Wear resistance testing method and device for surface coating

Abstract

The present invention relates to a method and device for detecting the wear resistance of a surface coating. The method: based on different combinations of multiple surface coating process parameters, a substrate material is subjected to surface coating deposition treatment to obtain multiple coating samples; multi-parameter testing and 3D scanning are performed to obtain wear resistance performance data and 3D surface morphology data; digital blanket calculations of different scales and directions are performed on the 3D surface morphology data to obtain a multi-scale wear resistance index; fractal analysis is performed on the wear resistance performance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set; the fractal wear resistance feature set is input into a double-layer spatiotemporal graph convolutional network model to perform wear resistance detection to obtain a wear resistance evaluation index; particle swarm optimization is performed on multiple groups of surface treatment process parameters according to the wear resistance evaluation index to obtain an optimal coating process parameter combination. The implementation of the present invention improves the efficiency and quality of solving the optimal coating process parameters.

Description

Technical Field

[0001] The invention relates to the technical field of surface coating detection, and in particular to a method and device for detecting the wear resistance of a surface coating. Background Art

[0002] Surface coating technology plays a vital role in modern industry. It can significantly improve the surface properties of materials, especially wear resistance. However, with the increasing demand for industrial applications, traditional coating wear resistance testing methods have been unable to meet the increasingly complex evaluation needs. These methods usually rely on single or limited parameter tests and cannot fully reflect the wear resistance of the coating, resulting in a large deviation between the evaluation results and the actual application situation.

[0003] In addition, existing detection methods often ignore the effect of coating surface morphology on wear resistance, while the multi-scale characteristics of surface morphology are actually closely related to the wear resistance of coatings. Traditional methods are difficult to effectively capture these complex multi-scale information, resulting in inaccurate and incomplete evaluation of wear resistance. Summary of the invention

[0004] The main purpose of the present invention is to provide a method and device for detecting the wear resistance of a surface coating, so as to improve the efficiency and quality of solving the optimal coating process parameters.

[0005] To achieve the above object, the present invention provides a method for detecting the wear resistance of a surface coating, comprising the following steps:

[0006] Based on different combinations of multiple surface coating process parameters, surface coating deposition treatment is performed on the substrate material to obtain multiple coating samples;

[0007] Performing multi-parameter testing and 3D scanning on the multiple coating samples respectively to obtain wear resistance performance data and 3D surface morphology data of each coating sample;

[0008] Performing digital blanket calculations of different scales and directions on the 3D surface topography data to obtain a multi-scale wear resistance index for each coating sample;

[0009] Performing fractal analysis on the wear resistance performance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set for each coating sample;

[0010] Inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtaining a wear resistance evaluation index for each coating sample;

[0011] The plurality of groups of surface treatment process parameters are subjected to particle swarm optimization processing according to the wear resistance evaluation index of each coating sample to obtain an optimal coating process parameter combination.

[0012] The present invention also provides a surface coating wear resistance detection device, comprising:

[0013] A processing module, used for performing surface coating deposition processing on the substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples;

[0014] A testing module, used to perform multi-parameter testing and 3D scanning on the plurality of coating samples respectively, to obtain wear resistance performance data and 3D surface morphology data of each coating sample;

[0015] A calculation module, used for performing digital blanket calculations of different scales and directions on the 3D surface topography data to obtain a multi-scale wear resistance index of each coating sample;

[0016] An analysis module, used for performing fractal analysis on the wear resistance performance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set of each coating sample;

[0017] A detection module, used for inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtain a wear resistance evaluation index for each coating sample;

[0018] The optimization module is used to perform particle swarm optimization on the plurality of groups of surface treatment process parameters according to the wear resistance evaluation index of each coating sample to obtain an optimal coating process parameter combination. The present invention also provides an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of any of the above methods are implemented.

[0019] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0020] In summary, the technical solution provided by the present invention, the method comprehensively obtains the wear resistance data and 3D surface morphology data of the coating sample through multi-parameter testing and 3D scanning, realizes the multi-dimensional evaluation of the wear resistance characteristics of the coating, and improves the comprehensiveness and accuracy of the evaluation. The digital blanket calculation and fractal analysis method are used to analyze the 3D surface morphology data at different scales and directions, and obtain the multi-scale wear resistance index and fractal wear resistance feature set, which effectively captures the complex multi-scale characteristics of the coating surface and enhances the in-depth understanding of the wear resistance. The double-layer spatiotemporal graph convolution network model is introduced for wear resistance detection, which realizes the intelligent evaluation of the wear resistance characteristics of the coating and improves the efficiency and accuracy of the evaluation. The surface treatment process parameters are optimized by the particle swarm optimization algorithm to obtain the optimal coating process parameter combination, which provides strong support for the precise control and optimization of the coating process. The real-time thickness monitoring and adaptive adjustment mechanism are introduced in the coating deposition process to ensure the thickness uniformity of the coating sample and improve the quality and consistency of sample preparation. By using multiple fractal analysis and complex network analysis methods, wear-resistant features are deeply integrated to construct a more comprehensive and accurate description of wear-resistant features, thereby improving the reliability of wear-resistant performance evaluation. The spatial attention, temporal attention, and self-attention mechanisms are introduced into the double-layer spatiotemporal graph convolutional network to enhance the model's ability to capture key wear-resistant features and improve the accuracy of wear resistance prediction. Dynamic mutation operations and diversity analysis are introduced into the particle swarm optimization process to enhance the algorithm's global search capabilities and improve the efficiency and quality of solving the optimal coating process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 1 is a schematic diagram of the steps of a method for detecting the wear resistance of a surface coating in one embodiment of the present invention;

[0022] Figure 2 It is a structural block diagram of a wear resistance detection device for a surface coating in one embodiment of the present invention;

[0023] Figure 3 It is a schematic block diagram of the structure of an electronic device according to an embodiment of the present invention.

[0024] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0026] Reference Figure 1 This embodiment provides a method for detecting the wear resistance of a surface coating, comprising the following steps:

[0027] S1, performing surface coating deposition treatment on the substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples;

[0028] Among them, the surface roughness of the substrate material is measured in detail to obtain the surface roughness data of the substrate. According to the surface roughness data of the substrate, the substrate material is subjected to plasma cleaning treatment so that the surface of the substrate material is activated to obtain a surface-activated substrate material. The coating process parameter space is sampled using the Latin hypercube sampling method to obtain N groups of candidate process parameter combinations, ensuring the uniformity and representativeness of the sampling and better covering the entire process parameter space. The N groups of candidate process parameter combinations are subjected to dimensionality reduction analysis to reduce the complexity and redundancy of the parameters and obtain M target process parameters. Based on the M target process parameters, a response surface model is constructed, and multiple surface coating process parameter combinations are optimized and screened out through simulation analysis. The optimized parameter combination can maximize the performance of the coating. According to the multiple surface coating process parameter combinations screened out, the parameters of the magnetron sputtering equipment are set to form different coating deposition conditions. During the coating deposition process, the thickness change of the coating is monitored in real time to obtain the growth curve of the coating. By analyzing the coating growth curve, the deposition process is adaptively adjusted to ensure uniform thickness of the coating. During the deposition process, any anomalies can be detected and adjusted in time through real-time thickness monitoring, ensuring that the final coating has uniform thickness and excellent performance. After the coating deposition is completed, the coating samples are quickly heat treated to improve the physical and chemical properties of the coating and obtain multiple coating samples with high wear resistance.

[0029] S2, performing multi-parameter testing and 3D scanning on multiple coating samples respectively to obtain wear resistance data and 3D surface morphology data of each coating sample;

[0030] Among them, nanoindentation tests were performed on multiple coating samples to obtain hardness data and elastic modulus data of the coating samples. Ball-disc friction and wear tests were performed on multiple coating samples respectively, and the friction coefficient data and wear rate data of each coating sample were obtained by simulating the friction conditions in the actual working environment. The friction coefficient data can reflect the resistance encountered by the coating during the friction process, while the wear rate data can directly show the material loss of the coating during the friction process. Nano scratch test was used to gradually increase the load until the coating failed, and the critical load data of each coating sample was obtained, reflecting the adhesion and scratch resistance of the coating. The hardness data, elastic modulus data, friction coefficient data, wear rate data and critical load data were comprehensively calculated to obtain the wear resistance data of each coating sample. White light interferometric 3D scanning technology was used to scan multiple coating samples to obtain the initial surface morphology data of each coating sample, showing the surface characteristics of the coating, including surface roughness, waviness, etc. The initial surface morphology data was subjected to noise filtering to obtain more accurate 3D surface morphology data.

