A continuous painting method for an automatic spraying equipment for rail-mounted steel structures

Abstract

This invention relates to the field of steel structure spraying technology and discloses a continuous painting method for an automatic spraying equipment for rail-mounted steel structures. The method involves obtaining the target structural specifications of the target steel structure, retrieving and determining a set of historical spraying records from a historical spraying database based on these specifications, differentiating the surface types of the historical spraying records to obtain multiple sets of spraying records differentiated by surface type, traversing these sets to perform spraying feature extraction operations, obtaining multiple sets of spraying feature values ​​and multiple spraying feature correlation scales, using the correlation scales as the analysis scales of multiple spraying processing network layers, and performing multi-scale feature analysis on the spraying monitoring data sequence within a preset spraying cycle using the configured network layers to obtain multiple sets of spraying monitoring features, and performing cross-scale interactive fusion on these sets to obtain a target interactive fused spraying monitoring feature set, which is then output as the continuous painting control result.

Description

Technical Field

[0001] This invention relates to the field of steel structure spraying technology, specifically a continuous spraying method for an automatic spraying equipment for rail-mounted steel structures. Background Technology

[0002] Steel structures, due to their high strength and stability, are widely used in fields such as railway engineering, and surface coating is a crucial means of ensuring structural durability. Railway-type steel structures exhibit diverse forms, with surfaces including welded seams, smooth surfaces, and corners. Different surfaces have varying requirements for coating material adhesion and thickness. Existing automated coating equipment often uses preset parameters when handling these steel structures. These parameters are frequently based on general structures and do not fully consider the unique specifications of specific steel structures, making it difficult to provide targeted treatment for different surface types. While a wealth of data accumulated from historical coating processes includes experience in handling different structural specifications and surface types, current technology often simply stores this data without systematic organization and utilization, failing to provide effective reference for new coating tasks.

[0003] During the spraying process, the equipment collects real-time monitoring data such as flow rate, pressure, and moving speed. These data exhibit different characteristics as the spraying stage changes. Existing technologies often analyze monitoring data using a single scale or focus only on the characteristics of local stages, making it difficult to comprehensively reflect the overall state of the spraying process. When the structural specifications of the object being sprayed change, or when switching between different surface types, the results obtained from single-scale analysis often cannot adapt to the new spraying requirements in a timely manner, leading to disjointed spraying processes.

[0004] Continuous painting of rail-mounted steel structures requires equipment to maintain stable processing results during long-term operation. However, existing methods, due to their neglect of differences in surface types, insufficient utilization of historical data, and limitations in monitoring data processing, are prone to fluctuations in painting quality and cannot meet the continuous processing needs of complex steel structures. Summary of the Invention

[0005] The purpose of this invention is to provide a continuous painting method for an automatic spraying equipment for rail-mounted steel structures, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a continuous painting method for an automatic spraying equipment for rail-mounted steel structures, the method comprising:

[0007] Obtain the target structural specifications of the target steel structure, and search the historical spraying database based on the target structural specifications to determine the set of historical spraying records;

[0008] The historical spraying record set is differentiated by surface type to determine multiple spraying record sets that differentiate by surface type;

[0009] The multiple sets of spraying records that distinguish surface types are traversed to perform spraying feature extraction operations, thereby obtaining multiple spraying feature set values ​​and multiple spraying feature correlation scales;

[0010] The multiple spraying feature association scales are used as multiple analysis scales of multiple spraying processing network layers. The configured multiple spraying processing network layers are used to perform multi-scale feature analysis processing on the spraying monitoring data sequence of the target steel structure within the preset spraying cycle to obtain multiple spraying monitoring feature sets.

[0011] Perform a cross-scale interactive fusion operation on the multiple spraying monitoring feature sets to obtain a target interactive fusion spraying monitoring feature set, and output the target interactive fusion spraying monitoring feature set as the continuous spraying control result.

[0012] Preferably, the multiple sets of spraying records distinguishing surface types are traversed to perform spraying feature extraction operations, obtaining multiple spraying feature set values ​​and multiple spraying feature correlation scales, including:

[0013] The multiple sets of spraying records that distinguish surface types are traversed to perform spraying monitoring data extraction operations, thereby obtaining multiple sets of spraying monitoring data sequences that distinguish surface types.

[0014] Perform optimal spraying time data extraction operations on the multiple sets of surface type distinguishing spraying monitoring data sequences respectively to determine multiple historical optimal spraying monitoring data sets, wherein each historical optimal spraying monitoring data corresponds to the monitoring data at the moment when the coating uniformity reaches the optimal state in each surface type distinguishing spraying monitoring data sequence;

[0015] Feature extraction processing is performed on the multiple historical best spraying monitoring data sets to obtain multiple historical best spraying feature sets, and centralized analysis processing is performed on the multiple historical best spraying feature sets to determine multiple spraying feature set values.

[0016] Using the multiple historical best spraying feature sets as indexes, a feature association scale diffusion identification operation is performed on the multiple surface type distinguishing spraying monitoring data sequence sets to determine multiple historical best spraying feature association scale sets, wherein each historical best spraying feature association scale reflects the duration of data associated with the best spraying time in the surface type distinguishing spraying monitoring data sequence;

[0017] The average value of the multiple historical best spraying feature association scale sets is calculated to obtain the multiple spraying feature association scales.

[0018] Preferably, feature extraction processing is performed on the multiple historical best spraying monitoring data sets to obtain multiple historical best spraying feature sets, and centralized analysis processing is performed on the multiple historical best spraying feature sets to determine multiple spraying feature set values, including:

[0019] The spraying feature extraction network layer is used to perform feature extraction processing on the multiple historical best spraying monitoring data sets to obtain multiple historical best spraying feature sets;

[0020] Iterate through multiple sets of historical best spraying features and perform feature mean calculation to determine the mean of multiple historical best spraying features;

[0021] According to a preset iteration scale, the average of the multiple historical best spraying features is used to perform iterative processing in the multiple historical best spraying feature sets to obtain multiple iterative historical best spraying features;

[0022] When the degree of aggregation of the multiple iterative historical best spraying features is less than or equal to the degree of aggregation of the average of the multiple historical best spraying features, the average of the multiple historical best spraying features is output as the set value of the multiple spraying features.

[0023] Preferably, the method includes:

[0024] When the degree of aggregation of the multiple iterative historical best spraying features is greater than the degree of aggregation of the average of the multiple historical best spraying features, it is determined whether the difference in the degree of aggregation between the multiple iterative historical best spraying features and the average of the multiple historical best spraying features is greater than or equal to a preset degree of aggregation difference threshold.

[0025] If the difference in clustering degree is greater than or equal to the preset threshold, the iterative process continues based on the multiple iterative history best spraying features until the preset maximum number of iterations is met, and the multiple iterative history best spraying features obtained in the last iteration are output as the set value of the multiple spraying features.

