Ship segmented coating quality evaluation and rework decision method and system

Abstract

The invention discloses a ship segment coating quality evaluation and rework decision-making method and system. The method comprises a ship segment coating quality evaluation method and a ship segment coating rework decision-making method. Key features are extracted from collected data of a sensor in the spraying process, feature vectors are formed, then the collected data and the feature vectors are combined to serve as input, and defect categories in sprayed segments are output through a random forest algorithm; according to the proportion area and the film thickness deviation of the defect areas, a dynamic priority ranking method is adopted to calculate and obtain the priority processing sequence of reworking and repairing of each defect area; in the supplementary spraying process, according to historical data and coating strategy data fed back in real time, a Bayesian optimization algorithm is adopted, and spraying parameters are automatically adjusted; the historical data comprises spraying parameters and coating quality data corresponding to different spraying parameters. According to the invention, real-time monitoring of the coating quality, automatic defect identification and intelligent optimization of the reworking sequence can be realized.

Description

Technical Field

[0001] The present invention relates to a ship section painting method, and in particular to a ship section painting quality assessment and rework decision-making method and system. Background Art

[0002] As a critical manufacturing process, ship section coating plays a vital role in multiple industries, including aerospace, automotive, marine, and electronics. The quality of the coating directly impacts the performance, durability, and appearance of the product, making optimization and quality control of the coating process crucial.

[0003] Currently, most spray trajectory planning relies on simulation software for design and optimization. However, while traditional collision and reachability simulations can detect potential problems, they fail to effectively consider the impact of coating thickness. Although a few trajectory planning methods based on film thickness simulation have improved spray quality to a certain extent, they have problems with boundary condition setting and low reliability in actual applications. In addition, using fluid simulation software such as Fluent to simulate the coating thickness of the entire segment has a long calculation cycle and high cost, which limits its breadth and efficiency of application in actual production. Summary of the Invention

[0004] Purpose of the invention: In response to the above problems, the present invention proposes a method and system for ship section painting quality assessment and rework decision-making, which can realize real-time monitoring of coating quality, automatic identification of defects and intelligent optimization of rework sequence, and efficiently and reliably perform quality monitoring and rework spraying of ship section painting.

[0005] Technical solution: The technical solution adopted by the present invention is a method for evaluating the quality of ship segmented painting and making decisions on rework, including: a method for evaluating the quality of ship segmented painting and a method for making decisions on ship segmented painting rework; the method for evaluating the quality of ship segmented painting includes: extracting key features from the data collected by the sensor during the spraying process to form a feature vector, then merging the collected data and the feature vector as input, and using a random forest algorithm to output the defect category in the segment that has been sprayed; the method for making decisions on ship segmented painting rework is based on the defect category output by the method for evaluating the quality of ship segmented painting, as well as the area ratio and film thickness deviation of the defect area, and using a dynamic priority sorting method to calculate the priority processing order of rework and repair of each defect area; in the process of re-spraying, the Bayesian optimization algorithm is used to automatically adjust the spraying parameters based on the historical data and the real-time feedback strategy data. The historical data includes the spraying parameters and coating quality data collected by the sensor during the first spraying process, and the real-time feedback strategy data includes the spraying parameters and coating quality data collected during the re-spraying process.

[0006] The data collected by the sensor include: spray gun nozzle position, spray gun flow, ambient temperature, ambient humidity, vibration acceleration and real-time spraying time; the key features include: spray film thickness, regularity of ship segment surface and spray pattern coverage; the defect categories include overspray, paint missing, orange peel and sagging.

[0007] According to the characteristics of the regularity of the ship segment surface and the coverage of the spray pattern, the categorical features of "regular / irregular" and "covered / uncovered" are converted into numerical values.

[0008] The spraying parameters include spray gun height, moving speed and spray gun flow rate, and the coating quality data adopts the mean square error between the actual film thickness and the expected film thickness.

[0009] Random forest uses CART decision tree as the base learner, and the number of decision trees is set to be no less than 100.

[0010] The calculation formula of the dynamic priority sorting method is:

[0011] S=α·MSE+β·R

[0012]

[0013] Where S is the comprehensive priority score of the defect area, α and β are weight coefficients set according to actual production experience, MSE is the film thickness deviation, which uses the mean square error between the actual film thickness and the expected film thickness, R is the area of the defect area, and A is the average value of the defect area. defect is the area of the defect region, A total is the area of the total monitoring area.

