Paint film thickness prediction method and system for ship spraying and automatic spraying system

Abstract

The invention discloses a paint film thickness prediction method and system for ship spraying and an automatic spraying system.According to the paint film thickness prediction method, a physical mechanism and data driving are combined to construct an online paint film thickness prediction model, and the dried paint film thickness can be accurately predicted; the automatic spraying system performs early warning based on online paint film thickness prediction and wet paint film defect detection, triggers feed-forward path adjustment, and changes quality control from post-event remediation into pre-event prediction and in-event control, so that reworking and material waste are fundamentally avoided, and one-time forming and high uniformity of coating quality are ensured.

Description

Technical Field

[0001] This invention relates to the field of automated spraying technology for ship surfaces, specifically to a method, system, and automated spraying system for predicting paint film thickness for ship spraying. Background Technology

[0002] The automated surface painting process for ships involves: establishing a three-dimensional digital surface model of the hull; dividing the surface into zones and planning the painting path based on the model; utilizing a painting robot to perform painting operations along the painting path; continuously predicting the paint film thickness during the operation; and optimizing the painting path based on the prediction results. The existing technical problem is that current paint film thickness prediction relies on a single physical mechanism formula, failing to consider the coupling relationship between painting process parameters and environmental factors, resulting in large prediction errors. Summary of the Invention

[0003] Objectives of the Invention: The first objective of this invention is to provide a method for predicting paint film thickness for ship painting with high prediction accuracy; the second objective of this invention is to provide a paint film thickness prediction system for ship painting; and the third objective of this invention is to provide an automated painting system incorporating the paint film thickness prediction system.

[0004] Technical solution: The present invention provides a method for predicting paint film thickness for ship painting, comprising:

[0005] Constructing an online paint film thickness prediction model:

[0006]

[0007] in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This represents a trained machine learning model; This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations.

[0008] Real-time acquisition of spraying process parameters and on-site environmental data is input into an online paint film thickness prediction model to obtain the predicted paint film thickness after drying. Spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed .

[0009] Furthermore, the machine learning model employs XGBoost, Random Forest, or LightGBM.

[0010] The present invention provides a paint film thickness prediction system for ship painting, comprising a detection unit and a prediction unit;

[0011] The detection unit is used to acquire real-time spraying process parameters and on-site environmental data. The spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed ;

[0012] The prediction unit is used to input the spraying process parameters and on-site environmental data into the built-in online paint film thickness prediction model to obtain the predicted value of the paint film thickness after drying; the online paint film thickness prediction model is as follows:

[0013]

[0014] in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This represents a trained machine learning model; This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations.

[0015] Furthermore, the detection unit includes a flow sensor, an encoder, an ultrasonic sensor, a joint angle sensor, a pressure sensor, and an environmental sensor. The flow sensor is used to acquire the spraying flow rate. The encoder is used to obtain the moving speed of the painting robot. Ultrasonic sensors are used to obtain the spraying distance. The joint angle sensor is used to obtain the spray gun angle. The pressure sensor is used to obtain the spray gun pressure. The environmental sensors include temperature sensors, humidity sensors, and wind speed sensors, which are used to acquire data at the work site. temperature ,humidity and wind speed .

[0016] Furthermore, the paint film thickness prediction system for ship painting also includes a quality early warning unit, which is used to issue an early warning when the predicted paint film thickness is not within the set dry film thickness range.

[0017] Furthermore, the detection unit also includes an industrial camera for acquiring wet paint film images; the quality warning unit is also used to identify defects based on the wet paint film images using a pre-trained convolutional neural network model, and to issue a warning when the proportion of any of the identified defects in the image area is greater than a set value.

[0018] An automated spraying system of the present invention includes a perception and modeling module, a path planning module, a spraying module, and the aforementioned paint film thickness prediction system;

[0019] The perception and modeling module is used to construct a three-dimensional digital surface model of the ship's hull;

[0020] The path planning module is used to divide the work area according to the three-dimensional digital surface model of the hull, overlay and generate a dynamic partition map, determine the path sequence and adjust the spraying path; the spraying path is optimized in real time according to the paint film thickness prediction value obtained by the paint film thickness prediction system, and fine-tuned when the quality warning unit issues a warning.