[0026] If the difference in aggregation degree is less than the preset threshold, the iterative processing is stopped, and the best spraying features from the multiple iteration history are output as the set value of the multiple spraying features.

[0027] Preferably, using the multiple historical best spraying feature sets as indexes, a feature association scale diffusion identification operation is performed on the multiple sets of surface type distinguishing spraying monitoring data sequences to determine multiple historical best spraying feature association scale sets, including:

[0028] The spraying feature extraction network layer is used to perform feature extraction processing on the multiple sets of spraying monitoring data sequences that distinguish surface types, to obtain multiple sets of spraying feature sequences that distinguish surface types.

[0029] One historical best spraying feature is randomly extracted from the multiple historical best spraying feature sets as a reference historical best spraying feature, and the corresponding reference surface type distinguishing spraying feature sequence is matched from the multiple surface type distinguishing spraying feature sequence sets.

[0030] According to a preset approximate association range, a nearest neighbor search operation is performed on the reference historical best spraying feature in the reference distinguishing surface type spraying feature sequence to obtain the neighborhood of the reference historical best spraying feature;

[0031] The duration of the neighborhood of the reference historical best spraying feature is statistically analyzed, and the statistical results are used as the correlation scale of the reference historical best spraying feature.

[0032] According to the preset nearest neighbor association range, the feature association scale diffusion identification operation is performed on the multiple historical best spraying feature sets in the corresponding multiple spraying monitoring data sequence sets that distinguish surface types, to determine multiple historical best spraying feature association scale sets.

[0033] Preferably, a cross-scale interactive fusion operation is performed on the plurality of spraying monitoring feature sets to obtain a target interactive fused spraying monitoring feature set, including:

[0034] An initial spraying monitoring feature set and subsequent spraying monitoring feature sets are randomly extracted from the plurality of spraying monitoring feature sets;

[0035] Calculate the similarity recognition results between the initial spraying monitoring feature set and the subsequent spraying monitoring feature set to determine the initial feature similarity set;

[0036] Perform a normalization operation on the initial feature similarity set to obtain an initial feature similarity normalized value set;

[0037] Perform a convolution operation on the initial feature similarity normalization value set and the subsequent spraying monitoring feature set to obtain an initial interactive fusion spraying monitoring feature set;

[0038] Randomly extract additional spraying monitoring feature sets from the multiple spraying monitoring feature sets, and perform cross-scale interactive fusion operation between the additional spraying monitoring feature sets and the initial interactive fusion spraying monitoring feature sets to obtain subsequent interactive fusion spraying monitoring feature sets;

[0039] After repeated cross-scale interactive fusion operations, until all spraying monitoring features in the multiple spraying monitoring feature sets are fused, the target interactive fused spraying monitoring feature set is obtained.

[0040] Preferably, the method further includes:

[0041] The training dataset is obtained by acquiring a set of normalized similarity values ​​of multiple sample features, a set of spraying monitoring features of multiple samples, and a set of interactive fusion spraying monitoring features of multiple samples.

[0042] Supervised training is performed on the fusion network layer built on the convolutional neural network using the training dataset until the training reaches a convergent state, thus obtaining the trained fusion convolutional network layer.

[0043] The initial interactive fused spraying monitoring feature set is obtained by performing convolution calculation operations on the initial feature similarity normalization value set and the subsequent spraying monitoring feature set using the trained fusion convolutional network layer.

[0044] Preferably, the historical spraying record set is differentiated by surface type to determine multiple spraying record sets that differentiate surface types, including:

[0045] Randomly select multiple historical spraying records from the aforementioned historical spraying record set;

[0046] Perform pairwise enumeration operations on the multiple historical spraying records to obtain multiple enumeration combinations;

[0047] Determine whether there is an enumeration combination whose record similarity is greater than a preset record similarity threshold among the multiple enumeration combinations;

[0048] If there is no enumerated combination greater than the preset record similarity threshold, then the multiple historical spraying records will be output as multiple distinguishing targets.

[0049] Based on the multiple distinguishing targets, a surface type difference distinguishing operation is performed on the historical spraying record set according to a preset record similarity threshold to obtain the multiple distinguishing surface type spraying record sets, wherein each distinguishing surface type spraying record set corresponds to a distinguishing target.

[0050] Preferably, the method further includes:

[0051] The surface vibration signal of the target steel structure during the spraying process is monitored to obtain vibration parameters. The running time and ambient humidity parameters of the spraying environment are collected, and the coating scaling prediction operation is performed to obtain predicted scaling parameters.

[0052] The surface image of the target steel structure is acquired by an industrial vision sensor. Based on the predicted scaling parameters, the surface image is input into multiple scaling parameter classifiers to identify the actual scaling parameters. Each scaling parameter classifier includes multiple scaling parameter classification paths.

[0053] Based on the actual scaling parameters, perform a coating deformation prediction operation to obtain the predicted coating deformation parameters;

[0054] The coating image of the target steel structure is acquired by an industrial vision sensor. Based on the predicted coating deformation parameters, the coating image is input into multiple coating parameter classifiers to identify and obtain the actual coating deformation parameters.

[0055] The continuous painting control results are adjusted based on the actual scaling parameters and actual coating deformation parameters.

[0056] Preferably, the method further includes:

[0057] Set the time window length, perform a sliding processing operation on the interference quantity according to the time window length, and obtain interference quantum data fragments, where the interference quantity includes environmental humidity parameters and vibration parameters;

[0058] Analyze and process the interfering quantum data fragments to obtain the periodic and probabilistic characteristics of each interference quantity;

[0059] Obtain the current time point, match the corresponding interference quantity based on the periodic characteristics of the interference quantity, and match the probability value of the occurrence of the interference quantity based on the probability characteristics.

[0060] The matched interference amount and probability value are input into the coating deformation prediction operation to optimize the predicted coating deformation parameters.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] This method obtains the target structural specifications of the target steel structure and retrieves historical spraying databases, enabling new spraying tasks to be linked to historical processing experience, thus making the spraying process targeted to the target structure from the outset. After differentiating the historical spraying records by surface type, the processing features of different surfaces are extracted independently, avoiding interference between different surface characteristics, and making subsequent feature extraction more closely aligned with the actual needs of the specific surface.

[0063] By traversing and differentiating the spraying record set, the concentrated values ​​of spraying features and correlation scales are extracted. Key information from historical data is transformed into quantifiable analytical basis. These feature values ​​and scales reflect the core characteristics of different surfaces during the spraying process, providing precise reference standards for subsequent multi-scale analysis. The correlation scale of spraying features is used as the analytical scale for multiple spraying processing network layers. This allows each network layer to analyze the monitoring data sequence within a preset spraying cycle at an appropriate scale. Different network layers capture features of different dimensions; some focus on subtle changes at the moment of spraying, while others focus on trend features over a longer period. This multi-dimensional analytical perspective allows for the full extraction of effective information from the monitoring data.