[0031] S3, digital blanket calculations of different scales and directions are performed on the 3D surface morphology data to obtain the multi-scale wear resistance index of each coating sample;

[0032] Among them, the 3D surface morphology data is decomposed at multiple scales to decompose the surface morphology data into surface morphology components of different scales. The surface morphology component of each scale provides details and contour information at that scale. Directional analysis is performed on the surface morphology components of different scales to obtain surface morphology features in multiple directions, reflecting the structural characteristics of the surface in different directions. By performing digital blanket coverage calculation on the surface morphology features in multiple directions, the blanket area data of each scale and direction are obtained. The fractal dimension of the surface morphology of each scale and direction is calculated using the blanket area data, and a multi-scale fractal dimension matrix is ​​formed. The multi-scale fractal dimension matrix is ​​subjected to singular value decomposition to extract the main eigenvectors of the surface morphology, which can represent the main information of the surface morphology. According to the main eigenvectors, the surface morphology is reconstructed to obtain the reconstructed surface morphology data. While retaining the main features, the reconstructed data removes redundant information, making the subsequent analysis more accurate. The reconstructed surface morphology data is subjected to local binary pattern analysis to obtain the surface texture feature descriptor, which can describe the microscopic texture features of the surface. Based on the texture feature descriptor, the surface morphology is anisotropically analyzed to obtain the surface anisotropy index, which reflects the uniformity and consistency of the surface in different directions. The preliminary wear resistance index is obtained by weighted fusion of the surface anisotropy index and the multi-scale fractal dimension matrix. The multi-scale and directional characteristics of the surface are considered comprehensively to make the wear resistance index more representative. According to the preliminary wear resistance index, the surface morphology is analyzed by autocorrelation function, the wear resistance evaluation is refined and adjusted, and the multi-scale wear resistance index of each coating sample is obtained.

[0033] S4, fractal analysis is performed on the wear resistance data and multi-scale wear resistance index to obtain a fractal wear resistance feature set for each coating sample;

[0034] Among them, the wear resistance data and multi-scale wear resistance index are normalized, and the data are adjusted to the same scale through standardization to eliminate the influence of different dimensions on the analysis results. A multidimensional phase space is constructed, and the time series data is mapped to a high-dimensional space through phase space reconstruction technology to obtain the phase space trajectory of the wear resistance characteristics. The phase space trajectory is reconstructed by applying the time delay embedding method to obtain a high-dimensional phase space. In the high-dimensional phase space, the correlation dimension and Lyapunov index are calculated to obtain preliminary fractal characteristics, which reflect the dynamic behavior of the wear resistance of the coating. The preliminary fractal characteristics are analyzed by multifractal spectrum to obtain the generalized dimension spectrum and singularity spectrum. The generalized dimension spectrum can describe the distribution of fractal dimensions at different scales, while the singularity spectrum can characterize the complexity of the coating surface morphology. By extracting the multifractal parameters of these spectra, the multifractal feature vectors are obtained to reflect the fractal characteristics of the wear resistance of the coating. The multifractal feature vectors are subjected to dimensionality reduction processing to extract the most representative characteristic principal components. Based on the characteristic principal components, a fractal network model is constructed, and the network topology is generated through the relationship between nodes and edges. The network topology is subjected to complex network analysis to obtain network characteristic parameters, such as node degree distribution, clustering coefficient, average path length, etc. These parameters can reflect the overall structure and local characteristics of the network. By fusion of network characteristic parameters and multiple fractal feature vectors, a variety of features are integrated to form a fractal wear-resistant feature set for each coating sample. The fractal wear-resistant feature set contains the fractal characteristics of the coating surface morphology and combines the structural information of the complex network, providing comprehensive quantitative indicators and theoretical support for the evaluation and optimization of the wear resistance of the coating.

[0035] S5, inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtaining a wear resistance evaluation index for each coating sample;

[0036] Among them, the fractal wear-resistant feature set is input into the preset two-layer spatiotemporal graph convolutional network model, and the spatiotemporal graph is constructed according to the fractal wear-resistant feature set to form a spatiotemporal graph sequence that can characterize the wear-resistant characteristics of the coating. By analyzing the spatiotemporal graph sequence, the wear-resistant characteristics of the coating at different time and space scales are captured. According to the spatiotemporal graph sequence, the input layer of the first layer of the spatiotemporal graph convolutional network is filled with data to ensure the integrity and continuity of the data, and the first layer initial feature map is obtained. Through the spatiotemporal graph convolution operation, the preliminary spatial and temporal features are extracted to form the first layer convolutional feature map. The first layer convolutional feature map is pooled to reduce the dimension and complexity of the data to obtain the downsampled feature map. The downsampled feature map is nonlinearly activated to enhance the feature expression ability and obtain the first layer output feature map. Based on the output feature map, the input layer of the second layer of the spatiotemporal graph convolutional network is filled with data again to obtain the second layer initial feature map, and the spatiotemporal graph convolution operation is performed to extract higher-level features to form the second layer convolutional feature map. The second layer of convolution feature map is subjected to global average pooling operation to reduce the feature map to a global feature vector, which effectively summarizes all feature information. The global feature vector is processed by the fully connected layer, and the features are processed and analyzed by the multi-layer perceptron to obtain the wear resistance prediction value. Through softmax classification, the prediction value is converted into a specific wear resistance evaluation index to clarify the wear resistance performance of each coating sample.

[0037] The spatiotemporal graph sequence is zero-filled, and the edges are expanded in the time and space dimensions by reflective filling to ensure the integrity and continuity of the edge information during the convolution operation, and the padded spatiotemporal graph sequence is obtained. The padded spatiotemporal graph sequence is input into the input layer of the first spatiotemporal graph convolution network, and the convolution operation is performed through 64 3x3x3 three-dimensional convolution kernels, and the convolution step is set to 1, so that the preliminary spatiotemporal features can be extracted to form the first layer of initial feature map. In order to ensure the stability and standardization of feature data, the first layer of initial feature map is batch normalized to make the feature distribution more uniform, and the first layer of normalized feature map is obtained. The normalized feature map is input into the first spatiotemporal convolution unit, and the spatiotemporal convolution operation is performed using 32 3x3x3 convolution kernels, and the convolution step is also set to 1. The ReLU activation function is applied for nonlinear transformation to obtain the first spatiotemporal feature map. In order to reduce the dimension and amount of calculation of the data, the first spatiotemporal feature map is subjected to the maximum pooling operation, using a pooling window of 2x2x2 with a step size of 2 to obtain a downsampled feature map, which effectively reduces the complexity of the data while retaining the key features. The downsampled feature map is input into the second spatiotemporal convolution unit, and 64 3x3x3 convolution kernels are used for spatiotemporal convolution operation, with a step size of 1, and the ReLU activation function is applied to obtain the second spatiotemporal feature map. In order to enhance the representation ability of the feature map, the second spatiotemporal feature map is subjected to the spatial attention mechanism analysis, and the attention weights on the spatial dimension are calculated to obtain the spatial weighted feature map. The spatial weighted feature map is subjected to the temporal attention mechanism analysis, and the attention weights on the temporal dimension are calculated to obtain the spatiotemporal weighted feature map. The spatiotemporal weighted feature map is subjected to the residual connection with the downsampled feature map, and the features of the two are fused by element-by-element addition operation to obtain the fused feature map. In order to avoid overfitting and enhance the generalization ability of the model, the fused feature map is subjected to the dropout operation, and a part of the neurons are randomly discarded to obtain the final first-layer convolution feature map.

[0038] The first layer output feature map is upsampled, and the bilinear interpolation method is used to expand the feature map size to the original input size. Upsampling helps to restore the spatial resolution of the feature map so that it can better match the details of the original input image. The upsampled feature map is input to the input layer of the second layer spatiotemporal graph convolution network, and convolution operation is performed through 128 3x3x3 three-dimensional convolution kernels with a step size of 1 to extract more detailed spatiotemporal features and form the second layer initial feature map. The second layer initial feature map is layer normalized to ensure that the feature distribution of each channel of each sample is uniform, and the second layer normalized feature map is obtained. Normalization makes network training more stable and efficient by adjusting the mean and variance of the feature map. The normalized feature map is input to the dilated convolution unit, and a 3x3x3 dilated convolution kernel with a dilation rate of 2 is used to perform spatiotemporal convolution. This convolution method can expand the receptive field of the convolution kernel, capture more spatiotemporal information, and obtain the dilated feature map. The Swish activation function is operated on the dilated feature map. Swish is a smooth and nonlinear activation function that can enhance the nonlinear expression ability of the network to obtain the second-layer activation feature map. The second-layer activation feature map is input into the self-attention module to calculate the self-attention weights in the spatiotemporal dimension. Through the self-attention mechanism, the network can automatically pay attention to the spatiotemporal region that is more important to the wear resistance characteristics to obtain the self-attention weighted feature map. The self-attention weighted feature map is processed by channel attention to calculate the importance weight of each channel to obtain the channel weighted feature map. The channel weighted feature map is fused with the second-layer initial feature map, and the information of different feature maps is integrated by weighted summation to obtain a multi-scale fused feature map. The fusion process can make full use of the advantages of different feature maps and improve the overall feature expression ability. The fused feature map is group normalized to divide the feature map into multiple groups, and then each group is normalized to obtain a group normalized feature map. Group normalization can reduce the amount of calculation and improve the calculation efficiency while maintaining the internal structure of the feature map. The group normalized feature map is input into the residual connection module, and the residual connection is performed with the upsampled feature map. The features of the two are fused through element-by-element addition operation to obtain the second-layer convolutional feature map.