[0021] The spraying module performs spraying operations according to the spraying path generated by the path planning module.

[0022] Furthermore, the perception and modeling module acquires point cloud data by scanning the hull with LiDAR to construct a 3D point cloud model, and simultaneously acquires 2D color images and videos with an industrial camera to give the 3D point cloud model color and texture. Then, a surface reconstruction algorithm is used to generate a 3D digital surface model of the hull with curvature information.

[0023] Furthermore, when dividing the work area, the path planning module identifies the planar area and the curvature area through a curvature analysis algorithm and performs curvature-accessibility dynamic partitioning; a bow-shaped spraying path is used for the planar area and the small curvature area, and a contour spraying path is used for the large curvature area.

[0024] Furthermore, the painting module includes a mobile platform and a height-adjustable high-altitude painting robot and a retractable low-altitude painting robot mounted side by side on the mobile platform to move closer to or further away from the hull. The low-altitude painting robot can also be raised and lowered to adjust its height. The path planning module divides the work area according to the working range of the high-altitude and low-altitude painting robots. The high-altitude and low-altitude painting robots automatically call the corresponding painting paths generated by the path planning module. The high-altitude and low-altitude painting robots do not perform painting operations at the same time.

[0025] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) The online paint film thickness prediction model integrates physical mechanism and data driving, and can accurately predict the paint film thickness after drying. (2) Based on online paint film thickness prediction and wet paint film defect detection, early warning is given, triggering feedforward path adjustment, changing quality control from post-remediation to pre-prediction and in-process control, fundamentally avoiding rework and material waste, and ensuring the one-time forming and high uniformity of coating quality. Attached Figure Description

[0026] Figure 1 This is a structural diagram and workflow diagram of an automated spraying system provided in an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the spraying module in an embodiment of the present invention. Detailed Implementation

[0028] The invention will now be further described with reference to the accompanying drawings.

[0029] Example 1: Example 1 provides a method for predicting paint film thickness for ship painting, including:

[0030] An online paint film thickness prediction model combining physical mechanisms and data-driven approaches is constructed. This model includes the physical formula for paint film formation and, based on this formula, incorporates a machine learning model to learn the complex mapping relationship between input and output.

[0031]

[0032] in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This refers to a trained machine learning model, which may employ XGBoost, Random Forest, or LightGBM. This embodiment will use XGBoost as an example. This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations.

[0033] Real-time acquisition of spraying process parameters and on-site environmental data is input into an online paint film thickness prediction model to obtain the predicted paint film thickness after drying. Spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed .

[0034] The specific architecture and implementation details of the online paint film thickness prediction model are as follows:

[0035] 【1】Derivation of basic formula: Based on the three-stage characteristics of coatings [laminar deposition - gravity leveling - drying and curing], and combining "momentum conservation" and "film thickness conservation", the basic formula for calculating the coating film thickness is derived:

[0036]

[0037] Wherein, paint volume deposition = spray flow rate ( ) × working time, spray coverage area = spray width ( ) × Robot movement speed ( Multiply the numerator by the work time, then remove the work time from both the numerator and denominator to obtain the basic relational expression:

[0038]

[0039] constant Integrating inherent properties of coatings (such as density) 1.2~1.5 Solid content (40%~70%) and equipment performance parameters (such as spray gun atomization efficiency) (85%~95%), calibrated through preliminary process experiments (such as fixing) , , Test the dry film thickness of different coatings and infer the results. (value), final The value ranges from 0.8 to 1.2 and needs to be calibrated according to the specific paint model.

[0040] Spraying width : is the spraying distance and spray gun angle A bivariate function, where the spraying distance It is a work site Vertical distance to the spray gun (unit: spray gun angle It is the spray gun axis and the working point The angle between the normal vectors (unit: °).

[0041] The relationship was derived through experimental fitting:

[0042] =0.3× +0.01×(90-| |)

[0043] Environmental impact factor :

[0044] Quantitative temperature (Unit: °C), Humidity (Unit: %RH) Effects on the drying process of paint film: Increased temperature will accelerate solvent evaporation, which may result in a slight decrease in dry film thickness; increased humidity will delay leveling, which may result in a slight increase in dry film thickness.