[0064] Cross-scale interactive fusion of multiple spraying monitoring feature sets breaks down the barriers between features at different scales, allowing subtle changes and trends to complement each other. The resulting target interactive fused spraying monitoring feature set more comprehensively reflects the true state of the spraying process. This fused feature set serves as the output of continuous painting control, enabling the equipment's control commands to simultaneously respond to spraying needs at different levels. When handling complex track-type steel structures, it can adapt to switching between different surface types and maintain processing continuity within continuous spraying cycles. This allows the automated spraying equipment to demonstrate stronger adaptability to diverse structural specifications, making the entire spraying process more aligned with the actual needs of the steel structure. This method eliminates the need for manual parameter adjustments. Through the combination of historical data and real-time monitoring, and the synergy of multi-scale analysis and cross-scale fusion, the spraying control can autonomously adapt to different spraying scenarios. During continuous processing, feature transfer and transformation at each stage are natural and smooth, reducing processing interruptions or adjustments caused by parameter mismatches, and making the operation of the automated spraying equipment for track-type steel structures more coordinated. Attached Figure Description

[0065] Figure 1 This is a schematic diagram illustrating the working principle of the continuous painting method for an automatic spraying equipment for rail-mounted steel structures according to the present invention.

[0066] Figure 2 A flowchart for extracting the concentrated values ​​and associated scales of spraying features;

[0067] Figure 3 A comparative analysis chart of spraying parameters for different surface types;

[0068] Figure 4 Optimization diagram for the spraying process of the target steel structure;

[0069] Figure 5 Flowchart for iterative determination of the concentrated values ​​of spraying features;

[0070] Figure 6 Feature correlation scale analysis diagram;

[0071] Figure 7 This is a correlation analysis chart of spraying parameters. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figure 1This invention provides a continuous painting method for an automatic spraying equipment for rail-mounted steel structures, the method comprising:

[0074] The target structural specifications of the target steel structure are obtained, and a historical spraying record set is determined by searching the historical spraying database based on the target structural specifications. The target structural specifications include the dimensional parameters of the steel structure (such as length, width, and height), surface curvature, and connection node type; the historical spraying database stores records of spraying parameters (such as paint type, spraying pressure, and spraying distance) and monitoring data (such as coating thickness and uniformity) for steel structures of different structural specifications.

[0075] The historical spraying record set is differentiated by surface type to identify multiple sets of spraying records that distinguish different surface types. Surface type differences include surface roughness, material density, and whether pretreatment (such as rust removal or polishing) has been performed. By analyzing these characteristics in the historical records, the records are divided into sets corresponding to different surface types.

[0076] The multiple sets of spraying records distinguishing surface types are traversed to extract spraying feature sets, resulting in multiple spraying feature set values ​​and multiple spraying feature correlation scales. The spraying feature set values ​​reflect the key feature data of the same surface type under optimal spraying conditions, while the spraying feature correlation scales reflect the data duration range related to the optimal spraying time.

[0077] The multiple spraying feature association scales are used as multiple analysis scales for multiple spraying processing network layers. Multi-scale feature analysis is then performed on the spraying monitoring data sequence of the target steel structure within a preset spraying cycle using the configured multiple spraying processing network layers to obtain multiple sets of spraying monitoring features. The preset spraying cycle is determined based on the spraying area of ​​the steel structure and the equipment operating speed. The multi-scale feature analysis extracts monitoring features at different time scales through different levels of network layers.

[0078] A cross-scale interactive fusion operation is performed on the multiple spraying monitoring feature sets to obtain a target interactive fused spraying monitoring feature set. This target interactive fused spraying monitoring feature set is then output as the continuous painting control result. This result is used to adjust the operating parameters of the automatic spraying equipment, such as nozzle movement speed and paint flow rate, to achieve precise control of continuous painting.

[0079] Example 1: See Figure 2The process involves iterating through multiple sets of spraying records differentiated by surface type to extract spraying monitoring data, resulting in multiple sets of spraying monitoring data sequences differentiated by surface type. Each set of spraying records corresponds to a specific steel structure surface type, such as a rough surface, a smooth surface, or a sandblasted surface. During data extraction, for each historical spraying record in each set, continuous monitoring data throughout the spraying process is extracted. This data includes, but is not limited to, parameters such as paint spraying pressure, paint flow rate, distance between the nozzle and the steel structure surface, spraying speed, ambient temperature, ambient humidity, and real-time detected coating thickness and uniformity. These parameters are arranged chronologically to form the monitoring data sequence corresponding to each record. Finally, all monitoring data sequences for the same surface type are integrated to obtain the spraying monitoring data sequence set corresponding to that surface type.

[0080] Optimal spraying time data extraction is performed on multiple sets of spraying monitoring data sequences differentiated by surface type to determine multiple historical optimal spraying monitoring data sets. For each monitoring data sequence in each set of spraying monitoring data sequences differentiated by surface type, the moment when the coating uniformity reaches its optimal state needs to be found. The criteria for judging coating uniformity are that the fluctuation of the detected coating thickness within a preset range is minimal at this moment, and the coating coverage is complete without omissions or accumulation. After finding this optimal moment, all monitoring data corresponding to this moment are extracted, including the spraying pressure, flow rate, nozzle distance, environmental parameters, etc., and these data are integrated to form the historical optimal spraying monitoring data for that sequence. This process is performed on all monitoring data sequences under the same surface type to collect all historical optimal spraying monitoring data, thus forming the historical optimal spraying monitoring data set corresponding to that surface type.

[0081] Feature extraction is performed on multiple historical best-performing spraying monitoring datasets to obtain multiple historical best-performing spraying feature sets. These feature sets are then subjected to centralized analysis to determine multiple concentrated values ​​for spraying features. During feature extraction, key features significantly impacting spraying performance are selected from the historical best-performing spraying monitoring data, such as the optimal spraying pressure range, optimal flow rate, optimal nozzle distance range, and optimal ambient temperature and humidity range. These key features are extracted from each historical best-performing spraying monitoring dataset to form a feature vector for each dataset. All feature vectors for the same surface type are integrated into a historical best-performing spraying feature set for that surface type. During centralized analysis, for each key feature, its mean, median, mode, and other statistical measures are calculated within the historical best-performing spraying feature sets. By comparing the stability of these statistical measures, the value that best represents the concentrated trend of the feature is selected as the concentrated value for the spraying feature. For example, for the feature of spraying pressure, if the pressure values ​​in multiple historical best-performing datasets are concentrated in a certain range, the median value of that range is taken as the concentrated value of the feature.

