Lateral closed-loop control method and system for coating machine based on fuzzy PID

Abstract

The invention relates to the technical field of automatic control, in particular to a coating machine transverse closed-loop control method and system based on fuzzy PID. The method comprises the following steps: arranging a plurality of pressure sensors on the inner side of the boundary of a coating head to collect the pressure value of each measuring point in real time, and calculating the pressure difference value between the adjacent sensors to form a pressure gradient sequence; the abnormal degree of the boundary flow state is recognized according to the distribution mode of the pressure gradient sequence, when the pressure gradient exceeds a normal fluctuation range, a boundary disturbance event is marked, and boundary state monitoring data is obtained; and measuring the actual coating thickness of each transverse position, and comparing the actual coating thickness with a preset target thickness to calculate a thickness deviation value. According to the method, hysteresis and rigid constraints of traditional PID control are broken through, restrictions of difficulty in boundary disturbance modeling, control feedback delay and frequent manual parameter adjustment are overcome, and transverse closed-loop control based on data perception and intelligent judgment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic control, and in particular to a coating machine transverse closed-loop control method and system based on fuzzy PID. BACKGROUND

[0002] The coating machine transverse closed-loop control is an industrial automation system that ensures the uniformity of the transverse distribution of coating materials on the substrate surface through real-time monitoring and adjustment. During the coating process, the substrate (such as film, paper, etc.) moves continuously along the machine direction (MD direction), while the coating head distributes the coating material in the transverse direction (CD direction) perpendicular to the machine direction. The transverse closed-loop control is achieved by real-time monitoring of the transverse coating thickness or weight distribution through sensors, and then adjusting the coating material flow of each region of the coating head according to the feedback signal to form a complete feedback control loop. The current problems of the coating machine control system include: due to the change of the fluid mechanics conditions at the boundary, the flow pattern of the coating material at the boundary region of the coating head is different from that in the middle region, and the coating quality at the boundary region of the coating head cannot reach the level of the middle region. When the production conditions change, such as changing different specifications of products or adjusting the production speed, the system cannot automatically adapt to the new working conditions, and the operator needs to re-adjust the control parameters, which not only consumes time, but also the adjustment effect is largely dependent on the experience level of the operator. In actual production, the operator has to adopt a conservative control strategy to ensure product quality by increasing the waste range of the boundary region. Although this method can ensure the product pass rate, the cost is significant material waste and cost increase. SUMMARY

[0003] Therefore, it is necessary to provide a coating machine transverse closed-loop control method and system based on fuzzy PID to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned purpose, the coating machine transverse closed-loop control method based on fuzzy PID comprises the following steps:

[0005] Step S1: Real-time acquisition of pressure values of each measurement point by arranging multiple pressure sensors inside the boundary of the coating head, and calculation of pressure difference values between adjacent sensors to form a pressure gradient sequence; according to the distribution pattern of the pressure gradient sequence, identifying the abnormal degree of the boundary flow state, marking as a boundary disturbance event when the pressure gradient exceeds the normal fluctuation range, and obtaining boundary state monitoring data;

[0006] Step S2: Measuring the actual coating thickness of each transverse position and comparing it with the preset target thickness to calculate the thickness deviation value; according to the boundary state monitoring data, time correlating the thickness deviation value to identify the thickness change pattern caused by the boundary disturbance event; based on the thickness change pattern, establishing the corresponding relationship between the boundary disturbance and the thickness influence, and generating boundary influence prediction data;

[0007] Step S3: constructing a fuzzy control rule base according to the boundary influence prediction data, wherein the fuzzy control rule base takes the intensity of the boundary disturbance and the degree of the thickness influence as fuzzy input variables, and takes the control adjustment amount as a fuzzy output variable; when a new boundary disturbance event is detected, a predictive control amount is generated by using the fuzzy control rule base; and a feedback control amount is obtained by performing PID calculation according to the thickness deviation value;

[0008] Step S4: performing real-time adjustment of the coating parameters according to the feedback control amount and the predictive control amount, continuously monitoring the adjustment effect and recording the processing result of the boundary disturbance event, and automatically adjusting the weight coefficient of the corresponding rule in the fuzzy control rule base when the processing effect is poor, so as to accumulate and optimize to form an adaptive control strategy.

[0009] The application also provides a fuzzy PID-based coating machine transverse closed-loop control system for executing the fuzzy PID-based coating machine transverse closed-loop control method described above, and the fuzzy PID-based coating machine transverse closed-loop control system comprises:

[0010] A boundary monitoring module is configured to collect pressure values of each measurement point in real time by arranging a plurality of pressure sensors inside the boundary of the coating head, to calculate pressure difference values between adjacent sensors to form a pressure gradient sequence, to identify the abnormal degree of the boundary flow state according to the distribution mode of the pressure gradient sequence, to mark a boundary disturbance event when the pressure gradient exceeds the normal fluctuation range, and to obtain boundary state monitoring data.

[0011] An association analysis module is configured to measure the actual coating thickness of each transverse position and compare it with the preset target thickness to calculate a thickness deviation value, to perform time association on the thickness deviation value according to the boundary state monitoring data to identify a thickness change mode caused by the boundary disturbance event, and to establish a corresponding relationship between the boundary disturbance and the thickness influence based on the thickness change mode to generate boundary influence prediction data.

[0012] An intelligent decision-making module is configured to construct a fuzzy control rule base according to the boundary influence prediction data, wherein the fuzzy control rule base takes the intensity of the boundary disturbance and the degree of the thickness influence as fuzzy input variables, and takes the control adjustment amount as a fuzzy output variable; when a new boundary disturbance event is detected, a predictive control amount is generated by using the fuzzy control rule base; and a feedback control amount is obtained by performing PID calculation according to the thickness deviation value.

[0013] An adaptive optimization module is configured to perform real-time adjustment of the coating parameters according to the feedback control amount and the predictive control amount, to continuously monitor the adjustment effect and record the processing result of the boundary disturbance event, and to automatically adjust the weight coefficient of the corresponding rule in the fuzzy control rule base when the processing effect is poor, so as to accumulate and optimize to form an adaptive control strategy.

[0014] The application realizes efficient detection, accurate identification and intelligent response to boundary disturbance in the coating process by constructing a fuzzy PID-based coating machine transverse closed-loop control system, significantly improving the uniformity of coating thickness and product quality stability. First, by arranging pressure sensor arrays at equal intervals along the inside of the coating head boundary, real-time multi-point pressure data can be collected, accurately reflecting the subtle changes in the boundary flow state. Using digital filtering technology, high-frequency noise interference is effectively suppressed, ensuring the stability and reliability of the pressure data, providing a solid foundation for subsequent pressure gradient calculation and anomaly detection. By calculating the pressure difference between adjacent sensors to form a pressure gradient sequence, and combining statistical characteristic parameters for anomaly pattern recognition, the coating head boundary can be monitored in real time, ensuring the sensitivity and accuracy of the boundary state monitoring. Second, this method uses a pressure gradient benchmark mode based on historical data to determine the normal fluctuation range by calculating the mean value distribution and coefficient of variation, and then realizes dynamic comparison of real-time pressure gradient and automatic judgment of boundary disturbance events. This dynamic threshold setting based on statistics effectively avoids the misjudgment problem caused by fixed threshold, improving the robustness and adaptability of anomaly identification. Through the quantitative calculation of the boundary anomaly intensity index, the disturbance events are classified and managed, realizing the differentiated response to disturbances of different intensities, and strengthening the system's ability to identify the diversity of boundary anomalies. Further, the structured arrangement of boundary disturbance event information provides efficient and standardized data support for subsequent analysis and control decisions. For monitoring the coating thickness, the method selects a thickness detection sensor arranged 500mm to 1500mm downstream of the coating head to realize the synchronous collection of transverse thickness. Through fine temperature compensation and zero correction of the original thickness signal, the high precision and consistency of thickness measurement are guaranteed. Combined with process parameters, the target thickness corresponding to each measurement point is accurately calculated using cubic spline interpolation, meeting the target setting requirements under complex transverse gradient conditions. The thickness deviation is calculated by difference and smoothed by five-point moving average method, effectively filtering out measurement fluctuations and improving the stability of deviation data. The deviation distribution characteristics are evaluated by statistical indicators such as mean and standard deviation, and the abnormal position is judged by combining the deviation gradient, realizing accurate positioning of thickness anomalies. At the same time, the dynamic confidence evaluation mechanism based on signal-to-noise ratio ensures the reliability of the data and the timeliness of the reacquisition, significantly reducing the risk of misjudgment caused by measurement errors. The time series analysis of thickness deviation, combined with the calculation of deviation rate and periodicity, can warn of the thickness change trend, support early anomaly intervention, and ensure production continuity and product consistency. In terms of time synchronization, the method corrects the timestamp of the thickness deviation data forward by calculating the material transmission delay, and aligns the time base with the boundary state monitoring data, ensuring the accurate alignment of the two types of data on the time axis. Using the double sequence time comparison table for event matching, the potential correlation between boundary disturbance and thickness anomaly can be accurately captured, improving the timeliness and accuracy of correlated event identification.Candidate correlation events are screened based on statistical correlations to identify valid correlation events, enhancing the system's understanding of disturbance-impact mechanisms and its feedback capabilities. Thickness variation patterns are characterized in multiple dimensions through a comprehensive analysis of amplitude, spatial distribution, and temporal evolution. These patterns are subdivided into large-scale, medium-scale, and localized variation patterns, as well as transient, medium-sustained, and persistent variation patterns. This enriches the characterization of disturbance consequences and lays a solid foundation for the development of differentiated control strategies. This classification of thickness variation patterns and statistical analysis of disturbance characteristics enables the system to quantitatively map disturbance intensity, type, and thickness impact based on historical and real-time data. By fitting the established prediction model and incorporating correction coefficients for operating parameters such as coating speed, coating viscosity, and ambient temperature, the system achieves condition-adaptive thickness impact prediction. This dynamic adaptability enhances the system's generalization performance, enabling it to maintain high control accuracy across diverse production conditions. Based on this, the constructed boundary impact prediction data not only covers disturbance type and intensity but also accurately predicts the expected thickness impact pattern, degree, and spatial extent, providing a scientific input basis for fuzzy control. The fuzzy control module establishes a highly adaptable fuzzy variable definition system by classifying disturbance intensity, thickness impact, and control variables into fuzzy levels. The membership function design fully incorporates statistical distribution characteristics, using triangular functions to achieve continuous transitions between fuzzy levels and ensure smooth mapping of input variables. The accuracy and stability of fuzzy reasoning are ensured by adjusting the overlap of adjacent membership functions and verifying function integrity. The fuzzy control rule base is established based on the combination of disturbance intensity and impact. The number of rules is simplified and merged, reducing rule complexity and improving system response speed. Real-time fuzzy reasoning outputs predictive control variables through weighted activation rules. This, combined with the feedback control variable calculated using the PID algorithm, achieves an integrated integration of prediction and feedback, effectively improving control accuracy and robustness. Actuator control converts the integrated control variable into a drive signal, which, after signal conditioning and power amplification, drives the actuator, ensuring accurate execution and timely response. Real-time monitoring of actuator response data enables the system to quantify parameter improvements before and after control, distinguishing effective from ineffective control, and forming a closed feedback loop. The establishment of a disturbance event processing results archive enables historical tracking of control strategies and statistical analysis of their effectiveness, providing a detailed data foundation for rule performance analysis. Through multi-dimensional analysis of rule success rates, degree of improvement, response time, and stability indicators, the system can identify underperforming rules and generate targeted optimization recommendations based on operating conditions and environmental changes. The adaptive rule adjustment mechanism continuously optimizes and upgrades the fuzzy control rule base during operation, gradually forming an intelligent control strategy that adapts to changing operating conditions.Overall, this approach overcomes the limitations of traditional coating machine control, which relies on fixed thresholds and a single feedback mechanism, and implements data-driven multi-sensor fusion and intelligent reasoning, effectively improving the detection sensitivity of boundary disturbances and the accuracy of thickness control. By dynamically adjusting fuzzy control rules in real time, the system possesses excellent adaptability and learning capabilities, capable of coping with complex operating conditions and ensuring high-quality and stable output of coated products. Furthermore, the integrated application of innovative technologies such as time synchronization correction, pattern recognition, operating condition adaptability correction, and multi-dimensional performance evaluation provides strong support for the intelligent upgrade of industrial coating processes, promoting increased production efficiency and a significant reduction in scrap rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0016] Figure 1 This is a schematic flow chart of the steps of a lateral closed-loop control method for a coating machine based on fuzzy PID of the present invention;