[0045] The results were obtained through fitting data from over 100 sets of environmental operating condition experiments:

[0046] =1+0.005×(25- )+0.002×( -60)

[0047] In summary, the final output of the physical mechanism layer is:

[0048]

[0049] [2] Data-driven layer: Machine learning based on XGBoost

[0050] Data-driven learning layers learn multi-parameter coupling relationships that are difficult to quantify in terms of physical mechanisms (such as...) and Interaction and influence and The synergistic effect of [the system] and random interference factors (such as paint fog rebound and local airflow fluctuations) are considered. The specific implementation steps are as follows:

[0051] [A] Data Acquisition: Multi-dimensional data collection across all scenarios and at high frequency is the foundation for model training. This data collection needs to cover typical ship painting scenarios (planar / curved surfaces, high / low altitudes, different environments). The specific data collection plan is as follows:

[0052]

[0053] The purpose of offline acquisition of the final dry film thickness using a coating thickness gauge is to provide a real label for training and validation of the online paint film thickness prediction model. On the one hand, the offline measured dry film thickness is key data during the model training phase, used to allow the machine learning of complex coupling relationships not covered by the physical mechanism layer, and to calibrate the model's prediction bias. On the other hand, after the model is put into use, the offline measured values ​​can be compared with the model's predicted values ​​periodically to verify the model's accuracy. For example, when the deviation exceeds a threshold, the model can be fine-tuned to ensure prediction reliability.

[0054] An industrial camera is mounted behind the painting robot to acquire images of the wet paint film. A trained convolutional neural network model can then be used to detect defects (such as missed areas, runs, and orange peel). The goal is to achieve early correlation control between the wet film state and the dry film quality: defects in the wet paint film directly lead to uneven thickness and poor appearance in the final dry film. For example, even if the predicted dry film thickness meets the standards, defects in the wet film stage will still cause the final coating to fail to meet quality requirements. Therefore, the detection of wet paint film defects is necessary, allowing adjustments to be made before these defects affect the dry film and avoiding subsequent rework. The image recognition method for wet paint film defects is existing technology.

[0055] Data volume and hull area: A total of 1,200 sets of valid historical data were collected, including 300 sets of side plane data, 250 sets of data from the bulbous bow high curvature area, 250 sets of data from the low-altitude area of ​​the deck, 200 sets of data from the complex corners of the hatches, and 200 sets of data from the high-altitude area of ​​the superstructure.

[0056] Environmental conditions: normal temperature and humidity ( =20~25℃, =50~60%RH, 400 groups), low temperature and high humidity ( =5~10℃, =80~90%RH, 200 groups), high temperature and low humidity ( =35~40℃, =30~40%RH, 200 groups), windy environment ( =2~5m / s, 200 sets), complex mixed environment (such as high temperature + wind, 200 sets).

[0057] Process parameter combinations: (100 / 150 / 200 / 250 / 300mL / min) × (0.1 / 0.2 / 0.3 / 0.4 / 0.5m / min) × (0.2 / 0.3 / 0.4m) × A full combination of (-30° / -15° / 0° / 15° / 30°) ensures data diversity.

[0058]

B

[0059] The raw data contains noise (such as instantaneous fluctuations in sensor data) and missing values ​​(such as brief periods of sensor offlineness), and needs to be processed through the following steps:

[0060] Outlier removal: using " The principle of handling continuous data (such as...) , , Data exceeding the mean ± 3 standard deviations were marked as outliers and replaced with the moving average of the five adjacent data points (e.g., ...). (Instantaneous jump to 500 mL / min, replaced with an average of 150 mL / min over the previous 5 seconds); for categorical data (such as defect type), samples with "fuzzy identification" (e.g., confidence level < 80%) are removed.

[0061] Missing value imputation: For data with short-term missing values ​​(≤5 seconds), linear interpolation is used to imput them. For samples with long-term missing values ​​(>5 seconds), they are directly removed (ensuring that the missing value rate of a single sample is <5%).