[0014] The dynamic priority sorting method is used to calculate the priority processing order of rework and repair of each defective area, including: sorting all defective areas from high to low according to the comprehensive priority score as the priority processing order of rework and repair of each defective area; when a defective area is reworked, the film thickness deviation of the defective area is recalculated to determine whether it has reached the expected standard. If not, the comprehensive priority score is recalculated according to the film thickness deviation data to update the priority sorting.

[0015] The Bayesian optimization algorithm includes the following steps:

[0016] Step 1, assume the spraying parameter vector is:

[0017] x=[h,v,q]

[0018] Where x is the spraying parameter vector, h is the spray gun height; v is the moving speed; q is the spray gun flow rate;

[0019] Assume that the coating quality data is the film thickness deviation, and define the objective function as:

[0020] f(x)=-MSE(x)

[0021] Where f(x) is the objective function, MSE(x) is the mean square error between the actual film thickness and the expected film thickness under parameter x;

[0022] The coating quality data under different spraying parameters in the historical data are used to form the initial sample set:

[0023]

[0024] D0 is the initial sample set, x i is the i-th spraying parameter vector, n0 represents the number of samples contained in the initial sample set;

[0025] Using the current sample data set D t Construct the agent model and Gaussian process model to model the objective function f(x):

[0026] f(x)~GP(μ(x),k(x,x′))

[0027] The above formula indicates that f(x) statistically obeys the Gaussian process (GP) model, where μ(x) is the mean function, which is estimated by the sample mean; k(x, x′) is the covariance kernel function, which describes the correlation between any two input points x and x′, that is, the covariance between the function values f(x) and f(x′);

[0028] Step 2: The next parameter point is obtained by solving the following equation in the parameter space:

[0029]

[0030] Among them, X next represents the next parameter point, EI(x) represents the acquisition function, and argmax represents the search for the parameter combination that maximizes the acquisition function in the parameter space;

[0031] Step 3, take the next parameter point X next Perform re-spraying, obtain the coating thickness mean square deviation measurement data in real time, and calculate the actual objective function value f(X next ); the newly obtained sample (X next ,f(X next ))Add sample dataset and update proxy model;

[0032] Step 4: Repeat the iterative process. If the objective function value does not improve after several consecutive iterations, stop the iteration and output the parameter set X that obtains the best objective function value among all samples. * .

[0033] The acquisition function of the Bayesian optimization algorithm is:

[0034]

[0035] σ(x)>0, otherwise EI is 0;

[0036] Where, is the optimal value of the objective function in the current sample; μ(x) and σ(x) are the predicted mean and standard deviation of GP at point x, respectively; Φ(Z) and φ(Z) are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0037] A ship section painting quality assessment and rework decision-making system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the ship section painting quality assessment and rework decision-making method is implemented.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0039] 1. Accurate defect identification and classification capabilities. The ship section coating quality assessment and rework decision-making system utilizes advanced algorithms such as random forests to accurately identify various defects that may occur during the coating process, such as overspray, underspray, bubbles, and sags. This high-precision defect classification capability ensures targeted treatment of different defect types during rework, thereby improving the consistency and reliability of coating quality.

[0040] 2. Dynamic Prioritization. The system uses dynamic prioritization to intelligently generate a rework priority matrix based on defect severity and impact. This matrix effectively guides multi-objective path optimization, planning the optimal re-spraying path with the goal of minimizing paint consumption and rework time. As a result, the system not only improves rework efficiency but also avoids duplicate coating and wasteful resources, saving costs.

[0041] 3. Real-time Bayesian optimization for spray parameter adjustment. Using Bayesian optimization methods combined with historical data and real-time feedback, the system dynamically adjusts spray parameters such as gun height, travel speed, and flow rate to optimize film thickness uniformity and coating quality. This real-time adjustment ensures high-quality coating results even under changing process conditions, improving production efficiency and product quality stability.

[0042] 4. Full-process, closed-loop intelligent control. The system implements automated control of the entire process, from quality assessment and defect identification to rework decisions. By integrating an intelligent decision-making engine, it continuously collects and analyzes data to optimize production processes and automate decision-making, reducing the need for manual intervention and improving overall production efficiency and quality management.

[0043] 5. Highly adaptable to different ship types and painting scenarios. The system is designed with high flexibility and adaptability, enabling rapid adaptation and optimization for different ship types and complex painting scenarios. Whether it's adapting to different paint types, hull structures, or changes in process standards, the system can provide customized solutions, ensuring consistent and stable painting quality, further promoting the intelligent development of the shipbuilding industry.

