Method and system for machining control of water turbine graphene ceramic coating based on image analysis

Abstract

The application provides a water turbine graphene ceramic coating processing control method and system based on image analysis, and relates to the technical field of water turbine coating processing.The method comprises the following steps: obtaining high-definition microscopic imaging data of a coating surface of a water turbine flow part; determining a plurality of graphene ceramic coating failure areas and coating failure information of each graphene ceramic coating failure area based on the high-definition microscopic imaging data of the coating surface of the water turbine flow part; obtaining a first coating peeling operation video; determining a coating removal second sand blasting pressure of each test coating peeling area based on the first coating peeling operation video; obtaining a second coating peeling operation video; and determining a coating removal target sand blasting pressure of each residual coating peeling area based on the first coating peeling operation video and the second coating peeling operation video.The method can accurately identify the graphene ceramic coating failure areas of the water turbine and reasonably determine the coating peeling sand blasting pressure.

Description

Technical Field

[0001] This invention relates to the field of turbine coating processing technology, specifically to a processing control method and system for graphene ceramic coatings for turbines based on image analysis. Background Technology

[0002] The protection and repair of graphene ceramic coatings on hydro turbines is a core aspect of the daily operation and maintenance of hydroelectric generator units, directly affecting the service life of flow-through components and the operational stability of the unit. Graphene ceramic coatings possess excellent wear resistance, corrosion resistance, and structural stability, making them a mainstream protective measure for key hydro turbine components against high-speed water flow erosion and sediment wear. After long-term service, coatings may experience failures such as localized peeling, thinning due to wear, and crack propagation, necessitating precise peeling and recoating to regenerate the components. Traditional coating repair work relies primarily on technicians' manual experience to determine the location and peeling strength, using uniformly set sandblasting pressures for batch processing. However, this method has significant limitations. Manual judgment struggles to accurately identify the type, degree, and spatial distribution of coating failures, and cannot differentiate pressure matching for different areas of coating condition, easily leading to over-peeling that damages the substrate or incomplete peeling. This standardized operation mode lacks refined condition perception and hierarchical control capabilities, resulting in unstable coating repair quality and a high rework rate. Meanwhile, manual operation and maintenance relies on the experience level of on-site personnel, resulting in poor consistency and weak controllability. This makes it difficult to adapt to the high-precision and high-reliability intelligent processing and standardized repair requirements of turbine coatings, further restricting the operation and maintenance efficiency and long-term safety of generator sets.

[0003] Therefore, accurately identifying the failure areas of the graphene ceramic coating on the turbine and reasonably determining the coating peeling sandblasting pressure are urgent problems to be solved. Summary of the Invention

[0004] The main technical problem solved by this invention is how to accurately identify the failure area of ​​the graphene ceramic coating of a water turbine and reasonably determine the coating peeling sandblasting pressure.

[0005] According to a first aspect, the present invention provides a processing control method for graphene ceramic coatings on a water turbine based on image analysis, comprising: acquiring high-resolution microscopic imaging data of the coating surface of a water turbine flow passage component; determining multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow passage component; determining multiple test coating peeling regions and a first blasting pressure for coating removal in each test coating peeling region based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow passage component, the multiple graphene ceramic coating failure regions, and the coating failure information in each graphene ceramic coating failure region; and applying the first blasting pressure for coating removal to each test coating peeling region. The system performs a first coating stripping operation and acquires a video of the first coating stripping operation. Based on the video of the first coating stripping operation, it determines a second sandblasting pressure for coating removal in each test coating stripping area. Based on the second sandblasting pressure for coating removal in each test coating stripping area, it performs a second coating stripping operation in each test coating stripping area and acquires a video of the second coating stripping operation. Based on the first coating stripping operation video and the second coating stripping operation video, it determines a target sandblasting pressure for coating removal in each remaining coating stripping area. Based on the target sandblasting pressure for coating removal in each remaining coating stripping area, it performs a coating stripping operation in each remaining coating stripping area. After the stripping is completed, it performs a recoating operation on the multiple graphene ceramic coating failure areas.

[0006] In one possible implementation, determining multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region based on high-resolution microscopic imaging data of the turbine flow-through component coating surface includes: generating a failure risk distribution map of the in-service graphene ceramic coating of the turbine based on the high-resolution microscopic imaging data of the turbine flow-through component coating surface; determining multiple suspected failure coating feature point information based on the high-resolution microscopic imaging data of the turbine flow-through component coating surface and the failure risk distribution map of the in-service graphene ceramic coating of the turbine; clustering the multiple suspected failure coating feature point information to obtain K failure coating feature clusters; acquiring coating surface damage detection data for each failure coating feature cluster; and determining multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region based on the coating surface damage detection data for each failure coating feature cluster.

[0007] In one possible implementation, determining the target sandblasting pressure for removing the coating in each remaining coating stripping area based on the first and second coating stripping operation videos includes: determining the initial sandblasting pressure for removing the coating in each remaining coating stripping area based on the coating failure information of each graphene ceramic coating failure area, the first sandblasting pressure for removing the coating in each test coating stripping area, and the second sandblasting pressure for removing the coating in each test coating stripping area; generating a coating stripping operation simulation video for each remaining coating stripping area based on the first, second, and initial sandblasting pressures for removing the coating in each remaining coating stripping area; constructing a sandblasting pressure map, which includes multiple remaining coating stripping area nodes and edges between multiple nodes, wherein the node characteristics of each remaining coating stripping area node are the coating failure information of each remaining coating stripping area and the coating stripping operation simulation video of each remaining coating stripping area, and the edges between nodes are the positional relationship information between the remaining coating stripping areas; and processing the sandblasting pressure map based on the sandblasting pressure determination model to obtain the target sandblasting pressure for removing the coating in each remaining coating stripping area.

[0008] In one possible implementation, the sandblasting pressure determination model is a graph neural network model.

[0009] According to a second aspect, the present invention provides a processing control system for graphene ceramic coatings of a water turbine based on image analysis, comprising: an imaging data acquisition module for acquiring high-resolution microscopic imaging data of the coating surface of a water turbine flow-through component; a failure analysis module for determining multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow-through component; a first sandblasting pressure determination module for determining multiple test coating peeling areas and a first sandblasting pressure for removing coatings in each test coating peeling area based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow-through component, the multiple graphene ceramic coating failure areas, and the coating failure information of each graphene ceramic coating failure area; and a first peeling execution module for performing peeling execution on each test coating peeling area based on the first sandblasting pressure for removing coatings in each test coating peeling area. The system comprises the following modules: a first coating stripping operation and acquisition of a video of the first coating stripping operation; a second sandblasting pressure determination module for determining the second sandblasting pressure for removing the coating in each test coating stripping area based on the video of the first coating stripping operation; a second stripping execution module for performing a second coating stripping operation on each test coating stripping area based on the second sandblasting pressure for removing the coating in each test coating stripping area and acquisition of a video of the second coating stripping operation; a target sandblasting pressure determination module for determining the target sandblasting pressure for removing the coating in each remaining coating stripping area based on the video of the first coating stripping operation and the video of the second coating stripping operation; and a final coating application module for performing a coating stripping operation on each remaining coating stripping area based on the target sandblasting pressure for removing the coating in each remaining coating stripping area, and performing a recoating operation on the multiple graphene ceramic coating failure areas after the stripping is completed.

[0010] In one possible implementation, the failure analysis module is further configured to: generate a failure risk distribution map of the in-service graphene ceramic coating of the turbine based on high-resolution microscopic imaging data of the coating surface of the turbine flow-through component; determine multiple suspected failure coating feature point information based on the high-resolution microscopic imaging data of the coating surface of the turbine flow-through component and the failure risk distribution map of the in-service graphene ceramic coating of the turbine; cluster the multiple suspected failure coating feature point information to obtain K failure coating feature clusters; acquire coating surface damage detection data for each failure coating feature cluster; and determine multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area based on the coating surface damage detection data for each failure coating feature cluster.