[0017] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0018] Figure 3 for Figure 1 Detailed step flow chart of step S3 in FIG. DETAILED DESCRIPTION

[0019] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0020] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0021] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0022] To achieve this, please refer to Figures 1 to 3 The present invention provides a lateral closed-loop control method for a coating machine based on fuzzy PID, the method comprising the following steps:

[0023] Step S1: Multiple pressure sensors are arranged inside the coating head boundary to collect the pressure value of each measurement point in real time, and the pressure difference between adjacent sensors is calculated to form a pressure gradient sequence; the abnormality of the boundary flow state is identified based on the distribution pattern of the pressure gradient sequence, and when the pressure gradient exceeds the normal fluctuation range, it is marked as a boundary disturbance event, thereby obtaining boundary state monitoring data;

[0024] In this embodiment of the present invention, 16 MEMS-based differential pressure sensors are evenly embedded within the coating head boundary of the coating machine. These sensors are numbered P1 to P16 from left to right in the transverse direction. These sensors are installed at equal intervals of 10 mm, forming a pressure measurement array along the boundary. The system collects instantaneous pressure values ​​at each measurement point in real time at a sampling frequency of 50 Hz and, combined with the sampling time, forms a time-series pressure matrix. To suppress high-frequency interference and system noise, the collected pressure signals are first processed through a second-order low-pass digital filter with a cutoff frequency set to 5 Hz, outputting a stable pressure value. Based on this, the system calculates the difference between the pressure values ​​of each pair of adjacent sensors, generating 15 pressure gradient values ​​(e.g., P2 minus P1, P3 minus P2, and so on), forming a pressure gradient sequence. This gradient sequence spatially reflects airflow disturbances or coating flow non-uniformity at the boundary. The control system's built-in boundary reference library stores a large number of historical pressure gradient sequences and their statistical fluctuation ranges under normal coating head operation, with the mean ±2 times the standard deviation as the normal range. When the current pressure gradient exceeds this normal range at any location, the system identifies a local disturbance and marks it as a "boundary disturbance event" if it persists for more than two seconds without recovery. Each event is recorded with its start time, sensor location, and anomaly intensity, generating structured boundary state monitoring data for subsequent control steps.

[0025] Step S2: measure the actual coating thickness at each transverse position and compare it with the preset target thickness to calculate the thickness deviation value; time correlate the thickness deviation value according to the boundary state monitoring data to identify the thickness change pattern caused by the boundary disturbance event; establish a corresponding relationship between the boundary disturbance and the thickness influence based on the thickness change pattern to generate boundary influence prediction data;

[0026] In the discharging direction of the coating machine, a set of laser triangulation type online thickness detection sensors are arranged about 1000mm away from the outlet of the coating head, with a transverse interval of 20mm, a total of 25 measuring points are installed, which are used to synchronously detect the transverse distribution of the film thickness, and the system sampling frequency is 20Hz. The sensor first compensates the original thickness signal for temperature drift of ±0.2℃ and zero point drift correction of ±0.1μm, and binds the transverse position coordinates with each measuring point to form a thickness data matrix. The target transverse thickness distribution value is provided by the process setting, and the theoretical thickness value of the measuring point position is completed by the cubic spline interpolation algorithm in the gradient change case, and then the difference value is obtained by subtracting the measured data. Further, the system uses five-point moving average smoothing processing of the deviation value to exclude the interference of accidental abnormal fluctuations. When the absolute value of the thickness deviation of a certain measuring point exceeds 3 times the standard deviation of the transverse deviation distribution, or the thickness deviation gradient between adjacent measuring points exceeds 2μm / cm, it is regarded as a local thickness abnormal area. In order to enhance the response ability of the control system to abnormal trends, the system organizes the thickness deviation data into a 10-second time window sequence, and combines the time stamp of the boundary disturbance event identified in S1 and the transmission delay (about 2.4 seconds calculated at the coating line speed of 25m / min and the detection point spacing) for forward time correction. The system compares the time difference between the time-corrected boundary disturbance event and the thickness abnormality, and when it matches successfully within ±2 seconds, it is identified as a thickness change caused by the boundary event. The correlation statistics between the intensity level and the deviation amplitude of all matching events are further carried out, and the system automatically selects high correlation events with a correlation coefficient greater than 0.6, and classifies the spatial variation range and variation duration, and outputs the thickness change pattern data, which is prepared for fuzzy control.

[0027] Step S3: construct a fuzzy control rule library according to the boundary influence prediction data, wherein the fuzzy control rule library takes the intensity of the boundary disturbance and the degree of the thickness influence as fuzzy input variables, and takes the control adjustment as fuzzy output variable; when a new boundary disturbance event is detected, a predictive control amount is generated by using the fuzzy control rule library; a feedback control amount is obtained by PID calculation according to the thickness deviation value;

[0028] After completing thickness variation pattern recognition, the control system in this embodiment of the present invention enters the fuzzy control rule construction phase. First, the intensity of boundary disturbance events is categorized into five levels: extremely weak, slight, moderate, significant, and severe. The thickness variation amplitude is categorized into four levels: small fluctuation, moderate anomaly, severe deviation, and continuous drift. The control adjustment variable is then set to five output levels based on the response sensitivity of the coating machine's drive control system: fine adjustment, light adjustment, medium adjustment, emphasized adjustment, and limited adjustment. Based on previous training data and current prediction data, the control system generates an initial fuzzy control rule base by statistically analyzing the mapping between disturbance level and thickness variation. For example, "When the disturbance is moderate and the thickness variation is severe, the control variable output is medium adjustment." A total of 25 initial rules are established. During the inference execution phase, when a new boundary disturbance event arrives, the system first uses the real-time identified disturbance level and the predicted thickness impact amplitude as fuzzy input variables. It then searches for fuzzy rules that meet the requirements and calculates their activation levels. The fuzzy output variable is then calculated through a weighted average. The final predictive control variable is then determined through a defuzzification process. At the same time, the system calculates the feedback control variable of the thickness deviation value through the conventional PID control module. The proportional component is based on the current deviation, the integral component considers the accumulated deviation over the past 10 seconds, and the differential component calculates the trend component based on the rate of change over the past 3 seconds. The three are superimposed to output the feedback control value. Ultimately, the system integrates the predictive control variable with the feedback control variable as the input for coating parameter adjustment.

[0029] Step S4: Adjust the coating parameters in real time according to the feedback control amount and the predictive control amount, continuously monitor the adjustment effect and record the processing results of the boundary disturbance events, and automatically adjust the weight coefficient of the corresponding rule in the fuzzy control rule library when the processing effect is not good, so as to accumulate and optimize to form an adaptive control strategy.

[0030] The embodiment of the present application converts the fused control quantity into a control signal of a coating machine driving mechanism in real time, drives a transverse adjustment execution mechanism through an analog signal conditioning module and a servo power amplifier, realizes operations such as transverse displacement of a coating head, fine adjustment of an electric pressure roller pressure, and supply pump flow adjustment. After adjustment, the system monitors thickness deviation values and pressure gradient sequences of the next cycle in real time, and compares the data before adjustment, judges whether the deviation is reduced and the pressure distribution is returned to the normal interval, if both are improved, it is determined as effective control, if there is no improvement or abnormal aggravation, it is recorded as invalid control. The control strategy module structures and archives the processing strategy and adjustment response result of each round of boundary disturbance event, and counts the success rate and average improvement degree of different fuzzy rules under different disturbance scenarios. For the rules with a success rate lower than 70% and an improvement effect less than 10% of the historical average, the system automatically marks them as to-be-optimized rules. Then, the fuzzy rule library optimization process is entered: the system analyzes the applicability of each rule under different external conditions such as ambient temperature, coating speed and coating viscosity according to historical control records, and automatically fine tunes the weight value or output level of the rule according to the best control strategy under similar disturbance, so as to continuously optimize and form adaptive fuzzy control strategy. In this embodiment, the typical production line speed of the coating machine is 20-30 m / min, the target value of the coating thickness is about 20 μm, and the allowed deviation is controlled within ±1.5 μm. Through the self-optimizing control strategy, the system can continuously maintain the thickness stability under the boundary working condition with frequent disturbances, and significantly improve the product yield.