[0044] This invention combines digital twin technology with advanced machine learning algorithms to enable real-time monitoring of coating quality, automatic defect identification, and intelligent optimization of rework paths. By leveraging multi-source data, optimization algorithms, and real-time control strategies, the system will significantly improve the efficiency and quality stability of the coating process, bringing significant technological advancements and economic benefits to the shipbuilding industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the present invention for training and using random forest to evaluate the quality of ship coating in sections;

[0046] Figure 2 is a flow chart of the dynamic priority sorting according to the present invention;

[0047] Figure 3 This is the Bayesian optimization flow chart of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] The ship segment painting quality assessment and rework decision-making method described in the present invention includes a ship segment painting quality assessment method and a ship segment painting rework decision-making method. The ship segment painting quality assessment method is to use a random forest algorithm to classify the defects in the segments that have been sprayed. The ship segment painting rework decision-making method is to dynamically adjust the priority processing order of each defect area according to the defect category using a dynamic priority sorting method; at the same time, the existing multi-objective path optimization algorithm is used to plan the optimal re-spraying path. During the re-spraying process, the Bayesian optimization method is used to combine historical data and real-time feedback to automatically adjust the spraying parameters.

[0050] like Figure 1 The following is a flowchart for training and utilizing random forest models to assess the quality of ship section painting. This system determines the quality of the painting in real time, identifying and classifying paint defects to facilitate rework decisions. Defect classification is primarily performed using a random forest model. The following is the main implementation process.

[0051] The following parameters are used as the original input sample data based on the data collected by the sensor:

[0052] Spray gun nozzle position (continuous value, including the distance d between the spray gun and the workpiece surface, and the position x, y, and z of the spray gun in space, in cm);

[0053] Spray gun flow rate (continuous value, unit such as mL / min, indicating the paint spraying rate);

[0054] Ambient temperature (continuous value, °C);

[0055] Ambient humidity (continuous value, %);

[0056] Vibration acceleration (continuous value, unit m / s 2 , indicating the vibration intensity of the spraying equipment);

[0057] Real-time spraying time (continuous value, unit s).

[0058] In feature engineering, three key features are derived based on the eight sample data items mentioned above:

[0059] 1. Spray Film Thickness: A continuous numerical characteristic representing the thickness of the coating formed after spraying (e.g., in microns). This characteristic is determined by the spraying process and sensor measurements, and is determined by the distance between the spray gun and the workpiece surface, the spray gun flow rate, the ambient temperature, humidity, and vibration acceleration parameters. Higher thickness may indicate a risk of overspray.

[0060] 2. Surface regularity of ship sections: This classification characteristic ("regular" / "irregular") indicates whether the surface shape of the hull section being sprayed is regular and smooth. This is determined by spray gun flow rate, ambient temperature, humidity, and vibration acceleration parameters. Surface irregularities often lead to uneven coating thickness or difficulty in coverage during spraying.

[0061] 3. Spray pattern coverage: This binary feature indicates whether the area has been previously sprayed. This feature is determined by the spray gun's spatial position, vibration acceleration, and the actual spraying time. This feature reflects the coverage of the area; for example, spraying an already coated area may result in overspray.

[0062] By integrating all the above features, each sample forms a feature vector. The target variable is the defect type, which has four categories: overspray, paint missing, orange peel, and sagging. The judgment rules for the above four defects are:

[0063] Overspray: The workpiece surface is regular, the spray pattern has been covered, and the film thickness is uniform and thicker than other sprayed areas;

[0064] Paint missing: The workpiece surface is regular, the spray pattern has been covered, and the film thickness is uniform and thinner than other sprayed areas;

[0065] Orange peel: The spray pattern has been covered, the workpiece surface is irregular, and the film thickness in the irregular area is uneven;

[0066] Sagging: The spray pattern has been covered, the workpiece surface is irregular, and the film thickness is thick in the irregular areas.

[0067] The judgment rules can be adjusted as needed.

[0068] Before building the random forest model, the data is preprocessed to ensure the correct operation of the calculation. For the existing data set and possible problems, the data is processed as follows:

[0069] 1. Data cleaning: Check and handle missing values or outliers. For numerical features, fill missing values with the mean or discard missing samples. For categorical features, set missing values as a special category.

