Real-time prediction based marine structure anticorrosion spraying thickness control method and device and medium

Abstract

The application relates to a marine structure anticorrosion spraying thickness control method and device based on real-time prediction, and a medium, wherein the method comprises the following steps: S1, a spraying state matrix is constructed; S2, the spraying state matrix is input into a trained spraying thickness prediction model to obtain a thickness prediction value; S3, actual coating thicknesses at each moment are collected in real time, and the actual coating thicknesses at each moment are compared with the thickness prediction values at the corresponding moments to obtain thickness prediction deviations; S4, thickness prediction deviations at multiple continuous moments are combined into a deviation sequence, and statistical features of the deviation sequence are calculated as deviation statistical features based on the deviation sequence; S5, a feature vector is generated by using the deviation statistical features, and adjustment amounts of PID controller control parameters are respectively obtained based on the adjustment vector; and S6, spraying pressure and nozzle moving speed are corrected based on the adjustment amounts of the control parameters and the real-time deviation. Compared with the prior art, the application has the advantages of improving thickness control precision and the like.

Description

Technical Field

[0001] This invention relates to the field of coating thickness control, and in particular to a method, apparatus and medium for controlling the thickness of anti-corrosion spray coatings on marine structures based on real-time prediction. Background Technology

[0002] Offshore structures such as offshore wind turbine foundations, platform jackets, and wharf steel pipe piles are subjected to harsh corrosive environments including high salt, high humidity, and ultraviolet radiation for extended periods. Their anti-corrosion coatings are crucial for ensuring structural safety and extending service life. Currently, anti-corrosion spraying operations for offshore structures mainly rely on manual experience, with operators setting spraying equipment parameters based on environmental conditions and their own experience. Furthermore, some existing technologies, such as CN120764226A, disclose an intelligent planning and construction method for multi-layer spraying of heavy-duty anti-corrosion coatings for concrete water tanks. This method determines the spraying sequence parameters, material combinations, and basic spraying parameter set for each coating based on the structural parameters and anti-corrosion grade of the water tank; it generates a spraying path planning set by combining equipment attributes and environmental data and clarifies the execution priority; it generates a spraying parameter correction list and a quality verification scheme list based on the construction cycle and drying rate; and it forms an intelligent overall construction plan by linking real-time data. It also covers the determination of regional process boundaries, construction mode switching, path optimization, quality verification, and construction data management. Through multi-factor fusion and intelligent algorithms, the method achieves precise control and quality optimization of multi-layer spraying, improving construction efficiency and anti-corrosion performance, and is suitable for heavy-duty anti-corrosion construction of concrete water tanks.

[0003] However, in the existing anti-corrosion spraying process of marine engineering structures, the complex surface structure of the sprayed material, including welds, bends, and nodes, leads to poor thickness uniformity. This makes it difficult to ensure the uniformity of coating thickness by manual spraying, and problems such as localized excessively thin or thick coatings are likely to occur. When a local area is too thin, it will lead to premature corrosion, and when a local area is too thick, it will cause problems such as sagging and paint waste. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the prior art by providing a method, device, and medium for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction, comprising: Step S1: Obtain the time sequence of each spraying state parameter and construct the spraying state matrix. The spraying state parameters include operating parameters, environmental parameters and basic parameters. The operating parameters include at least spraying pressure and nozzle moving speed. Step S2: Input the spraying state matrix into the trained spraying thickness prediction model to obtain the thickness prediction value; Step S3: Collect the actual coating thickness at each moment in real time and compare it with the thickness prediction value at the corresponding moment to obtain the thickness prediction deviation. Step S4: Combine the thickness prediction deviations at multiple consecutive time points into a deviation sequence, and calculate the statistical characteristics of the deviation sequence as the deviation statistical characteristics based on the deviation sequence. Step S5: Utilize the deviation statistical characteristics to generate a feature vector, and obtain the adjustment amount of each control parameter of the PID controller based on the adjustment vector; Step S6: Correct the spraying pressure and nozzle movement speed based on the adjustment amount and real-time deviation of each control parameter.

[0006] The operating parameters also include the distance between the nozzle and the workpiece surface, the paint flow rate, and the nozzle angle. The environmental parameters include ambient temperature, air humidity, wind speed, and salt spray concentration. The basic parameters include surface roughness, existing coating thickness, and material type.

[0007] The deviation statistical characteristics include the deviation mean, deviation variance, maximum deviation value, and deviation change rate.

[0008] Step S5 includes: Step S5-1: Compare the mean deviation with each first mean interval, and take the first index of the first mean interval containing the mean deviation as the first feature; Step S5-2: Compare the deviation variance with each second mean interval, and take the second index of the second mean interval containing the deviation variance as the second feature; Step S5-3: Compare the maximum deviation value with each third mean interval, and take the third index of the third mean interval containing the maximum deviation value as the third feature; Step S5-4: Compare the rate of change of deviation with each fourth mean interval, and take the fourth index of the fourth mean interval containing the rate of change of deviation as the fourth feature; Step S5-5: Concatenate the first feature, second feature, third feature, and fourth feature to obtain the feature vector; Step S5-6: Based on feature vector reasoning, obtain the adjustment amount of each control parameter of the PID controller, wherein the control parameters of the PID controller include proportional coefficient, integral coefficient and derivative coefficient.

[0009] Steps S5-6 include: Obtain the probability distribution of the adjustment amounts of each control parameter corresponding to the feature vector; The adjustment amount of each control parameter is obtained by sampling according to the probability distribution of the adjustment amount of each control parameter.

[0010] Steps S5-6 are obtained using fuzzy PID rule reasoning. The first mean interval consists of seven intervals, with each first index corresponding to negative large, negative medium, negative small, zero, positive small, positive medium, and positive large, respectively. There are three second mean intervals, with each second index corresponding to small, medium, and large, respectively. There are a total of 5 third mean intervals, with each third index corresponding to negative large, negative small, zero, positive small, and positive large, respectively. There are a total of 5 fourth mean intervals, with each fourth index corresponding to negative large, negative small, zero, positive small, and positive large, respectively; Fuzzy PID rules should include at least the following: Rule 1: If the first index corresponding to the mean deviation is positive and large, the second index corresponding to the variance deviation is large, the third index corresponding to the maximum deviation value is positive and large, and the fourth index corresponding to the rate of change of deviation is positive and small, then the adjustment amount of the proportional coefficient is positive and large, the adjustment amount of the integral coefficient is positive and small, and the adjustment amount of the derivative coefficient is moderate. Rule 2: If the first index corresponding to the mean deviation is negative and the second index corresponding to the variance deviation is small, the third index corresponding to the maximum deviation value is negative and the fourth index corresponding to the rate of change of deviation is negative and large, then the adjustment amount of the proportional coefficient is negative and small, the adjustment amount of the integral coefficient is negative and large, and the adjustment amount of the differential coefficient is zero. Steps S5-6 include: Based on the feature vector verification of each fuzzy PID rule, the possible ranges of the control vector adjustment amount and the number of hits of each possible range are obtained; By using the number of hits in each possible interval as a weight and combining it with the boundary values ​​of each possible interval, the precise value of the adjustment amount of each control vector is calculated.

[0011] In step S2, the thickness prediction value at time t is predicted by the spraying state matrix from time tk to time t, wherein the data from time tk to time t-1 in the spraying state matrix are actual measured values.

[0012] In the spraying state matrix, the ambient temperature, air humidity, wind speed, and salt spray concentration at time t are predicted values, the spraying pressure and nozzle moving speed at time t are calculated using the target thickness, and the distance between the nozzle and the workpiece surface and the nozzle angle at time t are calculated based on spatial calculation.

[0013] A device for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction includes a memory, a processor, and a program stored in the memory. When the processor executes the program, it implements the method described above.

[0014] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In addition to collecting operating parameters and environmental parameters of the spraying equipment, the system also collects basic information, including surface roughness, to construct a spraying state matrix. This matrix is ​​then input into a prediction model based on a temporal convolutional network and attention mechanism to output a future spraying thickness sequence. The coating thickness is monitored in real time and compared with the predicted sequence to obtain a thickness deviation sequence. Based on the deviation statistical characteristics, the system adaptively adjusts the PID controller parameters through fuzzy inference to generate adjustment commands for spraying pressure and nozzle movement speed. This enables intelligent control of the spraying equipment, ensuring uniform and stable coating thickness to adapt to the complex marine engineering environment and structural characteristics.

[0016] 2. By incorporating operating parameters such as nozzle distance, paint flow rate, and nozzle angle, as well as environmental parameters such as ambient temperature, humidity, and wind speed, the model comprehensively covers key factors affecting coating thickness, improving the accuracy and robustness of the prediction model. Due to the introduction of basic parameters such as surface roughness and existing coating thickness, the model can adapt to the heterogeneous characteristics of the workpiece surface, enhancing its adaptability to different marine engineering structure materials.

[0017] 3. Deviation statistical features, including mean, variance, maximum deviation, and rate of change, can comprehensively characterize the dynamic properties of the deviation, avoiding the limitations of a single deviation index. Based on multi-dimensional statistical features, control parameter adjustments are generated, making the PID controller response more precise and effectively suppressing system oscillations or overshoot.

[0018] 4. By mapping statistical features to discrete indices and concatenating them into feature vectors, the processing of complex data is simplified, and inference efficiency is improved. The structured design of feature vectors facilitates integration with fuzzy rules or probabilistic models, enabling the rapid generation of control parameter adjustment quantities and enhancing real-time performance.

[0019] 5. By extracting adjustment values ​​based on probability distribution, randomness is introduced to avoid the control strategy from getting trapped in local optima, thereby improving the system's adaptability in uncertain environments. The probability distribution method can smooth the control output, reduce equipment vibration caused by parameter mutations, and extend the life of the spraying equipment.