[0011] In one possible implementation, the target sandblasting pressure determination module is further configured to: determine the initial sandblasting pressure for removing the coating from each remaining coating peeling area based on the coating failure information of each graphene ceramic coating failure area, the first sandblasting pressure for removing the coating from each test coating peeling area, and the second sandblasting pressure for removing the coating from each test coating peeling area; generate a coating peeling operation simulation video for each remaining coating peeling area based on the first coating peeling operation video, the second coating peeling operation video, and the initial sandblasting pressure for removing the coating from each remaining coating peeling area; construct a sandblasting pressure map, the sandblasting pressure map including multiple remaining coating peeling area nodes and edges between multiple nodes, the node characteristics of each remaining coating peeling area node being the coating failure information of each remaining coating peeling area and the coating peeling operation simulation video of each remaining coating peeling area, and the edges between nodes being the positional relationship information between the remaining coating peeling areas; and process the sandblasting pressure map based on the sandblasting pressure determination model to obtain the target sandblasting pressure for removing the coating from each remaining coating peeling area.

[0012] In one possible implementation, the sandblasting pressure determination model is a graph neural network model.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: acquiring high-resolution microscopic imaging data of the surface of a coating of a turbine flow component; determining, based on the high-resolution microscopic imaging data of the surface of the coating of the turbine flow component, a plurality of graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region; determining, based on the high-resolution microscopic imaging data of the surface of the coating of the turbine flow component, the plurality of graphene ceramic coating failure regions, and the coating failure information for each graphene ceramic coating failure region, a plurality of test coating peeling regions and a first blasting pressure for coating removal of each test coating peeling region; and ... first blasting pressure for coating removal of each test coating; and determining, based on the first blasting pressure for coating removal of each test coating... The coating removal process involves applying a first sandblasting pressure to each test coating peeling area and acquiring a video of this process. Based on this video, a second sandblasting pressure is determined for each test coating peeling area. A second coating peeling operation is then performed on each test coating peeling area based on this second sandblasting pressure, and a video of this second peeling operation is acquired. Based on the first and second sandblasting operation videos, a target sandblasting pressure for removing the coating in each remaining coating peeling area is determined. Finally, a coating peeling operation is performed on each remaining coating peeling area based on this target sandblasting pressure. After peeling is complete, a recoating operation is performed on the multiple graphene ceramic coating failure areas.

[0014] According to a fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned processing control method for graphene ceramic coatings on a water turbine based on image analysis. The method includes: acquiring high-resolution microscopic imaging data of the coating surface of a water turbine flow-through component; determining multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow-through component; determining multiple test coating peeling areas and a first sandblasting pressure for coating removal in each test coating peeling area based on the high-resolution microscopic imaging data of the coating surface of the water turbine flow-through component, the multiple graphene ceramic coating failure areas, and the coating failure information in each graphene ceramic coating failure area; and determining multiple test coating peeling areas and a first sandblasting pressure for coating removal in each test coating peeling area. The process involves: applying a first blasting pressure to remove the coating in each test coating peeling area and acquiring a video of the first coating peeling operation; determining a second blasting pressure for coating removal in each test coating peeling area based on the first coating peeling operation video; performing a second coating peeling operation on each test coating peeling area based on the second blasting pressure for coating removal in each test coating peeling area and acquiring a video of the second coating peeling operation; determining a target blasting pressure for coating removal in each remaining coating peeling area based on the first and second coating peeling operation videos; performing a coating peeling operation on each remaining coating peeling area based on the target blasting pressure for coating removal in each remaining coating peeling area; and recoating the graphene ceramic coating in the multiple graphene ceramic coating failure areas after peeling is completed.

[0015] This invention provides a processing control method and system for graphene ceramic coatings on water turbines based on image analysis. The method includes acquiring high-resolution microscopic imaging data of the coating surface of the water turbine's flow-through components; determining multiple graphene ceramic coating failure areas and coating failure information for each failure area based on the high-resolution microscopic imaging data; determining multiple test coating peeling areas and a first blasting pressure for coating removal in each test coating peeling area based on the high-resolution microscopic imaging data, the multiple graphene ceramic coating failure areas, and the coating failure information in each failure area; performing a first coating peeling operation on each test coating peeling area based on the first blasting pressure for coating removal in each test coating peeling area, and acquiring a first coating... The method involves several steps: first, determining the coating removal second sandblasting pressure for each test coating peeling area based on the first coating peeling operation video; second, performing a second coating peeling operation on each test coating peeling area based on the second sandblasting pressure for each test coating peeling area, and acquiring a second coating peeling operation video; third, determining the target sandblasting pressure for coating removal for each remaining coating peeling area based on the first and second coating peeling operation videos; fourth, performing a coating peeling operation on each remaining coating peeling area based on the target sandblasting pressure for coating removal; and finally, recoating the graphene ceramic coating in the multiple graphene ceramic coating failure areas after peeling is completed. This method can accurately identify the failure areas of the graphene ceramic coating in the turbine and reasonably determine the coating peeling sandblasting pressure.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0017] 1) This invention uses high-definition microscopic imaging and image analysis technology, which can accurately identify the coating failure area, failure type and damage degree, replacing manual experience judgment and significantly improving the accuracy of failure location and condition assessment.

[0018] 2) This invention achieves differentiated pressure matching for different failure areas by using two-stage test peeling and video closed-loop feedback to dynamically optimize sandblasting pressure, thereby avoiding excessive peeling that damages the substrate or incomplete peeling, and improving repair quality and consistency.

[0019] 3) This invention introduces a graph neural network model to process spatial topological relationships. By combining the operation simulation video to generate the globally optimal target sandblasting pressure, it can achieve multi-region collaborative control to adapt to the intelligent repair of complex curved surface components of water turbines.

[0020] 4) This invention reduces reliance on human experience by automating perception, decision-making and execution throughout the entire process, thereby reducing rework rates and improving the efficiency of turbine coating repair and operation and maintenance safety.

[0021] 5) By precisely peeling off the coating before spraying, this invention can fully utilize the wear-resistant and corrosion-resistant advantages of graphene ceramic coating, thereby extending the service life of turbine flow components and reducing the total life-cycle maintenance cost of turbine flow components. Attached Figure Description

[0022] Figure 1 A schematic flowchart illustrating a processing control method for graphene ceramic coatings on water turbines based on image analysis, provided in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a water turbine provided in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of a process for determining multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region, provided as an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of a process for determining the target sandblasting pressure for removing coatings in each remaining coating peeling area, provided by an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of a processing control system for a water turbine graphene ceramic coating based on image analysis, provided as an embodiment of the present invention. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0028] In this embodiment of the invention, the following are provided: Figure 1 The method shown is a processing control method for graphene ceramic coatings on water turbines based on image analysis. The processing control method for graphene ceramic coatings on water turbines based on image analysis includes steps S1 to S8:

[0029] Step S1: Obtain high-resolution microscopic imaging data of the coating surface of the turbine flow-through components.

[0030] A water turbine is a rotary power machine that can convert the kinetic and potential energy of water flow into mechanical energy. Figure 2 This is a schematic diagram of a water turbine provided in an embodiment of the present invention.

[0031] Flow-through components refer to the key structural parts in a water turbine that are in direct contact with the water flow, participate in energy conversion, or guide the water flow path. Flow-through components include runner blades, guide vanes, volute, and draft tube.

[0032] In some embodiments, the coating of the turbine flow-through component is a graphene ceramic coating.

[0033] Graphene ceramic coating is a ceramic-based composite coating reinforced with graphene. It combines high hardness, high toughness, excellent resistance to cavitation, and chemical stability.

[0034] High-resolution microscopic imaging data of the coating surface of turbine flow passage components is obtained by using an industrial endoscope in conjunction with a high-resolution camera to capture images of the microscopic morphology of the coating surface of turbine flow passage components.