[0082] Using multiple historical best spraying feature sets as indexes, feature association scale diffusion identification is performed on multiple sets of surface type-differentiated spraying monitoring data sequences to determine multiple sets of historical best spraying feature association scales. Each historical best spraying feature reflects the state at the optimal spraying time. Feature association scale diffusion identification aims to find the duration of data in the monitoring data sequence that is associated with that optimal time and affects the optimal spraying effect. Specifically, starting from the optimal spraying time, the monitoring data is traced backward to find the starting time affecting the current optimal state, and simultaneously traced backward to find the ending time of the optimal state's duration. The time interval between these two times is the association scale corresponding to that optimal spraying feature. For example, if the optimal spraying time is at minute 10, and its preceding data begins to affect the optimal state from minute 8, with subsequent data continuing until minute 12, then the association scale is 4 minutes. This operation is performed on each historical best spraying feature to obtain multiple association scales. All association scales under the same surface type are integrated into a set of historical best spraying feature association scales for that surface type.

[0083] Calculate the average of multiple historical best spraying feature association scale sets to obtain multiple spraying feature association scales. For each surface type, calculate the arithmetic mean of all association scale values ​​in the historical best spraying feature association scale set. The resulting average value is the spraying feature association scale for that surface type. This value represents the average duration of effective data associated with the best spraying effect under that surface type, and can be used for subsequent multi-scale feature analysis.

[0084] See Figure 3 This paper presents a comparison of key coating parameters for five different surface types (rough surface, smooth surface, sandblasted surface, rusted surface, and galvanized surface). Through historical data analysis, we determined the optimal coating parameter settings for each surface type.

[0085] Rough and rusted surfaces require higher spraying pressure (0.65-0.7MPa) and a larger paint flow rate (1.4-1.5L / min) to ensure that the paint fully covers the uneven surface.

[0086] Smooth and galvanized surfaces require lower spraying pressure (0.5-0.55 MPa) and lower paint flow rate (1.0-1.1 L / min) to prevent paint buildup.

[0087] Regarding nozzle distance, rough surfaces require a greater distance (22-24cm), while smooth surfaces require a closer distance (18-20cm).

[0088] In terms of coating thickness, rusted surfaces require a thicker coating (85-90μm) to achieve adequate protection, while galvanized surfaces only require a thinner coating (55-60μm).

[0089] These differences reflect the specific requirements of different surface properties for the spraying process, providing a basis for parameter adjustment of automatic spraying equipment.

[0090] See Figure 4 This study demonstrates the changes in key parameters and optimization points of the target steel structure during the spraying process.

[0091] The coating thickness increases steadily over time, eventually reaching about 85 μm.

[0092] Coating uniformity reaches its optimal level (above 94%) during the 35-45 minute period, at which time the spraying pressure and flow rate remain stable.

[0093] The optimal time is at the 41st minute, when the spraying pressure is 0.62 MPa and the paint flow rate is 1.25 L / min.

[0094] Uniformity decreases significantly in the initial (0-10 minutes) and later (50-60 minutes) stages of spraying, requiring adjustment of equipment parameters.

[0095] Based on this analysis, we recommend increasing the spraying pressure by 0.05 MPa in the initial stage of spraying and reducing the paint flow rate by 0.1 L / min in the later stage to improve overall uniformity.

[0096] Example 2: See Figure 5 Feature extraction processing is performed on multiple historical best spraying monitoring datasets to obtain multiple historical best spraying feature sets. Then, centralized analysis is performed on these multiple historical best spraying feature sets to determine multiple concentrated values ​​of the spraying features. The specific process is as follows:

[0097] A feature extraction network layer is used to perform feature extraction processing on multiple historical best spray monitoring datasets to obtain multiple historical best spray feature sets. This network layer consists of an input layer, convolutional layers, pooling layers, and an output layer. The input layer receives raw data from the historical best spray monitoring datasets, including parameters such as coating thickness, spraying pressure, paint flow rate, and spraying distance. The convolutional layers perform sliding calculations on the input data using convolutional kernels of different sizes to extract local features, such as the variation trend of coating thickness in different areas and the fluctuation characteristics of spraying pressure. The pooling layers perform dimensionality reduction on the features output by the convolutional layers, retaining key features while reducing the amount of data. The output layer outputs the processed features as vectors, forming the feature vector corresponding to each historical best spray monitoring dataset. All feature vectors in the same dataset are integrated into the historical best spray feature set.

[0098] The process iterates through multiple historical best spraying feature sets, performing feature mean calculations to determine the average values ​​of these historical best spraying features. For each feature dimension (such as average coating thickness, average spraying pressure, etc.) in each historical best spraying feature set, the values ​​of all feature vectors under that dimension are collected, and the arithmetic mean of these values ​​is calculated. For example, for the feature dimension of coating thickness, if the set contains 100 feature vectors, and the coating thickness values ​​of each vector are 12μm, 13μm, 11μm, etc., then the sum of these 100 values ​​is calculated and divided by 100 to obtain the historical best spraying feature mean for that dimension. This process is performed for all feature dimensions, forming a set containing multiple means, i.e., the historical best spraying feature mean.

[0099] The process iterates through multiple historical best spraying features using a preset iteration scale, yielding multiple iteratively best historical spraying features. The preset iteration scale is the adjustment step size for each feature dimension; for example, the iteration scale for coating thickness is set to 0.5 μm, and the iteration scale for spraying pressure is set to 0.2 MPa. During iteration, the average of the historical best spraying features is used as the initial value. Feature vectors whose differences from the initial value are within the iteration scale range are selected from the historical best spraying feature set, and the average of these vectors is calculated as the new iteration value. This process is repeated, with each iteration using the new iteration value as a benchmark to select closer feature vectors and calculate their average, until the change in the iteration value is less than 1 / 10 of the iteration scale, thus obtaining the iteratively best historical spraying features.

[0100] When the clustering degree of multiple iterated historical best spraying features is less than or equal to the clustering degree of the mean of multiple historical best spraying features, the mean of multiple historical best spraying features is output as the clustered value of multiple spraying features. The clustering degree is measured by the standard deviation of the feature vector. The standard deviation of all iterated historical best spraying features in each dimension is calculated. If the standard deviation is less than or equal to the standard deviation of the mean of historical best spraying features, it indicates that the feature distribution after iteration is more concentrated or has a similar degree of concentration to the initial mean. In this case, there is no need to continue iterating, and the initial mean of historical best spraying features is directly used as the clustered value of spraying features.

[0101] When the clustering degree of multiple iterative best spraying features is greater than the clustering degree of the average of multiple best spraying features, it is determined whether the difference in clustering degree between the multiple iterative best spraying features and the average of multiple best spraying features is greater than or equal to a preset clustering degree difference threshold. The difference value is the difference between the standard deviation of the iterative features and the standard deviation of the initial mean. The preset clustering degree difference threshold is set according to different feature dimensions. For example, the threshold for coating thickness is set to 0.3 μm, and the threshold for spraying pressure is set to 0.1 MPa.

[0102] If the difference in clustering degree is greater than or equal to a preset threshold, iterative processing continues based on multiple historical best spraying features until the preset maximum number of iterations is met. The multiple historical best spraying features obtained in the last iteration are then output as the aggregate value of multiple spraying features. The preset maximum number of iterations is set according to the data size, typically 20. Each time an iteration continues, the iteration scale remains unchanged. Feature vectors are selected based on the current historical best spraying features, and a new mean is calculated until the maximum number of iterations is reached, ensuring that the central tendency in the data is fully exploited.