[0020] 6. Fuzzy PID rule-based reasoning, combined with human experience, makes control decisions more aligned with actual spraying scenarios, improving the intuitiveness and interpretability of control. Adjustments are calculated by weighting the number of rule hits, and multiple rule outputs are integrated to enhance control flexibility and precision, especially in nonlinear systems. Interval partitioning (e.g., negative large, positive large) simplifies the handling of continuous values, reduces computational complexity, and is suitable for embedded deployments.

[0021] 7. Historical measured data is used to predict the current thickness, and temporal correlation is utilized to improve prediction accuracy and reduce the impact of instantaneous errors. The sliding window design from time tk to time t takes into account both recent data and long-term dependencies, enabling the model to respond quickly to changes while maintaining stability.

[0022] 8. The environmental parameters at time t are predicted values, which overcomes the sensor delay problem and realizes true real-time forward control. The spraying pressure and calculation speed are based on dynamic programming of the target thickness and combined with the nozzle parameters calculated in space to ensure the coating uniformity of complex geometric surfaces. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the main steps of the method of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0025] A method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction, such as Figure 1 As shown, it includes: Step S1: Obtain the time sequence of each spraying state parameter and construct the spraying state matrix. The spraying state parameters include operating parameters, environmental parameters and basic parameters. The operating parameters include at least spraying pressure and nozzle moving speed. In this embodiment, the operating parameters also include the distance between the nozzle and the workpiece surface, the paint flow rate and the nozzle angle, the environmental parameters include the ambient temperature, air humidity, wind speed and salt spray concentration, and the basic parameters include the surface roughness, the existing coating thickness and the material type.

[0026] Because marine structures are exposed to harsh marine environments characterized by high salinity, high humidity, and abundant salt spray, the uniformity of the anti-corrosion coating thickness directly determines the structure's corrosion resistance and service life. Too thin a coating leads to premature corrosion, while too thick a coating easily causes sagging and wastes the coating. Furthermore, the surface of marine structures contains complex structures such as welds, bends, and joints. Dynamic changes in equipment operation, environmental conditions, and the initial state of the structure during the spraying process directly affect the coating thickness. Therefore, this application obtains the operating parameters and environmental parameters of the spraying equipment during the anti-corrosion spraying operation on the surface of marine structures. This data forms the foundation for the entire control method. Operating parameters are collected through sensors built into the spraying equipment, including dynamic data generated during equipment operation such as spraying pressure, nozzle movement speed, nozzle distance from the workpiece surface, real-time paint flow rate, and nozzle angle. Environmental parameters are obtained through environmental monitoring equipment deployed in the spraying operation area, covering meteorological and corrosion-related data specific to marine environments, such as ambient temperature, air humidity, wind speed, and salt spray concentration. At the same time, the structural curvature radius of the current spraying location is obtained by combining 3D scanning equipment, the solid content and volatile organic compound content of the spraying material are retrieved from the material property database, and the cumulative operating time of the spraying equipment is obtained from the equipment operation log. These data together constitute the original data source for subsequent analysis, providing a comprehensive information foundation for accurate feature extraction and avoiding control deviations caused by data loss. Based on operating parameters, the spraying pressure, nozzle movement speed, and nozzle-to-workpiece surface distance are obtained, and further, equipment operation characteristic sub-vectors are acquired. From a technical implementation perspective, firstly, the spraying pressure is logarithmically transformed to eliminate the influence of nonlinear distribution in the pressure data, resulting in a standardized spraying pressure. Secondly, the nozzle movement speed is piecewise linearly normalized to adapt to the speed variation range of different spraying stages, resulting in a standardized movement speed. Thirdly, the nozzle-to-workpiece surface distance is normalized to unify data dimensions, resulting in a standardized distance. Simultaneously, real-time paint flow rate and nozzle angle data of the spraying equipment are acquired. The paint flow rate is processed using a moving average filter to filter out interference from instantaneous fluctuations, resulting in a stable flow rate value that ensures the flow rate data reflects the true spraying state. Finally, the nozzle angle is sinusoidally transformed to accurately reflect its actual projection relationship with the workpiece surface, resulting in an effective angle value, as the tilt of the nozzle angle directly affects the paint coverage and thickness distribution. Furthermore, based on ambient temperature data, a temperature-viscosity relationship model was established using the Arrhenius equation to perform real-time compensation calculations on the coating viscosity, yielding corrected dynamic viscosity parameters. This is because temperature changes significantly affect coating viscosity, thereby altering the atomization effect and adhesion thickness during spraying. Finally, standardized spraying pressure, standardized moving speed, standardized distance, stable flow rate, effective angle value, and dynamic viscosity parameters were multi-dimensionally vectorized to form an enhanced equipment operation characteristic sub-vector. This vector comprehensively integrates the key factors affecting coating thickness during equipment operation, achieving a quantitative characterization of the equipment's operating status and providing accurate equipment-level feature support for subsequent thickness prediction. For example, when the nozzle moving speed of the spraying equipment fluctuates slightly due to mechanical vibration, piecewise linear normalization can incorporate it into a unified analysis dimension. Combined with the synergistic analysis of stable flow rate and dynamic viscosity parameters, the potential impact of such fluctuations on coating thickness can be accurately captured. The system acquires initial state data for ambient temperature, air humidity, wind speed, and the surface of marine structures based on environmental parameters, and obtains feature vectors of environmental influencing factors. Technically, the ambient temperature is first processed in multiple dimensions to obtain multiple ambient temperatures within a preset time period and calculate the average ambient temperature to obtain a standard value. Simultaneously, historical temperature data is retrieved to obtain historical standard values. Comparison between the two yields the current standardized temperature, considering both real-time temperature fluctuations and the reference value of historical data, avoiding the random influence of temperature data at a single point in time. A logarithmic transformation is performed on the air humidity to obtain standardized humidity, adapting to the distribution characteristics of humidity data. A square root transformation is performed on the wind speed to obtain standardized wind speed, effectively reducing the interference of large wind speed fluctuations on the analysis results. To address the unique characteristics of the marine environment, salt spray concentration data was acquired and subjected to exponential decay transformation to obtain an effective salt spray concentration value, as salt spray concentration affects the coating curing rate and thus the final thickness. The structural curvature characteristics of the current spraying location were calculated using 3D scanning data to obtain standardized curvature parameters, adapting to the impact of curvature variations on the complex surfaces of marine structures on the spraying thickness. Based on a material property database, the solid content and volatile organic compound content of the spraying material were normalized to obtain material property coefficients, quantifying the material's influence on coating formation. Based on an equipment wear model and combined with the cumulative operating time of the spraying equipment, time decay compensation calculations were performed to obtain an equipment condition correction factor, considering the impact of long-term equipment wear on spraying parameters. Finally, the standardized temperature, humidity, wind speed, effective salt spray concentration value, standardized curvature parameter, material property coefficient, and equipment condition correction factor were fused into a multi-dimensional vector to form an enhanced environmental impact factor feature sub-vector. This vector comprehensively covers external influencing factors such as environment, structure, materials, and equipment condition, achieving precise quantification of external spraying conditions and solving the problem of traditional methods neglecting the synergistic influence of multiple environmental factors. For example, in a marine environment with high salt spray and high humidity, the synergistic effect of effective salt spray concentration and standardized humidity can accelerate coating curing. Through feature extraction in this step, this synergistic effect can be quantified as part of the feature vector, providing an accurate basis for thickness prediction. The surface roughness, existing coating thickness, and material type are obtained based on initial state data, and basic feature sub-vectors are acquired. Initial state data is obtained through multiple detection methods: surface roughness is measured using a roughness meter; existing coating thickness is obtained using a coating thickness gauge; material type is determined through structural design documents or material testing; surface cleanliness is obtained by analyzing surface contaminant coverage using image recognition technology; surface energy parameters of the substrate material are obtained through contact angle measurement; stress concentration factors in key structural areas are obtained through finite element analysis; and historical corrosion rate data is accumulated through electrochemical detection. Technically, surface roughness is logarithmically transformed to obtain standardized roughness, eliminating the nonlinear effects of roughness data; existing coating thickness is linearly scaled to obtain standardized thickness, unifying the dimensions of existing coating thickness in different structural regions; and material type is uniquely encoded, converting categorical variables into binary vector representations to obtain encoded material types, facilitating quantitative analysis of the influence of different materials by the algorithm. Meanwhile, in some embodiments, surface contaminant coverage is analyzed using image recognition technology to obtain a standardized cleanliness index, as surface cleanliness affects coating adhesion and thickness uniformity. The coating adhesion potential is calculated based on contact angle measurement data to obtain a surface energy characteristic coefficient, quantifying the substrate material's adsorption capacity for the coating. Maximum stress location data is extracted based on finite element analysis results to obtain stress distribution characteristic values, as stress concentration in critical structural areas affects the coating's stress state and thickness requirements. A corrosion development trend model is established based on electrochemical detection results to obtain a corrosion risk index, quantifying the corrosion risk in different structural areas and providing a targeted basis for coating thickness control. Finally, standardized roughness, standardized thickness, coded material type, standardized cleanliness index, surface energy characteristic coefficient, stress distribution characteristic value, and corrosion risk index are fused into a multi-dimensional vector to form an enhanced basic feature sub-vector. This vector comprehensively characterizes the initial state and corrosion-related properties of the marine structure, enabling subsequent thickness prediction and control to adapt to individual structural differences. For example, for structural areas with high surface roughness and a high corrosion risk index, this feature extraction step quantifies their specific coating thickness requirements, providing a basis for subsequent precise control. A spraying state matrix is ​​constructed based on equipment operation feature vectors, environmental influencing factor feature vectors, and basic feature vectors, and a predicted spraying thickness sequence is obtained. Technically, the three types of feature vectors from multiple consecutive time steps are first horizontally concatenated to obtain a joint feature vector for each time step. Each joint feature vector integrates comprehensive information about the equipment, environment, and structure at the corresponding time point. Then, the joint feature vectors from multiple time steps are arranged chronologically to construct a spraying state matrix with rows equal to the number of time steps and columns equal to the feature dimensions. This matrix visually presents the dynamic characteristics of multiple factors changing over time during the spraying process.