[0035] High-resolution microscopic imaging data of the coating surface of turbine flow components can clearly show the microscopic morphology of the coating surface, including details such as coating smoothness, cracks, peeling, and bulging.

[0036] Step S2: Based on the high-definition microscopic imaging data of the coating surface of the turbine flow component, determine multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area.

[0037] In some embodiments, Figure 3 This is a schematic flowchart illustrating a process for determining multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region, as provided in an embodiment of the present invention. The determination of multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region includes steps S21-S25:

[0038] Step S21: Generate a failure risk distribution map of the graphene ceramic coating of the turbine in service based on the high-definition microscopic imaging data of the coating surface of the turbine flow component.

[0039] In some embodiments, a coating failure risk analysis model can be used to generate a failure risk distribution map of the graphene ceramic coating currently in service with the turbine. The coating failure risk analysis model is a deep neural network model. The input to the coating failure risk analysis model is high-resolution microscopic imaging data of the coating surface of the turbine's flow-through components, and the output of the coating failure risk analysis model is a failure risk distribution map of the graphene ceramic coating currently in service with the turbine.

[0040] Deep neural network models include deep neural networks (DNNs). A deep neural network is an artificial neural network containing multiple hidden layers that can automatically extract high-order abstract features from raw data layer by layer. Deep neural networks possess powerful learning and generalization capabilities, enabling them to handle large amounts of multimodal data tasks with subtle differences.

[0041] The failure risk distribution map of graphene ceramic coatings in active hydro turbines is a visual distribution map used to represent the failure probability and spatial distribution of graphene ceramic coatings at various locations on the surface of hydro turbine flow-through components. Each location in the distribution map corresponds to the actual spatial location of the coating surface on the hydro turbine flow-through component and can reflect the degree of failure risk of the graphene ceramic coating at that location.

[0042] High-resolution microscopic imaging data of the coating surface of turbine flow components contains direct visual evidence of microscopic damage to the coating. This data records the true morphology of the coating surface with high spatial resolution and clearly reveals microscopic features closely related to failure, such as microcracks, cavitation pits, spalling edges, pores and voids, and areas of abnormal surface roughness. These features are physical precursors to coating failure or impending failure. Models can automatically extract failure-related features and quantify risks from this high-resolution microscopic imaging data of turbine flow component coating surfaces.

[0043] Deep neural networks can perform layer-by-layer local perception and global feature extraction from high-resolution microscopic imaging data of the coating surface of turbine flow components. In shallow layers, the model can identify edge features in the high-resolution microscopic imaging data, such as the contours of microcracks on the coating surface and the edge lines of spalling areas. As the data is passed to deeper layers, the model begins to abstract more complex geometric patterns and compares irregular texture variations with known coating fatigue and wear characteristics. Deep neural networks can learn the mapping relationship between microscopic morphology and macroscopic failure risk through backpropagation. When processing the current high-resolution microscopic imaging data, the model can map the extracted multi-layer feature data into a risk assessment space, thereby calculating the failure probability corresponding to each pixel region. For areas with dense microcracks and abnormal textures, the model outputs a higher failure probability value, while for areas with smooth and intact surfaces, the model outputs a lower value. Deep neural networks can map these spatial failure probability values ​​to actual coordinate positions, thereby generating a complete two-dimensional spatial mapping result, ultimately accurately deriving and outputting a failure risk distribution map of the graphene ceramic coating currently in service with the turbine.

[0044] Step S22: Based on the high-definition microscopic imaging data of the coating surface of the turbine flow component and the failure risk distribution map of the existing graphene ceramic coating of the turbine, determine the feature point information of multiple suspected failure coatings.

[0045] In some embodiments, a feature point determination model can be used to determine multiple suspected failure coating feature point information. The feature point determination model is a convolutional neural network model. The input to the feature point determination model is high-resolution microscopic imaging data of the coating surface of the turbine's flow-through components and a failure risk distribution map of the turbine's existing graphene ceramic coating. The output of the feature point determination model is multiple suspected failure coating feature point information.

[0046] Convolutional Neural Network (CNN) models are a type of deep learning model used to process data with a grid-like structure. A CNN consists of alternating layers of convolutional layers, pooling layers, and fully connected layers. CNNs can progressively abstract high-level global semantic features from low-level local features.

[0047] Multiple suspected failure coating feature point information identifies specific locations on the surface of the graphene ceramic coating of the turbine flow components where failure may occur. Each suspected failure coating feature point information includes local texture feature information, suspected defect type, and failure probability value for that suspected failure coating feature point.

[0048] Local texture feature information refers to the surface topography of the coating centered on suspected failure points. Local texture feature information includes the number of texture pixels, the texture orientation coordinate sequence, and the texture grayscale intensity sequence.

[0049] Suspected defect types include microcracks, cavitation pits, coating peeling, coating blistering, and surface porosity.

[0050] The failure probability value is a numerical value used to indicate the likelihood of coating failure at a suspected failure point.

[0051] High-resolution microscopic imaging data of the coating surface of the turbine's flow-through components comprehensively records the microscopic geometry, microcrack propagation paths, and light and shadow variations on the material surface. This high-resolution microscopic imaging data reflects the continuous physical state of the coating surface and can characterize the integrity and surface uniformity of the coating structure of the flow-through components. The failure risk distribution map of the graphene ceramic coating currently in service with the turbine records the failure probability at various locations on the coating surface in numerical form. The failure probability values ​​at different locations differ spatially, and the failure risk distribution map reflects the spatial variation of the coating's failure probability.

[0052] Convolutional neural networks (CNNs) can perform overall feature analysis on high-resolution microscopic imaging data of the coating surface of turbine flow components and the failure risk distribution map of existing graphene ceramic coatings in turbines. The CNN can extract surface texture variations and defect-related features from the high-resolution microscopic imaging data of the turbine flow component coating surface and identify points with physical anomalies on the coating surface. The model can extract the distribution features of failure probability from the failure risk distribution map of existing graphene ceramic coatings in turbines, thereby identifying points with a high failure probability. Then, the CNN can align and match microscopic defect features with spatial features of failure probability and determine whether microscopic anomaly points and high failure probability points are in the same spatial location. For points with consistent spatial locations, the CNN will mark them as points with a failure tendency. The CNN can traverse and filter all points within the entire map to retain points that meet the anomaly criteria. The model can then perform numerical processing and format standardization on the retained points, thereby identifying multiple suspected failure coating feature points.

[0053] Step S23: Cluster the multiple suspected failure coating feature point information to obtain K failure coating feature clusters.

[0054] The clustering method used is K-means clustering. K-means clustering is an unsupervised learning iterative algorithm whose goal is to divide a dataset into K non-overlapping subsets. K-means clustering achieves data classification by minimizing the sum of squared distances between each sample point and the center of its cluster.

[0055] The K failure coating feature clusters are K feature sets obtained by dividing multiple suspected failure coating feature points according to feature similarity using the K-means clustering algorithm. Each failure coating feature cluster contains multiple suspected failure coating feature points that are close in location, have similar local texture features, consistent suspected defect types, and similar failure probabilities. The suspected failure coating feature points in different failure coating feature clusters have obvious feature differences.

[0056] In some embodiments, the value of K can be determined by a preset relationship table between the value of K and the spatial distribution density of suspected failed coating feature points on the surface of the turbine flow-through components. The higher the spatial distribution density of suspected failed coating feature points on the surface of the turbine flow-through components, the larger the value of K. The preset relationship table between the value of K and the spatial distribution density of suspected failed coating feature points on the surface of the turbine flow-through components is artificially constructed in advance.