[0031] Preferably, step S1 comprises the following steps:

[0032] Step S11: synchronously collecting the instantaneous pressure values of each measurement point through the pressure sensor array arranged equidistantly inside the coating head boundary, and marking according to the sensor number and sampling time to obtain time sequence pressure data;

[0033] The embodiment of the present application is installed with an array of high-sensitivity miniature pressure sensors inside the lateral boundary of the coating head of the coating machine, each sensor adopts capacitive sensing technology, the range is ±100Pa, and the response time is less than 5ms. The sensor array is composed of 12 measuring points, which are uniformly distributed along the coating width direction, the spacing between every two sensors is 15mm, and the whole covers a boundary area of 180mm. The sensor numbers are sequentially set as P1 to P12, and the numbers increase from left to right. The system controller adopts a 32-bit data acquisition unit to synchronously sample all the sensors, the sampling period is set to once every 20 milliseconds, so that the instantaneous pressure values of all points are collected at the same time, and the phase error caused by asynchronous sampling is prevented. All the collected data are transmitted to the data buffer area of the boundary monitoring module in real time, the system organizes the sampling values in time sequence to form a two-dimensional data record table, taking the sensor number as the row and the sampling time as the column, records the historical pressure change of each measuring point, and forms time sequence pressure data for subsequent dynamic analysis.

[0034] Step S12: the time sequence pressure data is subjected to digital filtering processing to remove high-frequency noise interference, and stable pressure measurement data is obtained;

[0035] In order to improve the stability and usability of the pressure data, the system performs digital filtering processing on the time sequence pressure data obtained in S11. Considering the factors such as fan vibration, air pressure fluctuation and electrical interference in the industrial field, the filtering algorithm adopts a five-order low-pass FIR (finite impulse response) digital filter, and the cutoff frequency is set to 2.5Hz to filter out high-frequency components higher than the normal boundary disturbance range. In the signal processing module embedded in the control system, the time sequence pressure value corresponding to each sensor is taken as the input, and the sliding window processing is performed on the continuous 5 data points, and the smoothed pressure value at the current time is output, so that the interference caused by instantaneous fluctuation is suppressed. The filtered data not only eliminates the disturbance of instantaneous peak value, but also retains the real trend of the change of the boundary pressure with time, forms stable pressure measurement data, and participates in subsequent spatial analysis processing on the basis, and avoids false disturbance identification.

[0036] Step S13: the pressure gradient values of each position are obtained through the pressure difference calculation between adjacent sensors according to the pressure measurement data, the pressure gradient values are arranged in the order of spatial positions to form a pressure gradient distribution sequence, and the statistical characteristic parameters including the mean value, the standard deviation and the maximum gradient value of the sequence are calculated to obtain the pressure gradient sequence;

[0037] After obtaining stable pressure measurement data of each pressure sensor point, the embodiment of the present application immediately performs spatial difference calculation between adjacent sensors to form pressure gradient values. The pressure gradient is a key physical quantity reflecting the trend of pressure change in the boundary region, which is expressed as the degree of pressure difference per unit distance. In the present embodiment, the system calculates 11 pressure gradient values in turn according to P2 minus P1, P3 minus P2, and so on, which correspond to the midpoint positions between adjacent sensors, respectively. For example, the gradient between P1 and P2 represents the change of airflow driving force at the position of 15 mm on the left side of the boundary. All gradient values are arranged in order of spatial position to form a "pressure gradient distribution sequence", which is used to identify whether the boundary has abnormal disturbance. In order to further extract features from the sequence, the system performs statistical analysis on multiple gradient snapshots per second for each group of sequences to extract the mean value (reflecting the overall flow trend), the standard deviation (reflecting the degree of fluctuation) and the maximum gradient value (locating the position of strong disturbance center), which constitute a complete "pressure gradient sequence" data structure, having spatial and quantitative features for identifying boundary dynamic abnormalities.

[0038] Step S14: identifying the degree of abnormality of the boundary flow state according to the distribution pattern of the pressure gradient sequence, marking a boundary disturbance event when the pressure gradient exceeds the normal fluctuation range, and obtaining boundary state monitoring data.

[0039] The embodiment of the present application uses the pressure gradient sequence obtained in S13 to identify the abnormal state of the airflow stability of the coating head boundary. The system first calls the "boundary reference model" stored in advance, which is built by long-time data acquisition when the coating system is in a state without abnormal disturbance, and defines the normal fluctuation range of the pressure gradient at each spatial position, including the mean value ± two times the standard deviation as the upper and lower limits of normality. The system compares the pressure gradient sequence generated in real time with the reference model point by point, and if the pressure gradient value of any position exceeds the corresponding normal range, it is recorded as an "abnormal point", and the abnormal amplitude is taken as the "over-limit intensity". When two or more adjacent positions are simultaneously over-limited, or the single-point over-limit amplitude reaches the set severe level (for example, exceeding 3 times the standard deviation), the system judges that the current boundary flow state has been disturbed, and constructs a "boundary disturbance event" combining the spatial range and duration of the disturbance. Each event is assigned a unique ID, and the occurrence time, maximum disturbance intensity, transverse position center and coverage range are recorded, and finally arranged into structured "boundary state monitoring data" for driving the input end of the fuzzy control rule engine. This process is completely automatic and does not require manual judgment, achieving high-precision, low-delay and fully-automatic identification of boundary disturbance, and providing a reliable basis for subsequent predictive control strategies.

[0040] Preferably, step S14 comprises the following steps:

[0041] Step S141: Determine the upper and lower limits of the normal fluctuation range of the pressure gradient at each position by collecting the pressure gradient sequence when the coating machine is running normally, and calculating the mean distribution and coefficient of variation distribution of the pressure gradient sequence, thereby obtaining the pressure gradient reference mode;

[0042] In the initial debugging of the coating machine, the system continuously collects pressure gradient data for at least 10 minutes in a stable running condition, without applying any intentional interference, to ensure that the data can truly reflect the normal boundary flow state. The data is collected at a frequency of 50 frames per second, and each frame contains 11 gradient values, corresponding to the middle points between the 11 sensors of the coating head boundary. The system performs statistical modeling on this batch of data: first, calculate the gradient mean and standard deviation at each spatial position for all sampling times, and based on this, calculate the "coefficient of variation", which is the ratio of the standard deviation to the mean, to describe the stability of the pressure gradient at that point. The system sets the mean plus or minus twice the standard deviation as the "upper and lower limits of the normal fluctuation range" at that position, that is, when the real-time measured gradient value is within this range, it is considered normal. The upper and lower limits of all positions form a complete "pressure gradient reference mode", which is stored in the boundary monitoring module in matrix form for subsequent real-time comparison and calling.

[0043] Step S142: Compare the real-time measured pressure gradient sequence with the pressure gradient reference mode point by point, and when the pressure gradient value at any position exceeds the corresponding normal fluctuation range, record the exceeding position and the exceeding degree to obtain the pressure exceeding range data;

[0044] In the actual operation of the coating machine, the embodiment of the present application obtains the pressure gradient sequence in real time, and compares the gradient values of the current frame one by one with the pressure gradient reference mode established in step S141 in a spatial position corresponding to each point, to determine whether there is an over-limit phenomenon at each position. The comparison method is as follows: if the real-time pressure gradient at a certain position is higher than the corresponding upper limit or lower than the lower limit, mark the point as an "out-of-range point", and record the number of the point, the actual value, and the deviation amount of the upper and lower limits. The system arranges the position information and deviation degree of all out-of-range points according to the time stamp to form a record, forming "pressure exceeding range data" for measuring the spatial extent and intensity distribution of the current boundary disturbance. In order to reduce the influence of incidental errors, the system sets up a "weak disturbance filtering mechanism", and only when the over-limit amplitude exceeds the threshold value of 0.1 Pa or the number of over-limit points is greater than 3, the record is retained for subsequent analysis, otherwise it is determined as invalid fluctuation.

[0045] Step S143: Calculate the boundary abnormal intensity index according to the number of exceeding positions and the size of the exceeding degree in the pressure exceeding range data;

[0046] The embodiment of the present invention inputs the pressure out-of-range data generated by S142 into the disturbance intensity assessment model to quantitatively calculate the overall impact of the anomaly. First, the system calculates the number of out-of-limit points at the current moment, and performs a weighted summation according to the degree of deviation of each point to obtain a "boundary anomaly integral value"; this integral value reflects the abnormal intensity of the current boundary state. Subsequently, the integral value is divided by the total number of sensors and normalized to obtain a "boundary anomaly intensity index", which usually ranges from 0 to 1. The meaning of this index is: 0 means completely normal, and 1 means that all points are significantly out of limit. Taking the actual coating scenario as an example, when the boundary anomaly intensity index exceeds 0.4, the coating edge thickness often has obvious fluctuations, so the control system needs to respond immediately. This index will serve as the core indicator for the determination of subsequent disturbance events.

[0047] Step S144: Disturbance events are identified based on the boundary anomaly intensity index. An anomaly intensity threshold is set to classify the boundary status into normal, slightly abnormal, moderately abnormal, and severely abnormal. When the anomaly intensity index exceeds the set threshold continuously for a preset period of time, a boundary disturbance event is confirmed to have occurred and the start time, duration, and intensity level of the event are recorded to obtain boundary disturbance event characterization data.

[0048] The embodiment of the present invention performs sliding window monitoring and analysis on the boundary anomaly intensity index within a continuous time period, and the window length is set to 3 seconds (corresponding to 150 frames) to identify the persistence of the disturbance. When the average intensity index in the window exceeds a preset threshold (for example, 0.3) and the duration reaches at least 2 seconds, it is determined that a "boundary disturbance event" has occurred. At the same time, the system classifies the event level according to the maximum intensity index: an index of 0.30.45 is a slight abnormality, 0.450.6 is a moderate abnormality, and above 0.6 is a severe abnormality. The system records the starting frame number, the number of continuous frames, the peak intensity, the corresponding spatial center position (that is, the weighted center position of all exceeding limit points) and the event level of the event, and encapsulates it into a "boundary disturbance event characterization data". On the one hand, this characterization data will be sent to the fuzzy control rule engine for control strategy triggering, and on the other hand, it will be used as historical sample data for later model iterative optimization.

[0049] Step S145: Integrate the boundary disturbance event information identified in the boundary disturbance event characterization data, including event type, intensity level, spatial location and time characteristics, into a standardized data format to obtain structured boundary state monitoring data.

[0050] The multi-dimensional attributes of the identified boundary disturbance event are integrated into structured data by the embodiment of the application, which is written into an event data table in a standardized format, and each data record contains: event type (mapped by level), intensity level (enumeration value), spatial position (sensor array transverse coordinate), start and end time (in frame number units), and duration (calculated in milliseconds). All fields have uniform data formats and field naming rules, facilitating fast calling and indexing by subsequent modules. The system also assigns a unique event number to each event to support tracking. The structured "boundary state monitoring data" will be one of the core inputs of the fuzzy control module, used to decide the predicted control amount and adjust the rule weight. At the same time, this data is also used as an abnormal knowledge sample for offline data analysis and disturbance model regression training, providing stable basic data support for the control system self-learning mechanism.