[0027] The specific process for obtaining the feature sub-vectors is as follows: Step S1-A-1: Perform a logarithmic transformation on the spraying pressure to obtain a standardized spraying pressure; Step S1-A-2: Perform piecewise linear normalization on the nozzle moving speed to obtain a standardized moving speed; Step S1-A-3: Normalize the distance between the nozzle and the workpiece surface to obtain a standardized distance; Step S1-A-4: Obtain real-time paint flow rate and nozzle angle data of the spraying equipment; perform moving average filtering on the paint flow rate to eliminate instantaneous fluctuations and obtain a stable flow rate value; and perform sine transformation on the nozzle angle to reflect its actual projection relationship with the workpiece surface and obtain an effective angle value. Step S1-A-5: Perform real-time compensation calculation on the coating viscosity based on ambient temperature data, establish a temperature-viscosity relationship model based on the Arrhenius equation, and obtain the corrected dynamic viscosity parameters; Step S1-A-6: Perform multi-dimensional vector splicing based on the standardized spraying pressure, the standardized moving speed, the standardized distance, the stable flow rate, the effective angle value, and the dynamic viscosity parameter to obtain the enhanced equipment operation feature vector.

[0028] As described in steps S1-A-1-S1-A-6 above, the operating status of the spraying equipment directly determines the atomization effect, adhesion efficiency, and final coating thickness of the coating. Spraying pressure, nozzle movement speed, and nozzle distance from the workpiece surface are the core parameters affecting coating thickness, while coating flow rate, nozzle angle, and coating viscosity further alter the thickness distribution through synergistic effects. In marine engineering structure anti-corrosion spraying operations, equipment operating parameters are easily affected by mechanical vibration, component wear, and other factors, resulting in fluctuations. Different parameters have different dimensions and distribution characteristics; directly using them for analysis can lead to distorted feature representation, thus affecting the accuracy of thickness prediction. This invention performs a logarithmic transformation on the spraying pressure to obtain a standardized spraying pressure. The raw data of the spraying pressure often exhibits a non-linear distribution, and the degree of influence of different pressure ranges on coating thickness varies. Directly using the raw data can lead to unreasonable weight allocation in subsequent model analysis. The logarithmic transformation process converts the non-linearly distributed pressure data into a form closer to a normal distribution through mapping relationships, while simultaneously compressing the influence range of extreme values, thus achieving standardization of the pressure data. The parameter settings for this processing method need to be combined with the commonly used pressure range for marine spraying (usually 0.3-0.8 MPa) to determine the base and value range of the logarithmic transformation. This ensures that the transformed data can both retain the original pressure variation trend and eliminate analytical interference caused by dimensional differences. For example, when the spraying pressure experiences an instantaneous peak of 0.2 MPa due to gas source fluctuations, the logarithmic transformation can effectively reduce the interference of this extreme value on the overall eigenvector, making the standardized pressure data more reflective of the true stable state of the spraying pressure and providing a more reliable pressure characteristic basis for thickness prediction. The nozzle movement speed is piecewise linearly normalized to obtain a standardized movement speed. The reasonable range of nozzle movement speed varies in different spraying stages (e.g., start-up, constant speed, and finish-up), and the impact of speed changes on thickness differs across these ranges. A single linear normalization method cannot accurately adapt to these segmented characteristics. Based on the actual process requirements of marine structure spraying, piecewise linear normalization divides the speed range into three intervals: the start-up interval (0-0.5 m / s), the constant speed interval (0.5-1.5 m / s), and the finish-up interval (1.5-2.0 m / s). Within each interval, an independent linear mapping function is used to normalize the speed value to the [0,1] interval. This method more accurately preserves the variation characteristics of different speed intervals and avoids speed characteristic distortion caused by a single normalization method. For example, in the uniform speed range (0.5-1.5m / s), the effect of a speed change of 0.1m / s on the coating thickness is relatively stable. By using linear normalization within this range, the contribution of this change to the thickness can be accurately quantified. However, the speed changes in the starting and ending speed ranges are more sensitive. Independent normalization processing can better capture this sensitive characteristic and improve the response accuracy of the feature vector to speed changes. The distance between the nozzle and the workpiece surface is normalized to obtain a standardized distance. The distance directly affects the coating coverage and adhesion thickness. The original data has dimensions inconsistent with other parameters, and its reasonable variation range is narrow (typically 15-30cm). Normalization eliminates these dimensional differences, allowing this parameter to participate in subsequent analysis in conjunction with other standardized core parameters. This step uses the min-max normalization method, taking the commonly used range of nozzle distance in marine spraying as a benchmark, mapping the original distance value to the [0,1] interval. The mapping function is: Standardized distance = (Original distance - Minimum distance threshold) / (Maximum distance threshold - Minimum distance threshold), where the minimum distance threshold is set to 15cm and the maximum distance threshold is set to 30cm. For example, when the nozzle distance changes from 20cm to 25cm due to robotic arm vibration, the normalized data change can intuitively reflect the degree of influence of the distance change on the coating thickness, ensuring that this parameter is fused with core parameters such as spraying pressure and moving speed under the same dimensions, improving the consistency and reliability of the feature vector. Real-time paint flow rate and nozzle angle data from the spraying equipment are acquired. The paint flow rate is processed using a moving average filter to obtain a stable flow rate value, and the nozzle angle is processed using a sine transform to obtain an effective angle value. Real-time paint flow rate is susceptible to instantaneous fluctuations due to pump fluctuations. Direct use of this data would lead to unstable characteristic values. The moving average filter, by selecting a fixed time window (typically 0.5-1s), calculates the arithmetic average of the flow rate data within the window, filtering out instantaneous fluctuations and obtaining a stable flow rate value that reflects the true spraying flow rate. The window size setting needs to balance response speed and anti-interference capability, ensuring that meaningless fluctuations are eliminated while capturing flow trend changes. The original nozzle angle data is the angle between the nozzle and the workpiece surface (0-90°). Its impact on coating thickness is essentially determined by the projection relationship between the nozzle and the workpiece surface. The sine transform process calculates the sine value of the angle, converting the angle parameter into an effective angle value that reflects the actual projection coverage efficiency. When the nozzle angle is 90°, the effective angle value is 1, corresponding to the highest projection coverage efficiency. When the angle deviates from 90°, the effective angle value decreases with the change in the sine value, accurately quantifying the impact of angle changes on coating thickness. For example, when the paint flow rate suddenly changes by 0.1 L / min due to instantaneous pressure fluctuations in the pump body, the moving average filtering process can quickly filter out this fluctuation and output a stable flow rate value, avoiding distortion of the feature vector caused by instantaneous fluctuations; when the nozzle angle is adjusted from 90° to 60°, the effective angle value after sine transformation changes from 1 to 0.866, accurately reflecting the decrease in projection coverage efficiency and providing accurate angular feature basis for thickness prediction; Real-time compensation calculations were performed on the coating viscosity based on ambient temperature data. A temperature-viscosity relationship model was established based on the Arrhenius equation to obtain the corrected dynamic viscosity parameters. Coating viscosity is a key factor affecting atomization effect and coating adhesion thickness. Changes in ambient temperature directly lead to significant changes in viscosity. In marine spraying operations, the ambient temperature fluctuates significantly (5-35℃). Ignoring the influence of temperature on viscosity would result in inaccurate characterization of equipment operation. The core form of the Arrhenius equation is η=η0exp (E a / (RT)), where η is the dynamic viscosity, η0 is the viscosity constant at the reference temperature, and E a The activation energy is set to 40-60 kJ / mol based on commonly used anti-corrosion coatings for marine engineering, R is the gas constant (8.314 J / (mol·K)), and T is the absolute temperature (in K). This step obtains the real-time ambient temperature using environmental monitoring equipment and substitutes it into the above model to calculate the corrected dynamic viscosity parameters, achieving real-time compensation for viscosity changes with temperature. For example, when the ambient temperature rises from 25℃ to 35℃, according to the Arrhenius equation, the coating viscosity will decrease by approximately 20%-30%. Real-time correction of the dynamic viscosity parameters accurately reflects the impact of this change on the spraying effect, making the equipment operating characteristic vector more closely match the actual working environment and improving the accuracy of subsequent thickness prediction. An enhanced equipment operating feature vector is obtained by multi-dimensional vector splicing based on standardized spraying pressure, standardized moving speed, standardized distance, stable flow rate, effective angle value, and dynamic viscosity parameters. After the aforementioned steps, the parameters possess the characteristics of unified dimensions, stability, reliability, and accurate representation. Multi-dimensional vector splicing arranges the parameters in a fixed order to form a 6-dimensional enhanced equipment operating feature vector. Each dimension of the vector corresponds to an optimized equipment operating parameter, ensuring comprehensive quantification of the equipment's operating status. The splicing order is set as [standardized spraying pressure, standardized moving speed, standardized distance, stable flow rate, effective angle value, dynamic viscosity parameter], which combines the importance of the parameters' influence on coating thickness, facilitating rapid extraction of key features by the subsequent prediction model. For example, when the standardized spraying pressure is 0.6, the standardized moving speed is 0.5, the standardized distance is 0.4, the stable flow rate is 0.7, the effective angle is 0.9, and the dynamic viscosity parameter is 0.3, the feature vector formed after splicing is [0.6, 0.5, 0.4, 0.7, 0.9, 0.3]. This fully integrates the core features of equipment operation and provides structured and accurate equipment-level data support for subsequent spraying state matrix construction and thickness prediction.