[0057] The process of clustering multiple suspected failed coating feature points using the K-means clustering algorithm is as follows: First, K points are randomly selected from the suspected failed coating feature point dataset as initial cluster centers. Next, for each suspected failed coating feature point in the dataset, Euclidean distance is used to measure and calculate its distance to these K initial cluster centers, and the suspected failed coating feature point is assigned to the corresponding cluster according to the principle of closest proximity. After all suspected failed coating feature points have been partitioned, the average value of each feature point within each cluster is recalculated, and the cluster center of each cluster is updated accordingly. This process of partitioning and updating cluster centers is repeated until the change in cluster centers is minimal. At this point, the clustering process is considered to have converged, and K-means clustering is complete.

[0058] By clustering multiple suspected coating failure feature points into K failure coating feature clusters, it is possible to group suspected failure coating feature points that are close in location, have similar local texture features, consistent suspected defect types, and similar failure probabilities into the same cluster, thereby achieving structured grouping of scattered failure points. This grouping method can reduce the interference from single-point data and highlight the overall characteristics of similar damage areas, facilitating subsequent damage detection and failure area determination at the cluster level, thus improving the efficiency and accuracy of coating failure analysis.

[0059] Step S24: Obtain the coating surface damage detection data for each failed coating feature cluster.

[0060] The surface damage detection data for each failed coating feature cluster is obtained by detecting the corresponding area of ​​each failed coating feature cluster using an integrated ultrasonic thickness gauge and laser profilometry probe. This data reflects the degree of coating damage for each failed coating feature cluster. The surface damage detection data for each failed coating feature cluster includes surface roughness distribution, average damage depth, remaining coating thickness, and the percentage of damaged area.

[0061] Step S25: Based on the coating surface damage detection data of each failed coating feature cluster, determine multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area.

[0062] In some embodiments, a coating failure region determination model can be used to determine multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region. The coating failure region determination model is a Transformer model. The input to the coating failure region determination model is the coating surface damage detection data of each failed coating feature cluster, and the output of the coating failure region determination model is the multiple graphene ceramic coating failure regions and coating failure information for each graphene ceramic coating failure region.

[0063] The Transformer model is a neural network model based on the self-attention mechanism. It consists of an encoder and a decoder. The encoder uses self-attention to capture dependencies between different positions in the input sequence, while the decoder generates the target sequence based on the encoder's output. The Transformer model can process sequence data in parallel and effectively capture long-range dependencies, demonstrating excellent performance in sequence modeling and feature fusion tasks.

[0064] Multiple graphene ceramic coating failure areas were determined by the coating failure area determination model, which shows that multiple areas on the surface of the turbine flow components have experienced graphene ceramic coating failure.

[0065] The coating failure information for each graphene ceramic coating failure region is specific information about the damage status of each failure region determined by the coating failure region determination model. The coating failure information for each graphene ceramic coating failure region includes the spatial coordinates of the region, the region extent, the failure type, the degree of damage, and the remaining coating thickness.

[0066] Failure type is a classification of the actual failure modes of graphene ceramic coatings. Failure types include microcrack failure, cavitation failure, coating peeling failure, coating blistering failure, and surface porosity failure.

[0067] The surface damage detection data for each failed coating feature cluster provides a deep physical indicator of the health status of the graphene ceramic coating and is an important supplement to the visual information presented by high-resolution microscopic imaging data of the coating surface of turbine flow components. The average damage depth in the surface damage detection data is directly related to the severity of the graphene ceramic coating failure and can correct for potential errors in visual identification based on high-resolution microscopic imaging data. Through the surface damage detection data, the coating failure area determination model can obtain qualitative and quantitative evidence of whether permanent physical damage has occurred to the internal structure of the graphene ceramic coating. This makes the final determination of the graphene ceramic coating failure area and the coating failure information within that area more objective and accurate, ensuring that subsequent repair work does not overlook hidden damage.

[0068] The Transformer model, leveraging its powerful multi-head self-attention mechanism, can perform multi-dimensional feature extraction and correlation analysis on the surface damage detection data of each failed coating feature cluster. The model can use an attention matrix to calculate the correlation weights between various physical indicators in the surface damage detection data of the failed coating feature cluster. Through an encoding layer, the Transformer model can perform high-order modeling of these physical features to uncover the potential nonlinear relationship between coating damage degree and coating failure risk. During decoding, the model can re-evaluate the failure risk degree of each feature cluster by combining pre-defined failure judgment criteria. The model compares the damage degree between different clusters, and for clusters with damage depth exceeding a set threshold and a certain proportion of coating damage area, the Transformer model will classify them as true failure areas. Simultaneously, the model can deduce the specific failure type and damage degree of the failure area based on the distribution characteristics of the coating surface damage detection data, for example, by analyzing the distribution characteristics of damage depth and surface roughness to distinguish between abnormal surface damage and structural failure. Finally, the model maps these logical judgment results back to physical space, delineates the precise area range, and generates corresponding attribute information, thereby accurately determining multiple graphene ceramic coating failure areas and corresponding coating failure information.

[0069] Step S3: Based on the high-definition microscopic imaging data of the coating surface of the turbine flow component, the multiple graphene ceramic coating failure areas, and the coating failure information of each graphene ceramic coating failure area, determine multiple test coating peeling areas and the first sandblasting pressure for coating removal in each test coating peeling area.

[0070] In some embodiments, a test region determination model can be used to determine multiple test coating peeling regions and a first blasting pressure for coating removal in each test coating peeling region. The test region determination model is a deep neural network. The inputs to the test region determination model are high-resolution microscopic imaging data of the coating surface of the turbine flow component, the multiple graphene ceramic coating failure regions, and coating failure information for each graphene ceramic coating failure region. The outputs of the test region determination model are the multiple test coating peeling regions and the first blasting pressure for coating removal in each test coating peeling region.

[0071] Multiple test coating peeling areas are representative areas selected from multiple graphene ceramic coating failure areas by the test area determination model for testing coating peeling sandblasting.

[0072] The initial blasting pressure for removing the coating from each test coating peel area is an initial conservative blasting pressure value set by the test area determination model for each test coating peel area for the initial coating removal operation.

[0073] The coating removal blasting pressure refers to the working pressure of the abrasive used by the blasting equipment when performing coating peeling operations on the flow parts of a water turbine. The blasting pressure is a key parameter for controlling the coating peeling strength.

[0074] If the sandblasting pressure for coating removal is too high, it can easily cause scratches, deformation, or surface damage to the substrate material of the turbine's flow components, resulting in irreversible structural damage. If the sandblasting pressure for coating removal is too low, the graphene ceramic coating cannot be effectively peeled off, leading to incomplete removal, excessive residue, and affecting subsequent recoating operations of the graphene ceramic coating.

[0075] The initial conservative blasting pressure was used to remove the coating from the peeling area during testing. This initial blasting pressure was relatively gentle, ensuring it wouldn't damage the substrate material while still producing an observable peeling response. This allowed for obtaining the true state and peeling characteristics of the coating under mild blasting conditions, and provided a safe and reliable initial test basis for subsequent optimization and adjustment of the blasting pressure.

[0076] High-resolution microscopic imaging data of the coating surface of turbine flow components allows the test area to accurately identify the microscopic damage state of the coating. The model can determine the fragility of the coating through microscopic texture and crack density. The coating failure information of the graphene ceramic coating failure area can directly reflect the actual strength and adhesion level of the coating, thus providing a core reference for setting a conservative sandblasting pressure for the model, avoiding damage to the base material due to excessive sandblasting pressure and failure to peel off the coating due to insufficient sandblasting pressure.

[0077] Different failure areas have different tolerances to sandblasting pressure. By combining high-definition microscopic imaging data of the coating surface of the turbine flow components, multiple graphene ceramic coating failure areas, and coating failure information of each graphene ceramic coating failure area, the model can avoid the weak or stress-sensitive parts of the substrate according to the actual physical state of the coating, and thus select representative and safe locations for testing. In this way, the model can set an initial conservative sandblasting pressure value that can achieve coating peeling without damaging the base material.