[0051] Preferably, the actual coating thickness at each transverse position is measured in step S2 and compared with the preset target thickness to calculate the thickness deviation value, which includes:

[0052] The thickness detection sensor arranged at a position 500-1500 mm downstream of the coating head synchronously collects the original thickness signal of the transverse distribution;

[0053] The original thickness signal is subjected to ±0.2℃ temperature compensation and ±0.1μm zero-point correction processing to obtain corrected thickness data;

[0054] The target thickness value corresponding to the transverse position is obtained according to the process setting, and when there is a transverse gradient in the target thickness value, the accurate target thickness value corresponding to the measurement point is calculated by cubic spline interpolation;

[0055] The corrected thickness data is subtracted from the accurate target thickness value, and a five-point moving average method is used for smoothing processing to obtain stable thickness deviation data;

[0056] The mean and standard deviation of the thickness deviation distribution are calculated according to the stable thickness deviation data, and when the absolute value of the single-point deviation exceeds three times the standard deviation or the deviation gradient of adjacent points exceeds 2μm / cm, the abnormal position is marked to obtain thickness deviation distribution characteristic data;

[0057] A data confidence evaluation mechanism is established based on the thickness deviation distribution characteristic data, and when the signal-to-noise ratio of the measurement data is less than 20dB, the original thickness signal is re-collected at the corresponding position to obtain the thickness deviation data;

[0058] The thickness deviation data is subjected to time series analysis with a time window of 10 seconds, and the deviation change rate and periodic characteristics of each position are calculated, and when the deviation change rate exceeds 0.5μm / s, a trend warning is marked to obtain the thickness deviation value containing position coordinates, deviation value, change trend, confidence, and warning mark.

[0059] The embodiment of the application arranges a set of high-precision coating thickness detection sensors in the transverse direction at the outlet position of the coating machine. These sensors use the laser triangulation principle and can measure the surface thickness variation without contacting the coating. According to experience and experimental verification, the coating is preliminarily stable between 500mm and 1500mm after completion, so the sensor array is installed about 1000mm downstream of the coating head, covering the entire area of the coating width in the transverse direction, usually 11 to 21 equally spaced measurement points are configured, corresponding to one sample every 50mm-100mm. Each sensor synchronously collects the original thickness signal of the transverse position at a frequency of 100Hz, and binds it with the timestamp and spatial coordinates to form a two-dimensional thickness data frame, providing basic data for subsequent correction and analysis. Since the coating thickness measurement is easily affected by environmental temperature drift and equipment initial deviation, the system performs two-level correction on the original thickness data. First, introduce a temperature sensor to monitor the environmental temperature change in the measurement area in real time, and establish a linear compensation model between temperature and thickness error according to the device calibration data: every 1℃ offset will cause about ±0.1μm measurement error, the system automatically adjusts the data of each sampling point with ±0.2℃ as the compensation threshold. Secondly, zero-point correction is performed, using the substrate thickness value under no coating condition as the reference standard, the system collects the initial zero-point data before each operation and establishes the corresponding zero-bias coefficient for each sensor, the system real-time subtracts the zero-bias from all original thickness values to obtain the corrected thickness data. These data have consistent reference and comparability, which guarantees the data consistency for subsequent comparison with target values. According to the current process parameter setting, the target thickness curve is extracted, which may be a constant value (such as 12μm) or have a transverse gradient characteristic (for example, the central part is thicker and the edge is slightly thinner). When there is a gradient change, the system uses a cubic spline interpolation algorithm to refine the key point target thickness data set by the process, so that each measurement point has an accurate target thickness value. The advantage of spline interpolation is that it can smoothly transition between different set values, avoiding sudden changes that disturb the control system. Finally, a target thickness array corresponding to each actual thickness point is formed, providing a high-precision reference baseline for difference operation. The corrected thickness data of each position is compared with the corresponding target thickness value point by point, and the current original thickness deviation data is obtained. Considering that slight fluctuations in the sampling process may produce false peak deviations, a five-point moving average filter is applied to smooth the difference results, specifically, the average deviation value of each position and its adjacent two positions is taken to suppress local noise and enhance trend consistency. After smoothing, a set of stable thickness deviation data is obtained, each value representing the true thickness deviation state of the transverse position, laying the foundation for judging abnormalities and predicting trends. Statistical analysis is performed on the smoothed thickness deviation data, first calculating the mean and standard deviation of the overall deviation sequence to measure the overall state of coating consistency.On this basis, the system determines point by point whether the thickness deviation value exceeds the statistical upper and lower limits of "mean plus or minus three times the standard deviation". If it exceeds, it is marked as an outlier. In addition, the system calculates the deviation gradient value between adjacent positions, that is, the deviation difference between two positions divided by their distance. When this gradient value exceeds 2μm / cm, it is also marked as an outlier. The position, deviation value and deviation direction of all outliers are summarized to form "thickness deviation distribution characteristic data", which is used to identify the specific range and intensity of lateral unevenness. To avoid misjudgment due to signal noise or light interference, the system performs a confidence assessment on the data of each measurement point, mainly based on the signal-to-noise ratio, that is, the ratio of the effective signal amplitude to the high-frequency fluctuation amplitude. When the signal-to-noise ratio of a measurement point falls below the set lower limit (such as 20 decibels), the system considers the current thickness data unreliable and triggers the data re-sampling mechanism for that point. The re-sampling method is to re-sampling 5 frames of thickness raw data continuously in a very short time (about 200ms) and re-perform the correction and difference processing, retaining only the frame with the highest data consistency as the final thickness value. This can eliminate instantaneous anomalies and improve the validity and robustness of thickness deviation data. A time series analysis is performed on the thickness deviation data in the past 10 seconds, 100 frames of data are collected per second, and the analysis window is 1000 frames in total. The system uses each horizontal measuring point as the analysis unit to extract the deviation change rate (i.e., the deviation increment per unit time) and periodic characteristics (the main frequency component is estimated by fast Fourier transform). If the change rate of a certain measuring point continues to exceed 0.5μm / s, the system will mark it as a "trend warning", indicating that there may be a boundary disturbance or a precursor to process fluctuations. Finally, the system integrates the spatial coordinates, deviation value, deviation change trend, data confidence and warning status of each measuring point into a structured output to form a "thickness deviation value" set at the current moment, providing key input variables for the next stage of fuzzy-PID control quantity calculation.

[0060] Preferably, in step S2, performing time correlation on the thickness deviation values ​​according to the boundary state monitoring data to identify the thickness variation pattern caused by the boundary disturbance event includes:

[0061] The material transmission delay time is calculated based on the coating line speed and the distance from the boundary monitoring position to the thickness detection position to obtain the time synchronization correction parameter;

[0062] The timestamp of the thickness deviation value is corrected forward based on the time synchronization correction parameter and kept consistent with the time reference of the boundary status monitoring data to obtain the time synchronized thickness deviation data;

[0063] The boundary disturbance events in the boundary state monitoring data are arranged in chronological order to form an event time series, and the time-synchronized thickness deviation data are arranged in chronological order to form a deviation time series, and a double-sequence time comparison table is established;

[0064] According to the double sequence time control table, event matching search is performed, and when the time difference between the occurrence time of the boundary disturbance event and the occurrence time of the thickness anomaly is within ±2 seconds, it is marked as a potential associated event pair to obtain candidate associated event data;

[0065] Based on the candidate associated event data, the correlation coefficient between the intensity level of the boundary disturbance event and the corresponding thickness anomaly degree is calculated, and when the correlation coefficient is greater than 0.6, it is confirmed as an effective associated event pair to obtain verified associated event data.

[0066] According to the verified associated event data, the thickness change mode characteristics are extracted to obtain the thickness change mode.

[0067] The embodiment of the application considers that there is a spatial distance between the coating head and the thickness detection sensor, and the material needs to undergo a certain transmission delay from the boundary disturbance to the thickness change. If not corrected, it will cause the time dislocation of events and results, affecting the subsequent causal analysis. The system calculates the transmission time delay of the material on this path according to the current set coating line speed (for example, 20 meters / minute) and the actual distance between the boundary monitoring point and the thickness detection point (for example, 1000 millimeters), that is, 3 seconds, which is the "time synchronization correction parameter". This parameter does not depend on manual estimation, but is automatically calculated by the system through the line speed sensor data collected every minute, and is updated regularly to ensure the dynamic accuracy of time alignment. Based on the above transmission delay time, the timestamp of the thickness deviation value is adjusted forward by the delay value, that is, the timestamp of each deviation record is reduced by 3 seconds, so that it corresponds to the actual upstream state occurrence time that causes the deviation. Through this "forward correction" operation, the occurrence time of the coating head boundary disturbance event and the detection time of the thickness anomaly can be unified and standardized, forming the "time-synchronized thickness deviation data" after alignment, ensuring that the subsequent matching search is carried out under a unified time reference, and fundamentally solving the misidentification problem caused by delay of cross-position data. The "boundary state monitoring data" and "time-synchronized thickness deviation data" are recorded continuously at a fixed refresh frequency (for example, 100 frames are collected per second). The starting time of all boundary disturbance events is extracted, arranged in time sequence to form an "event time sequence"; at the same time, the time nodes marked as abnormal in all thickness deviation values are extracted to form a "deviation time sequence". The system binds the two sequences in a two-dimensional table through the creation of a "time comparison table", each row of which represents a boundary event and a thickness state corresponding to a time point, forming a structured time indexing relationship, laying a data foundation for subsequent automatic event matching and verification. The time comparison table is executed by a bidirectional traversal algorithm, that is, whether there is a thickness abnormal point within ±2 seconds from the starting time of each boundary disturbance event is searched forward and backward. If there is, the event pair is recorded as "potential associated event pair", and its related fields (boundary event number, disturbance intensity level, abnormal thickness value, time difference between them) are recorded together to generate "candidate associated event data". The setting of the 2-second threshold value here is based on the stable fluctuation range of the coating line speed and the experience analysis of the viscosity characteristics of the coating material, and allows the system to adaptively adjust in actual operation, ensuring the sensitivity and accuracy. The candidate associated event data contains multiple disturbance and thickness anomaly pair records, and the system extracts the disturbance intensity level (such as medium abnormality, severe abnormality corresponding to 1-3 levels) and the thickness abnormality degree (i.e. absolute value of deviation) of the corresponding position to form a numerical variable pair. The system calculates the linear correlation between the variable pair by using the Pearson correlation coefficient. If a pair of events shows strong correlation in multiple samplings, and the overall correlation coefficient exceeds 0.6 (i.e. the relationship between them is more significant), it is marked as "effective associated event pair".The system retains the correlation coefficient values and event information to form the final "validation correlation event data", which truly reflects the causal relationship between the disturbance and the thickness anomaly. For all validation correlation event data, the corresponding time-synchronous thickness deviation sequence is analyzed by window slicing. Taking each disturbance event occurrence time as the center, the deviation value sequence within 5 seconds before and after is taken, and its change trend (such as rising, falling or fluctuating), deviation peak position, duration and recovery speed and other key characteristic parameters are extracted. These characteristics constitute the "thickness change pattern" which can be used to describe the specific way in which different types of boundary disturbances affect the thickness. The system outputs the thickness change pattern in a standardized template manner, such as "strong boundary disturbance + right deviation + slow recovery" type, for use in fuzzy control rule modeling, providing a basis for subsequent predictive adjustment.