[0029] The construction of the feature sub-vectors of environmental impact factors includes: Step S1-B-1: Obtain multiple ambient temperatures within the current preset time period, obtain the average ambient temperature based on the multiple ambient temperatures, and obtain the standard ambient temperature value based on the multiple ambient temperatures and the average ambient temperature. Step S1-B-2: Obtain historical temperature data, obtain historical temperature standard values ​​based on the historical temperature data, and obtain the current standardized temperature based on the historical temperature standard values ​​and the ambient temperature standard values; Step S1-B-3: Perform a logarithmic transformation on the air humidity to obtain a standardized humidity; Step S1-B-4: Perform a square root transformation on the wind speed to obtain a standardized wind speed; Step S1-B-5: Obtain salt spray concentration data specific to the marine environment, perform exponential decay transformation on the salt spray concentration to reflect its influence on the coating curing rate, and obtain an effective salt spray concentration value. Step S1-B-6: Obtain the structural curvature radius at the current spraying position, calculate the surface curvature characteristics based on the three-dimensional scanning data, and obtain the standardized curvature parameters; Step S1-B-7: Obtain the solid content and volatile organic compound content of the spraying material, and perform normalization processing based on the material property database to obtain the material property coefficient; Step S1-B-8: Obtain the cumulative running time of the spraying equipment, perform time decay compensation calculation based on the equipment wear model, and obtain the equipment state correction factor; Step S1-B-9: Perform multi-dimensional vector fusion based on the standardized temperature, standardized humidity, standardized wind speed, effective salt spray concentration value, standardized curvature parameter, material property coefficient and equipment state correction factor to obtain an enhanced environmental impact factor feature vector.

[0030] As described in steps S1-B-1 to S1-B-9 above, this invention acquires multiple ambient temperatures within a preset time period, calculates the average ambient temperature, and obtains a standard ambient temperature value. Ambient temperature is a key factor affecting the viscosity and curing rate of coatings. Temperature data at a single point in time is easily affected by instantaneous fluctuations and cannot reflect the true temperature state. This step uses environmental monitoring equipment to collect multiple ambient temperature data within a preset time window (e.g., 5 minutes) at fixed time intervals (usually 10-30 seconds). The arithmetic mean method is used to calculate the average ambient temperature within this window. Then, by calculating the average of the sum of squares of the deviations of each temperature data point from the mean and taking the square root, the standard ambient temperature value is obtained. This standard value reflects both the average temperature level and the degree of temperature fluctuation. For example, if 10 temperature data points are collected within 5 minutes, ranging from 20-25℃, the calculated average is 22.5℃, and the standard value is further calculated to be 1.2℃. This standard value comprehensively reflects the temperature state within this time period, avoiding the random influence of single temperature data and providing a reliable basis for subsequent temperature standardization processing. Historical temperature data is obtained to determine historical temperature standard values, which are then combined with ambient temperature standard values ​​to arrive at the current standardized temperature. Temperature variations in marine spraying operations exhibit certain periodicity and regularity; relying solely on real-time temperature data cannot reflect the differences between these variations and historical temperatures during the same period or under similar conditions. Historical temperature data is retrieved from the environmental history database of the work area. Historical data similar to the current work period and season (e.g., data from the same period last three months) is selected, and historical temperature standard values ​​are calculated using the same method as S1-B-1. The current standardized temperature is then obtained by normalizing the ratio of the current ambient temperature standard value to the historical temperature standard value. This value typically ranges from 0.8 to 1.2, directly reflecting the degree of deviation of the current temperature from historical levels. For example, if the current ambient temperature standard value is 1.2℃ and the historical temperature standard value is 1.0℃, then the current standardized temperature is 1.2, indicating that the current temperature fluctuation is higher than the historical average. This standardization process gives the temperature data historical comparative significance, improving the accuracy of temperature characteristic representation. Logarithmic transformation of air humidity is performed to obtain standardized humidity. Air humidity affects the drying speed and adhesion of coatings, and its raw data usually exhibits a non-linear distribution, which can lead to distortion of feature representation if used directly. Logarithmic transformation transforms the non-linear humidity data into a form closer to a normal distribution by selecting the natural logarithm or a commonly used logarithm (natural logarithm is used in this step), while compressing the influence range of extreme values, thus standardizing the humidity data. The transformation formula is: Standardized humidity = ln(actual humidity value + 1), where adding 1 is to avoid the logarithm being meaningless when the humidity value is 0. For example, when the air humidity is 80%, the standardized humidity is ln(80+1)≈4.39. This processing method preserves the trend of humidity data change and eliminates the analytical interference caused by dimensional differences, allowing humidity features to participate in subsequent analysis in conjunction with other standardized parameters. Standardized wind speed is obtained by performing a square root transformation on the wind speed. Wind speed affects the atomization range and adhesion efficiency of coatings. High wind speeds can cause coating dispersion, reducing coating thickness. The original wind speed data has a large variance, and extreme values ​​have a significant impact. The square root transformation reduces the dispersion of the data by calculating the square root of the wind speed, making the wind speed data closer to a normal distribution, thus achieving standardization. The transformation formula is: Standardized wind speed = The actual wind speed values ​​are obtained through wind speed sensors deployed in the work area, and the unit is m / s. For example, when the actual wind speed is 4 m / s, the standardized wind speed is 2, and when the actual wind speed is 9 m / s, the standardized wind speed is 3. This processing method effectively compresses the extreme effects of high wind speed data, making the changes in wind speed characteristics more stable and improving the stability of subsequent analysis. This involves acquiring salt spray concentration data specific to the marine environment and performing exponential decay transformation to obtain the effective salt spray concentration value. Salt spray is a corrosive factor unique to the marine environment, accelerating coating aging and affecting the coating curing rate. Its impact increases with increasing salt spray concentration, but the relationship is not linear. Salt spray concentration data is acquired using a dedicated salt spray sensor. The formula for exponential decay transformation is: Effective salt spray concentration value = 1 - exp(-k × actual salt spray concentration), where k is the decay coefficient, set to 0.001-0.005 (unit: m³) depending on the type of marine anti-corrosion coating. 2 / mg). This transformation makes the effective salt spray concentration value range between 0 and 1, which can accurately reflect the nonlinear effect of salt spray concentration on the coating curing rate. The higher the concentration, the closer the effective salt spray concentration value is to 1, and the more significant the effect on curing. For example, when the actual salt spray concentration is 300mg / m³, 2When k is 0.002, the effective salt spray concentration value is 1-exp (-0.002×300)=1-exp (-0.6)≈0.451, which accurately quantifies the degree of influence of the salt spray concentration on the coating curing and makes up for the shortcomings of traditional methods that ignore the influence of salt spray. The structural curvature radius at the current spraying location is obtained, and standardized curvature parameters are calculated based on 3D scanning data. Marine structures have complex structures such as welds, bends, and nodes, with varying curvature radii at different locations, affecting the distance between the nozzle and the workpiece surface and the coating coverage. The structural curvature radius is obtained using a 3D laser scanning device. Surface fitting is performed on the obtained 3D point cloud data to calculate the curvature radius at each spraying location. Then, the curvature radius is mapped to the [0,1] interval using a min-max normalization method to obtain standardized curvature parameters. The normalization baseline value is set according to the common curvature range of marine structures, with a minimum curvature radius threshold of 0.5m and a maximum curvature radius threshold of 10m. The mapping formula is: Standardized curvature parameter = (Actual curvature radius - Minimum threshold) / (Maximum threshold - Minimum threshold). For example, the actual radius of curvature at the current spraying location is 5m, and the standardized curvature parameter = (5-0.5) / (10-0.5) = 4.5 / 9.5≈0.474. This parameter accurately quantifies the curvature characteristics of the structural surface and provides environmental correlation characteristics at the structural level for subsequent thickness prediction. The solid content and volatile organic compound (VOC) content of the spraying material were obtained, and normalized based on a material property database to obtain material property coefficients. The solid content of the spraying material directly affects the dry film thickness of the coating, while the VOC content affects the drying speed of the coating. These two indicators vary significantly among different materials. Data was retrieved from a pre-set material property database, which stores key parameters such as solid content and VOC content of commonly used marine anti-corrosion coatings. The solid content and VOC content of the current material were normalized (mapped to [0,1]), and then a weighted average method (with weights set to 0.6 and 0.4 respectively) was used to calculate the material property coefficients. These coefficients comprehensively reflect the degree of influence of the material on the coating thickness. For example, if the normalized value of the solid content of a certain spraying material is 0.8 and the normalized value of the volatile organic compound content is 0.6, the material characteristic coefficient = 0.8×0.6+0.6×0.4=0.48+0.24=0.72, which accurately quantifies the influence of the material's characteristics on the coating formation, enabling the characteristic vector of environmental impact factors to integrate key information at the material level. The cumulative operating time of the spraying equipment is obtained, and a time decay compensation calculation is performed based on the equipment wear model to obtain the equipment condition correction factor. After long-term operation, components such as the pump body and nozzles of the spraying equipment will experience wear, leading to deviations between the actual output and theoretical values ​​of parameters such as spraying pressure and flow rate, affecting the coating thickness. The cumulative operating time of the equipment is obtained from the equipment operation log. The equipment wear model adopts a linear decay model, and the formula is: Equipment condition correction factor = 1 - λ × Cumulative operating time, where λ is the wear coefficient, set to 0.0001-0.0005 (unit: 1 / h) according to the equipment type and maintenance records. This factor ranges between 0.7 and 1, and the closer the value is to 1, the better the equipment condition. For example, if the cumulative operating time of the equipment is 1000h, λ is taken as 0.0002, and the equipment condition correction factor = 1 - 0.0002 × 1000 = 0.8. This factor quantifies the impact of equipment wear on the spraying effect, providing a correction basis at the equipment condition level for subsequent feature fusion, and avoiding feature representation distortion caused by equipment wear. A multi-dimensional vector fusion is performed based on standardized temperature, standardized humidity, standardized wind speed, effective salt spray concentration, standardized curvature parameter, material property coefficient, and equipment condition correction factor to obtain an enhanced environmental impact factor feature vector. After the aforementioned steps, the parameters possess uniform dimensions and precise characterization. Multi-dimensional vector fusion arranges these parameters in a fixed order to form a 7-dimensional enhanced environmental impact factor feature vector. The order is set as [standardized temperature, standardized humidity, standardized wind speed, effective salt spray concentration, standardized curvature parameter, material property coefficient, equipment condition correction factor]. This order reflects the importance of the parameters to the coating thickness, facilitating rapid extraction of key features by the subsequent prediction model. For example, when the parameters are 1.2, 4.39, 2, 0.451, 0.474, 0.72, and 0.8 respectively, the fused feature vector is [1.2, 4.39, 2, 0.451, 0.474, 0.72, 0.8]. This fully integrates the influencing factors from multiple dimensions such as environment, structure, materials, and equipment, providing structured and accurate environmental and correlation-level data support for the subsequent construction of the spraying state matrix and thickness prediction.