[0078] Deep neural networks, through a multilayer perceptron structure, can fuse visual features extracted from high-resolution microscopic imaging data of the coating surface of turbine flow components, spatial geometric features of graphene ceramic coating failure areas, and physical property features from coating failure information. The model utilizes the learned mapping relationship between graphene ceramic coating peeling pressure, damage depth, and failure type to assess the risk of candidate test sites. First, the deep neural network searches for geometrically regular areas with moderate damage levels within each graphene ceramic coating failure area as test coating peeling regions to ensure the generalizability of the test results. Next, the model identifies the microstructural features of the coating surface based on high-resolution microscopic imaging data, and combines this with key parameters such as remaining coating thickness and damage level from the coating failure information to determine the difficulty of coating peeling. For areas with loose structures and high damage levels, the model predicts lower energy requirements for coating peeling and selects a lower value from a preset safe pressure range as the first blasting pressure for coating removal. Simultaneously, the model integrates information such as remaining coating thickness and failure type, and uses nonlinear mapping to calculate the minimum pressure limit that allows the coating to peel off in layers without damaging the base material. Ultimately, the model can output the corresponding spatial location and sandblasting pressure parameters, thereby determining multiple test coating peeling areas and the first sandblasting pressure for coating removal in each test coating peeling area.

[0079] Step S4: Perform a first coating peeling operation on each test coating peeling area based on the first sandblasting pressure for coating removal in each test coating peeling area, and obtain a video of the first coating peeling operation.

[0080] The first coating peeling operation video is video data recording the process of peeling the coating off each test coating peeling area using the first sandblasting pressure. The first coating peeling operation video can be obtained in real-time by a high-definition camera mounted on the sandblasting equipment.

[0081] The first coating peeling operation video records real-time footage of the coating peeling off, the real-time characteristics of the impact between the abrasive and the coating surface, dust diffusion, and the color change process of the exposed substrate. This video reflects the actual peeling response of the coating under initial blasting pressure.

[0082] Step S5: Determine the second sandblasting pressure for coating removal in each test coating peeling area based on the first coating peeling operation video.

[0083] In some embodiments, a blasting pressure correction model can be used to determine the second blasting pressure for coating removal in each test coating peeling area. The blasting pressure correction model is a Transformer model. The input to the blasting pressure correction model is the first coating peeling operation video, and the output of the blasting pressure correction model is the second blasting pressure for coating removal in each test coating peeling area.

[0084] The second blasting pressure for removing the coating in each test coating peeling area is an optimized blasting pressure value that is obtained by adjusting the blasting pressure correction model based on the actual coating peeling response state reflected in the first coating peeling operation video, and can achieve safe and efficient coating peeling.

[0085] The first coating stripping operation video visually demonstrates the actual working effect under the initial sandblasting pressure, covering the coating removal rate and surface residue state. The time series information in the video reflects the coating cracking frequency and whether there is incomplete coating removal due to insufficient pressure or substrate discoloration due to excessive pressure. The dynamic feedback information contained in the first coating stripping operation video provides an objective basis for the model to adjust the sandblasting pressure, enabling the sandblasting pressure correction model to adjust the sandblasting pressure parameters based on the actual working effect. Through closed-loop control, a second sandblasting pressure for coating removal that can completely remove the coating while taking into account the working efficiency is calculated.

[0086] The Transformer model, through its self-attention mechanism, enables frame-by-frame analysis and temporal correlation analysis of the first coating peeling operation video. The model first extracts the area growth rate of the peeled area and the edge morphology features of the residual coating in the video frames. Using its self-attention mechanism, the Transformer model calculates the changing trend of the peeling effect at different time points in the video and identifies the actual removal state of the coating during the sandblasting process. In the encoding layer, the model transforms the peeling speed, substrate surface state changes, and abrasive motion observed in the video into feature sequences. Since the initial sandblasting pressure for coating removal is a conservative pressure, the model, through analysis of the first coating peeling operation video, determines that the coating peeling speed is below a set threshold and that coating residue exists. Based on this, the model calculates the corresponding pressure increase value according to the remaining thickness and density of the residual coating, and then determines the second sandblasting pressure for coating removal in each test coating peeling area, thereby improving the coating peeling effect while ensuring substrate safety.

[0087] In some embodiments, determining the second blasting pressure for coating removal in each test coating peeling area based on the first coating peeling operation video includes steps S51-S53:

[0088] Step S51: Based on the first coating peeling operation video, determine the remaining coating thickness distribution data after the first coating peeling operation in each test coating peeling area, the coating peeling amount per unit time in each test coating peeling area, and the current sandblasting pressure adaptability.

[0089] In some embodiments, the Transformer model can be used to determine the remaining coating thickness distribution data after the first coating peeling operation in each test coating peeling area, the coating peeling amount per unit time in each test coating peeling area, and the current sandblasting pressure fit.

[0090] The remaining coating thickness distribution data after the first coating peeling operation in each test coating peeling area refers to the statistical distribution data of the remaining coating thickness at different spatial locations within each test coating peeling area after the first coating peeling operation is completed.

[0091] The coating peeling amount per unit time for each test coating peeling area refers to the total thickness of the graphene ceramic coating removed from the corresponding test coating peeling area per unit time during the first coating peeling operation.

[0092] The current sandblasting pressure fit of each test coating peeling area refers to the quantitative value of the degree of matching between the first sandblasting pressure used for coating removal in the first coating peeling operation and the actual coating peeling requirements of the corresponding test coating peeling area.

[0093] The Transformer model possesses powerful temporal feature modeling capabilities and a global attention mechanism, enabling frame-by-frame analysis and global correlation analysis of the first coating stripping operation video. Through a multi-head self-attention structure, the Transformer model can capture the location, surface texture, grayscale changes, and boundary contour information of the coating stripping area in video frames, thereby accurately identifying the spatial extent of each test coating stripping area. Based on the visual features corresponding to the coating thickness in the video frames, the model can calculate the remaining coating thickness distribution data after the first coating stripping operation. Simultaneously, the Transformer model can perform temporal comparisons of consecutive video frames and statistically analyze the area and thickness changes of the coating removed per unit time, thus quantifying the coating stripping amount per unit time. Building upon this, the Transformer model can fuse the remaining coating thickness distribution and the coating stripping amount per unit time, then comprehensively evaluate the matching degree between the currently used first blasting pressure for coating removal and the actual stripping requirements, ultimately determining the suitability of the current blasting pressure.

[0094] Step S52: Based on the remaining coating thickness distribution data after the first coating peeling operation in each test coating peeling area, the coating peeling amount per unit time in each test coating peeling area, and the current sandblasting pressure adaptability, determine the coating peeling completion score and the sandblasting pressure safety upper limit value for each test coating peeling area.

[0095] In some embodiments, the Transformer model can be used to determine the coating peeling completion score and the safe upper limit value of the sandblasting pressure for each test coating peeling area.

[0096] The coating peeling completion score in the test area refers to a comprehensive quantitative score indicating whether the coating peeling effect has achieved the expected removal target after the first coating peeling operation. The coating peeling completion score can objectively reflect the degree to which the coating has been removed; the higher the score, the more thorough the peeling.

[0097] The safe upper limit of the sandblasting pressure for the tested coating peeling area refers to the maximum sandblasting pressure that the tested coating peeling area can withstand without damaging the substrate of the turbine's flow-through components.

[0098] The Transformer model can perform global correlation analysis and deep feature mining on the remaining coating thickness distribution data after the first coating peeling operation, the coating peeling amount per unit time, and the current sandblasting pressure adaptability. The Transformer model can adaptively weight the importance of the remaining coating thickness distribution, coating peeling amount per unit time, and sandblasting pressure adaptability through a self-attention mechanism, and accurately learn the inherent logic between thickness distribution uniformity, peeling rate rationality, and pressure adaptability, thereby comprehensively evaluating and outputting a coating peeling completion score for each test coating peeling area. By combining the inherent correlation between the remaining coating thickness distribution data after the first coating peeling operation, the coating peeling amount per unit time, and the current sandblasting pressure adaptability, the Transformer model can infer the safe upper limit value of the sandblasting pressure for each test coating peeling area.