[0062] Data normalization: Min-Max normalization is performed on the input parameters (mapped to the [0,1] interval) to avoid data loss due to differences in units (e.g., Unit: mL / min (Unit: °C) This causes a shift in model weights.

[0063] Data alignment and annotation: Downsample high-frequency data to 1Hz, and... , Align 1Hz data; use the dry film thickness measured offline as the sample label to ensure that the input parameters and output labels correspond one-to-one.

[0064] [C] Feature Engineering: Constructing effective features that the model can recognize.

[0065] Directly inputting raw parameters into the model makes it difficult to capture complex relationships; feature mapping is necessary. Constructing higher-order features specifically includes:

[0066] Basic features: Preserve preprocessed data (7).

[0067] Interactive features: Constructing product / ratio features between parameters, such as (Flow rate to velocity ratio, reflecting the amount of paint deposited per unit distance) (Distance-angle interaction reflects the comprehensive impact of swath width) (Temperature and humidity work together to reflect the overall environmental impact) (3 items).

[0068] Time series characteristics: Calculate the moving average of parameters for the first 3 seconds / 5 seconds (e.g., The 5-second average reflects speed stability), and the rate of change of parameters (such as... The instantaneous rate of change, reflecting flow fluctuations (6).

[0069] Physical residual characteristics: Calculate the difference between the "physical mechanism layer predicted value" and the "historical measured value" (i.e., residual = physical predicted value - historical measured value), which serves as the core learning objective (1) of the data-driven layer.

[0070] Ultimately, feature engineering resulted in 17 effective features, significantly improving the model's ability to capture complex relationships.

[0071] [D] Model Training and Optimization: Iterative Tuning Based on XGBoost

[0072] XGBoost was chosen as the machine learning model because of its advantages, including strong resistance to overfitting, high efficiency in handling nonlinear relationships, and support for feature importance analysis. The specific training process is as follows:

[0073] Data partitioning: The 1200 sets of data were divided into a training set (840 sets, used for model fitting), a validation set (240 sets, used for hyperparameter tuning), and a test set (120 sets, used for final performance evaluation) in a ratio of 7:2:1.

[0074] Hyperparameter initialization: Initial parameters are set as follows: learning rate = 0.1, tree depth = ... =5, estimators=100, subsample=0.8, column sample=0.8.

[0075] Hyperparameter tuning: Parameters are optimized using a grid search + 5-fold cross-validation approach, iterating through the following parameter combinations:

[0076] Learning rate (lr): 0.05, 0.1, 0.2;

[0077] Tree depth : 3, 5, 7;

[0078] Estimators: 100, 200, 300;

[0079] Using the mean absolute error of the validation set as the optimization objective, the optimal parameters were finally determined as: learning rate lr = 0.1, tree depth... Estimators, subsample proportion = 0.8, column sample proportion = 0.8.

[0080] Model training: Fit the model to the training set with the goal of minimizing the residual. This allows the data-driven layer to learn the bias of the physical mechanism, while avoiding overfitting through an early stopping mechanism (stop training when the MAE of the validation set does not decrease for 10 consecutive rounds).

[0081] Performance evaluation: Validate the model performance on the test set, and finally achieve:

[0082] Mean absolute error (MAE) = 3.2 μm (≤ ±5 μm);

[0083] The root mean square error (RMSE) is 4.1 μm.

[0084] The prediction accuracy (percentage of samples with an error ≤ ±5μm) is 98.3%; compared with the traditional single physical model (MAE=12.5μm), the prediction accuracy is improved by 74.4%, which fully meets the requirements of industrial-grade spraying quality.

[0085] Model Fusion: Synergistic Output of Physical Mechanisms and Data-Driven Approaches

[0086] Final predicted paint film thickness This is achieved through a combination of "physical values" and "data correction values".

[0087] =Predicted value from physical mechanism layer + Residual correction value from data-driven layer output +

[0088] in, This is a random error term, caused by uncontrollable factors such as slight rebound of paint mist, and is derived through statistical analysis of numerous experiments. ∈[-2μm,+2μm], which can be ignored. The residual correction value output by the data-driven layer is essentially a dynamic compensation for the deviation of the physical model (such as when...). =200mL / min At 35℃, the physical model's predicted value may be too low. The data-driven layer will output a correction value of +2μm to make the final prediction more accurate.