[0039] S6, performing particle swarm optimization on multiple groups of surface treatment process parameters according to the wear resistance evaluation index of each coating sample to obtain the optimal coating process parameter combination.

[0040] Among them, multiple groups of surface treatment process parameters are encoded, and each group of process parameters is converted into a multidimensional vector to form an initial particle swarm. By constructing a fitness function, the wear resistance evaluation index is used as the optimization target, and the fitness of each particle is evaluated to obtain the initial fitness value. In order to initialize the speed of the particle, the initial speed vector is randomly generated using Gaussian distribution to ensure that the particle swarm has a variety of search directions and speeds. According to the initial fitness value, the particle with the highest fitness is selected as the global optimal solution, and the initial position of each particle is set as the individual optimal solution to form an initial optimal solution set. In the iterative update process, according to the current speed and position of the particle, the new speed and position are calculated, and the speed is adjusted by applying the inertia weight, individual cognitive coefficient and social cognitive coefficient to ensure that the particle swarm conducts effective exploration and development in the search space. According to the updated particle swarm, the fitness value of each particle is recalculated, and the individual optimal solution and the global optimal solution are updated according to the new fitness value to form a new optimal solution set. In order to prevent the particle swarm from falling into the local optimum, the new optimal solution set is analyzed for diversity, the aggregation degree of the particle swarm is calculated, and the population diversity index is obtained. According to the population diversity index, the particle swarm is dynamically mutated. The positions of some particles are randomly disturbed by Gaussian mutation, uniform mutation and other methods to increase the diversity of the population and obtain the mutated particle swarm. For the mutated particle swarm, boundary processing is performed to ensure that the positions of all particles are within the feasible solution space, avoid exceeding the design parameter range, and form an effective particle swarm. After all iterations and mutation operations are completed, the particle position corresponding to the global optimal solution is selected and decoded to obtain the optimal coating process parameter combination. In this way, the surface treatment process parameters can be effectively optimized to obtain a parameter combination that can maximize the wear resistance of the coating.

[0041] In one example, a surface coating deposition process is performed on a substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples, including:

[0042] The surface roughness of the substrate material is measured to obtain substrate surface roughness data, and according to the substrate surface roughness data, the substrate material is subjected to plasma cleaning treatment to obtain a surface-activated substrate material;

[0043] Latin hypercube sampling is performed on the coating process parameter space to obtain N groups of candidate process parameter combinations, and dimension reduction analysis is performed on the N groups of candidate process parameter combinations to obtain M target process parameters;

[0044] According to M target process parameters, a response surface model is constructed, and the response surface model is simulated and analyzed to obtain multiple surface coating process parameter combinations;

[0045] According to the combination of multiple surface coating process parameters, the magnetron sputtering equipment is parameterized to obtain different coating deposition conditions, and the thickness of each coating deposition condition is monitored in real time to obtain the coating growth curve;

[0046] According to the coating growth curve, the deposition process is adaptively adjusted to obtain a coating sample with uniform thickness, and the coating sample is rapidly heat treated to finally obtain multiple coating samples.

[0047] In this example, the surface roughness of the substrate material is measured using precision surface roughness measurement equipment, such as white light interferometer or atomic force microscope. These devices can provide nanometer-level surface roughness data to ensure the accuracy and reliability of the measurement. The obtained substrate surface roughness data can be expressed as Ra, which is the arithmetic average roughness, and its formula is:

[0048] ;

[0049] in, is the measured length, is a height function of the surface profile. The Ra value calculated by this formula can quantify the roughness of the substrate surface. According to the substrate surface roughness data, the substrate material is plasma cleaned to remove pollutants and oxide layers on the substrate surface to obtain a clean and activated surface. Plasma cleaning usually uses oxygen plasma or argon plasma to bombard the substrate surface with high-energy ions under vacuum conditions, so that the surface atoms are in a highly active state. This treatment method can significantly improve the adhesion between the coating and the substrate. Latin hypercube sampling is performed on the coating process parameter space to obtain N groups of candidate process parameter combinations. Latin hypercube sampling is a statistical method used to uniformly extract sample points from a high-dimensional parameter space to ensure that the sample coverage of each parameter dimension is as wide as possible. Assuming that the coating process parameters include temperature, pressure, gas flow, etc., each parameter is divided into several intervals. Through Latin hypercube sampling, N groups of candidate process parameter combinations are obtained. Dimensionality reduction analysis is performed on the N groups of candidate process parameter combinations, usually using methods such as principal component analysis or linear discriminant analysis. Through dimensionality reduction analysis, the high-dimensional parameter space is mapped to a low-dimensional space to obtain M target process parameters. Based on M target process parameters, a response surface model is constructed. A response surface model is a mathematical model used to describe the relationship between input parameters and response variables. Commonly used model forms include polynomial models and Gaussian process models. Assume that the input parameters are , the response variable is , then the response surface model can be expressed as:

[0050] ;

[0051] in, is an unknown real function, is the error term. By fitting the experimental data, an approximate response surface model is obtained. The response surface model is simulated and analyzed to predict the coating performance under different parameter combinations, and multiple surface coating process parameter combinations are obtained. According to the multiple surface coating process parameter combinations obtained, the magnetron sputtering equipment is parameterized to obtain different coating deposition conditions. During the magnetron sputtering process, parameter settings include target power, gas flow rate, deposition time, etc. Real-time thickness monitoring is an important means to ensure coating uniformity. An optical thickness monitor or a crystal thickness monitor can be used to adjust the deposition parameters by monitoring the coating growth curve. The coating growth curve can be expressed as a function of thickness change over time:

[0052] ;

[0053] in, It's time The coating thickness at is the deposition rate, is the material density, By monitoring and analyzing the coating growth curve, the deposition process is automatically adjusted to ensure uniform coating thickness.

[0054] In one example, multiple coating samples are subjected to multi-parameter testing and 3D scanning to obtain wear resistance data and 3D surface morphology data of each coating sample, including:

[0055] Perform nanoindentation tests on multiple coating samples respectively to obtain hardness data of each coating sample;

[0056] Performing nanoindentation tests on multiple coating samples respectively to obtain elastic modulus data of each coating sample;

[0057] Conduct ball-disc friction and wear tests on multiple coating samples respectively to obtain friction coefficient data of each coating sample;

[0058] Conduct ball-on-disk friction and wear tests on multiple coating samples respectively to obtain wear rate data of each coating sample;

[0059] Conduct nano scratch tests on multiple coating samples respectively to obtain critical load data of each coating sample;

[0060] According to the hardness data, elastic modulus data, friction coefficient data, wear rate data and critical load data, the wear resistance data of each coating sample is calculated;

[0061] Perform white light interferometric 3D scanning on multiple coating samples respectively to obtain initial surface morphology data of each coating sample;

[0062] The initial surface morphology data were subjected to noise filtering to obtain the 3D surface morphology data of each coating sample.

[0063] In this example, nanoindentation tests were performed on multiple coating samples to obtain hardness data and elastic modulus data for each coating sample. Nanoindentation testing is an accurate test method that can measure the mechanical properties of materials at a microscale. Using a nanoindenter, a gradually increasing load is applied to the sample surface and the indentation depth is recorded. Based on the relationship between the indentation depth and the load, the hardness (H) and elastic modulus (E) of the material are calculated. The hardness calculation formula is:

[0064] ;

[0065] in, is the maximum applied load, is the contact area of ​​the indentation. The elastic modulus is calculated by the slope of the unloading phase of the indentation curve. Ball-disc friction and wear tests were performed on multiple coating samples to obtain the friction coefficient data and wear rate data of each coating sample. The ball-disc friction and wear test is a test in which a rotating steel ball rubs against the coating surface to simulate the wear process in actual applications. The friction coefficient (μ) can be calculated by recording the ratio of the friction force (F) to the normal force (N):

[0066] ;

[0067] The wear rate is calculated by measuring the volume loss of the wear scar using the formula:

[0068] ;

[0069] in, is the wear rate, is the wear volume, is the wear length. These two data are used to evaluate the performance of the coating under friction and wear conditions. Nano scratch tests are performed on multiple coating samples to obtain the critical load data of each coating sample. The nano scratch test applies a gradually increasing load on the coating surface, scratches the surface, and records the load (Lc) when the coating fails. The critical load is an important indicator of coating adhesion and scratch resistance. Based on the above test data, the wear resistance data of each coating sample is calculated according to the hardness data, elastic modulus data, friction coefficient data, wear rate data and critical load data. These performance indicators are combined into a wear resistance score through a weighted comprehensive formula:

[0070] ;

[0071] in, It is the weight of each performance index, which is set according to the actual application requirements. White light interferometer 3D scanning is performed on multiple coating samples to obtain the initial surface morphology data of each coating sample. White light interferometer is a non-contact measurement device that can accurately obtain three-dimensional morphology data of the sample surface and provide nanometer-level surface height distribution information. These data can be used to analyze the surface characteristics of the coating, such as roughness and uniformity. The initial surface morphology data is subjected to noise filtering processing to obtain the 3D surface morphology data of each coating sample. Noise filtering processing can adopt a variety of methods, such as Gaussian filtering, median filtering, etc., to remove the random noise generated during the measurement process and retain the true morphological characteristics of the surface.