[0099] Step S53: Based on the current sandblasting pressure adaptation, coating peeling completion score, and sandblasting pressure safety limit value of each test coating peeling area, determine the second sandblasting pressure for coating removal in each test coating peeling area.

[0100] In some embodiments, a deep neural network can be used to determine the second blasting pressure for coating removal in each test coating peeling area.

[0101] Deep neural networks possess intelligent decision-making and numerical fitting capabilities under multiple constraints, and can globally integrate and analyze the current sandblasting pressure adaptation, coating peeling completion score, and safe upper limit of sandblasting pressure for each test coating peeling area. Through a multi-layered perceptual structure, deep neural networks can learn the collaborative constraint rules between the current sandblasting pressure adaptation, coating peeling completion score, and safe upper limit of sandblasting pressure for each test coating peeling area, comprehensively balancing the actual needs and safety boundary limitations of the coating peeling operation. Under the premise of satisfying the safe upper limit of sandblasting pressure for each test coating peeling area, the model can combine the current sandblasting pressure adaptation and coating peeling completion score for each test coating peeling area to perform accurate numerical derivation, ultimately determining the second sandblasting pressure for coating removal in each test coating peeling area.

[0102] Step S6: Perform a second coating peeling operation on each test coating peeling area based on the second sandblasting pressure of each test coating peeling area, and obtain a video of the second coating peeling operation.

[0103] The second coating peeling operation video is video data recording the process of supplementing the coating peeling operation on each test coating peeling area using the second sandblasting pressure for coating removal.

[0104] The second video recording of the coating removal operation shows the removal state of the coating and the surface texture of the substrate after exposure under optimized and more precise sandblasting pressure.

[0105] Step S7: Based on the first coating stripping operation video and the second coating stripping operation video, determine the target sandblasting pressure for coating removal in each remaining coating stripping area.

[0106] In some embodiments, Figure 4 This is a schematic flowchart illustrating the process of determining the target sandblasting pressure for removing coatings in each remaining coating peeling area, as provided in an embodiment of the present invention. The determination of the target sandblasting pressure for removing coatings in each remaining coating peeling area includes steps S71 to S74:

[0107] Step S71: Based on the coating failure information of each graphene ceramic coating failure area, the first sandblasting pressure for coating removal of each test coating peeling area, and the second sandblasting pressure for coating removal of each test coating peeling area, determine the initial sandblasting pressure for coating removal of each remaining coating peeling area.

[0108] In some embodiments, an initial blasting pressure determination model can be used to determine the initial blasting pressure for coating removal in each remaining coating peeling area. The initial blasting pressure determination model is a deep neural network. The inputs to the initial blasting pressure determination model are coating failure information for each graphene ceramic coating failure area, a first blasting pressure for coating removal in each tested coating peeling area, and a second blasting pressure for coating removal in each tested coating peeling area. The output of the initial blasting pressure determination model is the initial blasting pressure for coating removal in each remaining coating peeling area.

[0109] Multiple test coating peeling areas and multiple remaining coating peeling areas together constitute multiple graphene ceramic coating failure areas. Each test coating peeling area and each remaining coating peeling area corresponds to one graphene ceramic coating failure area.

[0110] The initial blasting pressure for removing the coating from each remaining coating stripping area is the initial blasting pressure value set for each remaining coating stripping area that has not undergone coating stripping operations.

[0111] For each test coating peeling area, the first blasting pressure for coating removal is an initial conservative blasting pressure, and the second blasting pressure is an optimized blasting pressure to achieve complete peeling. Coating failure information, the first blasting pressure for coating removal, and the second blasting pressure for coating removal together constitute a mapping sample between the coating's physical state and the operating pressure. Since the remaining coating peeling areas and the test coating peeling areas belong to the same turbine flow component, their coating materials, failure mechanisms, and bonding strengths are consistent. By learning the correlation patterns in the samples, the model can accurately and safely extrapolate the initial blasting pressure for coating removal for each remaining coating peeling area.

[0112] Deep neural networks (DNNs) utilize multi-layer nonlinear mapping to extract deep features from input data. The DNN establishes a correspondence between the physical state of the coating and the intensity of the blasting operation through the weight distribution of hidden layers. The DNN can fit the coating failure information of each tested coating peeling area to the second blasting pressure for coating removal, thereby identifying the energy threshold required for complete coating peeling under different damage states. Simultaneously, by analyzing the difference between the first and second blasting pressures for coating removal, the model can quantify the sensitivity of different failure levels to pressure adjustments. When processing the remaining coating peeling areas, the DNN can extract the coating failure information of the corresponding graphene ceramic coating failure areas and map it to a high-dimensional feature space. Based on the learned coating peeling response features, the model performs similarity matching between the physical features of the remaining coating peeling areas and known test samples. Simultaneously, it refers to the pressure correction rules of similar damage areas and calculates the appropriate initial pressure through nonlinear transformation. The DNN can compensate for adhesion differences caused by region location and failure type through activation function operations, and can integrate global failure distribution and local test experience to output an initial blasting pressure for coating removal that achieves effective peeling without damaging the turbine's flow component substrate.

[0113] Step S72: Based on the first coating peeling operation video, the second coating peeling operation video, and the initial sandblasting pressure for coating removal in each remaining coating peeling area, generate a coating peeling operation simulation video for each remaining coating peeling area.

[0114] In some embodiments, a peeling operation simulation model can be used to generate a coating peeling operation simulation video for each remaining coating peeling area. The peeling operation simulation model is a generative adversarial network (GAN). The inputs to the peeling operation simulation model are the first coating peeling operation video, the second coating peeling operation video, and the initial blasting pressure for coating removal in each remaining coating peeling area. The output of the peeling operation simulation model is a coating peeling operation simulation video for each remaining coating peeling area.

[0115] Generative Adversarial Networks (GANs) are deep learning models that consist of a generator and a discriminator that compete against each other and evolve collaboratively. The generator is responsible for producing pseudo-data from the latent space that approximates the real distribution, aiming to make it indistinguishable to the discriminator. The discriminator is responsible for distinguishing whether the input data comes from the real sample set or the generator's output, thereby improving the discrimination accuracy. The generator and discriminator continuously compete during training, eventually enabling the generator to produce complex samples with extremely high realism that conform to the underlying physical laws or data distribution characteristics. GANs have shown great potential in fields such as image generation, style transfer, and video prediction.

[0116] The simulation video of the coating peeling operation for each remaining coating peeling area is simulated and output by the peeling operation simulation model. It is a simulation video of the coating peeling operation process performed on each remaining coating peeling area using the initial sandblasting pressure for coating removal.

[0117] The first and second coating peeling operation videos contain realistic temporal characteristics of the coating peeling process, surface deformation, failure state, and operation effect under actual sandblasting pressure. These videos provide a realistic physical process basis and dynamic change reference for model simulation video construction. The initial sandblasting pressure for each remaining coating peeling area serves as the initial operation parameter for that area, characterizing the sandblasting intensity suitable for the coating under specific failure states. This provides accurate pressure input conditions for the model to simulate the coating peeling operation process. Based on the features of the real operation videos and the initial sandblasting pressure parameters, the model can reconstruct the actual peeling response law of the coating under the corresponding pressure, thereby accurately generating a coating peeling operation simulation video for each remaining coating peeling area that fits the real operation scenario and conforms to the physical peeling characteristics of the coating.

[0118] The generator of the Generative Adversarial Network (GAN) learns the dynamic features and coating change patterns in real sandblasting operation videos by taking as input a first coating peeling operation video, a second coating peeling operation video, and the initial sandblasting pressure for each remaining coating peeling area. Based on the coating failure information of the remaining areas and the initial sandblasting pressure, the generator can simulate the coating peeling process under sandblasting and generate corresponding simulation videos. The model's discriminator can distinguish between the simulated videos generated by the generator and the real operation videos. Then, through adversarial training, the generator's parameters are continuously optimized, making the generated simulation videos more realistic and believable. Finally, the model outputs a simulated coating peeling operation video for each remaining coating peeling area.