[0068] Especially important is that the thickness change pattern feature extraction according to the validation correlation event data is specifically:

[0069] By analyzing the performance rules of thickness change caused by different types of boundary disturbances in amplitude characteristics, spatial distribution characteristics and time evolution characteristics in the validation correlation event data, when the influence range exceeds the inside of the boundary by 30 mm, it is marked as a large range change pattern, when the influence range is 10-30 mm, it is marked as a medium range change pattern, and when the influence range is less than 10 mm, it is marked as a local change pattern. According to the change duration, the pattern is divided into instantaneous change pattern, medium duration change pattern and persistent change pattern, to obtain the thickness change pattern.

[0070] The embodiment of the present application performs structured analysis on the thickness variation pattern in the verified correlation event data of each group extracted in the previous stage, specifically from three dimensions: amplitude characteristics, spatial distribution characteristics and time evolution characteristics. For amplitude characteristics, the system extracts the maximum deviation value from each time-synchronous thickness deviation data and takes it as the peak value index of disturbance influence; for example, after a serious boundary disturbance, the coating layer has a maximum thickness deviation of 2.8 microns in the right 25mm area, which will be used as the amplitude characteristic quantity of the event. For spatial distribution characteristics, the system counts the total number of transverse measurement points whose thickness deviation exceeds the judgment threshold (such as 1 micron) in each disturbance influence, and converts the physical space width according to the sensor spacing (such as one measurement point every 5mm); if the width exceeds 30mm, it is marked as “large range variation pattern”, otherwise if it is between 10mm and 30mm, it is marked as “medium range variation pattern”, and less than 10mm is “local variation pattern”. This classification method can materialize the disturbance affected area, so that the fuzzy control rule can perceive the width of the transversely affected area. In terms of time evolution characteristic extraction, the system automatically labels the “change duration” based on the time when each thickness deviation occurs and the time length required for its recovery to the stable range. For example, if the time required from the occurrence of the disturbance event to the falling of the deviation value to the set tolerance range (±0.5 microns) is less than 2 seconds, it is classified as “instantaneous change pattern”; if it is between 2 seconds and 5 seconds, it is classified as “medium duration change pattern”; if it exceeds 5 seconds and still has not recovered to the stable state, it is identified as “persistent change pattern”. This classification is generated continuously by the control system using time stamp and deviation sequence change rate without human intervention. In this way, the system not only identifies whether the disturbance exists, but also quantifies the time dimension of the “disturbance consequences”, which helps the subsequent fuzzy control system to judge whether to apply immediate large adjustment or delayed fine tuning. Each thickness variation pattern is packaged in the form of a triple, i.e. “spatial pattern category + amplitude peak + duration level”, such as “medium range variation pattern + deviation 2.3 microns + medium duration change pattern”, and stored in the thickness variation pattern database. This database serves as the basis for training fuzzy control rules, and the subsequent system will continuously induce the fuzzy correspondence between disturbance types and control quantities based on the database, thereby improving the dynamic adaptation capability of the closed-loop control system in complex disturbance scenarios. The whole process is automatically executed on the basis of boundary disturbance event monitoring and thickness deviation analysis, ensuring high precision, real-time and stability.

[0071] Preferably, the step S3 of establishing the correspondence between the boundary disturbance and the thickness influence based on the thickness variation pattern comprises:

[0072] By calculating the Euclidean distance between the feature vectors of different thickness variation patterns, similar patterns are classified into the same category when the distance is less than a set threshold, obtaining pattern classification data;

[0073] Based on the mode classification data, the boundary disturbance feature distribution corresponding to the thickness change mode of each category is counted, the average disturbance intensity, disturbance type and occurrence frequency corresponding to each change mode are calculated, and disturbance feature statistical data is obtained;

[0074] According to the disturbance feature statistical data, a quantitative mapping relationship between the disturbance intensity and the influence degree is established, a function relationship between the boundary disturbance intensity index and the thickness change amplitude is fitted, when the fitting correlation coefficient is greater than 0.8, the mapping relationship is confirmed to be effective, and quantitative prediction model data is obtained;

[0075] Based on the quantitative prediction model data, the correction effect of the change of coating speed, coating viscosity and environmental temperature on the disturbance influence degree is analyzed, and working condition adaptability correction coefficients are generated;

[0076] According to the quantitative prediction model data and the working condition adaptability correction coefficients, the disturbance type, intensity level and working condition parameters are taken as query conditions, and the expected thickness influence mode, influence degree and spatial range are taken as prediction results, so as to construct boundary influence prediction data.

[0077] In an embodiment of the present invention, the thickness variation pattern triples formed in the previous step are vectorized, i.e., the spatial range (e.g., in millimeters), maximum deviation amplitude (in microns), and duration (in seconds) of each pattern are arranged in a fixed order to form a three-dimensional feature vector. The system then uses the Euclidean distance algorithm to calculate the distance between all feature vectors. The distance calculation is based on standard normalized data to avoid dimensional inconsistency affecting the clustering effect. When the Euclidean distance between two vectors is lower than a set threshold (e.g., 1.5 unit lengths), the patterns are considered to be highly similar in terms of spatial, amplitude, and temporal performance and can be classified into the same category. The entire classification process is automatically performed using a density-based spatial clustering algorithm (e.g., DBSCAN), without the need to manually set the number of categories. After classification, the results are annotated as pattern category labels to obtain structured pattern classification data. Based on the above pattern classification data, the boundary disturbance event information contained in each category is then counted. The boundary disturbance event intensity level, event type (e.g., pressure rise type, turbulence type, pressure drop type, etc.), and occurrence frequency corresponding to the same type of thickness variation pattern are integrated and analyzed. For example, a certain type of thickness variation pattern was associated with 85 disturbance events, with an average disturbance intensity index of 3.7. These disturbances were primarily concentrated in the "mild anomaly" and "moderate anomaly" categories, with high-frequency disturbances occurring most frequently during night shifts. The system converts these statistical results into a table of disturbance characteristic statistics, providing a foundation for quantitative modeling. Based on this, the system attempts to establish a mathematical mapping between disturbance intensity and the resulting thickness variation. Using the intensity index of each type of disturbance event as the input variable and the corresponding thickness deviation peak as the output variable, the system uses the least squares method to fit a polynomial or piecewise linear function to identify the model that best reflects the relationship between the two. When the correlation coefficient after fitting is greater than 0.8, the functional relationship is confirmed to be statistically significant, and valid quantitative prediction model data is established. For example, if the thickness deviation corresponding to a moderate abnormal disturbance remains within a linear range (e.g., each 1-point increase in disturbance intensity causes an average deviation of 0.9 microns), the system incorporates this function into the prediction model database. Considering that the coating process is significantly affected by varying operating conditions, the system further performs a multi-factor correction analysis on the quantitative prediction model. By tracing back historical data, the mapping deviation rate between disturbance intensity and thickness deviation is recalculated under different coating line speeds (such as from 0.8m / s to 1.5m / s), different coating viscosities (such as the range of 200 to 800mPa·s) and different ambient temperatures (such as 20 to 30°C), and the sensitivity of the working condition to the disturbance response is extracted. Finally, the system constructs a three-dimensional correction factor table, taking speed, viscosity and temperature as the main variables, and performs weighted correction on the output of the original mapping function to generate a working condition adaptability correction coefficient for dynamic adaptation. A complete set of boundary influence prediction models is established, taking the disturbance type, intensity level and current working condition parameters as input variables, and querying the matching influence pattern and expected deviation range.For example, under the conditions of "moderate anomaly + sudden pressure drop + viscosity of 300 mPa·s + temperature of 25°C + speed of 1.0 m / s," the system predicts a thickness anomaly of "moderate variation + deviation of 2.1 microns + duration of 4 seconds," providing direct basis for the fuzzy control system to subsequently set adjustment strategies. This predictive mechanism not only enhances proactive response but also offers good scalability, making it applicable to various coating systems.

[0078] Preferably, step S3 includes the following steps:

[0079] Step S31: performing fuzzy variable definition processing according to the boundary influence prediction data, performing fuzzy level classification on the boundary disturbance intensity, thickness influence degree, and control adjustment amount, and obtaining fuzzy variable definition data;

[0080] The embodiment of the present invention performs fuzzy processing on the input and output variables, wherein "boundary disturbance intensity" represents the degree of pressure anomaly, corresponding to the input variable, and the numerical range is defined as 0 to 10 based on the common range of the disturbance intensity index; "thickness influence degree" represents the coating deviation amplitude, corresponding to the input variable, and the range is set to 0 to 5 microns; "control adjustment amount" is the output variable, indicating the lateral compensation instruction value required for the system to adjust, ranging from -3 to 3 units of adjustment voltage value. During the fuzzification process, the system divides the disturbance intensity into five levels: very weak, weak, medium, strong, and very strong; the thickness influence degree is divided into five levels: slight, medium-low, medium, medium-high, and severe; and the control adjustment amount is divided into seven levels: large negative, medium negative, small negative, no adjustment, small positive, medium positive, and large positive, thus obtaining complete fuzzy variable definition data.

[0081] Step S32: performing membership function design processing based on the fuzzy variable definition data, and using the triangular membership function to define the numerical range of each fuzzy level according to the statistical distribution characteristics in the boundary impact prediction data, to obtain membership function parameter data;

[0082] In order to realize the numerical operation of fuzzy reasoning, the embodiment of the present invention designs a membership function based on the divided fuzzy levels. The system uses a triangular membership function to express the numerical coverage interval of each fuzzy level on its domain of definition. For example, the triangular function of the "medium" level of disturbance intensity is centered on the intensity value of 5, and extends to 3.5 and 6.5 on both sides, forming a set of ladder-shaped transition zones to ensure a smooth transition between adjacent levels. The system determines the overlapping range of membership of each level by analyzing the data distribution density of historical disturbance intensity and thickness influence, so that the function is more in line with the disturbance behavior characteristics in actual operation. The membership function parameters of all variables eventually form a standardized data table for subsequent reasoning.