[0089] Example 2: Example 2 provides a paint film thickness prediction system for ship painting, including a detection unit, a prediction unit and a quality early warning unit.

[0090] The detection unit is used to acquire real-time spraying process parameters and on-site environmental data. The spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed .

[0091] Specifically, the detection unit includes a flow sensor, an encoder, an ultrasonic sensor, a joint angle sensor, a pressure sensor, an environmental sensor, and an industrial camera. The flow sensor is used to acquire the spraying flow rate. The encoder is used to obtain the moving speed of the painting robot. Ultrasonic sensors are used to obtain the spraying distance. The joint angle sensor is used to obtain the spray gun angle. The pressure sensor is used to obtain the spray gun pressure. The environmental sensors include temperature sensors, humidity sensors, and wind speed sensors, which are used to acquire data at the work site. temperature ,humidity and wind speed Industrial cameras are used to acquire images of wet paint films.

[0092] The prediction unit is used to input the spraying process parameters and on-site environmental data into the built-in online paint film thickness prediction model to obtain the predicted value of the paint film thickness after drying; the online paint film thickness prediction model is as follows:

[0093]

[0094] in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This represents a trained machine learning model; This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations.

[0095] The quality early warning unit is used to issue warnings when the predicted paint film thickness is outside the set dry film thickness range, and to identify defects based on the wet paint film image using a pre-trained convolutional neural network model. Warnings are issued when the proportion of any identified defect in the image area exceeds a set value. For example, the model predicts the paint film thickness for an area 1 meter ahead (calculated at a robot speed of 0.3 m / min, corresponding to the work area 2 seconds later). An alert will be triggered immediately if the dry film thickness is outside the set range (e.g., 195~205μm) or if wet film defects are present (e.g., run-off accounts for more than 80% of the image area).

[0096] Determining the area 1 meter ahead allows sufficient response time for the system's "prediction-adjustment" process, while also matching the actual efficiency of the spraying operation. Assuming the robot's movement speed is 0.3 m / min (0.005 m per second), 2 seconds corresponds to an actual movement distance of only 0.01 m. A 1-meter warning distance can completely cover the width of a single spraying pass, ensuring that adjustment commands cover the entire area the current spray gun is about to work on, avoiding missed adjustments due to an insufficient warning range. Simultaneously, this distance allows sufficient time for path planning to generate a fine-tuned path, ensuring parameter adjustments are completed before the spraying robot reaches the area, enabling pre-emptive intervention.

[0097] The quality assessment configuration is as follows: the painting is deemed satisfactory only when the predicted paint film thickness meets the requirements and the wet paint film is free of defects (tolerable wet paint film defects are considered defect-free). When the painting fails to meet the standards, no overlay spraying is performed. Instead, the path is dynamically fine-tuned based on the reason for the failure: in flat / small curvature areas, the path spacing is reduced and the speed of the painting robot is decreased to achieve high overlap rate respraying; in large curvature areas, the dwell time of the contour trajectory and the spray gun angle are adjusted for targeted local respraying.

[0098] Example 3: As Figure 1 As shown, Embodiment 3 provides an automated spraying system, including a perception and modeling module, a path planning module, a spraying module, and the paint film thickness prediction system described in Embodiment 2.

[0099] (a) Perception and Modeling Module

[0100] The perception and modeling module uses LiDAR to scan the hull to acquire point cloud data to construct a 3D point cloud model. It uses an industrial camera to simultaneously acquire 2D color images and videos to give the 3D point cloud model color and texture. Finally, a surface reconstruction algorithm is used to generate a 3D digital surface model of the hull with curvature information.

[0101] (ii) Path Planning Module

[0102] The path planning module divides the work area based on the 3D digital surface model of the hull, overlays it to generate a dynamic partition map, determines the path sequence, and adjusts the spraying path. When dividing the work area, the module uses a curvature analysis algorithm to identify planar and curvature areas, performing dynamic curvature-accessibility partitioning. A bow-shaped spraying path is used for planar and low-curvature areas, while a contour-following spraying path is used for high-curvature areas. The spraying path is optimized in real-time based on the predicted paint thickness obtained from the paint thickness prediction system, and fine-tuned when a quality warning unit issues a warning.