[0031] For the basic feature vectors, including: Step S1-C-1: Perform a logarithmic transformation on the surface roughness to obtain the standardized roughness; Step S1-C-2: Perform linear scaling on the existing coating thickness to obtain a standardized thickness; Step S1-C-3: Perform one-hot encoding on the material type, convert the category variable into a binary vector representation, and obtain the encoded material type; Step S1-C-4: Obtain the cleanliness level of the marine structure surface, analyze the surface contaminant coverage based on image recognition technology, and obtain standardized cleanliness indicators; Step S1-C-5: Obtain the surface energy parameters of the substrate material, calculate the coating adhesion potential value based on the contact angle measurement data, and obtain the surface energy characteristic coefficient; Step S1-C-6: Obtain the stress concentration factor in the key areas of the structure, extract the maximum stress location data based on the finite element analysis results, and obtain the stress distribution characteristic value; Step S1-C-7: Obtain historical corrosion rate data, establish a corrosion development trend model based on electrochemical detection results, and obtain the corrosion risk index; Step S1-C-8: Perform multi-dimensional vector fusion based on the standardized roughness, the standardized thickness, the encoded material type, the standardized cleanliness index, the surface energy characteristic coefficient, the stress distribution characteristic value, and the corrosion risk index to obtain an enhanced basic feature vector.

[0032] As described in steps S1-C-1 to S1-C-8 above, this invention performs a logarithmic transformation on surface roughness to obtain standardized roughness. Surface roughness is a key parameter affecting coating adhesion. Its original data typically exhibits a non-linear distribution, and different roughness ranges have varying degrees of influence on coating thickness. Direct use of this data can lead to distortion in feature representation. Surface roughness data is obtained by multi-point detection on the surface of marine structures using a roughness meter. The average value is taken as the original data, and the logarithmic transformation uses the natural logarithm. The transformation formula is: Standardized Roughness = ln(Original Roughness Value + 1), where adding 1 is to avoid the logarithm being meaningless when the roughness value is 0. This processing method can transform the non-linearly distributed roughness data into a form closer to a normal distribution, while compressing the influence range of extreme values, thus achieving standardization. For example, when the original surface roughness is 5 μm, the standardized roughness = ln(5+1) = ln6≈1.79. This processing preserves the trend of roughness data variation and eliminates analytical interference caused by dimensional differences, allowing roughness features to participate in subsequent analysis in conjunction with other standardized parameters. The existing coating thickness is linearly scaled to obtain a standardized thickness. The existing coating thickness determines the required thickness of the new coating. The existing coating thickness may vary significantly across different structural regions; linear scaling unifies these dimensions for easier subsequent analysis. The existing coating thickness is obtained using a coating thickness gauge. The linear scaling employs a min-max normalization method, mapping the original thickness value to the [0,1] interval. The mapping formula is: Standardized Thickness = (Original Existing Coating Thickness - Minimum Thickness Threshold) / (Maximum Thickness Threshold - Minimum Thickness Threshold). The minimum thickness threshold is set to 50 μm based on the minimum effective thickness of anti-corrosion coatings for marine structures, and the maximum thickness threshold is set to 300 μm based on the common maximum existing coating thickness. For example, when the existing coating thickness in a certain area is 150 μm, the standardized thickness = (150-50) / (300-50) = 100 / 250 = 0.4. This parameter accurately quantifies the relative level of the existing coating thickness, providing a reliable basis for predicting the thickness of subsequent new coatings. One-hot encoding is used to convert categorical variables into binary vector representations, resulting in encoded material types. Marine structures utilize diverse material types (such as carbon steel, stainless steel, and aluminum alloys), and different materials exhibit varying compatibility with coatings, making them unsuitable for direct processing by the prediction model when used as categorical variables. Material types are determined through structural design documents or material testing. One-hot encoding assigns a binary vector to each material type, with the vector length equal to the total number of material types. Positions corresponding to the material type are set to 1, and all other positions are set to 0. For example, if a marine structure includes carbon steel, stainless steel, and aluminum alloy, the encoded vector for carbon steel is [1,0,0], for stainless steel it is [0,1,0], and for aluminum alloy it is [0,0,1]. This method transforms categorical variables into numerical vectors recognizable by the model, accurately quantifying the differences in material types and providing material-level feature support for subsequent thickness prediction. The cleanliness level of marine engineering structure surfaces is obtained by analyzing surface contaminant coverage using image recognition technology to derive a standardized cleanliness index. Surface cleanliness directly affects the adhesion of coatings; contaminant coverage leads to poor adhesion between the coating and the substrate, thus affecting the uniformity of coating thickness. Surface cleanliness data is obtained by capturing images of the marine engineering structure surface using a high-definition camera. Image recognition technology is used to analyze the proportion of contaminant coverage area in the images to obtain the surface contaminant coverage rate. The standardized cleanliness index adopts a reverse mapping method, i.e., standardized cleanliness index = 1 - surface contaminant coverage rate. The closer the index value is to 1, the higher the surface cleanliness. For example, if the image recognition analysis shows a surface contaminant coverage rate of 10%, then the standardized cleanliness index = 1 - 0.1 = 0.9. This index accurately quantifies the surface cleanliness level, overcoming the shortcomings of traditional methods that ignore the influence of surface cleanliness. The surface energy parameters of the substrate material are obtained, and the coating adhesion potential value is calculated based on the contact angle measurement data to obtain the surface energy characteristic coefficient. The surface energy of the substrate material directly affects the wetting effect and adhesion of the coating. The higher the surface energy, the better the wetting effect and the stronger the adhesion, which in turn affects the stability of the coating thickness. The surface energy parameter is obtained by measuring the contact angle of the liquid on the surface of the substrate material using a contact angle meter and calculating it based on the Yang-Laplace equation. The calculation formula is: Surface Energy = γ_LV × (1 + cosθ) / 2, where γ_LV is the surface tension of the liquid and θ is the contact angle. The surface energy characteristic coefficient is linearly scaled to map the surface energy parameter to the [0,1] interval. The mapping formula refers to the min-max normalization method of S42. The threshold is set according to the surface energy range of common marine engineering materials (the minimum surface energy threshold is 20 mN / m, and the maximum surface energy threshold is 80 mN / m). For example, when the surface energy of the substrate material is 50 mN / m, the surface energy characteristic coefficient = (50-20) / (80-20) = 30 / 60 = 0.5. This coefficient accurately quantifies the surface energy level of the substrate material, providing a characteristic basis for the adhesion layer for subsequent thickness prediction. The stress concentration factor in key areas of the structure is obtained, and the location data of the maximum stress is extracted based on the finite element analysis results to obtain the stress distribution characteristic value. Stress concentration in key areas of marine engineering structures (such as welds and joints) affects the stress state of the coating, thus affecting the stability of the coating thickness. The larger the stress concentration factor, the greater the stress on the coating in that area, and the higher the requirement for thickness uniformity. The stress concentration factor is obtained through mechanical simulation analysis of the marine engineering structure using finite element analysis software. The location data of the maximum stress in key areas is extracted, and the stress distribution characteristic value is linearly scaled to map the stress concentration factor to the [0,1] interval. The mapping formula is: Stress distribution characteristic value = (Stress concentration factor - Minimum stress concentration factor) / (Maximum stress concentration factor - Minimum stress concentration factor). The minimum stress concentration factor is set to 1.0 (no obvious stress concentration), and the maximum stress concentration factor is set to 5.0 (severe stress concentration). For example, when the stress concentration factor of a certain critical area is 3.0, the stress distribution characteristic value = (3.0-1.0) / (5.0-1.0) = 2.0 / 4.0 = 0.5. This characteristic value accurately quantifies the stress distribution state of the critical area of ​​the structure, providing mechanical characteristic support for subsequent thickness prediction. Historical corrosion rate data is acquired, and a corrosion development trend model is established based on electrochemical detection results to obtain a corrosion risk index. Historical corrosion rate reflects the corrosion characteristics of marine structures in the marine environment; the faster the corrosion rate, the higher the requirement for coating thickness. The corrosion risk index quantifies the corrosion risk level of the structure. Historical corrosion rate data is accumulated through long-term monitoring of marine structures using electrochemical detection equipment and stored in a corrosion database. Based on this data, a linear regression model is used to establish a corrosion development trend model to predict the corrosion rate over a future period. The corrosion risk index = predicted corrosion rate / maximum allowable corrosion rate, where the maximum allowable corrosion rate is set to 0.1 mm / year based on the design service life of the marine structure. The index ranges from 0 to 1; the closer to 1, the higher the corrosion risk. For example, if the predicted future corrosion rate is 0.05 mm / year based on historical corrosion rate data, then the corrosion risk index = 0.05 / 0.1 = 0.5. This index accurately quantifies the corrosion risk level of the structure, providing corrosion-related characteristic basis for subsequent thickness prediction. An enhanced basic feature vector is obtained by multi-dimensional vector fusion based on standardized roughness, standardized thickness, encoded material type, standardized cleanliness index, surface energy characteristic coefficient, stress distribution characteristic value, and corrosion risk index. The parameters processed in the preceding steps possess uniform dimensions and precise representation. Multi-dimensional vector fusion forms the enhanced basic feature vector by arranging the parameters in a fixed order. The arrangement order is set as [standardized roughness, standardized thickness, encoded material type, standardized cleanliness index, surface energy characteristic coefficient, stress distribution characteristic value, corrosion risk index], where the encoded material type is a binary vector. The overall vector dimension is determined based on the total number of material types. If there are 3 material types, the vector dimension is 1 (standardized roughness) + 1 (standardized thickness) + 3 (encoded material type) + 1 (standardized cleanliness index) + 1 (surface energy characteristic coefficient) + 1 (stress distribution characteristic value) + 1 (corrosion risk index) = 9. For example, when the parameters are 1.79, 0.4, [1,0,0], 0.9, 0.5, 0.5, and 0.5 respectively, the fused feature vector is [1.79, 0.4, 1, 0, 0, 0.9, 0.5, 0.5, 0.5]. This fully integrates the key features of the initial state and related characteristics of the marine structure, providing structured and accurate structural foundation data support for the subsequent construction of the spraying state matrix and thickness prediction.