[0072] In one example, digital blanket calculations at different scales and directions were performed on the 3D surface topography data to obtain a multi-scale wear resistance index for each coating sample, including:

[0073] Perform multi-scale decomposition on 3D surface morphology data to obtain surface morphology components of different scales, and perform directional analysis on the surface morphology components of each scale based on the surface morphology components of different scales to obtain surface morphology features in multiple directions;

[0074] Perform digital blanket coverage calculation on the surface morphology features in multiple directions to obtain blanket area data of each scale and direction, and perform fractal dimension calculation on the surface morphology of each scale and direction based on the blanket area data to obtain a multi-scale fractal dimension matrix;

[0075] Performing singular value decomposition on the multi-scale fractal dimension matrix to obtain the main eigenvectors of the surface morphology, and reconstructing the surface morphology according to the main eigenvectors to obtain reconstructed surface morphology data;

[0076] Performing local binary pattern analysis on the reconstructed surface morphology data to obtain a surface texture feature descriptor, and performing anisotropy analysis on the surface morphology based on the surface texture feature descriptor to obtain a surface anisotropy index;

[0077] The surface anisotropy index and the multi-scale fractal dimension matrix are weightedly fused to obtain the preliminary wear resistance index. Based on the preliminary wear resistance index, the surface morphology is analyzed by autocorrelation function to obtain the multi-scale wear resistance index of each coating sample.

[0078] In this example, 3D surface topography data is decomposed into multiple scales by wavelet transform, which is a commonly used signal processing technique that decomposes data into components of different scales. For 3D surface topography data, wavelet transform can decompose it into detail components and approximation components of different scales, expressed as:

[0079] ;

[0080] in, Represents 3D surface topography data, is the wavelet basis function, are wavelet coefficients. By adjusting the scale of the wavelet basis function, surface morphology components of different scales are obtained. Directional analysis is performed on surface morphology components of different scales. Directional analysis can be performed using Fourier transform or structural tensor analysis. Assuming that Fourier transform is used, a two-dimensional Fourier transform is performed on the surface morphology components of each scale to obtain a frequency domain image. Then different directions are selected in the frequency domain image for analysis to obtain the surface morphology features in each direction. Digital blanket coverage is calculated for the surface morphology features in multiple directions. Digital blanket coverage is a method for calculating surface complexity. A virtual blanket is covered on the surface and the area of ​​the blanket is calculated to measure the complexity of the surface. Assume that the size of the blanket is The surface morphology characteristics are , then the carpet area The calculation formula is:

[0081] ;

[0082] By changing the size of the blanket , and obtain the carpet area data of different scales and directions. According to the carpet area data, the fractal dimension of the surface morphology of each scale and direction is calculated. The fractal dimension is a method to describe complex morphology. Commonly used fractal dimension calculation methods include box counting method and fractal box dimension. Assuming the box counting method is used, the fractal dimension The calculation formula is:

[0083] ;

[0084] in, is the number of boxes needed to cover the surface, is the size of the box. By calculating the fractal dimensions at different scales and directions, the multi-scale fractal dimension matrix is ​​obtained. Singular value decomposition is performed on the multi-scale fractal dimension matrix. Singular value decomposition is a matrix decomposition technique that decomposes the matrix into the product of three sub-matrices to extract the main eigenvectors. Assume that the multi-scale fractal dimension matrix is , its singular value decomposition is expressed as:

[0085] ;

[0086] in, and is an orthogonal matrix, Is a diagonal matrix whose diagonal elements are singular values. By selecting the eigenvector corresponding to the largest singular value, the main eigenvector of the surface morphology is obtained. According to the main eigenvector, the surface morphology is reconstructed to obtain the reconstructed surface morphology data. The reconstructed surface morphology data is subjected to local binary pattern (LBP) analysis. LBP is a texture analysis method that compares each pixel in the image with its neighboring pixels to obtain a binary pattern code. Assume that the reconstructed surface morphology data is , then the LBP code of a pixel can be expressed as:

[0087] ;

[0088] in, is the neighborhood pixel value, is a sign function. By calculating the LBP code of each pixel, the surface texture feature descriptor is obtained. According to the surface texture feature descriptor, the surface morphology is anisotropically analyzed. The anisotropic analysis can be achieved by calculating the texture intensity in different directions. Assume that the surface texture feature descriptor is , then the surface anisotropy index The calculation formula is

[0089] ;

[0090] in, is the number of directions, It is The texture strength in each direction, is the average texture intensity. By calculating the anisotropy index, the surface anisotropy characteristics are obtained. The surface anisotropy index and the multi-scale fractal dimension matrix are weighted and fused to obtain the preliminary wear resistance index. The weighted fusion can be achieved through linear combination. Assuming that the anisotropy index is , the fractal dimension matrix is , then the initial wear resistance index The calculation formula is:

[0091] ;

[0092] in, and is the weight coefficient. According to the preliminary wear resistance index, the surface morphology is analyzed by autocorrelation function. The autocorrelation function is a tool to describe the self-similarity of the signal. By calculating the similarity of the surface morphology data under different displacements, the multi-scale wear resistance index of each coating sample is obtained.

[0093] In one example, fractal analysis is performed on the wear resistance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set for each coating sample, including:

[0094] Normalizing the wear resistance data and the multi-scale wear resistance index to obtain a standardized wear resistance data set, and constructing a multi-dimensional phase space based on the standardized wear resistance data set to obtain the phase space trajectory of the wear resistance characteristics;

[0095] The phase space trajectory is embedded with time delay to obtain the reconstructed high-dimensional phase space, and the correlation dimension and Lyapunov index are calculated based on the high-dimensional phase space to obtain the preliminary fractal characteristics;

[0096] Perform multifractal spectrum analysis on the preliminary fractal features to obtain generalized dimension spectrum and singularity spectrum, and extract multifractal parameters based on the generalized dimension spectrum and singularity spectrum to obtain multifractal feature vectors;

[0097] Perform dimensionality reduction processing on the multi-fractal feature vectors to obtain the characteristic principal components after dimensionality reduction, and build a fractal network model and generate a network topology structure based on the characteristic principal components;

[0098] The network topology is subjected to complex network analysis to obtain network characteristic parameters, and the network characteristic parameters and multi-fractal feature vectors are fused to obtain the fractal wear-resistant feature set of each coating sample.

[0099] In this example, the wear resistance data and multi-scale wear resistance index are normalized to eliminate the influence of different dimensions and obtain a standardized wear resistance data set. Normalization usually uses a standardization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Based on the standardized wear resistance data set, a multidimensional phase space is constructed. Phase space reconstruction is a nonlinear time series analysis method that maps a one-dimensional time series to a high-dimensional phase space through a time delay coordinate. Assume that the standardized wear resistance data set is , the reconstructed phase space trajectory is ,in, is the time delay, is the embedding dimension. By selecting appropriate time delay and embedding dimension, the phase space trajectory of wear resistance is obtained. After time delay embedding of the phase space trajectory, the reconstructed high-dimensional phase space is obtained. In the high-dimensional phase space, the correlation dimension and Lyapunov index are calculated to quantify the complexity and dynamic characteristics of the phase space trajectory. The correlation dimension is a method to measure the distribution complexity of points in the phase space, and its calculation formula is:

[0100] ;

[0101] in, Is the distance less than The ratio of the number of point pairs to the total number of point pairs. The Lyapunov index is used to describe the sensitivity of the system trajectory to the initial conditions, and its calculation formula is:

[0102] ;

[0103] By calculating the correlation dimension and Lyapunov index, the preliminary fractal characteristics are obtained. The preliminary fractal characteristics are subjected to multifractal spectrum analysis to obtain the generalized dimension spectrum and singularity spectrum. Multifractal spectrum analysis is a method to describe the multi-scale characteristics of complex systems. The multifractal characteristics of the system are extracted by calculating the generalized dimension spectrum and singularity spectrum. The calculation formula of the generalized dimension spectrum is:

[0104] ;

[0105] in, is the order, It is The probability distribution of the points in the box. The calculation formula of the singularity spectrum is:

[0106] ;