[0083] Step S33: constructing a fuzzy control rule base according to the membership function parameter data and the boundary influence prediction data, establishing reasoning rules by analyzing the combination of the disturbance intensity and the influence degree, simplifying the rule base by merging similar rules when the total number of rules exceeds 15, and obtaining the fuzzy control rule base;

[0084] In the embodiment of the application, the aforementioned membership function is used in combination with the prediction data to construct a fuzzy control rule base. By analyzing the combination of different disturbance intensities and thickness influence levels, for example, "if the disturbance intensity is strong and the thickness influence is medium-high, then the control adjustment amount is medium-positive", each possible combination is mapped to a control rule. Since the combination of two inputs of five levels and one output of seven levels can generate 25 initial rules, when the total number of rules exceeds 15, in order to improve the reasoning efficiency of the system, the system merges rules with slightly different input levels but the same or less than one level of output control amount by using a similarity merging algorithm, and adjusts the subsequent reasoning response by using an average weight method, so as to finally form a simplified fuzzy control rule base of about 15 rules, which takes into account the response accuracy and improves the real-time calculation capability.

[0085] Step S34: when a new boundary disturbance event is detected, real-time fuzzy reasoning processing is performed according to the fuzzy control rule base, the intensity level of the current disturbance and the predicted influence degree are taken as input variables, and the activated reasoning rules are found by rule matching, the weights of each activated rule are calculated and weighted average is performed to obtain the predictive control amount.

[0086] In the embodiment of the application, when a new boundary disturbance event is detected during operation, the current disturbance intensity level and the predicted thickness deviation level are taken as inputs and transmitted into the fuzzy control rule base for real-time reasoning. The fuzzy reasoning process uses a rule activation mechanism based on the maximum membership method to find all rules with high membership for the current input values, and assigns each rule a corresponding activation weight. Then, a weighted average method is used to synthesize the output control amounts of all activated rules to calculate a final predictive control amount, which represents a preventive adjustment instruction before the disturbance affects the actual thickness, and can realize feedforward control of the system.

[0087] Step S35: calculating an immediate control component corresponding to the current deviation through a proportional link according to the thickness deviation value, calculating a steady-state control component corresponding to the historical deviation accumulation through an integral link, and calculating a predictive control component corresponding to the deviation trend through a differential link to obtain a feedback control amount.

[0088] The embodiment of the present application is to enhance the steady-state accuracy and response sensitivity of the control system. The system is based on the traditional PID feedback control superimposed on the basis of fuzzy control quantity. Specifically, the proportional element takes the current collected thickness deviation value as the input, and directly calculates the adjustment component according to the proportional factor; the integral element accumulates the integral results of the deviation value in a period of time to form a long-term deviation correction quantity for system steady-state adjustment; the derivative element analyzes the rate of change of the deviation value, identifies the trend deviation and makes a quick response. The above three parts are automatically executed by the built-in algorithm of the controller according to the set period without manual intervention. Finally, the feedback control quantity is synthesized, and the prediction control quantity generated by the fuzzy reasoning is combined to form the output signal of the control system, realizing the intelligentization and high responsiveness of the coating machine transverse closed-loop regulation control.

[0089] Preferably, step S32 comprises the following steps:

[0090] Step S321: According to the fuzzy variable definition data and the boundary influence prediction data, the actual value range, mean value and distribution density corresponding to each fuzzy level are counted, the concentration trend and dispersion degree of the data are identified, and the distribution characteristic analysis data is obtained;

[0091] The embodiment of the present application takes the defined fuzzy variable level as the classification standard, and statistically processes the historical numerical samples of "disturbance intensity", "thickness influence degree" and "control adjustment quantity" in the boundary influence prediction data. Taking the disturbance intensity variable as an example, the system clusters all prediction data according to the fuzzy level (such as "weak", "medium", "strong", etc.) to which it belongs, and respectively counts the data range, average value and frequency distribution of each level. In this process, the system uses the probability density estimation method to model the data distribution, identifies the concentration trend (i.e. mainly concentrated around what value) and dispersion degree (i.e. whether the data in this fuzzy level is too dispersed) of the data in each fuzzy level, and forms the distribution characteristic analysis data, which provides a basis for subsequent function shape design.

[0092] Step S322: Based on the distribution characteristic analysis data, the triangular function core parameter determination processing is performed, the statistical mean value of each fuzzy level is taken as the triangular vertex position, and the width of the bottom of the triangle is determined according to the standard deviation of the data, and the function core parameter data is obtained;

[0093] The embodiment of the application designs an initial shape of a corresponding triangular membership function for each fuzzy grade based on the distribution feature analysis data obtained in the previous step. Specifically, the system takes the statistical mean of the data of each fuzzy grade as the vertex position of the triangle, i.e., the maximum membership point of the fuzzy grade; and the width of the bottom side of the function is determined by the standard deviation of the grade, the greater the standard deviation, the wider the bottom side of the triangle, indicating that the range of values allowed in the grade is wider. For example, for the "medium" grade of disturbance intensity, if the mean is 5 and the standard deviation is 1, the system constructs a triangular membership function with the vertex at 5 and the bottom side covering 3 to 7, and records the core parameter data of all functions, including the vertex position, the left and right bottom side endpoints, and the function slope coefficient.

[0094] Step S323: Adjust the overlap of adjacent functions according to the function core parameter data, calculate the intersection position and the overlapping area of adjacent triangular functions, and adjust the bottom side range of the triangle when the overlap deviates from the target range to obtain overlap adjustment parameter data.

[0095] To ensure smooth transition between fuzzy variable grades and avoid discontinuity in the output of the controller, the system adjusts the overlap of adjacent triangular functions. Specifically, the intersection position and the overlapping area of adjacent two triangular functions on the numerical axis are calculated, and the results are compared with the preset target overlap range (generally 25% to 40%). When the intersection of two functions is too close, causing insufficient overlap area, or the intersection is too far, causing excessive overlap area, the system automatically adjusts the bottom side range by narrowing or expanding the left and right endpoints of the bottom side to make the function overlap area return to a reasonable range. For example, for the "medium" grade and the "strong" grade, the initial intersection is 6.5, and the overlap area accounts for only 15% of the sum of the areas of the two functions. The system widens the right boundary of "medium" to 7.2 and moves the left boundary of "strong" to the left to 6.2 to improve the overlap area ratio, and finally forms overlap adjustment parameter data.

[0096] Step S324: Verify the integrity of the function based on the overlap adjustment parameter data, check the coverage of all membership functions on the entire definition domain, and fine-tune the boundaries when gaps or excessive overlaps are found to obtain membership function parameter data.

[0097] The embodiment of the present application performs integrity verification on all membership functions that have completed overlap adjustment. The specific method is to traverse the entire interval of the definition domain of each fuzzy variable (for example, disturbance intensity is 0 to 10), verify whether any value point in the interval is covered by at least one membership function, and check whether more than two functions have membership close to 1 at a value point, causing excessive overlap. If it is found that a value point is not effectively covered by any function, it means that there is a gap between functions, and the system will automatically fine-tune the bottom edge of the adjacent function to extend the range to fill the gap; if it is found that multiple functions have maximum membership at a value point, the system fine-tunes the distance between the vertices of the functions to reduce the excessive overlap. After the above integrity correction, the final complete membership function parameter data meeting the accuracy requirements of closed-loop control is generated, providing a stable, continuous and unambiguous input-output mapping basis for fuzzy reasoning.

[0098] Preferably, step S4 comprises the following steps:

[0099] Step S41: Perform actuator control signal conversion according to the feedback control quantity and the predictive control quantity, and drive the actuator to move through signal conditioning and power amplification, to obtain actuator response data;

[0100] The embodiment of the present application generates the final control target value by weighting and synthesizing the predictive control quantity output by the fuzzy controller and the feedback control quantity output by the PID controller, and converts it into a control signal suitable for the actuator. The signal is first processed by a signal conditioning circuit for level matching, filtering and anti-interference, and then converted into a current or voltage signal with actual driving capability through a power amplification circuit. Taking the driving of the horizontal scraper fine adjustment mechanism as an example, the control signal is finally used to control the micro-displacement of the stepper motor or piezoelectric driver, adjust the horizontal distance between the scraper and the substrate, and realize fine adjustment of the local coating thickness. The system synchronously collects parameters such as actuator position feedback, action response time and adjustment amplitude to form structured actuator response data for subsequent effect evaluation.

[0101] Step S42: Perform real-time effect monitoring processing based on the actuator response data, calculate the improvement degree of the parameters before and after the control action, record as effective control when the thickness deviation decreases or the boundary disturbance intensity decreases, record as invalid control when the parameters deteriorate, and obtain control effect evaluation data;

[0102] The embodiment of the present invention uses the actuator response data as a benchmark, calls the real-time thickness deviation data collected by the thickness sensor and the boundary disturbance intensity data fed back by the pressure sensor, extracts the key parameter values ​​before and after the control execution, and calculates the improvement range. The improvement range can be expressed as the absolute reduction in the thickness deviation value or the decrease in the disturbance intensity index. For example, if the thickness deviation at a certain position decreases from +4μm to +1.2μm, it is considered positive control; if the deviation deteriorates from -2μm to -5μm, it is determined to be a control failure. The system labels each control operation as "valid" or "invalid", and outputs control effect evaluation data based on the action time and response speed, providing a basis for subsequent strategy judgment.

[0103] Step S43: Record the disturbance event processing results based on the control effect evaluation data, create a corresponding file for the type, intensity, adopted control strategy and final processing effect of each boundary disturbance event, and calculate the success rate and effect index of different types of disturbances under different control strategies to obtain processing result statistics;

[0104] The embodiment of the present invention records the processing status of each boundary disturbance event one by one based on the control effect evaluation data. The record content includes: the type of disturbance event (such as instantaneous disturbance, boundary backflow fluctuation, etc.), the disturbance intensity level, the number of the called fuzzy control rule and its specific control strategy (such as executing +10-step adjustment when the pressure difference at the left boundary increases), the change value of the thickness deviation and the disturbance index after control, etc. The system classifies each control strategy by disturbance type and counts the number of executions, the number of successes and the average effect improvement rate. For example, the combination strategy of "medium disturbance + local thickness anomaly" was executed 24 times, 18 times were effectively controlled, and the average deviation was improved by 3.1μm, forming the processing result statistical data, which will be used to dynamically optimize the control strategy.