[0033] Subsequently, a spraying state matrix is ​​constructed based on the equipment operation feature vector, the environmental influencing factor feature vector, and the basic feature vector. A predicted spraying thickness sequence is then obtained based on the spraying state matrix, specifically including: The joint feature vector for each time step is obtained by horizontally concatenating the equipment operation feature vector, environmental influencing factor feature vector, and basic feature vector from multiple consecutive time steps. Arrange the joint feature vectors of multiple time steps in chronological order to construct a spraying state matrix with the number of rows equal to the number of time steps and the number of columns equal to the feature dimension; The spraying state matrix is ​​input into a pre-trained spraying thickness prediction model, and time series features are extracted through a temporal convolutional network to obtain a high-dimensional feature representation. The high-dimensional feature representation is weighted and weighted summed using an attention mechanism to obtain a context vector. The context vector is input into a fully connected regression layer, which outputs a sequence of predicted coating thicknesses for multiple future time steps.

[0034] As described above, this application horizontally concatenates the three types of feature vectors from multiple consecutive time steps to obtain a joint feature vector for each time step. The equipment operation feature vector, environmental impact factor feature vector, and basic feature vector are all derived from the results processed in the preceding steps. While the basic feature vector is relatively stable, it may fluctuate during long-term operation due to changes in the surface condition of the structure (such as slight changes in cleanliness), requiring inclusion in time-series analysis. This step selects consecutive time steps (the time step length is set to 0.5-1 seconds based on the response speed of the spraying operation, and the number of consecutive time steps is set to 10-20, covering 5-20 seconds of historical data). The three types of feature vectors corresponding to each time step are horizontally concatenated sequentially, that is, all dimensions of the equipment operation feature vector, all dimensions of the environmental impact factor feature vector, and all dimensions of the basic feature vector are sequentially linked to form a joint feature vector for a single time step. For example, if the equipment operation feature vector is 6-dimensional, the environmental impact factor feature vector is 7-dimensional, and the basic feature vector is 9-dimensional, then the joint feature vector for each time step has 6+7+9=22 dimensions. This stitching process achieves comprehensive integration of multi-dimensional features within a single time step, enabling the feature vector of each time step to fully reflect the comprehensive state of the equipment, environment, and structure at that moment, laying the foundation for subsequent time-series feature extraction. A spraying state matrix is ​​constructed by arranging the joint feature vectors of multiple time steps in chronological order. The joint feature vector of each time step is a quantitative representation of the overall state at that moment. Arranging the feature vectors of multiple consecutive time steps in chronological order forms a temporally sequenced state matrix. The number of rows in the matrix equals the number of consecutive time steps (e.g., 15), and the number of columns equals the dimension of the joint feature vectors (e.g., 22 dimensions), resulting in a 15×22 spraying state matrix. Each row of this matrix corresponds to the overall state of a time step, and each column corresponds to the temporal change of a feature, intuitively presenting the dynamic evolution of multiple factors over time during the spraying process. For example, a column in the matrix corresponding to the time-series data of standardized spraying pressure can clearly reflect the fluctuation trend of spraying pressure over the past 15 time steps. This temporally sequenced matrix form can be effectively recognized by deep learning models, providing structured data support for extracting long-term dependent features. The spraying state matrix is ​​input into a pre-trained spraying thickness prediction model. A temporal convolutional network (TCN) is used to extract time-series features, resulting in a high-dimensional feature representation. The training data for the spraying thickness prediction model comes from measured data of historical spraying operations, including continuous time-step feature vectors and corresponding measured thickness data collected during the spraying of the same or similar marine structures and the same type of coating. After processing as described above, training samples are constructed, and mean squared error is used as the loss function for model training. The structure of the temporal convolutional network (TCN) is adapted to the temporal characteristics of marine spraying. Its convolutional kernel size is set to 3-5 (corresponding to the feature associations of 3-5 consecutive time steps). Causal convolution is used to ensure that predictions rely only on historical data, avoiding the leakage of future information. The network has 2-3 convolutional layers, with batch normalization layers and ReLU activation functions added after each convolutional layer to improve the model's nonlinear fitting ability and training stability. Finally, dilated convolution is used to expand the receptive field, enabling the model to capture long-term dependencies within 5-20 seconds. For example, when the spraying pressure continuously increases over three consecutive time steps, while the ambient temperature also gradually rises, the temporal convolutional network, through kernel sliding calculation, can identify the cumulative effect of the combined changes of these two factors on the thickness and encode it as a specific dimension in the high-dimensional feature representation. This process realizes the transformation from raw temporal features to high-dimensional abstract features, effectively extracting key information about the temporal correlation of multiple factors. The dimension of the high-dimensional feature representation is usually set to 2-4 times the dimension of the joint feature vector (e.g., 22-dimensional joint features correspond to 44-88-dimensional high-dimensional features), ensuring that temporal correlation information is fully preserved. The high-dimensional feature representation is weighted and summed using an attention mechanism to obtain a context vector. The high-dimensional feature representation extracted by the temporal convolutional network contains correlation information between different time steps and features, but the importance of this information for thickness prediction varies. For example, changes in spraying pressure in recent time steps have a much greater impact on the current thickness than in earlier time steps, and temporal changes in certain features (such as dynamic viscosity parameters) also have a more significant impact on thickness. The attention mechanism calculates the similarity between each high-dimensional feature vector and the learnable query vector to obtain attention weights. The weight values ​​are between 0 and 1, and the sum of all weights is 1. The similarity calculation uses a dot product attention method, with the formula: Attention weight = softmax(high-dimensional feature representation × query vector^T / √d), where d is the dimension of the high-dimensional feature representation used to normalize the similarity value. Subsequently, the high-dimensional feature representation and the corresponding attention weight are weighted and summed to obtain the context vector, which has the same dimension as the high-dimensional feature representation (e.g., 64 dimensions). For example, if the high-dimensional features of the three most recent time steps have a high similarity to the query vector, the resulting attention weights are 0.3, 0.25, and 0.2, respectively, while the weights of earlier time steps are only 0.01-0.05. In this case, the weighted summed context vector will emphasize the key features of the most recent time steps and weaken irrelevant or less influential earlier features. This process achieves adaptive filtering of high-dimensional temporal features, enabling the model to focus on temporal information crucial for thickness prediction and improving prediction accuracy. The context vector is input into a fully connected regression layer, which outputs a predicted coating thickness sequence for multiple future time steps. The fully connected regression layer is designed with 2-3 fully connected layers. The number of neurons in the first fully connected layer is set to half the dimension of the context vector (e.g., 32 neurons for a 64-dimensional context vector), the second layer has 16 neurons, and the last layer has the same number of neurons as the number of future prediction time steps (3-5 based on control response requirements, i.e., predicting thickness data for 1.5-5 seconds in the future). A Dropout layer (with a dropout rate of 0.2) is added after each fully connected layer to prevent overfitting. The output layer uses a linear activation function and directly outputs the predicted thickness value (in μm). For example, if the future prediction time steps are set to 3, the predicted thickness values ​​for 3 consecutive time steps are output, forming a predicted coating thickness sequence [200μm, 205μm, 203μm]. This regression process maps the high-dimensional contextual features filtered by the attention mechanism into specific thickness prediction values, realizing the transformation from abstract features to actual physical quantities, and providing a clear prediction basis for subsequent comparison with real-time thickness sequences.

[0035] Step S2: Input the spraying state matrix into the trained spraying thickness prediction model to obtain the thickness prediction value; The spraying state matrix is ​​input into a pre-trained spraying thickness prediction model. The model extracts time-series features through a temporal convolutional network (TCNN), which has the ability to capture long-term dependencies and effectively extract the correlation between features at different time steps, resulting in a high-dimensional feature representation. Then, an attention mechanism is used to weight and sum the high-dimensional feature representation to obtain a context vector. The attention mechanism highlights features that are more critical to coating thickness, improving prediction accuracy. Finally, the context vector is input into a fully connected regression layer, outputting the predicted spraying thickness sequence for multiple future time steps. This step, through the combination of matrix modeling and deep learning models, achieves forward-looking prediction of spraying thickness, solving the problem that traditional methods cannot predict thickness changes in advance, and providing lead time for subsequent real-time control. For example, when the ambient temperature gradually increases, causing the coating viscosity to decrease, the TCNN can capture the temporal correlation between temperature changes and viscosity parameters and spraying pressure. The attention mechanism strengthens the weights of these key factors, enabling the prediction model to accurately output the coating thickness change trend for subsequent time steps. In particular, in some embodiments, in step S2, the thickness prediction value at time t is predicted by the spraying state matrix from time tk to time t, wherein the data from time tk to time t-1 in the spraying state matrix are actual measured values.