[0107] in, is the singularity intensity, is a scaling function. Through the generalized dimension spectrum and singularity spectrum, the multifractal parameters are extracted to obtain the multifractal eigenvector. The multifractal eigenvector is subjected to dimensionality reduction processing, usually using principal component analysis or linear discriminant analysis to reduce the dimension and redundancy of the features. Through dimensionality reduction processing, the principal components of the features after dimensionality reduction are obtained. Assume that the multifractal eigenvector is , the principal component of the feature after dimensionality reduction is , the conversion formula is:

[0108] ;

[0109] in, is a dimensionality reduction matrix. Based on the characteristic principal components after dimensionality reduction, a fractal network model is constructed and a network topology is generated. The fractal network model is a model based on complex network theory, which can describe the multi-scale characteristics of the system through the relationship between nodes and edges. Assuming that the network nodes represent the characteristic principal components after dimensionality reduction and the edges represent the relationship between nodes, the network topology can be obtained through the adjacency matrix Description. Perform complex network analysis on the network topology to obtain network characteristic parameters, such as node degree distribution, clustering coefficient, average path length, etc. Node degree distribution is used to describe the number of connections between nodes, clustering coefficient is used to describe the degree of aggregation between nodes, and average path length is used to describe the overall connectivity of the network. Assume that the node degree of the network is , the clustering coefficient is The average path length is , and its calculation formula is:

[0110] ;

[0111] ;

[0112] ;

[0113] in, are the elements of the adjacency matrix, is the number of nodes in the network, Is a node and The shortest path length between them. The network feature parameters and multi-fractal feature vectors are fused to obtain the fractal wear-resistant feature set of each coating sample. Feature fusion can be achieved by weighted average or multimodal learning methods, and finally the fractal wear-resistant feature set of each coating sample is obtained.

[0114] In one example, the fractal wear-resistant feature set is input into a preset two-layer spatiotemporal graph convolutional network model for wear resistance detection, and the wear resistance evaluation index of each coating sample is obtained, including:

[0115] The fractal wear-resistant feature set is input into a preset double-layer space-time graph convolutional network model, and a space-time graph is constructed for the fractal wear-resistant feature set to obtain a space-time graph sequence that characterizes the wear-resistant characteristics of the coating;

[0116] According to the spatiotemporal graph sequence, data is filled in the input layer of the first spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model to obtain the first-layer initial feature map, and the spatiotemporal graph convolution operation is performed on the first-layer initial feature map to obtain the first-layer convolutional feature map;

[0117] Perform a pooling operation on the first-layer convolution feature map to obtain a downsampled feature map, and perform a nonlinear activation on the downsampled feature map to obtain the first-layer output feature map;

[0118] According to the output feature map of the first layer, the input layer of the second-layer spatiotemporal graph convolution network in the double-layer spatiotemporal graph convolution network model is filled with data to obtain the second-layer initial feature map, and the spatiotemporal graph convolution operation is performed on the second-layer initial feature map to obtain the second-layer convolution feature map;

[0119] Perform global average pooling on the second-layer convolutional feature map to obtain a global feature vector, and perform full-connection layer processing on the global feature vector to obtain a wear resistance prediction value;

[0120] The wear resistance prediction value is classified by softmax to obtain the wear resistance evaluation index of each coating sample.

[0121] In this example, the fractal wear-resistant feature set is input into the preset two-layer spatiotemporal graph convolutional network model. The fractal wear-resistant feature set usually contains multi-dimensional features, such as multi-scale fractal dimension, anisotropy index, network characteristic parameters, etc. After preprocessing, these feature data are constructed into a spatiotemporal graph sequence, which can characterize the wear resistance of the coating. In the process of constructing the spatiotemporal graph, it is assumed that the feature vector at each moment is , then the space-time graph sequence can be expressed as:

[0122] ;

[0123] in, Indicates the length of the time series. Contains multiple feature dimensions, such as fractal dimension, anisotropy index, etc. According to the spatiotemporal graph sequence, the input layer of the first spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model is filled with data to obtain the first-layer initial feature map. Data filling can be done by zero filling or edge filling to ensure the integrity of the data during the convolution operation. Assume that the feature map of the input layer is , then the initial feature map of the first layer can be expressed as:

[0124] ;

[0125] The spatiotemporal graph convolution operation is performed on the initial feature map of the first layer, and the size of the convolution kernel is , with a step size of 1. The convolution operation can extract local spatiotemporal features. Assuming the convolution kernel is , the feature map after convolution is , then the convolution operation formula is:

[0126] ;

[0127] Among them, * represents the convolution operation. Through the convolution operation, the first layer of convolution feature map is extracted. The first layer of convolution feature map is pooled. The pooling operation usually adopts maximum pooling or average pooling, which can reduce the dimension and computational complexity of the feature map. Assume that the pooling window size is , the step size is 2, and the feature map after pooling is , then the pooling operation formula is:

[0128] ;

[0129] Perform nonlinear activation on the downsampled feature map. Common activation functions include ReLU, Sigmoid, etc. Assuming the ReLU activation function is used, the feature map after activation is , then the activation function formula is:

[0130] ;

[0131] Among them, ReLU(x)=max(0,x). According to the output feature map of the first layer, the input layer of the second layer of the spatiotemporal graph convolution network in the two-layer spatiotemporal graph convolution network model is filled with data to obtain the initial feature map of the second layer. The method of data filling is the same as that of the first layer. The spatiotemporal graph convolution operation is performed on the initial feature map of the second layer, and the convolution kernel size is , the step size is 1, assuming the convolution kernel is , the feature map after convolution is , then the convolution operation formula is:

[0132] ;

[0133] Through the convolution operation, the second layer convolution feature map is extracted. The second layer convolution feature map is subjected to global average pooling operation, and the feature map after pooling is , the global average pooling operation can reduce the feature map to a global feature vector, the formula is:

[0134] ;

[0135] in, Represent the height, width and depth of the feature map respectively. Through global average pooling, the global feature vector is obtained The global feature vector is processed by the fully connected layer and mapped into the wear resistance prediction value. Assume that the weight of the fully connected layer is , the bias is , then the predicted value of wear resistance is The calculation formula is:

[0136] ;

[0137] The wear resistance prediction value is subjected to softmax classification to obtain the wear resistance evaluation index of each coating sample. Softmax classification can convert the prediction value into a probability distribution, and the formula is:

[0138] ;

[0139] in, It is Through softmax classification, the wear resistance evaluation index of each coating sample is obtained.

[0140] In one example, according to the spatiotemporal graph sequence, data is filled in the input layer of the first layer of the spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model to obtain the first layer initial feature map, and the spatiotemporal graph convolution operation is performed on the first layer initial feature map to obtain the first layer convolutional feature map, including:

[0141] Zero-fill the space-time graph sequence, and use the reflection filling method to expand the edge in the time and space dimensions to obtain the filled space-time graph sequence;

[0142] The padded spatiotemporal graph sequence is input into the input layer of the first spatiotemporal graph convolutional network, and convolution operation is performed through 64 3x3x3 three-dimensional convolution kernels with a step size of 1 to obtain the first layer initial feature map;

[0143] Perform batch normalization on the initial feature map of the first layer, adjust the feature distribution, and obtain the normalized feature map of the first layer;

[0144] The first layer normalized feature map is input into the first spatiotemporal convolution unit, and spatiotemporal convolution is performed through 32 3x3x3 convolution kernels with a step size of 1, and the ReLU activation function is applied to obtain the first spatiotemporal feature map;

[0145] Perform a maximum pooling operation on the first spatiotemporal feature map, using a 2x2x2 pooling window with a step size of 2 to obtain a downsampled feature map;

[0146] The downsampled feature map is input into the second spatiotemporal convolution unit, and spatiotemporal convolution is performed through 64 3x3x3 convolution kernels with a step size of 1, and the ReLU activation function is applied to obtain the second spatiotemporal feature map;

[0147] Performing a spatial attention mechanism analysis on the second spatiotemporal feature map, calculating the attention weight in the spatial dimension, and obtaining a spatial weighted feature map; and performing a temporal attention mechanism analysis on the spatial weighted feature map, calculating the attention weight in the temporal dimension, and obtaining a spatiotemporal weighted feature map;

[0148] The spatiotemporal weighted feature map is residually connected with the downsampled feature map, and a fused feature map is obtained by element-by-element addition operation. The fused feature map is dropout-operated to obtain the first-layer convolutional feature map.