[0103] If the difference in clustering degree is less than the preset threshold, the iterative process stops, and the best spraying features from multiple iterations are output as the set of multiple spraying features. At this point, although the clustering degree of the iterative features is slightly higher than the initial mean, the difference is small, and continuing the iteration will have limited improvement on the results. Therefore, the current best spraying features from the iteration history are used as the final set of spraying features to balance computational efficiency and result accuracy.

[0104] See Figure 6 This demonstrates the duration of optimal coating conditions for different surface types (feature-related scale), reflecting the ability of each surface type to maintain a stable coating condition.

[0105] The galvanized surface has the longest stable coating time (approximately 14.5 minutes), which is related to its surface uniformity and corrosion resistance.

[0106] The stabilization time for sandblasted and smooth surfaces was the second longest (12-13 minutes), indicating that these surface conditions are beneficial for maintaining coating quality.

[0107] Rough and rusted surfaces have the shortest stabilization time (8-9 minutes) and require more frequent parameter adjustments.

[0108] Feature-related scale data provides key parameters for multi-scale feature analysis, guiding automatic spraying equipment to adjust parameters at different time scales.

[0109] See Figure 7 This demonstrates the correlation between various parameters for different surface types under optimal spraying conditions.

[0110] Spraying pressure and paint flow rate are positively correlated, but the slope varies for different surface types.

[0111] There is a non-linear relationship between coating thickness and uniformity: the uniformity is optimal when the thickness is 60-80μm.

[0112] The nozzle distance has a weak negative correlation with the coating thickness, but no significant correlation with uniformity.

[0113] Different surface types form distinct clusters in the parameter space, validating the importance of distinguishing surface type differences.

[0114] These correlations provide a theoretical basis for multi-scale feature analysis and cross-scale interactive fusion.

[0115] Example 3: Using multiple historical best spraying feature sets as indexes, a feature association scale diffusion identification operation is performed on multiple sets of spraying monitoring data sequences that distinguish surface types to determine multiple sets of historical best spraying feature association scales. The specific process is as follows: A spraying feature extraction network layer is used to perform feature extraction processing on multiple sets of spraying monitoring data sequences that distinguish surface types to obtain multiple sets of spraying feature sequences. This network layer includes an input layer, a convolutional layer, and an activation layer. The input layer receives the raw data from the sets of spraying monitoring data sequences that distinguish surface types, which covers the changes of parameters such as pressure, flow rate, and distance during the spraying process over time. The convolutional layer performs sliding processing on the input data through a set convolutional kernel to extract feature information at different time points. The activation layer uses the ReLU function to perform nonlinear transformation on the convolution results, and finally outputs the feature vector corresponding to each time point. These vectors are arranged in chronological order to form a set of spraying feature sequences that distinguish surface types.

[0116] One historical best spraying feature is randomly extracted from multiple sets of historical best spraying features as a reference historical best spraying feature. Then, the corresponding reference surface-type distinguishing spraying feature sequence is matched against multiple sets of surface-type distinguishing spraying feature sequences. The reference historical best spraying feature is a vector containing multiple parameter features. During the matching process, this vector is compared with the feature vectors at each time point in the surface-type distinguishing spraying feature sequence set, and the feature sequence with the highest similarity is selected as the reference surface-type distinguishing spraying feature sequence.

[0117] According to a preset approximate association range, a nearest neighbor search operation is performed on the reference historical best spraying feature in the reference surface type spraying feature sequence to obtain the neighborhood of the reference historical best spraying feature. The preset approximate association range is an interval on the time axis, for example, set to the range of 10 data points before and after the time point corresponding to the reference historical best spraying feature. Within this range, the similarity between each feature vector and the reference historical best spraying feature is calculated, and the time points of feature vectors with similarity higher than a set threshold are included in the neighborhood, forming the neighborhood of the reference historical best spraying feature.

[0118] The duration of the neighborhood of the historical best spraying feature is statistically referenced, and the statistical results are used as the correlation scale for the historical best spraying feature. The duration is calculated as the time difference between the last time point and the first time point in the neighborhood, and this value is in seconds, reflecting the length of data duration associated with the best spraying time.

[0119] Based on a preset nearest neighbor association range, multiple historical best spraying feature sets are analyzed. Then, feature association scale diffusion identification is performed on the corresponding multiple sets of surface type-differentiated spraying monitoring data sequences to determine multiple historical best spraying feature association scale sets. For each historical best spraying feature, the above processing steps for referencing historical best spraying features are repeated: feature extraction, sequence matching, neighborhood retrieval, and duration statistics. The association scales corresponding to each historical best spraying feature are then integrated to form a historical best spraying feature association scale set.

[0120] A cross-scale interactive fusion operation is performed on multiple spraying monitoring feature sets to obtain a target interactive fused spraying monitoring feature set. The specific process is as follows: An initial spraying monitoring feature set and subsequent spraying monitoring feature sets are randomly extracted from the multiple spraying monitoring feature sets. The initial set can be selected as the set containing the most basic features, and the subsequent sets are feature sets of other different scales. The features in these sets reflect the state information at different time granularities during the spraying process.

[0121] The similarity recognition results between the initial spraying monitoring feature set and the subsequent spraying monitoring feature set are calculated to determine the initial feature similarity set. The similarity calculation uses the cosine similarity formula, as follows:

[0122]

[0123] in, Represents the feature vector in the initial spray monitoring feature set. Feature vectors in the subsequent spraying monitoring feature set Similarity; The dimension of the feature vector; Representing the eigenvector The One component; Representing the eigenvector The There are several components. The similarity value calculated by this formula ranges from 0 to 1, with a higher value indicating a higher degree of similarity between the two feature vectors.

[0124] Normalization is performed on the initial feature similarity set to obtain the initial feature similarity normalized value set. During normalization, each similarity value in the initial feature similarity set is mapped to the interval between 0 and 1. Specifically, the maximum and minimum values ​​in the set are first found, and then each similarity value is subtracted from the minimum value and divided by the difference between the maximum and minimum values ​​to obtain the normalized value.

[0125] The initial set of normalized feature similarity values ​​and the subsequent set of spraying monitoring features are subjected to convolution to obtain the initial interactive fused spraying monitoring feature set. The convolution calculation uses a convolution kernel of a set size, such as a 3×3 convolution kernel. The normalized feature similarity values ​​of the initial feature are used as weights and multiplied with the corresponding elements in the subsequent set of spraying monitoring features. The product results are then summed to obtain the fused feature values. These feature values ​​are arranged in their original sequence to form the initial interactive fused spraying monitoring feature set.