[0036] In the spraying state matrix, the ambient temperature, air humidity, wind speed and salt spray concentration at time t are predicted values, the spraying pressure and nozzle moving speed at time t are calculated using the target thickness, and the distance between the nozzle and the workpiece surface and the nozzle angle at time t are calculated based on spatial calculation.

[0037] Step S3: Collect the actual coating thickness at each moment in real time and compare it with the thickness prediction value at the corresponding moment to obtain the thickness prediction deviation. The system acquires a real-time coating thickness sequence within a preset time period and compares it with a predicted coating thickness sequence to obtain a thickness deviation sequence. The real-time coating thickness sequence is acquired through online thickness monitoring equipment deployed on the spraying equipment or work area, and arranged chronologically to form a continuous thickness data sequence. The real-time sequence is then compared point-by-point with the predicted coating thickness sequence obtained in step S5, i.e., the real-time thickness at each corresponding time point is subtracted from the predicted thickness, resulting in a thickness deviation sequence. This sequence directly reflects the deviation between the actual coating thickness and the predicted thickness, providing direct feedback for subsequent equipment control. For example, if the real-time thickness at a certain time point is 0.2 mm less than the predicted thickness, this deviation value will be included in the thickness deviation sequence, becoming an important reference for adjusting equipment parameters and ensuring that deviations are captured and processed in a timely manner. Step S4: Combine the thickness prediction deviations at multiple consecutive time points into a deviation sequence, and calculate the statistical characteristics of the deviation sequence as the deviation statistical characteristics based on the deviation sequence. Intelligent control of the spraying equipment based on the thickness deviation sequence is the core step in achieving precise thickness control. Technically, this involves first extracting multiple thickness deviation values ​​from the sequence and calculating the mean thickness deviation to reflect the overall deviation level; then, calculating the thickness variance based on the multiple thickness deviation values ​​and the mean thickness deviation, and combining this with the deviation values ​​themselves to obtain the statistical characteristics of the deviation, comprehensively characterizing the magnitude, fluctuation range, and other properties of the deviation.

[0038] Step S5: Utilize the deviation statistical characteristics to generate a feature vector, and obtain the adjustment amount of each control parameter of the PID controller based on the adjustment vector; By inputting the statistical characteristics of the deviation into the fuzzy inference rules, the control parameter adjustment of the PID controller is generated, including the adjustment of the proportional coefficient, integral coefficient and derivative coefficient. The fuzzy inference rules can adaptively generate a control parameter adjustment scheme for the PID controller according to different characteristics of the deviation, avoiding control lag or overshoot problems caused by fixed parameters. Specifically, step S5 includes the following: Step S5-1: Compare the mean deviation with each first mean interval, and take the first index of the first mean interval containing the mean deviation as the first feature; Step S5-2: Compare the deviation variance with each second mean interval, and take the second index of the second mean interval containing the deviation variance as the second feature; Step S5-3: Compare the maximum deviation value with each third mean interval, and take the third index of the third mean interval containing the maximum deviation value as the third feature; Step S5-4: Compare the rate of change of deviation with each fourth mean interval, and take the fourth index of the fourth mean interval containing the rate of change of deviation as the fourth feature; Step S5-5: Concatenate the first feature, second feature, third feature, and fourth feature to obtain the feature vector; Step S5-6: Obtain the adjustment amounts of each control parameter of the PID controller based on feature vector inference. The control parameters of the PID controller include proportional coefficient, integral coefficient, and derivative coefficient. In one embodiment, step S5-6 includes: Obtain the probability distribution of the adjustment amounts of each control parameter corresponding to the feature vector; The adjustment amount of each control parameter is obtained by sampling according to the probability distribution of the adjustment amount of each control parameter.

[0039] In other embodiments, steps S5-6 are obtained using fuzzy PID rule reasoning. The first mean interval consists of seven intervals, with each first index corresponding to negative large, negative medium, negative small, zero, positive small, positive medium, and positive large, respectively. There are three second mean intervals, with each second index corresponding to small, medium, and large, respectively. There are a total of 5 third mean intervals, with each third index corresponding to negative large, negative small, zero, positive small, and positive large, respectively. There are a total of 5 fourth mean intervals, with each fourth index corresponding to negative large, negative small, zero, positive small, and positive large, respectively; Fuzzy PID rules should include at least the following: Rule 1: If the first index corresponding to the mean deviation is positive and large, the second index corresponding to the variance deviation is large, the third index corresponding to the maximum deviation value is positive and large, and the fourth index corresponding to the rate of change of deviation is positive and small, then the adjustment amount of the proportional coefficient is positive and large, the adjustment amount of the integral coefficient is positive and small, and the adjustment amount of the derivative coefficient is moderate. Rule 2: If the first index corresponding to the mean deviation is negative and the second index corresponding to the variance deviation is small, the third index corresponding to the maximum deviation value is negative and the fourth index corresponding to the rate of change of deviation is negative and large, then the adjustment amount of the proportional coefficient is negative and small, the adjustment amount of the integral coefficient is negative and large, and the adjustment amount of the differential coefficient is zero. Steps S5-6 include: Based on the feature vector verification of each fuzzy PID rule, the possible ranges of the control vector adjustment amount and the number of hits of each possible range are obtained; By using the number of hits in each possible interval as a weight and combining it with the boundary values ​​of each possible interval, the precise value of the adjustment amount of each control vector is calculated.

[0040] This application acquires real-time monitored coating thickness data and arranges it chronologically to form a real-time coating thickness sequence. The real-time coating thickness data is acquired through online thickness monitoring equipment (such as ultrasonic thickness gauges or laser thickness gauges) deployed next to the spraying equipment. The monitoring frequency is consistent with the time step of the aforementioned prediction model (set to 0.5-1 seconds / time) to ensure data time sequence alignment. The continuously collected thickness data are arranged chronologically to form a real-time coating thickness sequence, which visually reflects the dynamic changes in coating thickness during the current spraying process. For example, 10 thickness data points were collected at a frequency of 1 second per acquisition within 10 seconds, namely 200μm, 202μm, 198μm, 205μm, 203μm, 197μm, 201μm, 204μm, 199μm, and 202μm. After being arranged in chronological order, they formed a real-time coating thickness sequence [200, 202, 198, 205, 203, 197, 201, 204, 199, 202], which provided a structured real-time data foundation for subsequent comparison with the predicted sequence. The thickness deviation sequence is obtained by performing point-by-point difference calculation between the real-time coating thickness sequence and the predicted coating thickness sequence. The predicted coating thickness sequence comes from the output of step S55 above, and its time step is consistent with that of the real-time coating thickness sequence. The point-by-point difference calculation involves subtracting the thickness values ​​at corresponding time steps in the two sequences (real-time thickness value - predicted thickness value) to obtain the thickness deviation value for each time step. All deviation values ​​are arranged in chronological order to form the thickness deviation sequence. A positive deviation value indicates that the real-time thickness is greater than the predicted thickness, and a negative deviation value indicates that the real-time thickness is less than the predicted thickness. The absolute value of the deviation value reflects the degree of deviation. For example, if the real-time thickness at a certain time step is 202 μm and the corresponding predicted thickness is 200 μm, then the deviation value for that time step is +2 μm; if the real-time thickness is 198 μm and the predicted thickness is 200 μm, then the deviation value is -2 μm. By calculating point by point, a thickness deviation sequence with the same length as the real-time sequence can be obtained, clearly showing the dynamic change trend of the thickness deviation. Multiple thickness deviation values ​​are obtained from the thickness deviation sequence, and the average thickness deviation is calculated from these values. The average thickness deviation is obtained by averaging all deviation values ​​in the deviation sequence, using the formula: Average thickness deviation = (deviation value 1 + deviation value 2 + ... + deviation value n) / n, where n is the length of the deviation sequence (i.e., the number of thickness data points collected). This average reflects the overall level of thickness deviation over a period of time. If the average is positive and the absolute value is large, it indicates that the overall coating thickness is too thick; if the average is negative and the absolute value is large, it indicates that the overall coating thickness is too thin. For example, if a thickness deviation sequence is [+2,+4,-2,+5,+3,-3,+1,+4,-1,+2], then the average thickness deviation = (2+4-2+5+3-3+1+4-1+2) / 10 = 15 / 10 = +1.5μm, indicating that the overall coating thickness during this period is 1.5μm thicker than the predicted value, providing an overall basis for determining the subsequent control direction. The thickness variance is obtained based on multiple thickness deviation values ​​and the mean thickness deviation, and the deviation statistical characteristics are obtained based on these values ​​and the thickness variance. The thickness variance is calculated by averaging the sum of the squared differences between each deviation value and the mean deviation, using the formula: Thickness Variance = Σ(Deviation value i - Mean deviation). 2The parameter / n reflects the degree of fluctuation in thickness deviation. A larger variance indicates greater instability, requiring a stronger control response to suppress fluctuations. The deviation statistical characteristics integrate the mean thickness deviation, the thickness variance, and key information such as the maximum deviation value and the rate of change of the deviation sequence, forming a statistical feature vector with four dimensions: [mean deviation, variance deviation, maximum deviation value, rate of change of deviation]. This comprehensively characterizes the overall level, fluctuation range, extreme cases, and trends of the deviation. For example, based on the deviation sequence calculated above, if the mean deviation is 1.5 μm, the variance deviation is 5.25, the maximum deviation value is +5 μm, and the rate of change of deviation is 0.3 μm / step, then the deviation statistical characteristics are [1.5, 5.25, 5, 0.3], providing a comprehensive basis for subsequent PID parameter adjustments. The deviation statistical characteristics are used to generate the parameter adjustments for the PID controller through fuzzy inference rules, including proportional coefficient adjustment, integral coefficient adjustment, and derivative coefficient adjustment. The initial parameters of the PID controller are preset based on experience in marine spraying processes: the initial value of the proportional coefficient Kp is set to 0.8-1.2, the initial value of the integral coefficient Ki is set to 0.1-0.3, and the initial value of the derivative coefficient Kd is set to 0.2-0.5. The fuzzy inference rules are designed based on the deviation characteristics and control requirements of marine spraying. The mean deviation is divided into seven fuzzy subsets: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large; the variance deviation is divided into three fuzzy subsets: small, medium, and large; the maximum deviation value is divided into five fuzzy subsets: negative large, negative small, zero, positive small, and positive large; and the rate of change of deviation is divided into five fuzzy subsets: negative large, negative small, zero, positive small, and positive large. Each combination of input subsets corresponds to a specific PID parameter adjustment output. Fuzzy inference employs the Mamdani algorithm, which uses fuzzification, rule matching, and fuzzy decision-making (centroid method) to map deviation statistical characteristics to parameter adjustment amounts. For example, when the deviation statistical characteristics are [1.5, 5.25, 5, 0.3] (large mean deviation, large variance, large maximum deviation, and small rate of change), fuzzy inference rule matching generates parameter adjustment amounts of +0.2 for the proportional coefficient, +0.1 for the integral coefficient, and +0.15 for the derivative coefficient, achieving adaptive adjustment of PID parameters and making the controller more adaptable to the current deviation state. Step S6: Correct the spraying pressure and nozzle movement speed based on the adjustment amount and real-time deviation of each control parameter.