[0149] In this example, the spatiotemporal graph sequence is zero-filled. Zero-filling is a common data preprocessing method before convolution operation, which can maintain the integrity of image edge information. Based on zero-filling, the reflection filling method is used to further extend the edge in time and space dimensions. Reflection filling is a method of extending the image boundary by reflecting the edge pixels of the image. It not only retains the edge features, but also avoids excessive interference of the filling value on the image content. Assume that the original spatiotemporal graph sequence is ,in is the time dimension, and is the spatial dimension, and the filled space-time graph sequence obtained after reflection filling is The padded spatiotemporal graph sequence is input into the input layer of the first spatiotemporal graph convolutional network. The convolution operation is performed through 64 3x3x3 three-dimensional convolution kernels with a step size of 1 to obtain the first layer initial feature map. The formula for the convolution operation is:

[0150] ;

[0151] in, is the initial feature map of the first layer, is the convolution kernel weight, is the bias term. Through this formula, local features in the spatiotemporal graph sequence are extracted. Batch normalization is performed on the initial feature map of the first layer. Batch normalization is a method to accelerate deep network training and improve stability. Batch normalization adjusts the feature distribution of each batch of data so that its mean is close to 0 and its variance is close to 1. Assume that the input data is , the formula for batch normalization is:

[0152] ;

[0153] in, is the mean of the batch data, is the standard deviation of the batch data, is a very small constant used to prevent zero division errors. After batch normalization, the first layer of normalized feature maps is obtained. The first layer of normalized feature maps is input into the first spatiotemporal convolution unit, and spatiotemporal convolution is performed through 32 3x3x3 convolution kernels with a step size of 1, and the ReLU activation function is applied. The formula of the ReLU activation function is:

[0154] ;

[0155] The formula of the convolution operation is the same as before. After convolution, the first spatiotemporal feature map is obtained. By applying the ReLU activation function, nonlinearity is introduced to improve the expressiveness of the model. The first spatiotemporal feature map is subjected to the maximum pooling operation. Maximum pooling is a downsampling method that can reduce the dimension of the feature map while retaining important features. Assuming the pooling window size is 2x2x2 and the step size is 2, the formula of the maximum pooling operation is:

[0156] ;

[0157] in, Changes within the pooling window. After the maximum pooling operation, the downsampled feature map is obtained. The downsampled feature map is input into the second spatiotemporal convolution unit, and spatiotemporal convolution is performed through 64 3x3x3 convolution kernels with a step size of 1, and the ReLU activation function is applied to obtain the second spatiotemporal feature map. The formulas for convolution and activation functions are similar to the previous ones. Spatiotemporal features are extracted through this operation. The spatial attention mechanism is analyzed on the second spatiotemporal feature map, and the attention weights on the spatial dimension are calculated. The spatial attention mechanism can enhance important features while suppressing unimportant features. Assume that the input feature map is , the calculation formula of spatial attention weight is:

[0158] ;

[0159] in, is the spatial attention weight, and are weights and biases, is the Sigmoid activation function. Through the spatial attention mechanism, a spatial weighted feature map is obtained. The temporal attention mechanism is analyzed on the spatial weighted feature map to calculate the attention weight in the time dimension. The calculation method of the temporal attention mechanism is similar to the spatial attention mechanism, and the formula is:

[0160] ;

[0161] in, is the temporal attention weight, and are weights and biases, is the Sigmoid activation function. Through the temporal attention mechanism, the spatiotemporal weighted feature map is obtained. The spatiotemporal weighted feature map is residually connected with the downsampled feature map. The residual connection is a technology to prevent the degradation of deep networks by directly adding the input to the output. The formula of the residual connection is:

[0162] ;

[0163] in, is the fusion feature map after residual connection, is the spatiotemporal weighted feature map, is the downsampled feature map. By adding the elements one by one, the fused feature map is obtained. Dropout is performed on the fused feature map. Dropout is a technique to prevent overfitting by randomly discarding some neurons. Assume that the probability of dropout is , then the feature map after dropout is:

[0164] ;

[0165] in, is a random variable of Bernoulli distribution, with a value of 0 or 1. Through the dropout operation, the final first-layer convolution feature map is obtained.

[0166] In one example, according to the first layer output feature map, data is filled in the input layer of the second layer of the spatiotemporal graph convolution network in the double-layer spatiotemporal graph convolution network model to obtain the second layer initial feature map, and the spatiotemporal graph convolution operation is performed on the second layer initial feature map to obtain the second layer convolution feature map, including:

[0167] The first layer output feature map is upsampled and the feature map size is expanded to the original input size using the bilinear interpolation method to obtain the upsampled feature map;

[0168] The upsampled feature map is input into the input layer of the second-layer spatiotemporal graph convolutional network, and convolution operation is performed through 128 3x3x3 three-dimensional convolution kernels with a step size of 1 to obtain the second-layer initial feature map;

[0169] Perform layer normalization on the second layer initial feature map, and normalize each channel of each sample independently to obtain the second layer normalized feature map;

[0170] The second layer normalized feature map is input into the dilated convolution unit, and a 3x3x3 dilated convolution kernel with a dilation rate of 2 is used to perform a spatiotemporal convolution operation to obtain a dilated feature map.

[0171] Perform a Swish activation function operation on the dilated feature map to obtain the second-layer activated feature map, and input the second-layer activated feature map into the self-attention module to calculate the self-attention weights in the spatiotemporal dimension to obtain the self-attention weighted feature map;

[0172] Perform channel attention processing on the self-attention weighted feature map to obtain a channel weighted feature map, and perform feature fusion on the channel weighted feature map and the second-layer initial feature map, and use a weighted summation method to obtain a multi-scale fusion feature map;

[0173] The multi-scale fusion feature map is grouped and normalized to obtain a group normalized feature map. The group normalized feature map is input into the residual connection module and residually connected with the upsampled feature map. The second layer of convolutional feature map is obtained by element-by-element addition.

[0174] In this example, the first layer output feature map is upsampled to expand the size of the feature map to the same size as the original input. The bilinear interpolation method is used to effectively expand the size of the feature map. Assume that the first layer output feature map is , the upsampled feature map is , then the formula for bilinear interpolation is

[0175] ;

[0176] in, is the weight of bilinear interpolation, is the pixel value in the input feature map. The upsampled feature map is input into the input layer of the second-layer spatiotemporal graph convolution network, and convolution operation is performed through 128 3x3x3 three-dimensional convolution kernels with a step size of 1 to obtain the second-layer initial feature map. The convolution operation can extract higher-level features. The formula for the convolution operation is:

[0177] ;

[0178] in, is the initial feature map of the second layer, is the convolution kernel weight, is the bias term. The initial feature map of the second layer is normalized, and each channel of each sample is normalized independently. Layer normalization can speed up model training and stabilize the training process. The normalization formula is:

[0179] ;

[0180] in, is the mean value of the channel, is the standard deviation of the channel, is a small constant used to prevent division by zero errors. After normalization, the second layer normalized feature map is obtained. The second layer normalized feature map is input into the dilated convolution unit, and a 3x3x3 dilated convolution kernel with a dilation rate of 2 is used to perform spatiotemporal convolution. Dilated convolution (or hole convolution) can expand the receptive field of the convolution kernel without increasing the number of parameters. The formula for dilated convolution is:

[0181] ;

[0182] in, is the dilated convolution kernel weight, is the expansion rate (here is 2). Through the expansion convolution operation, the expansion feature map is obtained. The Swish activation function is operated on the expansion feature map. The Swish activation function is a new type of activation function, and the formula is:

[0183] ;

[0184] in, is the Sigmoid function, After being processed by the Swish activation function, the second-layer activation feature map is obtained. The second-layer activation feature map is input into the self-attention module to calculate the self-attention weights in the spatiotemporal dimension. The self-attention mechanism can capture global information and give higher weights to important features. Assume that the input of the self-attention module is , the output is , the formula for self-attention is:

[0185] ;

[0186] in, are query, key, and value matrices respectively, is an exponential function. Through the self-attention mechanism, a self-attention weighted feature map is obtained. Channel attention processing is performed on the self-attention weighted feature map to calculate the attention weight on the channel dimension. The channel attention mechanism can give higher weights to important channels. The formula for channel attention is:

[0187] ;

[0188] in, For the The attention weight of each channel, are the height, width and depth of the feature map respectively. Through the channel attention mechanism, the channel weighted feature map is obtained. The channel weighted feature map is fused with the second layer initial feature map, and a multi-scale fused feature map is obtained by weighted summation. The fusion formula is:

[0189] ;

[0190] in, and is the weight coefficient, and multi-scale features are fused in this way. The multi-scale fusion feature map is grouped and normalized to obtain a group normalized feature map. The formula for group normalization is:

[0191] ;

[0192] in, is the mean of the group, is the standard deviation of the group, is a small constant. After group normalization, a group normalized feature map is obtained. The group normalized feature map is input into the residual connection module, and residual connection is performed with the upsampled feature map. The second layer convolution feature map is obtained by element-by-element addition. The formula of residual connection is:

[0193] ;

[0194] in, is the feature map after residual connection, is the upsampled feature map. Through the above steps, the feature extraction and fusion of the double-layer spatiotemporal graph convolutional network model are completed.

