enter a forced cooling period. The lag compensation execution module is used to generate an advance offset timestamp based on the average historical command response delay, perform lag compensation on the target displacement command to offset transmission and mechanical lag, and execute the command according to the advance offset timestamp. The verification and write-back driver module is used to scan and obtain the actual thickness after the preset paint film stress relaxation time following instruction execution, and compare it with the expected drift value. If it is not qualified, a single micro-compensation is initiated. If it is still out of tolerance, the expected drift value is doubled or halved according to the positive or negative deviation. If it is qualified, the actual thickness data is written back to the three-layer loop memory stack for update, and the actual thickness field of this round is used as the initial thickness field snapshot of the next cycle input to the driver loop.

[0127] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0128] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0129] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0130] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0131] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0132] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method of paint film thickness fluctuation suppression regulation, characterized by, include: Continuously acquire surface data of the paint film during the coating process; Spatiotemporal alignment with the pulse signal of the position encoder of the coating machine is performed to generate an initial thickness field snapshot; Based on the initial thickness field snapshot, the coating area is discretized into virtual cells. Each virtual cell compares its own thickness difference with that of its neighboring cells through a short-term circular memory stack to determine whether it is a bulge, depression or steady state and generate a basic thickness evolution sequence. An anti-phase pre-compensation component opposite to the basic thickness evolution sequence is generated through a medium-term circular memory stack. Then, the drift vector synthesized from the basic thickness evolution sequence and the anti-phase pre-compensation component is scaled proportionally through a long-term circular memory stack combined with the current substrate roughness, thereby outputting the expected thickness drift value. Based on the vector direction similarity and spatial adjacency of the expected thickness drift value, virtual cells are aggregated into temporary control alliances. Alliance-level risk index and interference prediction matrix are calculated. Each temporary control alliance applies for a control token from the central scheduler based on the alliance-level risk index and interference prediction matrix. After a limited number of rounds of negotiation and concession based on the prediction matrix, the alliance that obtains the control token will fit and map the expected internal drift value to the target displacement command of the mold head fine-tuning screw or slit throttle. Alliances that do not obtain the control token will enter a forced cooling period. Generate an advance offset timestamp based on the average historical command response delay, perform lag compensation on the target displacement command to offset transmission and mechanical lag, and execute the command according to the advance offset timestamp. After the instruction is executed and a preset paint film stress relaxation time has elapsed, the actual thickness is scanned and compared with the expected drift value. If it is not up to standard, a single micro-compensation is initiated. If it is still out of tolerance, the expected drift value is doubled or halved depending on the deviation. If it is up to standard, the actual thickness data is written back to the three-layer loop memory stack for updating, and the actual thickness field of this round is used as the initial thickness field snapshot input for the next cycle to drive the loop.

2. The method for suppressing and controlling film thickness fluctuations according to claim 1, characterized in that, The method for generating the initial thickness field snapshot includes: An array-type optical interferometric thickness gauge is arranged behind the coating machine mold head along the width direction. An incremental rotary encoder is connected to the end of the coating machine conveyor roller shaft. A synchronous latching signal is sent to the thickness gauge data acquisition card and the encoder counting card with a unified clock source as a reference to obtain the one-dimensional linear array thickness value vector, the cumulative pulse count value of the encoder, and the corresponding high-precision timestamp output by the thickness gauge along the width direction. Based on the pre-calibrated physical installation distance from the die head lip exit plane to the thickness gauge measuring spot plane and the substrate travel speed calculated in real time by encoder pulses, the physical lag time experienced by the thickness data from the die head exit to the measuring point is calculated, and the original thickness value is assigned to the spatial grid point where the die head exit is located before the physical lag time. The scattered data obtained from the mapping is gridded and regularized. For the grid gaps where no data points fall, a linear interpolation algorithm based on Delaunay triangulation is used to fill them. For the regions where data points overlap, an inverse distance weighted average method is used to fuse them, generating an undistorted initial thickness field snapshot matrix.