[0103] Specifically, after reading the 3D digital surface model of the ship's hull, the path planning module first divides the area into accessibility zones based on the maximum working height of the high-altitude spraying robot (12 meters) and the working range of the low-altitude spraying robot (0-3 meters). Then, it calculates the Gaussian curvature of each point on the model surface, where the curvature threshold is... The value was determined to be 0.03 based on preliminary process experiments. The process involves dividing areas with an absolute curvature value <0.03 into planar and low-curvature zones, and areas with a curvature value ≥0.03 into high-curvature zones. Finally, the accessibility zones and curvature zones are overlaid to generate a dynamic zoning map (containing four sub-regions: "High-altitude Planar / Low-curvature Work Zone," "High-altitude High-curvature Work Zone," "Low-altitude Planar / Low-curvature Work Zone," and "Low-altitude High-curvature Work Zone") to guide the spraying operation. The path planning module plans paths based on the dynamic zoning map. For planar / low-curvature zones, a bow-shaped path with a spacing of 50cm is planned. For high-curvature zones, a contour trajectory is generated. This trajectory ensures that the spray gun tip maintains an optimal distance of 25cm from the hull surface, and that the angle between the spray gun axis and the normal vector of the spray point is always within ±10°, thus achieving high-quality, close-fitting spraying.

[0104] Fine-tuning of the spraying path: When the prediction model determines that the predicted thickness of the area 1 meter ahead is lower than the lower limit (e.g., 195μm), insert a "micro-re-spraying segment" in the path of that area: If it is a flat / small curvature area: maintain the bow-shaped path, but reduce the path spacing from 50cm to 30cm and reduce the moving speed by 20%~30% to achieve high overlap re-spraying; If it is a large curvature area: fine-tune the "dwell time" of the contour trajectory (dwell in the area for 0.5~1 seconds longer), and at the same time adjust the spray gun angle to a better value (e.g., from 10° to 5°) to ensure that the paint film thickness meets the standard.

[0105] The above-mentioned model building, spraying path planning, real-time optimization and fine-tuning are all existing technologies.

[0106] (III) Spraying Module

[0107] The spraying module performs the spraying operation according to the spraying path generated by the path planning module. Specifically, for example... Figure 2 As shown, the painting module includes a mobile platform and a height-adjustable high-altitude painting robot mounted side by side on the mobile platform, and a retractable low-altitude painting robot that can move closer to or further away from the hull (e.g., based on a telescopic screw). The low-altitude painting robot can also be raised and lowered to adjust its height. The path planning module divides the work area according to the working range of the high-altitude and low-altitude painting robots. The high-altitude and low-altitude painting robots automatically call the corresponding painting path generated by the path planning module.

[0108] High-altitude and low-altitude spraying robots do not perform spraying operations simultaneously because their movement speed is the same as that of the moving platform. High-altitude spraying robots operate on flat surfaces or areas with small curvatures, and their speed is typically set at 0.2~0.5 m / min to balance efficiency and uniformity. Low-altitude spraying robots operate on complex curved surfaces, and their speed is mostly controlled at 0.1~0.3 m / min to ensure precise application. Since their working speeds differ, a spraying sequence of high-altitude first, followed by low-altitude, can be adopted. After completing its spraying operation, the high-altitude spraying robot descends to lower its center of gravity, ensuring the overall stability of the spraying module when the low-altitude spraying robot extends.

Claims

1. A method for predicting paint film thickness for ship painting, characterized in that, include: Constructing an online paint film thickness prediction model: in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This represents a trained machine learning model; This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations. Real-time acquisition of spraying process parameters and on-site environmental data is input into an online paint film thickness prediction model to obtain the predicted paint film thickness after drying. Spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed .

2. The method for predicting paint film thickness for ship painting according to claim 1, characterized in that, The machine learning model used is XGBoost, Random Forest, or LightGBM.