[0126] Additional spraying monitoring feature sets are randomly extracted from multiple spraying monitoring feature sets. These additional feature sets are then combined with the initial interactive fusion spraying monitoring feature set through a cross-scale interactive fusion operation to obtain a subsequent interactive fusion spraying monitoring feature set. The operation process is the same as the initial fusion: first, the similarity between the additional and initial interactive fusion spraying monitoring feature sets is calculated and normalized; then, feature fusion is achieved through convolution to generate a new fused feature set.

[0127] After repeated cross-scale interactive fusion operations, until all spraying monitoring features from multiple spraying monitoring feature sets are fused, a target interactive fused spraying monitoring feature set is obtained. During each fusion, a spraying monitoring feature set that has not yet participated in the fusion is randomly selected and fused with the current interactive fused spraying monitoring feature set. The steps of calculating similarity, normalization, and convolution are repeated until all feature sets are included in the fusion process. The final target interactive fused spraying monitoring feature set contains feature information at different scales and can comprehensively reflect the state of the spraying process. The fused spraying monitoring feature set is then processed.

[0128] Example 4: Multiple sets of sample feature similarity normalized values, multiple sets of sample spraying monitoring features, and multiple sets of sample interactive fusion spraying monitoring features were obtained as training datasets. The sample feature similarity normalized value set was derived from the similarity calculation results between feature sets at different scales during the spraying process of steel structures with different surface types, and was formed after normalization. The sample spraying monitoring feature set included monitoring data of parameters such as spraying pressure, paint flow rate, and coating thickness at different time points, and these data were integrated according to feature type. The sample interactive fusion spraying monitoring feature set was a standard feature set obtained by manually annotating and fusing monitoring features at multiple scales according to the actual spraying effect.

[0129] Supervised training of the fusion network layer built on a convolutional neural network was performed using the training dataset until convergence, resulting in a trained fusion convolutional network layer. The input layer of this network layer receives a set of normalized similarity values ​​of sample features and a set of spraying monitoring features from the samples. The intermediate layers include convolutional layers and pooling layers. The convolutional layers extract the correlation information between features using a set of convolutional kernels, and the pooling layers compress the extracted features to reduce the amount of data. The output layer outputs the fused feature set, which is compared with the sample-interactive fused spraying monitoring feature set, and the network parameters are adjusted through backpropagation. During training, convergence is determined when the deviation between the output result and the sample remains stable within a certain range for several consecutive training epochs.

[0130] The trained fusion convolutional network layer performs convolutional calculations on the initial set of normalized feature similarities and the subsequent set of spray monitoring features to obtain an initial interactive fusion spray monitoring feature set. The initial normalized feature similarities and the subsequent spray monitoring features are input into the network in a fixed format, and after weighted calculations by the convolutional layers and processing by the pooling layers, the initial interactive fusion feature set integrating the two feature information is output.

[0131] The historical spraying record set is differentiated by surface type to determine multiple spraying record sets that differentiate surface types. The specific process is as follows: multiple historical spraying records are randomly selected from the historical spraying record set, and the number of selections is determined based on the total number of records.

[0132] Determine if the record similarity of each enumerated combination is greater than a preset record similarity threshold. Record similarity is calculated by comparing parameters such as surface roughness, material, and preprocessing method in the records. For example, a higher similarity is considered when the difference in surface roughness is within a certain range, the material is the same, and the preprocessing method is consistent. The preset record similarity threshold is set according to the actual spraying requirements, for example, 0.8.

[0133] If no enumerated combination with a similarity greater than the preset record similarity threshold exists, it indicates that the surface types corresponding to these sample records are significantly different and can be directly used as differentiation targets. Based on these differentiation targets, the historical spraying record set is classified according to the preset record similarity threshold. Records with a similarity greater than the threshold to each differentiation target are grouped into the same set, forming multiple spraying record sets that differentiate surface types.

[0134] Table 1 shows sample data from some historical spraying records, including parameters such as surface roughness, material, and pretreatment method.

[0135]

[0136] In the example in Table 1 above, if the preset record similarity threshold is 0.8, the surface roughness of record 1 and record 3 are similar, the materials are the same and the preprocessing methods are the same, and the similarity is greater than the threshold, so they can be classified into the same set of spraying records that distinguish the surface type; the parameter features of record 2 and record 4 are similar, so they are classified into another set; the parameters of record 5 are significantly different from other records, so it is classified into a separate set.

[0137] Example 5: Monitoring the surface vibration signals of the target steel structure during the spraying process yields vibration parameters. The running time and ambient humidity parameters of the spraying environment are also collected. A coating scaling prediction operation is then performed to obtain predicted scaling parameters. Vibration parameters, including frequency and amplitude, are collected by vibration sensors installed at different locations on the steel structure. The sensor sampling frequency is set according to the equipment's operating speed to ensure that subtle vibration changes during the spraying process are captured. The running time parameter is the cumulative running time of the automatic spraying equipment from startup to the current moment, recorded in minutes. The ambient humidity parameter is obtained from multiple humidity sensors deployed in the spraying workshop, and the average value of multiple sensors is taken as the current ambient humidity. The coating scaling prediction operation analyzes the correlation between vibration parameters, running time parameters, ambient humidity parameters, and the amount of coating scaling in historical data to generate predicted scaling parameters under the current spraying conditions. These scaling parameters include the thickness and distribution density of the scaling.

[0138] Industrial vision sensors are used to acquire surface images of the target steel structure. Based on predicted scaling parameters, these images are input into multiple scaling parameter classifiers to identify the actual scaling parameters. The industrial vision sensors are mounted on the moving track of the spraying equipment and move synchronously with the spray nozzle, capturing images of the steel structure surface from different angles. The image resolution is adjusted according to the size of the steel structure to ensure clear representation of surface details. Multiple scaling parameter classifiers correspond to different ranges of predicted scaling parameters; for example, one classifier targets a predicted scaling thickness of 0-5 μm, while another targets 5-10 μm. After the surface image is input into a classifier matching the predicted scaling parameters, the classifier identifies the grayscale changes and morphological features of the scaling areas in the image to output the actual scaling thickness and distribution density. Each scaling parameter classifier contains multiple scaling parameter classification paths, with different paths corresponding to different lighting conditions and shooting angles to adapt to various imaging environments.

[0139] Based on actual scaling parameters, a coating deformation prediction operation is performed to obtain predicted coating deformation parameters. This operation analyzes the influence of scaling distribution on stress distribution during coating curing, based on the correspondence between actual scaling parameters and historical coating deformation data, to predict the potential shrinkage and warpage of the coating. The predicted coating deformation parameters include the shrinkage rate and maximum warpage height in different regions.

[0140] Industrial vision sensors are used to acquire coating images of the target steel structure. Based on predicted coating deformation parameters, these images are input into multiple coating parameter classifiers to identify the actual coating deformation parameters. The coating images are acquired after spraying and initial curing, at which point the deformation trend of the coating is already apparent. Multiple coating parameter classifiers correspond to different ranges of predicted coating deformation parameters; for example, a classifier for predicted shrinkage rates of 0-2% and another for 2-4%. The coating parameter classifiers calculate the actual shrinkage rate and warpage height by comparing the actual contour in the coating image with a standard contour.