[0041] Finally, based on the adjustment of the control parameters of these PID controllers and the real-time thickness deviation, spraying pressure adjustment commands and nozzle movement speed adjustment commands are generated. Since spraying pressure directly affects the amount of paint sprayed, and nozzle movement speed affects the amount of paint adhered per unit area, their coordinated adjustment can quickly correct thickness deviations. For example, when the mean thickness deviation is positive and the variance is small, it indicates that the overall spraying thickness is too thick and the fluctuation is small. Fuzzy inference rules will generate commands to reduce the proportional coefficient and appropriately adjust the integral coefficient, thereby generating control commands to reduce the spraying pressure or increase the nozzle movement speed, making the subsequent spraying thickness approach the preset target value. This ensures the stability of the control effect. Furthermore, the construction of multi-dimensional feature vectors comprehensively covers the key factors affecting spraying thickness, avoiding control deviations caused by single-factor analysis. Simultaneously, the synergistic effect of various technical features jointly achieves precise, real-time, and intelligent control of the anti-corrosion spraying thickness of marine engineering structures.

[0042] The system generates spray pressure adjustment commands and nozzle movement speed adjustment commands based on proportional coefficient adjustment, integral coefficient adjustment, derivative coefficient adjustment, and real-time thickness deviation, and performs intelligent control of the spraying equipment. First, the initial PID parameters are superimposed with the adjustment values ​​to obtain the currently adapted PID parameters (adjusted Kp = initial Kp + proportional coefficient adjustment, adjusted Ki = initial Ki + integral coefficient adjustment, adjusted Kd = initial Kd + derivative coefficient adjustment). Then, the real-time thickness deviation is input into the adjusted PID controller, and the adjustment values ​​for spray pressure and nozzle movement speed are calculated through the PID control algorithm. The core formula of the PID control algorithm is: Δu(t) = Kp × e(t) + Ki × ∫e(t) dt + Kd × de(t) / dt, where Δu(t) is the control output adjustment value, e(t) is the real-time thickness deviation, ∫e(t) dt is the deviation integral, and de(t) / dt is the deviation derivative. Based on the parameter characteristics of the marine coating equipment, the adjustment step size for the spraying pressure is set to 0.02-0.05 MPa / time, and the adjustment step size for the nozzle movement speed is set to 0.05-0.1 m / s / time. The adjustment amount output by the PID is mapped to an adjustment command that conforms to the operating range of the equipment. For example, when the real-time thickness deviation is +2 μm, after adjustment Kp=1.0, Ki=0.2, Kd=0.35, the PID algorithm calculates an adjustment amount of -0.04 MPa for the spraying pressure and +0.08 m / s for the nozzle movement speed. This generates a control command to reduce the spraying pressure by 0.04 MPa and increase the nozzle movement speed by 0.08 m / s. The command is sent to the actuator of the coating equipment through the communication module to realize real-time parameter adjustment, so that the coating thickness approaches the preset target value.

[0043] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction, characterized in that, include: Step S1: Obtain the time sequence of each spraying state parameter and construct the spraying state matrix. The spraying state parameters include operating parameters, environmental parameters and basic parameters. The operating parameters include at least spraying pressure and nozzle moving speed. Step S2: Input the spraying state matrix into the trained spraying thickness prediction model to obtain the thickness prediction value; Step S3: Collect the actual coating thickness at each moment in real time and compare it with the thickness prediction value at the corresponding moment to obtain the thickness prediction deviation. Step S4: Combine the thickness prediction deviations at multiple consecutive time points into a deviation sequence, and calculate the statistical characteristics of the deviation sequence as the deviation statistical characteristics based on the deviation sequence. Step S5: Utilize the deviation statistical characteristics to generate a feature vector, and obtain the adjustment amount of each control parameter of the PID controller based on the adjustment vector; Step S6: Correct the spraying pressure and nozzle movement speed based on the adjustment amount and real-time deviation of each control parameter.

2. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction as described in claim 1, characterized in that, The operating parameters also include the distance between the nozzle and the workpiece surface, the paint flow rate, and the nozzle angle. The environmental parameters include ambient temperature, air humidity, wind speed, and salt spray concentration. The basic parameters include surface roughness, existing coating thickness, and material type.

3. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 1, characterized in that, The deviation statistical characteristics include the deviation mean, deviation variance, maximum deviation value, and deviation change rate.

4. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 3, characterized in that, Step S5 includes: Step S5-1: Compare the mean deviation with each first mean interval, and take the first index of the first mean interval containing the mean deviation as the first feature; Step S5-2: Compare the deviation variance with each second mean interval, and take the second index of the second mean interval containing the deviation variance as the second feature; Step S5-3: Compare the maximum deviation value with each third mean interval, and take the third index of the third mean interval containing the maximum deviation value as the third feature; Step S5-4: Compare the rate of change of deviation with each fourth mean interval, and take the fourth index of the fourth mean interval containing the rate of change of deviation as the fourth feature; Step S5-5: Concatenate the first feature, second feature, third feature, and fourth feature to obtain the feature vector; Step S5-6: Based on feature vector reasoning, obtain the adjustment amount of each control parameter of the PID controller, wherein the control parameters of the PID controller include proportional coefficient, integral coefficient and derivative coefficient.

5. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 4, characterized in that, Steps S5-6 include: Obtain the probability distribution of the adjustment amounts of each control parameter corresponding to the feature vector; The adjustment amount of each control parameter is obtained by sampling according to the probability distribution of the adjustment amount of each control parameter.

6. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 4, characterized in that, Steps S5-6 are obtained using fuzzy PID rule reasoning. The first mean interval consists of seven intervals, with each first index corresponding to negative large, negative medium, negative small, zero, positive small, positive medium, and positive large, respectively. There are three second mean intervals, with each second index corresponding to small, medium, and large, respectively. There are a total of 5 third mean intervals, with each third index corresponding to negative large, negative small, zero, positive small, and positive large, respectively. There are a total of 5 fourth mean intervals, with each fourth index corresponding to negative large, negative small, zero, positive small, and positive large, respectively; Fuzzy PID rules should include at least the following: Rule 1: If the first index corresponding to the mean deviation is positive and large, the second index corresponding to the variance deviation is large, the third index corresponding to the maximum deviation value is positive and large, and the fourth index corresponding to the rate of change of deviation is positive and small, then the adjustment amount of the proportional coefficient is positive and large, the adjustment amount of the integral coefficient is positive and small, and the adjustment amount of the derivative coefficient is moderate. Rule 2: If the first index corresponding to the mean deviation is negative and the second index corresponding to the variance deviation is small, the third index corresponding to the maximum deviation value is negative and the fourth index corresponding to the rate of change of deviation is negative and large, then the adjustment amount of the proportional coefficient is negative and small, the adjustment amount of the integral coefficient is negative and large, and the adjustment amount of the differential coefficient is zero. Steps S5-6 include: Based on the feature vector verification of each fuzzy PID rule, the possible ranges of the control vector adjustment amount and the number of hits of each possible range are obtained; By using the number of hits in each possible interval as a weight and combining it with the boundary values ​​of each possible interval, the precise value of the adjustment amount of each control vector is calculated.

7. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 2, characterized in that, In step S2, the thickness prediction value at time t is predicted by the spraying state matrix from time tk to time t, wherein the data from time tk to time t-1 in the spraying state matrix are actual measured values.

8. The method for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction according to claim 7, characterized in that, In the spraying state matrix, the ambient temperature, air humidity, wind speed, and salt spray concentration at time t are predicted values, the spraying pressure and nozzle moving speed at time t are calculated using the target thickness, and the distance between the nozzle and the workpiece surface and the nozzle angle at time t are calculated based on spatial calculation.

9. A device for controlling the thickness of anti-corrosion spraying on marine structures based on real-time prediction, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-8.

10. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-8.

Images