3. A paint film thickness prediction system for ship painting, characterized in that, Includes a detection unit and a prediction unit; The detection unit is used to acquire real-time spraying process parameters and on-site environmental data. The spraying process parameters include spraying flow rate. The moving speed of the painting robot Spraying distance Spray gun angle and spray gun pressure On-site environmental data includes work sites. temperature ,humidity and wind speed ; The prediction unit is used to input the spraying process parameters and on-site environmental data into the built-in online paint film thickness prediction model to obtain the predicted value of the paint film thickness after drying. The online paint film thickness prediction model is as follows: in, Indicates work point Predicted value of paint film thickness after drying; It is a constant; Indicates the spraying flow rate; This indicates the moving speed of the painting robot; This indicates the spray width, which is the spray distance. and spray gun angle The function; The environmental impact factor is represented by the work point. temperature and humidity Joint decision; This represents a trained machine learning model; This represents a feature mapping, which transforms the original input parameters into feature vectors suitable for processing by machine learning models. Indicates the pressure of the spray gun; Indicates the surface roughness of the workpiece; Indicates work point wind speed; This represents the random error in the model's prediction; the online paint film thickness prediction model is trained using data accumulated from historical operations.

4. The paint film thickness prediction system for ship painting according to claim 3, characterized in that, The detection unit includes a flow sensor, an encoder, an ultrasonic sensor, a joint angle sensor, a pressure sensor, and an environmental sensor. The flow sensor is used to acquire the spraying flow rate. The encoder is used to obtain the moving speed of the painting robot. Ultrasonic sensors are used to obtain the spraying distance. The joint angle sensor is used to obtain the spray gun angle. The pressure sensor is used to obtain the spray gun pressure. The environmental sensors include temperature sensors, humidity sensors, and wind speed sensors, which are used to acquire data at the work site. temperature ,humidity and wind speed .

5. The paint film thickness prediction system for ship painting according to claim 4, characterized in that, It also includes a quality warning unit, which is used to issue a warning when the predicted value of the paint film thickness is not within the set dry film thickness range.

6. The paint film thickness prediction system for ship painting according to claim 5, characterized in that, The detection unit also includes an industrial camera for acquiring wet paint film images; the quality warning unit is also used to identify defects based on the wet paint film images using a pre-trained convolutional neural network model, and to issue a warning when the proportion of any of the identified defects in the image area is greater than a set value.

7. An automated spraying system, characterized in that, It includes a perception and modeling module, a path planning module, a spraying module, and the paint film thickness prediction system as described in claim 6; The perception and modeling module is used to construct a three-dimensional digital surface model of the ship's hull; The path planning module is used to divide the work area according to the three-dimensional digital surface model of the hull, overlay and generate a dynamic partition map, determine the path sequence and adjust the spraying path; The spraying path is optimized in real time based on the predicted value of the paint film thickness obtained by the paint film thickness prediction system, and fine-tuned when the quality warning unit issues a warning. The spraying module performs spraying operations according to the spraying path generated by the path planning module.

8. The automated spraying system according to claim 7, characterized in that, The perception and modeling module uses LiDAR to scan the hull to acquire point cloud data to construct a 3D point cloud model. It uses an industrial camera to simultaneously acquire 2D color images and videos to give the 3D point cloud model color and texture. Finally, a surface reconstruction algorithm is used to generate a 3D digital surface model of the hull with curvature information.

9. The automated spraying system according to claim 7, characterized in that, When dividing the work area, the path planning module identifies the planar area and the curvature area through the curvature analysis algorithm and performs curvature-accessibility dynamic partitioning; a bow-shaped spraying path is used for the planar area and the small curvature area, and a contour spraying path is used for the large curvature area.

10. The automated spraying system according to claim 9, characterized in that, The painting module includes a mobile platform and two parallel high-altitude painting robots mounted on the platform, one that can be raised and lowered and the other that can be extended to approach or move away from the hull. The low-altitude painting robot can also be raised and lowered to adjust its height. The path planning module divides the work area according to the working range of the high-altitude and low-altitude painting robots. The high-altitude and low-altitude painting robots automatically call the corresponding painting paths generated by the path planning module. The high-altitude and low-altitude painting robots do not perform painting operations at the same time.

Images