[0195] In one example, particle swarm optimization was performed on multiple groups of surface treatment process parameters according to the wear resistance evaluation index of each coating sample to obtain the optimal coating process parameter combination, including:

[0196] Encoding multiple groups of surface treatment process parameters, converting each group of process parameters into a multi-dimensional vector, and obtaining an initial particle group;

[0197] According to the wear resistance evaluation index, a fitness function is constructed, and the fitness of each particle is evaluated to obtain the initial fitness value;

[0198] Initialize the velocity of the initial particle swarm and randomly generate the initial velocity vector of each particle using Gaussian distribution;

[0199] According to the initial fitness value, the particle with the highest fitness is selected as the global optimal solution, and the initial position of each particle is set as the individual optimal solution to obtain the initial optimal solution set;

[0200] Iteratively update the initial particle swarm to obtain an updated particle swarm;

[0201] According to the updated particle swarm, recalculate the fitness value of each particle, and update the individual optimal solution and the global optimal solution to obtain a new optimal solution set;

[0202] Perform diversity analysis on the new optimal solution set, calculate the aggregation degree of the particle swarm, and obtain the population diversity index;

[0203] According to the population diversity index, the particle swarm is dynamically mutated to obtain the mutated particle swarm;

[0204] The boundary of the mutated particle swarm is processed to ensure that the particle position is within the feasible solution space, to obtain a valid particle swarm, and to select the particle position corresponding to the global optimal solution, and to decode and obtain the optimal coating process parameter combination.

[0205] In this example, each set of surface treatment process parameters is encoded and converted into a multidimensional vector. Group of process parameters, each group of process parameters includes specific parameters, such as temperature, pressure, gas flow, etc. Each set of parameters is encoded as a dimensional vector, which can be expressed as:

[0206] ;

[0207] in, It is A vector of process parameters, It is process parameters. In this way, the initial particle swarm is obtained, and each particle in the particle swarm is a process parameter vector. According to the wear resistance evaluation index, a fitness function is constructed, and the fitness of each particle is evaluated to obtain the initial fitness value. The fitness function is constructed based on the influence of process parameters on the wear resistance of the coating. Assuming that the wear resistance evaluation index is , then the fitness function can be defined as:

[0208] Fitness ;

[0209] Among them, Fitness Represents particles The fitness value of each particle is calculated to evaluate its performance under the current parameter combination. The initial particle swarm is initialized with a velocity vector of each particle randomly generated using a Gaussian distribution. The purpose of velocity initialization is to provide the particle swarm optimization algorithm with an initial search direction and step size. Assume that the velocity vector is , the parameter of the Gaussian distribution is the mean and standard deviation , then the velocity vector of each particle can be expressed as:

[0210] ;

[0211] in, Represents Gaussian distribution. In this way, the initial velocity vector of each particle is obtained. According to the initial fitness value, the particle with the highest fitness is selected as the global optimal solution, and the initial position of each particle is set as the individual optimal solution to obtain the initial optimal solution set. Assume that the initial particle swarm is The particle with the highest fitness is , then the global optimal solution is:

[0212] ;

[0213] At the same time, the initial position of each particle is set as the individual optimal solution to obtain the initial optimal solution set ,in:

[0214] ;

[0215] The initial particle swarm is iteratively updated to obtain the updated particle swarm. The particle swarm optimization algorithm searches for the optimal solution by updating the position and velocity of the particles. The update formula for position and velocity is:

[0216] ;

[0217] in, is the inertia weight, and is the learning factor, and is a random number. Through these formulas, the position and speed of the particle are updated. According to the updated particle swarm, the fitness value of each particle is recalculated, and the individual optimal solution and the global optimal solution are updated to obtain a new optimal solution set. The fitness value of each particle is calculated through the fitness function. If the current fitness value is better than the fitness value of the individual optimal solution, the individual optimal solution is updated; if the current fitness value is better than the fitness value of the global optimal solution, the global optimal solution is updated. Perform diversity analysis on the new optimal solution set, calculate the aggregation degree of the particle swarm, and obtain the population diversity index. Diversity analysis can be achieved by calculating the Euclidean distance between particles. Assume that the particle swarm is , then the population diversity index The calculation formula is:

[0218] ;

[0219] in, Represents particles and particles The Euclidean distance between them. According to the population diversity index, the particle swarm is dynamically mutated to obtain the mutated particle swarm. The dynamic mutation operation can be implemented by Gaussian mutation or uniform mutation to increase the diversity of the population. Assume that the mutated particle is , the formula for mutation operation is:

[0220] ;

[0221] in, is the variation intensity, is a standard normal distribution. The mutated particle swarm is subjected to boundary processing to ensure that the particle position is within the feasible solution space and obtain an effective particle swarm. Boundary processing can be achieved by limiting the particle position within the parameter range. Assume that the parameter range is , then the boundary processing formula is:

[0222] ;

[0223] By processing the boundaries, the particle position is ensured to be within the feasible solution space. The particle position corresponding to the global optimal solution is selected and decoded to obtain the optimal coating process parameter combination.

[0224] Reference Figure 2 This embodiment provides a surface coating wear resistance detection device, comprising:

[0225] Processing module 1, used for performing surface coating deposition processing on the substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples;

[0226] Test module 2, used to perform multi-parameter testing and 3D scanning on multiple coating samples respectively, to obtain wear resistance data and 3D surface morphology data of each coating sample;

[0227] Calculation module 3, used to perform digital blanket calculations of different scales and directions on the 3D surface topography data to obtain a multi-scale wear resistance index for each coating sample;

[0228] Analysis module 4, used for fractal analysis of wear resistance performance data and multi-scale wear resistance index, to obtain a fractal wear resistance feature set of each coating sample;

[0229] Detection module 5, used for inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtain the wear resistance evaluation index of each coating sample;

[0230] The optimization module 6 is used to perform particle swarm optimization on multiple groups of surface treatment process parameters according to the wear resistance evaluation index of each coating sample to obtain the optimal coating process parameter combination.

[0231] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0232] Reference Figure 3 In an embodiment of the present invention, an electronic device is also provided. The electronic device may be a server, and its internal structure may be as follows: Figure 3 As shown. The electronic device includes a processor, a memory, a display screen, an input device, a network interface and a database connected via a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store the corresponding data in this embodiment. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0233] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the electronic device to which the solution of the present invention is applied.

[0234] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0235] In summary, a method for detecting the wear resistance of a surface coating is proposed in an embodiment of the present invention.

[0236] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0237] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0238] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for detecting the wear resistance of a surface coating, characterized in that: The following steps are involved: Based on different combinations of multiple surface coating process parameters, surface coating deposition treatment is performed on the substrate material to obtain multiple coating samples; Performing multi-parameter testing and 3D scanning on the multiple coating samples respectively to obtain wear resistance performance data and 3D surface morphology data of each coating sample; Performing digital blanket calculations of different scales and directions on the 3D surface topography data to obtain a multi-scale wear resistance index for each coating sample; Performing fractal analysis on the wear resistance performance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set for each coating sample; Inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtaining a wear resistance evaluation index for each coating sample; According to the wear resistance evaluation index of each coating sample, particle swarm optimization is performed on multiple groups of surface coating process parameters to obtain the optimal coating process parameter combination; Wherein, the surface coating deposition process is performed on the substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples, including: Measuring the surface roughness of the substrate material to obtain substrate surface roughness data, and performing plasma cleaning on the substrate material according to the substrate surface roughness data to obtain a surface-activated substrate material; Performing Latin hypercube sampling on the coating process parameter space to obtain N groups of candidate process parameter combinations, and performing dimensionality reduction analysis on the N groups of candidate process parameter combinations to obtain M target process parameters; According to the M target process parameters, a response surface model is constructed, and simulation analysis is performed on the response surface model to obtain multiple surface coating process parameter combinations; According to the plurality of surface coating process parameter combinations, the magnetron sputtering equipment is parameterized to obtain different coating deposition conditions, and the thickness of each coating deposition condition is monitored in real time to obtain a coating growth curve; According to the coating growth curve, the deposition process is adaptively adjusted to obtain a coating sample with uniform thickness, and the coating sample is rapidly heat treated to finally obtain a plurality of coating samples; Wherein, the digital blanket calculation of different scales and directions is performed on the 3D surface morphology data to obtain a multi-scale wear resistance index of each coating sample, including: Performing multi-scale decomposition on the 3D surface morphology data to obtain surface morphology components of different scales, and performing directional analysis on the surface morphology components of each scale based on the surface morphology components of different scales to obtain surface morphology features in multiple directions; Performing digital blanket coverage calculation on the surface morphology features in the multiple directions to obtain blanket area data of each scale and direction, and performing fractal dimension calculation on the surface morphology of each scale and direction based on the blanket area data to obtain a multi-scale fractal dimension matrix; Performing singular value decomposition on the multi-scale fractal dimension matrix to obtain main eigenvectors of the surface morphology, and reconstructing the surface morphology according to the main eigenvectors to obtain reconstructed surface morphology data; Performing local binary pattern analysis on the reconstructed surface topography data to obtain a surface texture feature descriptor, and performing anisotropy analysis on the surface topography based on the surface texture feature descriptor to obtain a surface anisotropy index; Performing weighted fusion on the surface anisotropy index and the multi-scale fractal dimension matrix to obtain a preliminary wear resistance index, and performing autocorrelation function analysis on the surface morphology based on the preliminary wear resistance index to obtain a multi-scale wear resistance index for each coating sample; Wherein, the particle swarm optimization process is performed on multiple groups of surface coating process parameters according to the wear resistance evaluation index of each coating sample to obtain the optimal coating process parameter combination, including: Encoding the multiple groups of surface coating process parameters, converting each group of process parameters into a multi-dimensional vector, and obtaining an initial particle group; According to the wear resistance evaluation index, a fitness function is constructed, and the fitness of each particle is evaluated to obtain an initial fitness value; Initializing the velocity of the initial particle group, and randomly generating an initial velocity vector of each particle using a Gaussian distribution; According to the initial fitness value, the particle with the highest fitness is selected as the global optimal solution, and the initial position of each particle is set as the individual optimal solution to obtain an initial optimal solution set; Iteratively updating the initial particle swarm to obtain an updated particle swarm; According to the updated particle swarm, the fitness value of each particle is recalculated, and the individual optimal solution and the global optimal solution are updated to obtain a new optimal solution set; Performing diversity analysis on the new optimal solution set, calculating the aggregation degree of the particle swarm, and obtaining a population diversity index; According to the population diversity index, a dynamic mutation operation is performed on the particle group to obtain a mutated particle group; The mutated particle swarm is subjected to boundary processing to ensure that the particle position is within the feasible solution space, thereby obtaining a valid particle swarm, and the particle position corresponding to the global optimal solution is selected and decoded to obtain the optimal coating process parameter combination.