[0119] Step S73: Construct a sandblasting pressure map. The sandblasting pressure map includes multiple nodes of remaining coating peeling areas and edges between multiple nodes. The node features of each node of remaining coating peeling area are coating failure information of each remaining coating peeling area and a simulation video of coating peeling operation of each remaining coating peeling area. The edges between nodes are the positional relationship information between the remaining coating peeling areas.

[0120] The sandblasting pressure map consists of multiple nodes and multiple edges. The multiple nodes represent multiple remaining coating peeling areas, and each edge corresponds to the connection relationship between the remaining coating peeling area nodes.

[0121] The remaining coating stripping area node represents the various areas where the coating removal operation is to be performed. The node features include coating failure information for each remaining coating stripping area and a simulation video of the coating stripping operation for each remaining coating stripping area.

[0122] Edges represent the spatial relationships between nodes in different remaining coating stripping areas, and edges represent the positional relationships between the remaining coating stripping areas.

[0123] Locational relationship information includes the orientation and distance between the remaining coating peeling areas.

[0124] Sandblasting pressure maps can systematically integrate single-point attributes and global topological constraints. The coating failure state and operation parameters in different remaining coating peeling areas are spatially correlated. By constructing sandblasting pressure maps, the coating failure characteristics, simulation operation process, and spatial distribution relationship of each area can be fully characterized.

[0125] Step S74: Based on the sandblasting pressure determination model, the sandblasting pressure spectrum is processed to obtain the target sandblasting pressure for coating removal in each remaining coating peeling area.

[0126] The sandblasting pressure determination model is a graph neural network (GNN) model. The GNN model includes a graph neural network (GNN) and fully connected layers. A GNN is a deep learning model that can run directly on a graph and can be used to capture structural information and dependencies between nodes. By aggregating and transmitting neighbor information in the graph, the GNN can update node features and perceive the global state of the graph. The GNN consists of graph convolutional layers, aggregation functions, and state update functions. GNNs can process data in non-Euclidean space.

[0127] The target blasting pressure for removing the coating in each remaining coating stripping area is determined by analyzing the blasting pressure spectrum using a blasting pressure determination model. It is the optimal blasting pressure value used for the final coating stripping operation in each remaining coating stripping area.

[0128] The target sandblasting pressure for removing the coating from each remaining area ensures structural safety, operational stability, and high efficiency throughout the entire process of coating removal and repair of the turbine's flow components.

[0129] The sandblasting pressure map can represent the physical state and operational simulation knowledge of the remaining coating peeling areas on the turbine's flow-through components, and can intuitively display the nodes of each remaining coating peeling area, their coating failure characteristics, and the simulation process of coating peeling operations. This structured representation facilitates the understanding and processing of complex spatial relationships between remaining coating peeling areas by graph neural networks. Through node features and edge information, graph neural networks can better capture the differences in coating failure states and operational parameter correlations among different remaining coating peeling areas. For example, the coating failure information and coating peeling operation simulation video in the node features provide the damage state of the coating in that area and the operational simulation process, while the positional relationship information on the edges reflects the spatial distribution and correlation patterns of these areas on the turbine's flow-through components.

[0130] Constructing a sandblasting pressure map allows for the effective organization of scattered remaining coating stripping areas and their associated operational information, facilitating aggregation calculations by graph neural networks. This organization reduces computational complexity and improves the training and inference efficiency of the model when optimizing sandblasting parameters and determining the coating stripping sequence. Because the sandblasting pressure map provides structured node and edge information, graph neural networks can better handle the multi-dimensional features of each remaining coating stripping area, thus avoiding data fragmentation and computational redundancy problems caused by the dispersion of information across regions in traditional methods. This structural advantage makes graph neural networks more effective in handling sandblasting pressure optimization problems involving multi-region collaborative operations.

[0131] By comprehensively analyzing the coating failure characteristics of nodes and the spatial positional relationship of edges in the sandblasting pressure spectrum, the graph neural network can accurately identify the operation priority and parameter adaptation requirements of each remaining coating peeling area. Thus, under the premise of ensuring complete coating peeling, a more reasonable sandblasting operation plan can be formulated, which can ensure operation efficiency and operation quality, while avoiding damage to the substrate of the turbine flow components.

[0132] Graph neural networks (GNNs) can perform multi-round message passing and feature interaction operations on the sandblasting pressure map. In each round of calculation, each node in the remaining coating peeling area absorbs coating failure features and operational simulation features from its neighboring nodes, and integrates the pressure trend of the neighborhood into its own state through an aggregation function. GNNs can analyze the topological connections between remaining regions through graph convolutional layers. If two nodes are physically adjacent on the turbine flow components and have consistent simulation results, the model will enhance the feature coupling between them. When updating node features, the model utilizes its deep perception capabilities to identify the hidden spatial distribution patterns in the coating failure information. For example, in the region near the blade edge of the turbine flow component, the model will collaboratively reduce the target sandblasting pressure of all nodes in that edge strip region based on the edge relationships in the sandblasting pressure map to avoid excessive sandblasting damage to the blade edge structure. Through global information passing, GNNs can effectively suppress pressure jumps caused by single-point detection errors. After processing through multiple layers of graph convolution, the hidden state of each node is decoded into the final target sandblasting pressure value. This value not only meets the efficiency requirements presented in the peeling simulation video of this point, but also conforms to the process logic of maintaining consistency with the surrounding area. Finally, the model can accurately infer and output the target sandblasting pressure for coating removal in each remaining coating peeling area.

[0133] Step S8: Perform coating stripping operation on each remaining coating stripping area based on the target sandblasting pressure for coating removal in each remaining coating stripping area. After the stripping is completed, perform graphene ceramic coating recoating operation on the multiple graphene ceramic coating failure areas.

[0134] Once the target sandblasting pressure for removing the coating in each remaining coating stripping area is determined, a precise and safe coating stripping operation is performed on each corresponding remaining coating stripping area based on the target sandblasting pressure for each remaining coating stripping area. This ensures complete removal of the original coating without damaging the substrate structure. Then, the entire graphene ceramic coating failure area is recoated, completing the overall repair treatment of the failure area.

[0135] Based on the same inventive concept Figure 5 This is a schematic diagram of a processing control system for graphene ceramic coatings on water turbines based on image analysis, provided in an embodiment of the present invention. The processing control system for graphene ceramic coatings on water turbines based on image analysis includes:

[0136] The imaging data acquisition module 91 is used to acquire high-definition microscopic imaging data of the coating surface of the turbine flow-through components;

[0137] The failure analysis module 92 is used to determine multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area based on high-definition microscopic imaging data of the coating surface of the turbine flow component.

[0138] The first sandblasting pressure determination module 93 is used to determine multiple test coating peeling areas and the first sandblasting pressure for coating removal of each test coating peeling area based on the high-definition microscopic imaging data of the coating surface of the turbine flow component, the multiple graphene ceramic coating failure areas, and the coating failure information of each graphene ceramic coating failure area.

[0139] The first peeling execution module 94 is used to perform a first coating peeling operation on each test coating peeling area based on the coating removal first sandblasting pressure of each test coating peeling area, and to acquire a video of the first coating peeling operation.

[0140] The second sandblasting pressure determination module 95 is used to determine the second sandblasting pressure for coating removal in each test coating peeling area based on the first coating peeling operation video.

[0141] The second peeling execution module 96 is used to perform a second coating peeling operation on each test coating peeling area based on the second sandblasting pressure for coating removal in each test coating peeling area, and to acquire a video of the second coating peeling operation.