[0070] 2. Category feature coding: Convert category features such as "regular / irregular" and "covered / uncovered" into numerical values, with coverage of the spray pattern being 1 and uncovered being 0; regular surface of the workpiece is assigned a value of 1 and irregular surface of the workpiece is assigned a value of 0.

[0071] 3. Feature combination: Combine the original features and derived feature columns into the model input matrix X, and the label column is y.

[0072] 4. Data Partitioning: Before modeling, the dataset was divided into training and test sets in an 8:2 ratio. The holdout method was used to evaluate the generalization ability of the model, which sets aside a portion of the data for testing to evaluate the model's performance on unseen data.

[0073] Random forests use CART decision trees as their base learner. Random forests are an ensemble model composed of multiple decision trees. By training each tree with a different data and feature subset and then voting on the results, the model's stability and generalization performance can be improved.

[0074] The Gini coefficient is used as the partitioning metric (by default) to select the optimal splitting features and thresholds. The Gini coefficient is defined as the impurity measure of the sample class in the node, and the calculation formula is:

[0075]

[0076] Where k is the number of defect categories, p i The Gini value ranges from 0 to 0.5, with smaller values indicating a purer node (0 means all node samples belong to the same class).

[0077] The following are the parameter settings for random forest training to classify defects.

[0078] Number of decision trees: In this model, the number of decision trees is set to 100. This achieves higher accuracy, more stable model, and lower generalization error.

[0079] Maximum depth of each tree: The maximum depth of each tree is set to 6. This is to limit the depth of tree growth, prevent overfitting due to too few leaf node samples, and ensure that the degradation of generalization performance caused by the noise of the training data is minimized.

[0080] Minimum number of node samples:

[0081] The minimum number of samples required to split an internal node is set to 5. As long as there are >= 5 samples, the split is attempted to avoid over-division of very few samples.

[0082] The minimum number of samples required for a leaf node is set to 2, which means that a leaf node is allowed to contain at least 2 samples, making the model more robust and preventing the model from overfitting individual samples.

[0083] Number of features considered per split: One of the random characteristics of random forests is that only a subset of all features is considered when selecting split features at each node. Setting the number of features considered per split to 2 means that two candidate features are randomly selected for each split, and the optimal feature split is chosen from these two. This random feature selection mechanism ensures diversity between trees, reduces correlation, and improves the overall generalization ability of the forest.

[0084] The random forest algorithm is used to train its ability to identify defects, and the new data from the spraying process is used to determine the defect type in real time, and then perform real-time rework on the defects.

[0085] The decision-making method for ship section painting rework adopts dynamic priority sorting to design the rework sequence. The flowchart of dynamic priority sorting is as follows: Figure 2 To improve rework efficiency and coating quality, a dynamic priority ranking method is used to determine the order of rework and repair spraying. After the rework order is determined, the system uses Bayesian optimization to optimize rework decisions, namely, to optimize variables such as spray gun height, movement speed, and flow rate, to achieve the goal of improving coating quality and efficiency.

[0086] The data of the defective area are organized into records, which include: area ID, measurement point set, defect category, defective area, total monitoring area, and expected film thickness.

[0087] Based on the severity and impact of the defects, the priority of rework for each defective area is dynamically adjusted, with severe defects being reworked first to ensure that high-priority defects are handled first. The severity quantification indicator is the mean square error between the film thickness of the defective area and the expected film thickness, and the formula is as follows:

[0088]

[0089] Where n is the number of sampling points in the defect area, T i is the actual film thickness at the i-th sampling point, T expected is the expected film thickness in the region. A larger MSE indicates a more severe film thickness deviation in the region, thus reflecting a higher defect severity.

[0090] The quantitative indicator of the impact range is the proportion of the defect area, which is the ratio of the area of the area to the total area of the entire segment or monitoring area. The formula is:

[0091]

[0092] Among them, A defect is the area of the defect region, A total is the area of the total monitoring area. The larger the proportion R is, the wider the area affected by the defect is, and the greater the impact on the overall coating quality is.

[0093] The severity of the defect and the scope of impact are weighted together to obtain a comprehensive priority score for each defect area. The comprehensive priority score formula is:

[0094] S=α·MSE+β·R

[0095] Here, α and β are weighting coefficients, set based on actual production experience, representing the respective influences of defect severity and impact area in the comprehensive priority score. This formula influences both film thickness deviation and area ratio. A higher score, S, indicates a higher priority for rework in the defective area.