3. The method for suppressing and controlling film thickness fluctuations according to claim 1, characterized in that, The method for generating the basic thickness evolution sequence of the short-term circular memory stack includes: Traverse all virtual cells, obtain the thickness value of the eight neighboring cells, and calculate the thickness difference between the current virtual cell thickness and the thickness of each neighboring cell. Set a bulge diffusion threshold and a depression convergence threshold. If there is a thickness difference in the neighborhood that exceeds a preset proportion and is positive and the amplitude exceeds the bulge diffusion threshold, the virtual cell is determined to be in a bulge diffusion state. If there is a thickness difference that exceeds a preset proportion and is negative and the absolute value of the amplitude exceeds the depression convergence threshold, the cell is determined to be in a depression convergence state. All other cases are determined to be in a steady state. For the bulging diffusion state, the thickness change rate is proportional to the Laplacian operator of the local thickness gradient; for the concave convergence state, the thickness change rate is proportional to the local thickness curvature; the short-term circular memory stack iteratively updates the thickness at each internal time step, generating a basic thickness evolution sequence that predicts the future set number of steps.

4. The paint film thickness fluctuation suppression regulation method according to claim 3, characterized by, The method for generating inverse pre-compensation components using the intermediate circular memory stack includes: Fast Fourier Transform is performed on the historical thickness evolution sequence stored in the intermediate circular memory stack. Peaks with significantly higher amplitudes than the noise floor and frequencies matching the known mechanical excitation frequencies are identified in the spectrum. The dominant frequency component and its corresponding amplitude and phase are extracted. For each extracted main frequency component, an anti-phase compensation signal with the same frequency, proportional amplitude, and 180° phase difference is generated. All anti-phase compensation signals are superimposed in the time domain to generate an anti-phase pre-compensation component sequence. Based on the actual thickness fluctuation suppression effect after the mid-term pre-compensation component is executed in advance according to recent historical data, the confidence weighting coefficient of the mid-term pre-compensation component is dynamically adjusted. If the recent pre-compensation success rate is higher than the preset first threshold, the confidence weighting coefficient is increased; if the pre-compensation success rate is lower than the preset second threshold, the confidence weighting coefficient is decreased.

5. The paint film thickness fluctuation suppression regulation method according to claim 4, wherein The method for scaling the long-term circular memory stack includes: Obtain the batch information of the substrate currently being coated and retrieve the corresponding surface roughness parameters from the process database; According to the formula Calculate the scaling factor, where This is the scaling factor. For the surface roughness of the substrate, denoted as the material's adsorption characteristic constant.

6. The paint film thickness fluctuation suppression regulation method according to claim 5, wherein The method for obtaining the expected thickness drift value includes: Substituting the evolution sequence of the basic thickness of the short-term circular stack output, the inverse pre-compensation component and its confidence weighting coefficient, and the scaling factor of the long-term circular stack output into the formula... The expected thickness drift vector of the virtual cell is calculated, where This is the expected vector of thickness drift. Based on the thickness evolution sequence components For the inverting pre-compensation component, These are confidence-weighted coefficients. The scaling factor is the direction of the expected thickness drift vector, which points in the direction of the expected thickness change. The modulus is... This represents the degree of drastic change expected.

7. The paint film thickness fluctuation suppression regulation method according to claim 1, characterized by, The aggregation method of the temporary regulatory alliance includes: Using virtual cells as nodes, if two virtual cells satisfy an eight-neighbor adjacency relationship in space, then an edge is created between the corresponding nodes in the graph structure. Calculate the angle between the expected values ​​of the virtual cell thickness drift at both ends of the connecting edge. If the angle is less than the preset angle similarity threshold, the weight of the edge is set to the first value. If the angle is greater than or equal to the angle similarity threshold, the weight of the edge is set to the second value. Starting from any unvisited edge with a weight of the first value, the virtual cell nodes connected by edges with the first weight are aggregated into an initial cluster using either breadth-first search or depth-first search. After traversing all edges, the entire field is divided into several temporary control alliances.

8. The method for suppressing and controlling film thickness fluctuations according to claim 7, characterized in that, The calculation methods for the alliance-level risk index and the intervention prediction matrix include: The arithmetic mean of the expected thickness drift values ​​of all virtual cells within the temporary control alliance is obtained. The current slope of the basic thickness evolution sequence output from the short-term cyclic memory stack of each virtual cell within the temporary control alliance is then obtained. The root mean square value of the slope sequence is calculated to obtain the dispersion. Combining the deviation weight and the dispersion weight, the alliance-level risk index is calculated using the following formula: ;in, It is an alliance-level risk index. This is the arithmetic mean of the expected drift values ​​of all cells within the temporary control alliance. This represents the root mean square value of the slope sequence within the alliance. and These are the deviation weight and the dispersion weight, respectively. The composite drift vector is obtained by summing and normalizing the expected thickness drift values ​​of all virtual cells within the temporary control alliance. For each pair of adjacent temporary control alliances with adjacent virtual cells on the spatial boundary, the formula is used... Calculate the elements of the interferometric prediction matrix, where For quantitative temporary control alliance Temporary regulatory alliance Matrix elements representing the expected level of interference. Drift vector With drift vector The included angle, Let Euclidean distance be the centroids of the boundaries of the two alliances. This is the lateral flow attenuation index.