[0142] The target sandblasting pressure determination module 97 is used to determine the target sandblasting pressure for coating removal in each remaining coating removal area based on the first coating peeling operation video and the second coating peeling operation video.

[0143] The final coating module 98 is used to perform coating stripping operations on each remaining coating stripping area based on the target sandblasting pressure for coating removal in each remaining coating stripping area, and to perform graphene ceramic coating recoating operations on the multiple graphene ceramic coating failure areas after the stripping is completed.

[0144] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0145] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A processing control method for graphene ceramic coatings on water turbines based on image analysis, characterized in that, include: Acquire high-resolution microscopic imaging data of the coating surface of the flow-through components of a water turbine; Based on high-definition microscopic imaging data of the coating surface of the turbine flow components, multiple graphene ceramic coating failure areas and coating failure information of each graphene ceramic coating failure area were determined. Based on the high-definition microscopic imaging data of the coating surface of the turbine flow component, the multiple graphene ceramic coating failure areas, and the coating failure information of each graphene ceramic coating failure area, multiple test coating peeling areas and the first sandblasting pressure for coating removal in each test coating peeling area are determined. Based on the first sandblasting pressure of the coating removal in each test coating peeling area, a first coating peeling operation is performed on each test coating peeling area, and a video of the first coating peeling operation is obtained; The second sandblasting pressure for removing the coating in each test coating peeling area is determined based on the first coating peeling operation video. A second coating removal operation is performed on each test coating removal area based on the second sandblasting pressure, and a video of the second coating removal operation is obtained. Based on the first coating stripping operation video and the second coating stripping operation video, determine the target sandblasting pressure for coating removal in each remaining coating stripping area; Based on the target sandblasting pressure for removing the coating in each remaining coating stripping area, a coating stripping operation is performed on each remaining coating stripping area. After the stripping is completed, a graphene ceramic coating recoating operation is performed on the multiple graphene ceramic coating failure areas.

2. The processing control method for graphene ceramic coatings for water turbines based on image analysis as described in claim 1, characterized in that, The determination of multiple graphene ceramic coating failure areas based on high-resolution microscopic imaging data of the coating surface of the turbine flow components, and the coating failure information of each graphene ceramic coating failure area include: A failure risk distribution map of the graphene ceramic coating of the turbine in service is generated based on high-definition microscopic imaging data of the coating surface of the turbine's flow-through components. Based on the high-resolution microscopic imaging data of the coating surface of the turbine flow-through components and the failure risk distribution map of the existing graphene ceramic coating of the turbine, information on multiple suspected failure coating feature points was determined. K clusters of failure coating feature points are obtained by clustering based on the information of the multiple suspected failure coating feature points; Acquire surface damage detection data for each cluster of failed coating features; Based on the coating surface damage detection data of each failed coating feature cluster, multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area are determined.

3. The processing control method for graphene ceramic coatings for water turbines based on image analysis as described in claim 1, characterized in that, The step of determining the target sandblasting pressure for removing the coating in each remaining coating removal area based on the first coating stripping operation video and the second coating stripping operation video includes: Based on the coating failure information of each graphene ceramic coating failure area, the first sandblasting pressure for coating removal of each tested coating peeling area, and the second sandblasting pressure for coating removal of each tested coating peeling area, the initial sandblasting pressure for coating removal of each remaining coating peeling area is determined; Based on the first coating peeling operation video, the second coating peeling operation video, and the initial sandblasting pressure for coating removal in each remaining coating peeling area, a coating peeling operation simulation video for each remaining coating peeling area is generated. A sandblasting pressure map is constructed, which includes multiple nodes of remaining coating peeling areas and edges between multiple nodes. The node features of each node of remaining coating peeling area are coating failure information of each remaining coating peeling area and a simulation video of coating peeling operation of each remaining coating peeling area. The edges between nodes are the positional relationship information between the remaining coating peeling areas. The sandblasting pressure spectrum is processed based on the sandblasting pressure determination model to obtain the target sandblasting pressure for coating removal in each remaining coating peeling area.

4. The processing control method for graphene ceramic coatings for water turbines based on image analysis as described in claim 3, characterized in that, The sandblasting pressure determination model is a graph neural network model.

5. A processing control system for graphene ceramic coatings on water turbines based on image analysis, characterized in that, include: The imaging data acquisition module is used to acquire high-resolution microscopic imaging data of the coating surface of the turbine's flow-through components; The failure analysis module is used to determine multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area based on high-definition microscopic imaging data of the coating surface of the turbine flow component. The first sandblasting pressure determination module is used to determine multiple test coating peeling areas and the first sandblasting pressure for coating removal in each test coating peeling area based on high-definition microscopic imaging data of the coating surface of the turbine flow component, the multiple graphene ceramic coating failure areas, and coating failure information of each graphene ceramic coating failure area. The first peeling execution module is used to perform a first coating peeling operation on each test coating peeling area based on the coating removal first sandblasting pressure of each test coating peeling area, and to acquire a video of the first coating peeling operation; The second sandblasting pressure determination module is used to determine the second sandblasting pressure for removing the coating in each test coating peeling area based on the first coating peeling operation video. The second peeling execution module is used to perform a second coating peeling operation on each test coating peeling area based on the second sandblasting pressure for coating removal in each test coating peeling area, and to acquire a video of the second coating peeling operation. The target sandblasting pressure determination module is used to determine the target sandblasting pressure for coating removal in each remaining coating removal area based on the first coating peeling operation video and the second coating peeling operation video. The final coating module is used to perform coating stripping operations on each remaining coating stripping area based on the target sandblasting pressure for coating removal in each remaining coating stripping area, and to perform graphene ceramic coating recoating operations on the multiple graphene ceramic coating failure areas after the stripping is completed.

6. The processing control system for graphene ceramic coatings for water turbines based on image analysis as described in claim 5, characterized in that, The failure analysis module is also used for: A failure risk distribution map of the graphene ceramic coating of the turbine in service is generated based on high-definition microscopic imaging data of the coating surface of the turbine's flow-through components. Based on the high-resolution microscopic imaging data of the coating surface of the turbine flow-through components and the failure risk distribution map of the existing graphene ceramic coating of the turbine, information on multiple suspected failure coating feature points was determined. K clusters of failure coating feature points are obtained by clustering based on the information of the multiple suspected failure coating feature points; Acquire surface damage detection data for each cluster of failed coating features; Based on the coating surface damage detection data of each failed coating feature cluster, multiple graphene ceramic coating failure areas and coating failure information for each graphene ceramic coating failure area are determined.

7. The processing control system for graphene ceramic coating of water turbines based on image analysis as described in claim 5, characterized in that, The target sandblasting pressure determination module is also used for: Based on the coating failure information of each graphene ceramic coating failure area, the first sandblasting pressure for coating removal of each tested coating peeling area, and the second sandblasting pressure for coating removal of each tested coating peeling area, the initial sandblasting pressure for coating removal of each remaining coating peeling area is determined; Based on the first coating peeling operation video, the second coating peeling operation video, and the initial sandblasting pressure for coating removal in each remaining coating peeling area, a coating peeling operation simulation video for each remaining coating peeling area is generated. A sandblasting pressure map is constructed, which includes multiple nodes of remaining coating peeling areas and edges between multiple nodes. The node features of each node of remaining coating peeling area are coating failure information of each remaining coating peeling area and a simulation video of coating peeling operation of each remaining coating peeling area. The edges between nodes are the positional relationship information between the remaining coating peeling areas. The sandblasting pressure spectrum is processed based on the sandblasting pressure determination model to obtain the target sandblasting pressure for coating removal in each remaining coating peeling area.

8. The processing control system for graphene ceramic coating of water turbine based on image analysis as described in claim 7, characterized in that, The sandblasting pressure determination model is a graph neural network model.

9. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the image analysis-based processing control method for graphene ceramic coatings for water turbines as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the processing control method for graphene ceramic coatings for water turbines based on image analysis as described in any one of claims 1 to 4.

Images