[0096] Sort all defective areas from high to low according to the comprehensive score S and generate a priority matrix. Each matrix item contains the area ID, defect category, MSE value, area ratio R, comprehensive score S and other related auxiliary information. For example:

[0097]

[0098] During the spraying process, the system continuously collects new film thickness data and updates the MSE and R-value of each defective area in real time. When a certain area undergoes rework, its film thickness data changes, and the system recalculates the MSE of that area to determine whether it has met the expected standard. Based on the new data, the comprehensive score S is recalculated, and the priority ranking is updated to form a new rework priority matrix. Based on historical data and real-time feedback, the system dynamically adjusts the weighting coefficients α and β to adapt to different coating processes in different batches or environments.

[0099] Bayesian optimization is a global optimization method whose core concept is to approximate the objective function using a surrogate model (Gaussian process). The method then designs an acquisition function to strike a balance between exploration (exploring poorly understood areas) and exploitation (refined search within known good areas) to find the optimal parameter combination. For the segmented re-spraying process for ship coating, the goal is to automatically adjust the spraying parameters to improve the film thickness qualification rate.

[0100] like Figure 3 The figure shows the Bayesian optimization flow chart of the present invention. The coating quality is mainly related to the spray gun state. The distance between the spray gun and the bus surface, the spray gun movement speed, and the spray gun flow rate are selected as the parameters for optimizing the coating quality. Based on this, the parameter vector is set as:

[0101] x=[h,v,q]

[0102] Where h: spray gun height; v: moving speed; q: spray gun flow rate.

[0103] The objective function f(x) is defined to represent the coating quality evaluation index after re-spraying, i.e., the film thickness uniformity. The goal is to maximize this index. The objective function is defined as:

[0104] f(x)=-MSE(x)

[0105] Where MSE(x) is the mean square error between the actual film thickness and the expected film thickness under parameter x. The goal is to minimize MSE, thereby maximizing f(x).

[0106] The coating quality data under different spraying parameters in the historical records are used to form the initial sample set.

[0107]

[0108] n0 = 10 to 20 points were selected from the historical data to ensure that the initial exploration of the parameter space was covered. During the re-spraying process, new film thickness measurement data and coating quality assessment results were collected in real time for subsequent updates.

[0109] Using the current sample data D t Construct the agent model and Gaussian process model to model the objective function f(x):

[0110] f(x)~GP(μ(x),k(x,x′))

[0111] Where μ(x) is the mean function, estimated from the sample mean, and k(x, x′) is the covariance kernel function (Radial Basis Function (RBF)). This surrogate model can predict the mean μ(x) and standard deviation σ(x) at any unsampled point, reflecting the model's uncertainty about the target function.

[0112] The expected improvement EI is the acquisition function, and its formula is:

[0113]

[0114] in: is the optimal value of the objective function in the current sample;

[0115] μ(x) and σ(x) are the predicted mean and standard deviation of GP at point x, respectively;

[0116] Φ(Z) and φ(Z) are the cumulative distribution function and probability density function of the standard normal distribution, respectively;

[0117] (σ(x)>0, otherwise EI is 0).

[0118] EI measures the expected value of an improvement at point x, taking into account both high prediction means and areas of high uncertainty.

[0119] Solving the next parameter point is done by solving the following equation in the parameter space:

[0120]

[0121] The particle swarm algorithm is used to find the parameter combination that maximizes EI. This point represents the parameter setting that is most likely to improve the objective function under this model.

[0122] With the selected parameter X next Perform re-spraying, obtain the coating thickness mean square deviation measurement data in real time, and calculate the actual objective function value f(X next ).

[0123] The newly obtained sample (X next ,f(X next ))Add to dataset:

[0124] D t+1 =D t ∪{(X next ,f(X next ))}

[0125] And update the proxy model (Gaussian process model) with the new data.

[0126] Repeat this iterative process. If the objective function value does not improve significantly after several consecutive iterations, stop the iteration. Finally, return the parameter set X that obtains the best objective function value among all samples. * .

Claims

1. A method for ship section painting quality assessment and rework decision-making, characterized by: The invention comprises a method for evaluating the quality of ship section painting and a method for making decisions on the rework of ship section painting; the method for evaluating the quality of ship section painting comprises: extracting key features from the data collected by the sensor during the spraying process to form a feature vector, then merging the collected data and the feature vector as input, and using a random forest algorithm to output the defect category in the sprayed section; the method for making decisions on the rework of ship section painting adopts a dynamic priority sorting method to calculate the priority processing order of rework and repair of each defect area based on the proportion of the defect area and the film thickness deviation; in the re-spraying process, the Bayesian optimization algorithm is adopted to automatically adjust the spraying parameters according to the historical data and the real-time feedback strategy data; the historical data includes the spraying parameters and coating quality data collected by the sensor during the first spraying process, and the real-time feedback strategy data includes the spraying parameters and coating quality data collected during the re-spraying process.

2. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: The data collected by the sensor include: spray gun nozzle position, spray gun flow, ambient temperature, ambient humidity, vibration acceleration and real-time spraying time; the key features include: spray film thickness, regularity of ship segment surface and spray pattern coverage; the defect categories include overspray, paint missing, orange peel and sagging.

3. The ship section painting quality assessment and rework decision-making method according to claim 2 is characterized by: For the characteristics of the regularity of the ship section surface and the coverage of the spray pattern, the categorical features of "regular / irregular" and "covered / uncovered" are converted into numerical values.

4. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: The spraying parameters include spray gun height, moving speed and spray gun flow rate, and the coating quality data adopts the mean square error between the actual film thickness and the expected film thickness.

5. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: Random forest uses CART decision tree as the base learner, and the number of decision trees is set to be no less than 100.

6. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: The calculation formula of the dynamic priority sorting method is: S=α·MSE+β·R Where S is the comprehensive priority score of the defect area, α and β are weight coefficients set according to actual production experience, MSE is the film thickness deviation, which uses the mean square error between the actual film thickness and the expected film thickness, R is the area of the defect area, and A is the average value of the defect area. defect is the area of the defect region, A total is the area of the total monitoring area.

7. The ship section painting quality assessment and rework decision-making method according to claim 6 is characterized by: The dynamic priority sorting method is used to calculate the priority processing order of rework and repair of each defective area, including: sorting all defective areas from high to low according to the comprehensive priority score as the priority processing order of rework and repair of each defective area; when a defective area is reworked, the film thickness deviation of the defective area is recalculated to determine whether it has reached the expected standard. If not, the comprehensive priority score is recalculated according to the film thickness deviation data to update the priority sorting.

8. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: Based on historical data and real-time feedback strategy data, the Bayesian optimization algorithm is used to automatically adjust the spraying parameters, including the following steps: Step 1, assume the spraying parameter vector is: x=[h,v,q] Where x is the spraying parameter vector, h is the spray gun height; v is the moving speed; q is the spray gun flow rate; Assume that the coating quality data is the film thickness deviation, and define the objective function as: f(x)=-MSE(x) Where f(x) is the objective function, MSE(x) is the mean square error between the actual film thickness and the expected film thickness under parameter x; The coating quality data under different spraying parameters in the historical data are used to form the initial sample set: D0 is the initial sample set, x i is the i-th spraying parameter vector, n0 represents the number of samples contained in the initial sample set; Using the current sample data set D t Construct the agent model and Gaussian process model to model the objective function f(x): f(x)~GP(μ(x),k(x,x′)) The above formula indicates that f(x) statistically obeys the Gaussian process GP model, where μ(x) is the mean function, which is estimated by the sample mean; k(x, x′) is the covariance kernel function, which describes the correlation between any two input points x and x′, that is, the covariance between the function values f(x) and f(x′); Step 2: The next parameter point is obtained by solving the following equation in the parameter space: Among them, X next represents the next parameter point, EI(x) represents the acquisition function, and argmax represents the search for the parameter combination that maximizes the acquisition function in the parameter space; Step 3, take the next parameter point X next Perform supplementary spraying, obtain the spray gun height, moving speed, spray gun flow rate and coating film thickness mean square deviation measurement data in real time, and calculate the actual objective function value f(X next ); the newly obtained sample (X next ,f(X next ))Add sample dataset and update proxy model; Step 4: Repeat the iterative process. If the objective function value does not improve after several consecutive iterations, stop the iteration and output the parameter set X that obtains the best objective function value among all samples. * .

9. The ship section painting quality assessment and rework decision-making method according to claim 1 is characterized by: The acquisition function of the Bayesian optimization algorithm is: Otherwise, EI is 0; Where, is the optimal value of the objective function in the current sample; μ(x) and σ(x) are the predicted mean and standard deviation of GP at point x, respectively; Φ(Z) and φ(Z) are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

10. A ship section painting quality assessment and rework decision-making system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the ship section painting quality assessment and rework decision-making method according to any one of claims 1 to 9 is implemented.

Images