9. The paint film thickness fluctuation suppression regulation method according to claim 8, wherein The method for requesting and mapping the control token to the target displacement command includes: Each temporary control alliance submits an application to the central scheduler based on the alliance-level risk index as the initial application weight. The central scheduler checks whether there are strong interference conflicts between the applying alliances based on the interference prediction matrix. For alliance pairs with strong interference conflicts, the alliance with the lower risk index is forced to reduce its application weight. At the end of each round of negotiation, the central scheduler will grant control tokens to the top few temporary control alliances that are not in the mandatory cooling period and have the highest application weight. The negotiation cycle will terminate when the total risk index of all alliances decreases by less than the preset threshold or when the maximum number of negotiation rounds is reached. For a temporary control alliance that has obtained a control token, the thickness drift expectation value of the internal virtual cell is weighted and accumulated in the direction of travel to obtain a comprehensive drift demand curve. The comprehensive drift demand curve is fitted with a cubic spline function using least squares to generate a smooth actuator displacement curve and discretely sampled to obtain the target displacement command of the mold head fine-tuning screw or slit throttle.

10. The paint film thickness fluctuation suppression regulation method according to claim 1, characterized by, The method for generating the advance offset timestamp includes: Maintain a fixed-length first-in-first-out delay history queue to record the time interval from the moment the controller generates the digital value of the target displacement command to the moment the displacement sensor sends back a confirmation signal that the actuator has actually reached the command position. According to the formula Calculate the lag-compensated average delay, where To compensate for the average delay, To maintain a fixed length for the delayed history queue, This represents the average latency in the historical latency queue. The standard deviation of the delayed data in the delayed history queue. For index value; The original planned execution time of the target displacement command is received. The average delay due to lag compensation is subtracted from the original planned execution time to generate an advance offset command timestamp. The target displacement command with the additional advance offset command timestamp is then loaded into the hardware timer trigger queue for execution.

11. The paint film thickness fluctuation suppression and regulation system of any one of claims 1-10, wherein, The system includes: The spatiotemporal alignment module is used to continuously acquire paint film surface data during the coating process; Spatiotemporal alignment with the pulse signal of the position encoder of the coating machine is performed to generate an initial thickness field snapshot; The drift expectation value output module is used to discretize the coating area into virtual cells based on the initial thickness field snapshot. Each virtual cell compares its own thickness difference with that of its neighboring cells through a short-term circular memory stack to determine whether it is a bulge, depression or steady state and generate a basic thickness evolution sequence. An anti-phase pre-compensation component opposite to the basic thickness evolution sequence is generated through a medium-term circular memory stack. Then, the drift vector synthesized by the basic thickness evolution sequence and the anti-phase pre-compensation component is scaled proportionally through a long-term circular memory stack combined with the current substrate roughness, thereby outputting the thickness drift expectation value. The alliance negotiation and command mapping module is used to aggregate virtual cells into temporary control alliances based on the vector direction similarity and spatial adjacency of the expected thickness drift value. It calculates the alliance-level risk index and interference prediction matrix. Each temporary control alliance applies for a control token from the central scheduler based on the alliance-level risk index and interference prediction matrix. After a limited number of rounds of negotiation and concession based on the prediction matrix, the alliance that obtains the control token will fit and map the expected internal drift value to the target displacement command of the mold head fine-tuning screw or slit throttle. The alliance that does not obtain the control token will enter a forced cooling period. The lag compensation execution module is used to generate an advance offset timestamp based on the average historical command response delay, perform lag compensation on the target displacement command to offset transmission and mechanical lag, and execute the command according to the advance offset timestamp. The verification and write-back driver module is used to scan and obtain the actual thickness after the preset paint film stress relaxation time following instruction execution, and compare it with the expected drift value. If it is not qualified, a single micro-compensation is initiated. If it is still out of tolerance, the expected drift value is doubled or halved according to the positive or negative deviation. If it is qualified, the actual thickness data is written back to the three-layer loop memory stack for update, and the actual thickness field of this round is used as the initial thickness field snapshot of the next cycle input to the driver loop.

Images