[0141] By combining actual scaling parameters and actual coating deformation parameters, the continuous painting control results are adjusted. For example, when the actual scaling thickness exceeds the predicted value and the shrinkage rate in a certain area is large, the spray flow rate and nozzle movement speed in that area are adjusted to reduce paint accumulation in that area; when the actual warpage height is high, the temperature parameters of subsequent sprays are adjusted to slow down the curing speed and release some stress.

[0142] The time window length is set, and a sliding processing operation is performed on the interference quantity according to the time window length to obtain interference quantum data segments. The time window length is set according to the operating speed of the spraying equipment, for example, set to 1 minute, with a sliding step size of 30 seconds, that is, an interference quantum data segment of 1 minute in length is extracted every 30 seconds. The interference quantity includes environmental humidity parameters and vibration parameters. Each sub-data segment contains the humidity change curve and the vibration frequency and amplitude change sequence within that time period.

[0143] The interfering quantum data fragments are analyzed to obtain the periodic and probabilistic characteristics of each interference quantity. Periodic characteristics are obtained by observing the repetitive changes in parameters within the interfering quantum data fragments; for example, ambient humidity tends to increase at a certain time each day, and vibration frequency fluctuates once per hour with the device's operating time. Probabilistic characteristics are obtained by statistically analyzing the percentage of occurrences of interference quantities within different numerical ranges in the sub-data fragments; for example, the probability of occurrence when ambient humidity is between 60-70% and the probability of occurrence when vibration amplitude is between 0.1-0.2 mm.

[0144] The system obtains the current time point, matches the corresponding interference quantity based on the periodic characteristics of the interference quantity, and matches the probability value of the interference quantity's occurrence based on the probability characteristics. For example, if the current time is 10:00 AM, the periodic characteristics indicate that the ambient humidity usually shows an upward trend during this period, so the system matches the common humidity range for this time period; at the same time, based on the probability characteristics, the probability value of this humidity range is 0.7.

[0145] The matched disturbance values ​​and probability values ​​are input into the coating deformation prediction operation to optimize the predicted coating deformation parameters. For example, by incorporating the matched ambient humidity and its corresponding probability value into the analysis, the prediction model can consider the influence of humidity on the coating curing speed when calculating the shrinkage rate, thereby more accurately predicting the coating deformation.

[0146] Multiple sets of sample feature similarity normalized values, multiple sets of sample spraying monitoring features, and multiple sets of sample interactive fusion spraying monitoring features were obtained as training datasets. The sample feature similarity normalized value set is the normalized result of the similarity between different sets of sample spraying monitoring features. The sample spraying monitoring feature set contains monitoring data such as pressure and flow rate during the sample steel structure spraying process. The sample interactive fusion spraying monitoring feature set is a standard set after manually integrating multi-scale features.

[0147] Supervised training of the fusion network layer built on a convolutional neural network was performed using the training dataset until convergence, resulting in the trained fusion convolutional network layer. The input to the fusion network layer consisted of a set of normalized similarity values ​​of sample features and a set of sample spraying monitoring features. The output was the fused feature set. By continuously adjusting the convolutional kernel parameters in the network, the difference between the output and the sample interaction fused spraying monitoring feature set was gradually reduced. Training stopped when the difference stabilized within a certain range.

[0148] The trained fusion convolutional network layer performs convolution operations on the initial set of normalized feature similarities and the subsequent set of spray monitoring features to obtain the initial interactive fusion spray monitoring feature set. After inputting the initial normalized feature similarities and the subsequent spray monitoring features according to the network's required format, the fusion convolutional network layer integrates the closely related parts of the two types of features through multiple convolution and pooling operations to form an initial interactive fusion feature set containing multi-dimensional information.

[0149] The historical spraying record set is differentiated by surface type to determine multiple sets of spraying records for distinguishing surface types. Multiple historical spraying records are randomly selected from these sets; the number is determined by the total number of records, for example, 30 records are selected if there are 200 records in total. These records are then paired to form multiple combinations. The similarity of records in each combination is calculated by comparing the degree of difference in parameters such as surface roughness and material. When the similarity of records in all combinations is not greater than a preset threshold, these records are used as distinction targets. Records in the historical record set that are similar to each distinction target are then grouped into one category according to the preset threshold, forming multiple sets of spraying records for distinguishing surface types.

[0150] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0151] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A continuous painting method for a rail-mounted steel structure automatic painting apparatus, characterized by, The method comprises: acquiring a target structure specification of a target steel structure, retrieving a historical spraying record set in a historical spraying database based on the target structure specification; distinguishing surface types of the historical spraying record set, and determining a plurality of distinguished surface type spraying record sets; performing spraying feature set extraction operations on the plurality of distinguished surface type spraying record sets to obtain a plurality of spraying feature set values and a plurality of spraying feature correlation scales; using the plurality of spraying feature correlation scales as a plurality of analysis scales of a plurality of spraying processing network layers, performing multi-scale feature analysis processing on spraying monitoring data sequences of the target steel structure within a preset spraying period by using the plurality of spraying processing network layers configured to be complete, and obtaining a plurality of spraying monitoring feature sets; performing cross-scale interactive fusion operations on the plurality of spraying monitoring feature sets to obtain a target interactive fusion spraying monitoring feature set, comprising: randomly extracting an initial spraying monitoring feature set and a subsequent spraying monitoring feature set from the plurality of spraying monitoring feature sets; calculating similarity recognition results of the initial spraying monitoring feature set and the subsequent spraying monitoring feature set to determine an initial feature similarity set; performing normalization processing operations on the initial feature similarity set to obtain an initial feature similarity normalization value set; performing convolution calculation operations on the initial feature similarity normalization value set and the subsequent spraying monitoring feature set to obtain an initial interactive fusion spraying monitoring feature set; randomly extracting an additional spraying monitoring feature set from the plurality of spraying monitoring feature sets, and performing cross-scale interactive fusion operations on the additional spraying monitoring feature set and the initial interactive fusion spraying monitoring feature set to obtain a subsequent interactive fusion spraying monitoring feature set; after a plurality of repeated cross-scale interactive fusion operations, until all spraying monitoring features in the plurality of spraying monitoring feature sets are fused and processed, the target interactive fusion spraying monitoring feature set is obtained; outputting the target interactive fusion spraying monitoring feature set as a continuous paint spraying control result.

2. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 1, wherein performing spraying feature set extraction operations on the plurality of distinguished surface type spraying record sets to obtain a plurality of spraying feature set values and a plurality of spraying feature correlation scales, comprising: performing spraying monitoring data extraction operations on the plurality of distinguished surface type spraying record sets to obtain a plurality of distinguished surface type spraying monitoring data sequence sets; performing best spraying time data extraction operations on the plurality of distinguished surface type spraying monitoring data sequence sets respectively to determine a plurality of historical best spraying monitoring data sets, wherein each historical best spraying monitoring data corresponds to monitoring data at a time when coating uniformity in each distinguished surface type spraying monitoring data sequence is optimal; performing feature extraction processing on the plurality of historical best spraying monitoring data sets to obtain a plurality of historical best spraying feature sets, and performing centralized analysis processing on the plurality of historical best spraying feature sets to determine a plurality of spraying feature set values; perform feature correlation scale diffusion identification operations on the multiple sets of surface type differentiated spraying monitoring data sequences with the multiple sets of historical best spraying features as indexes to determine a set of historical best spraying feature correlation scales, wherein each historical best spraying feature correlation scale reflects a data duration associated with a best spraying time in the surface type differentiated spraying monitoring data sequence; respectively calculate average values of the set of historical best spraying feature correlation scales to obtain the multiple spraying feature correlation scales.

3. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 2, wherein perform feature extraction processing on the multiple sets of historical best spraying monitoring data to obtain a set of historical best spraying features, and perform centralized analysis processing on the set of historical best spraying features to determine a set of spraying feature central values, including: perform feature extraction processing on the multiple sets of historical best spraying monitoring data using a spraying feature extraction network layer to obtain a set of historical best spraying features; perform feature mean value calculation operations on the set of historical best spraying features to determine a set of historical best spraying feature means; perform iteration processing on the set of historical best spraying feature means in the set of historical best spraying features according to a preset iteration scale to obtain a set of iterated historical best spraying features; when the aggregation degree of the set of iterated historical best spraying features is less than or equal to the aggregation degree of the set of historical best spraying feature means, output the set of historical best spraying feature means as the set of spraying feature central values.

4. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 3, wherein including: when the aggregation degree of the set of iterated historical best spraying features is greater than the aggregation degree of the set of historical best spraying feature means, determine whether a difference value of the aggregation degrees of the set of iterated historical best spraying features and the set of historical best spraying feature means is greater than or equal to a preset aggregation degree difference threshold value; if greater than or equal to the preset aggregation degree difference threshold value, continue to perform iteration processing based on the set of iterated historical best spraying features until a preset maximum iteration number is met, and output the set of iterated historical best spraying features obtained in the last iteration as the set of spraying feature central values; if less than the preset aggregation degree difference threshold value, stop the iteration processing, and output the set of iterated historical best spraying features as the set of spraying feature central values.

5. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 4, wherein perform feature correlation scale diffusion identification operations on the multiple sets of surface type differentiated spraying monitoring data sequences with the multiple sets of historical best spraying features as indexes to determine a set of historical best spraying feature correlation scales, including: perform feature extraction processing on the multiple sets of surface type differentiated spraying monitoring data sequences using the spraying feature extraction network layer to obtain a set of surface type differentiated spraying feature sequences; randomly extract a historical best spraying feature from the set of historical best spraying features as a reference historical best spraying feature, and match a corresponding reference surface type differentiated spraying feature sequence from the set of surface type differentiated spraying feature sequences; According to the preset approximate correlation range, a nearest neighbor search operation is performed on the reference historical best spraying feature in the reference surface type distinguishing spraying feature sequence, to obtain a reference historical best spraying feature neighborhood; The duration of the reference historical best spraying feature neighborhood is counted, and the counting result is taken as a reference historical best spraying feature correlation scale; According to the preset nearest neighbor correlation range, a feature correlation scale diffusion identification operation is performed on the plurality of historical best spraying feature sets in the corresponding plurality of surface type distinguishing spraying monitoring data sequence sets, to determine a plurality of historical best spraying feature correlation scale sets.

6. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 1, wherein Also includes: Obtain a plurality of sample feature similarity normalization value sets, a plurality of sample spraying monitoring feature sets and a plurality of sample interactive fusion spraying monitoring feature sets as training data sets; Using the training data set, a supervised training operation is performed on the fusion network layer based on the convolutional neural network, until the training is converged, and a trained fusion convolutional network layer is obtained; Using the trained fusion convolutional network layer, a convolution calculation operation is performed on the initial feature similarity normalization value set and the subsequent spraying monitoring feature set, to obtain the initial interactive fusion spraying monitoring feature set.

7. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 1, wherein The surface type difference of the historical spraying record set is distinguished, and a plurality of surface type distinguishing spraying record sets are determined, including: Randomly extracting a plurality of historical spraying records from the historical spraying record set; Performing a two-by-two enumeration operation on the plurality of historical spraying records to obtain a plurality of enumeration combinations; Judging whether there is an enumeration combination with a record similarity greater than a preset record similarity threshold in the plurality of enumeration combinations; If there is no enumeration combination with a record similarity greater than a preset record similarity threshold, the plurality of historical spraying records are output as a plurality of distinguishing targets; Based on the plurality of distinguishing targets, a surface type difference distinguishing operation is performed on the historical spraying record set according to the preset record similarity threshold, to obtain the plurality of surface type distinguishing spraying record sets, wherein each surface type distinguishing spraying record set corresponds to a distinguishing target.

8. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 1, wherein Also includes: Performing a monitoring operation on the surface vibration signal of the target steel structure in the spraying process to obtain vibration parameters, and collecting the running time parameters and environmental humidity parameters of the spraying environment to perform a coating fouling prediction operation to obtain a predicted fouling parameter; Through an industrial vision sensor, the surface image of the target steel structure is collected, and the surface image is input into a plurality of fouling parameter classifiers according to the predicted fouling parameter, to identify and obtain an actual fouling parameter, wherein each fouling parameter classifier includes a plurality of fouling parameter classification paths; According to the actual fouling parameter, a coating deformation prediction operation is performed to obtain a predicted coating deformation parameter; Through an industrial vision sensor, the coating image of the target steel structure is collected, and the coating image is input into a plurality of coating parameter classifiers according to the predicted coating deformation parameter, to identify and obtain an actual coating deformation parameter; The continuous paint spraying control result is adjusted in combination with the actual fouling parameter and the actual coating deformation parameter.

9. A continuous painting method for a rail-type steel structure automatic painting apparatus according to claim 8, wherein Also includes: A time window length is set, a sliding processing operation is performed on the interference quantity according to the time window length, and interference quantity sub-data segments are obtained, wherein the interference quantity includes an environmental humidity parameter and a vibration parameter; An analysis processing operation is performed on the interference quantity sub-data segments, and period characteristics and probability characteristics of each interference quantity are obtained; A current time point is obtained, a corresponding interference quantity is matched according to the period characteristics of the interference quantity, and a probability value of occurrence of the interference quantity is matched according to the probability characteristics; The matched interference quantity and the probability value are input into the coating deformation prediction operation, and the predicted coating deformation parameter is optimized.

Images