[0105] Step S44: Calculate the success rate and average effect improvement of each fuzzy control rule in actual application based on the statistical data of the processing results, identify the rules with poor performance and the rules that need to be strengthened, and mark a rule as a rule to be optimized when the success rate of a rule is lower than 70% or the effect improvement is lower than the expected value, and obtain the rule performance analysis data

[0106] The embodiment of the present invention performs a performance analysis on each rule in the fuzzy control rule library based on the statistical data of the processing results. The main analysis dimensions include the success rate after the rule is triggered (that is, the proportion of effective control times after the output control amount) and the average improvement amplitude (such as the average value of the decrease in disturbance intensity after control). When the historical triggering times of a certain rule exceed 10 times and the success rate is less than 70%, or the average thickness deviation improvement amplitude is lower than the minimum standard set by the system (such as less than 2μm), the rule will be automatically marked as "to be optimized" by the system. At the same time, the system also identifies which rules perform well or poorly under specific working conditions (such as high coating speed or low viscosity conditions), and forms rule performance analysis data as a basis for decision-making on the optimization strategy.

[0107] Step S45: Adaptively adjust the fuzzy control rule base according to the rule performance analysis data to obtain optimized fuzzy rule base data, and accumulate and optimize to form an adaptive control strategy.

[0108] The embodiment of the present invention executes an adaptive optimization mechanism of the fuzzy control rule base based on the rule performance analysis data. The optimization methods include: modifying the parameters of the rules with low success rates (such as reducing the range of the input fuzzy variables, adjusting the size of the output control quantity), deleting the rules with extremely poor performance, or generating new candidate rules by cross-fusion of high success rate rules. The optimization process is carried out dynamically during the operation of the control system through online learning, without relying on human intervention. After each optimization, the system performs version management on the rule adjustment results to form optimized fuzzy rule base data. With the accumulation of disturbance processing data, the controller continuously learns from the historical effective control experience, and gradually evolves to form an adaptive control strategy with high adaptability to working conditions and generalization capabilities, so that the entire coating machine lateral closed-loop control system has stronger stability and predictability when facing disturbances.

[0109] It is particularly important that step S44 includes the following steps:

[0110] Step S441: Based on the statistical data of the processing results, the integrity, consistency and time continuity of the statistical data are checked to identify and mark abnormal data points. When the data missing rate exceeds 10% or there is a significant data jump, data cleaning is performed to obtain verified and cleaned statistical data;

[0111] The embodiment of the present application performs data quality inspection on the existing processing result statistical data, specifically including verifying the completeness of the data, judging whether there is missing record or sampling discontinuity; detecting data consistency, checking the reasonable fluctuation range of each statistical index on the time sequence, avoiding abnormal jump; and evaluating time continuity, ensuring that the data covers the control period corresponding to all boundary disturbance events. When the missing rate is detected to be more than 10% or there is an abnormal jump point, the system automatically calls the interpolation method (such as linear interpolation or spline interpolation) and the outlier detection algorithm (such as Z-score or median absolute deviation method based on moving window) to clean and repair the abnormal and missing data, eliminate or correct unreasonable data, and finally output the verified and cleaned statistical data meeting the statistical analysis requirements, ensuring that the subsequent performance evaluation is based on high-quality data set.

[0112] Step S442: Based on the statistical data, the success rate, the average effect improvement degree, the response time and the stability index of each fuzzy control rule are calculated respectively, a multi-dimensional performance profile of each rule is established, and rule performance index data is obtained.

[0113] The embodiment of the present application utilizes the cleaned statistical data, and calculates multi-dimensional performance indexes for each rule in the fuzzy control rule library. Specifically, the success rate refers to the proportion of the number of effective control times after the rule is triggered to the total number of triggering times; the average effect improvement degree is the average decline amplitude of the thickness deviation or disturbance intensity before and after the rule control; the response time refers to the time difference from the issuance of the control signal to the feedback of the actuator, reflecting the response speed of the control system; and the stability index evaluates the fluctuation of the output control amount of the rule under different disturbance intensities and working conditions, which is usually obtained by calculating the standard deviation or coefficient of variation of the control output. All indexes are integrated into the rule performance profile, forming structured rule performance index data, which is used to describe the comprehensive performance of each rule and provide detailed basis for subsequent analysis.

[0114] Step S443: According to the rule performance index data, transverse comparison analysis is performed, the best applicable scene and performance weak link of each rule are identified, the correlation diagram of rule performance and environmental conditions is established, and applicable condition analysis data is obtained.

[0115] The embodiment of the present invention conducts a horizontal comparison and analysis of each fuzzy control rule based on the rule performance index data, and explores the best applicable scenarios and performance shortcomings of each rule. The success rate and effect improvement level of the rule under different working parameters (including coating speed, paint viscosity, ambient temperature, etc.) are displayed through multidimensional data visualization tools (such as radar charts and heat maps), and the working condition intervals where the rule performs well and poorly are identified. Furthermore, the machine learning cluster analysis method is used to establish a correlation between the rule performance and the environmental conditions, and a correlation relationship map is generated to reveal the regularity of how the rule performance is affected by changes in working conditions, providing a quantitative basis for targeted optimization, thereby obtaining systematic applicable condition analysis data.

[0116] Step S444: Generate rule optimization suggestions based on the applicable condition analysis data to obtain rule performance analysis data.

[0117] The embodiment of the present invention automatically generates rule optimization suggestions based on the applicable condition analysis data. The optimization suggestions include proposing specific improvement directions for poorly performing rules, such as adjusting the fuzzy membership function parameters to expand or narrow the range of input variables and enhance the response sensitivity of the rules to specific disturbance intensities; or recommending the merging of rules with overlapping functions and similar effects to simplify the control logic; at the same time, the system will also recommend the priority use of the best performance rule combination under the corresponding working conditions based on the performance of the rules under specific environmental conditions to improve the overall control effect. The suggestion is formatted into rule performance analysis data by the rule base management module, which serves as an important input for the adaptive optimization of the fuzzy PID controller, supporting subsequent dynamic adjustment and iterative upgrades.

[0118] The present invention also provides a coating machine transverse closed-loop control system based on fuzzy PID, which is used to execute the above-mentioned coating machine transverse closed-loop control method based on fuzzy PID. The coating machine transverse closed-loop control system based on fuzzy PID includes:

[0119] The boundary monitoring module is used to collect the pressure value of each measurement point in real time by arranging multiple pressure sensors inside the coating head boundary, and calculate the pressure difference between adjacent sensors to form a pressure gradient sequence. The abnormality of the boundary flow state is identified based on the distribution pattern of the pressure gradient sequence. When the pressure gradient exceeds the normal fluctuation range, it is marked as a boundary disturbance event, and the boundary state monitoring data is obtained.

[0120] The correlation analysis module is used to measure the actual coating thickness at each lateral position and compare it with the preset target thickness to calculate the thickness deviation value. The thickness deviation value is temporally correlated based on the boundary state monitoring data to identify the thickness variation pattern caused by the boundary disturbance event. Based on the thickness variation pattern, the corresponding relationship between the boundary disturbance and the thickness impact is established to generate boundary impact prediction data.

[0121] The intelligent decision module is used to construct a fuzzy control rule base according to the boundary influence prediction data, wherein the fuzzy control rule base takes the intensity of boundary disturbance and the degree of thickness influence as fuzzy input variables and takes the control adjustment as fuzzy output variable; when a new boundary disturbance event is detected, the predictive control amount is generated by using the fuzzy control rule base; and the feedback control amount is obtained by PID calculation according to the thickness deviation value.

[0122] The adaptive optimization module is used to perform real-time adjustment of coating parameters according to the feedback control amount and the predictive control amount, continuously monitor the adjustment effect and record the processing result of the boundary disturbance event, and automatically adjust the weight coefficient of the corresponding rule in the fuzzy control rule base when the processing effect is poor, so as to accumulate optimization and form an adaptive control strategy.

[0123] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application being defined by the appended claims and not by the above description, and it is intended to encompass all variations falling within the meaning and the scope of the equivalent elements of the application file. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features developed herein.

[0124] The above description is only a specific implementation of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features developed herein.

Claims

1. A lateral closed-loop control method for a coating machine based on fuzzy PID, characterized in that: The following steps are involved: Step S1: Multiple pressure sensors are arranged inside the coating head boundary to collect the pressure value of each measurement point in real time, and the pressure difference between adjacent sensors is calculated to form a pressure gradient sequence; the abnormality of the boundary flow state is identified based on the distribution pattern of the pressure gradient sequence, and when the pressure gradient exceeds the normal fluctuation range, it is marked as a boundary disturbance event, thereby obtaining boundary state monitoring data; Step S2: Measure the actual coating thickness at each lateral position and compare it with the preset target thickness to calculate the thickness deviation value; perform time correlation on the thickness deviation value based on the boundary state monitoring data to identify the thickness variation pattern caused by the boundary disturbance event; establish the corresponding relationship between the boundary disturbance and the thickness impact based on the thickness variation pattern to generate boundary impact prediction data; Step S3: A fuzzy control rule base is constructed based on the boundary impact prediction data, wherein the fuzzy control rule base uses the intensity of the boundary disturbance and the degree of thickness influence as fuzzy input variables and the control adjustment amount as the fuzzy output variable; when a new boundary disturbance event is detected, the fuzzy control rule base is used to generate a predictive control amount; and a PID calculation is performed based on the thickness deviation value to obtain a feedback control amount; Step S4: Adjust the coating parameters in real time according to the feedback control amount and the predictive control amount, continuously monitor the adjustment effect and record the processing results of the boundary disturbance events, and automatically adjust the weight coefficient of the corresponding rule in the fuzzy control rule library when the processing effect is not good, so as to accumulate and optimize to form an adaptive control strategy.

2. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: synchronously collecting the instantaneous pressure value of each measuring point through the pressure sensor array arranged at equal intervals along the inner side of the coating head boundary, and marking them according to the sensor number and sampling time to obtain time series pressure data; Step S12: digitally filter the time series pressure data to remove high-frequency noise interference and obtain stable pressure measurement data; Step S13: Based on the pressure measurement data, the pressure gradient value at each position is obtained by calculating the pressure difference between adjacent sensors, and the pressure gradient values ​​are arranged in spatial order to form a pressure gradient distribution sequence. The statistical characteristic parameters of the sequence, including the mean, standard deviation, and maximum gradient value, are calculated to obtain a pressure gradient sequence. Step S14: Identify the abnormality of the boundary flow state according to the distribution pattern of the pressure gradient sequence, and mark it as a boundary disturbance event when the pressure gradient exceeds the normal fluctuation range to obtain boundary state monitoring data.

3. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 2 is characterized in that: Step S14 includes the following steps: Step S141: by collecting the pressure gradient sequence during normal operation of the coating machine, and calculating the mean distribution and coefficient of variation distribution of the pressure gradient sequence, the upper and lower limits of the normal fluctuation range of the pressure gradient at each position are determined, thereby obtaining a pressure gradient reference pattern; Step S142: comparing the real-time measured pressure gradient sequence with the pressure gradient reference pattern point by point. When the pressure gradient value at any position exceeds the corresponding normal fluctuation range, the exceeding position and degree are recorded to obtain pressure out-of-range data. Step S143: calculating a boundary anomaly intensity index according to the number of exceeding positions and the degree of exceeding in the pressure exceeding range data; Step S144: Disturbance events are identified based on the boundary anomaly intensity index. An anomaly intensity threshold is set to classify the boundary status into normal, slightly abnormal, moderately abnormal, and severely abnormal. When the anomaly intensity index exceeds the set threshold continuously for a preset period of time, a boundary disturbance event is confirmed to have occurred and the start time, duration, and intensity level of the event are recorded to obtain boundary disturbance event characterization data. Step S145: Integrate the boundary disturbance event information identified in the boundary disturbance event characterization data, including event type, intensity level, spatial location and time characteristics, into a standardized data format to obtain structured boundary state monitoring data.

4. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 3 is characterized in that: In step S2, the actual coating thickness at each lateral position is measured and compared with the preset target thickness to calculate the thickness deviation value, which includes: The original thickness signal of the lateral distribution is synchronously collected by a thickness detection sensor arranged at a position 500mm-1500mm downstream of the coating head; The original thickness signal is subjected to ±0.2°C temperature compensation and ±0.1μm zero point correction to obtain the corrected thickness data; Obtain the target thickness value corresponding to the lateral position according to the process settings. When the target thickness value has a lateral gradient, the precise target thickness value corresponding to the measuring point is obtained by cubic spline interpolation calculation; The difference between the corrected thickness data and the precise target thickness value is calculated and smoothed using the five-point moving average method to obtain stable thickness deviation data; The mean and standard deviation of the thickness deviation distribution are calculated based on the stable thickness deviation data. When the absolute value of the single-point deviation exceeds three times the standard deviation or the deviation gradient of adjacent points exceeds 2μm / cm, it is marked as an abnormal position to obtain the thickness deviation distribution characteristic data; A data confidence assessment mechanism is established based on the thickness deviation distribution characteristic data. When the signal-to-noise ratio of the measured data is lower than 20dB, the original thickness signal at the corresponding position is re-collected to obtain the thickness deviation data. A time series analysis of the thickness deviation data with a time window of 10 seconds was performed to calculate the deviation change rate and periodic characteristics of each position. When the deviation change rate exceeded 0.5 μm / s, a trend warning was issued. The thickness deviation value including position coordinates, deviation value, change trend, confidence level and warning mark was obtained.

5. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 4 is characterized in that: In step S2, time correlation of the thickness deviation values ​​is performed based on the boundary state monitoring data to identify the thickness variation pattern caused by the boundary disturbance event, including: The material transmission delay time is calculated based on the coating line speed and the distance from the boundary monitoring position to the thickness detection position to obtain the time synchronization correction parameter; The timestamp of the thickness deviation value is corrected forward based on the time synchronization correction parameter and kept consistent with the time reference of the boundary status monitoring data to obtain the time synchronized thickness deviation data; The boundary disturbance events in the boundary state monitoring data are arranged in chronological order to form an event time series, and the time-synchronized thickness deviation data are arranged in chronological order to form a deviation time series, and a double-sequence time comparison table is established; An event matching search is performed based on the dual-sequence time comparison table. When the time difference between the boundary disturbance event and the thickness anomaly is within the range of ±2 seconds, it is marked as a potential correlation event pair, and candidate correlation event data is obtained. Based on the candidate correlation event data, the correlation coefficient between the intensity level of the boundary disturbance event and the corresponding thickness anomaly degree is calculated. When the correlation coefficient is greater than 0.6, it is confirmed as a valid correlation event pair to obtain the verified correlation event data; The thickness variation pattern feature is extracted based on the verification correlation event data to obtain the thickness variation pattern.

6. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 5 is characterized in that: In step S3, establishing a corresponding relationship between boundary disturbance and thickness influence based on the thickness variation pattern includes: By calculating the Euclidean distance between the feature vectors of different thickness change patterns, similar patterns are classified into the same category when the distance is less than a set threshold, and pattern classification data is obtained; Based on the pattern classification data, the boundary disturbance feature distribution corresponding to each category of thickness change pattern is statistically analyzed, and the average disturbance intensity, disturbance type and occurrence frequency corresponding to each change pattern are calculated to obtain disturbance feature statistics. A quantitative mapping relationship between disturbance intensity and impact degree is established based on disturbance characteristic statistical data. By fitting the functional relationship between the boundary disturbance intensity index and the thickness variation amplitude, the mapping relationship is confirmed to be valid when the fitting correlation coefficient is greater than 0.8, and the quantitative prediction model data is obtained. Based on the quantitative prediction model data, the correction effect of changes in coating speed, coating viscosity and ambient temperature on the degree of disturbance is analyzed to generate the working condition adaptability correction coefficient; According to the quantitative prediction model data and the working condition adaptability correction coefficient, the disturbance type, intensity level and working condition parameters are used as query conditions, and the expected thickness impact pattern, impact degree and spatial range are used as prediction results to construct the boundary impact prediction data.

7. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 6 is characterized in that: Step S3 includes the following steps: Step S31: performing fuzzy variable definition processing according to the boundary influence prediction data, performing fuzzy level classification on the boundary disturbance intensity, thickness influence degree, and control adjustment amount, and obtaining fuzzy variable definition data; Step S32: performing membership function design processing based on the fuzzy variable definition data, and using the triangular membership function to define the numerical range of each fuzzy level according to the statistical distribution characteristics in the boundary impact prediction data, to obtain membership function parameter data; Step S33: constructing a fuzzy control rule base based on the membership function parameter data and the boundary impact prediction data, establishing inference rules by analyzing the combination of disturbance intensity and impact degree, and simplifying the rule base by merging similar rules when the total number of rules exceeds 15 to obtain a fuzzy control rule base; Step S34: When a new boundary disturbance event is detected, real-time fuzzy inference processing is performed according to the fuzzy control rule library. The current disturbance intensity level and the predicted impact degree are used as input variables. The activated inference rules are found through rule matching. The weights of the activated rules are calculated and weighted averaged to obtain the predictive control quantity. Step S35: According to the thickness deviation value, the immediate control component corresponding to the current deviation of the proportional link is calculated, the steady-state control component corresponding to the historical deviation accumulation of the integral link is calculated, and the predictive control component corresponding to the deviation change trend of the differential link is calculated to obtain the feedback control amount.

8. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 7 is characterized in that: Step S32 includes the following steps: Step S321: Counting the actual numerical range, mean, and distribution density corresponding to each fuzzy level based on the fuzzy variable definition data and boundary impact prediction data, identifying the central tendency and dispersion of the data, and obtaining distribution characteristic analysis data; Step S322: performing triangle function core parameter determination processing based on the distribution feature analysis data, taking the statistical mean of each blur level as the triangle vertex position, and determining the base width of the triangle based on the standard deviation of the data to obtain the function core parameter data; Step S323: performing overlapping adjustment on adjacent functions based on the function core parameter data, by calculating the intersection position and overlapping area of ​​adjacent triangle functions, and adjusting the base range of the triangle when the overlap deviates from the target range, to obtain overlapping adjustment parameter data; Step S324: Perform function integrity verification based on the overlap adjustment parameter data, check the coverage of all membership functions on the entire definition domain, and perform boundary fine-tuning when coverage gaps or excessive overlap are found to obtain membership function parameter data.

9. The lateral closed-loop control method of a coating machine based on fuzzy PID according to claim 8, characterized in that: Step S4 includes the following steps: Step S41: converting the actuator control signal according to the feedback control variable and the predictive control variable, and driving the actuator to operate through signal conditioning and power amplification to obtain actuator response data; Step S42: Performing real-time effect monitoring based on the actuator response data to calculate the degree of parameter improvement before and after the control action is implemented. When the thickness deviation decreases or the boundary disturbance intensity decreases, it is recorded as effective control, and when the parameter deteriorates, it is recorded as ineffective control, thereby obtaining control effect evaluation data; Step S43: Record the disturbance event processing results based on the control effect evaluation data, create a corresponding file for the type, intensity, adopted control strategy and final processing effect of each boundary disturbance event, and calculate the success rate and effect index of different types of disturbances under different control strategies to obtain processing result statistics; Step S44: Calculate the success rate and average effect improvement of each fuzzy control rule in actual application based on the statistical data of the processing results, identify the rules with poor performance and the rules that need to be strengthened, and mark a rule as a rule to be optimized when the success rate of a rule is lower than 70% or the effect improvement is lower than the expected value, and obtain the rule performance analysis data Step S45: Adaptively adjust the fuzzy control rule base according to the rule performance analysis data to obtain optimized fuzzy rule base data, and accumulate and optimize to form an adaptive control strategy.

10. A lateral closed-loop control system for a coating machine based on fuzzy PID, characterized in that: Used to execute the coating machine transverse closed-loop control method based on fuzzy PID as claimed in claim 1, the coating machine transverse closed-loop control system based on fuzzy PID includes: The boundary monitoring module is used to collect the pressure value of each measurement point in real time by arranging multiple pressure sensors inside the coating head boundary, and calculate the pressure difference between adjacent sensors to form a pressure gradient sequence. The abnormality of the boundary flow state is identified based on the distribution pattern of the pressure gradient sequence. When the pressure gradient exceeds the normal fluctuation range, it is marked as a boundary disturbance event, and the boundary state monitoring data is obtained. The correlation analysis module is used to measure the actual coating thickness at each lateral position and compare it with the preset target thickness to calculate the thickness deviation value. The thickness deviation value is temporally correlated based on the boundary state monitoring data to identify the thickness variation pattern caused by the boundary disturbance event. Based on the thickness variation pattern, the corresponding relationship between the boundary disturbance and the thickness impact is established to generate boundary impact prediction data. The intelligent decision-making module is used to construct a fuzzy control rule base based on boundary impact prediction data. The fuzzy control rule base uses the intensity of boundary disturbance and the degree of thickness impact as fuzzy input variables and the control adjustment amount as fuzzy output variables. When a new boundary disturbance event is detected, the fuzzy control rule base is used to generate a predictive control amount. The feedback control amount is obtained by performing PID calculation based on the thickness deviation value. The adaptive optimization module is used to adjust the coating parameters in real time according to the feedback control quantity and the predictive control quantity, continuously monitor the adjustment effect and record the processing results of the boundary disturbance events. When the processing effect is not good, the weight coefficient of the corresponding rule in the fuzzy control rule library is automatically adjusted, thereby accumulating optimization to form an adaptive control strategy.

Images