2. The method for detecting the wear resistance of a surface coating according to claim 1, characterized in that: The multi-parameter test and 3D scanning are respectively performed on the plurality of coating samples to obtain wear resistance performance data and 3D surface morphology data of each coating sample, including: Performing nanoindentation tests on the multiple coating samples respectively to obtain hardness data of each coating sample; Performing nanoindentation tests on the multiple coating samples respectively to obtain elastic modulus data of each coating sample; Performing ball-disc friction and wear tests on the plurality of coating samples respectively to obtain friction coefficient data of each coating sample; Performing ball-on-disk friction and wear tests on the multiple coating samples respectively to obtain wear rate data of each coating sample; Performing nano scratch tests on the multiple coating samples respectively to obtain critical load data of each coating sample; According to the hardness data, elastic modulus data, friction coefficient data, wear rate data and critical load data, the wear resistance data of each coating sample is calculated; Performing white light interferometric 3D scanning on the plurality of coating samples respectively to obtain initial surface morphology data of each coating sample; The initial surface morphology data are respectively subjected to noise filtering processing to obtain 3D surface morphology data of each coating sample.

3. The method for detecting the wear resistance of a surface coating according to claim 1, characterized in that: The fractal analysis of the wear resistance performance data and the multi-scale wear resistance index is performed to obtain a fractal wear resistance feature set for each coating sample, including: Normalizing the wear resistance performance data and the multi-scale wear resistance index to obtain a standardized wear resistance data set, and constructing a multi-dimensional phase space based on the standardized wear resistance data set to obtain a phase space trajectory of wear resistance characteristics; Performing time-delay embedding on the phase space trajectory to obtain a reconstructed high-dimensional phase space, and calculating the correlation dimension and Lyapunov index based on the high-dimensional phase space to obtain preliminary fractal features; Performing multifractal spectrum analysis on the preliminary fractal features to obtain a generalized dimension spectrum and a singularity spectrum, and extracting multifractal parameters based on the generalized dimension spectrum and the singularity spectrum to obtain a multifractal feature vector; Performing dimensionality reduction processing on the multi-fractal feature vector to obtain the characteristic principal components after dimensionality reduction, and constructing a fractal network model and generating a network topology structure based on the characteristic principal components; A complex network analysis is performed on the network topology to obtain network characteristic parameters, and feature fusion is performed on the network characteristic parameters and the multi-fractal feature vectors to obtain a fractal wear-resistant feature set for each coating sample.

4. The method for detecting the wear resistance of a surface coating according to claim 1, characterized in that: The fractal wear-resistant feature set is input into a preset double-layer spatiotemporal graph convolutional network model for wear resistance detection to obtain the wear resistance evaluation index of each coating sample, including: Inputting the fractal wear-resistant feature set into a preset double-layer space-time graph convolutional network model, and constructing a space-time graph for the fractal wear-resistant feature set to obtain a space-time graph sequence characterizing the wear-resistant characteristics of the coating; According to the spatiotemporal graph sequence, data is filled in the input layer of the first layer of the spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model to obtain a first layer initial feature map, and a spatiotemporal graph convolution operation is performed on the first layer initial feature map to obtain a first layer convolutional feature map; Performing a pooling operation on the first-layer convolutional feature map to obtain a downsampled feature map, and performing a nonlinear activation on the downsampled feature map to obtain a first-layer output feature map; According to the first-layer output feature map, data is filled in the input layer of the second-layer spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model to obtain a second-layer initial feature map, and a spatiotemporal graph convolution operation is performed on the second-layer initial feature map to obtain a second-layer convolutional feature map; Performing a global average pooling operation on the second-layer convolutional feature map to obtain a global feature vector, and performing a fully connected layer processing on the global feature vector to obtain a wear resistance prediction value; The wear resistance prediction value is subjected to softmax classification to obtain the wear resistance evaluation index of each coating sample.

5. The method for detecting the wear resistance of a surface coating according to claim 4, characterized in that: The step of filling data on the input layer of the first layer of the space-time graph convolution network in the double-layer space-time graph convolution network model according to the space-time graph sequence to obtain a first layer of initial feature graphs, and performing a space-time graph convolution operation on the first layer of initial feature graphs to obtain a first layer of convolution feature graphs, including: Zero-filling the space-time graph sequence, and performing edge expansion in time and space dimensions using a reflection filling method to obtain a filled space-time graph sequence; Input the padded spatiotemporal graph sequence into the input layer of the first spatiotemporal graph convolutional network, perform convolution operation through 64 3x3x3 three-dimensional convolution kernels with a step size of 1, and obtain the first layer initial feature map; Performing batch normalization on the first layer of initial feature maps to adjust feature distribution and obtain a first layer of normalized feature maps; Input the first layer normalized feature map into the first spatiotemporal convolution unit, perform spatiotemporal convolution through 32 3x3x3 convolution kernels with a step size of 1, and apply the ReLU activation function to obtain the first spatiotemporal feature map; Performing a maximum pooling operation on the first spatiotemporal feature map, using a 2x2x2 pooling window with a step size of 2, to obtain a downsampled feature map; Input the downsampled feature map into the second spatiotemporal convolution unit, perform spatiotemporal convolution through 64 3x3x3 convolution kernels with a step size of 1, and apply a ReLU activation function to obtain a second spatiotemporal feature map; Performing a spatial attention mechanism analysis on the second spatiotemporal feature map, calculating the attention weight in the spatial dimension, and obtaining a spatial weighted feature map; and performing a temporal attention mechanism analysis on the spatial weighted feature map, calculating the attention weight in the temporal dimension, and obtaining a spatiotemporal weighted feature map; The spatiotemporal weighted feature map is residually connected with the downsampled feature map, an element-by-element addition operation is performed to obtain a fused feature map, and a dropout operation is performed on the fused feature map to obtain a first-layer convolutional feature map.

6. The method for detecting the wear resistance of a surface coating according to claim 4, characterized in that: The step of filling the input layer of the second-layer spatiotemporal graph convolutional network in the double-layer spatiotemporal graph convolutional network model with data according to the first-layer output feature map to obtain the second-layer initial feature map, and performing a spatiotemporal graph convolution operation on the second-layer initial feature map to obtain the second-layer convolutional feature map includes: Upsampling the output feature map of the first layer, and using a bilinear interpolation method to expand the feature map size to the original input size to obtain an upsampled feature map; Input the upsampled feature map into the input layer of the second-layer spatiotemporal graph convolutional network, perform convolution operation through 128 3x3x3 three-dimensional convolution kernels with a step size of 1, and obtain the second-layer initial feature map; Performing layer normalization processing on the second layer initial feature map, independently normalizing each channel of each sample, to obtain a second layer normalized feature map; The second layer normalized feature map is input into the dilated convolution unit, and a 3x3x3 dilated convolution kernel and a dilation rate of 2 are used to perform a spatiotemporal convolution operation to obtain a dilated feature map; Performing a Swish activation function operation on the dilated feature map to obtain a second-layer activated feature map, and inputting the second-layer activated feature map into a self-attention module to calculate the self-attention weights in the spatiotemporal dimension to obtain a self-attention weighted feature map; Performing channel attention processing on the self-attention weighted feature map to obtain a channel weighted feature map, and performing feature fusion on the channel weighted feature map and the second layer initial feature map in a weighted summation manner to obtain a multi-scale fused feature map; The multi-scale fusion feature map is subjected to group normalization processing, the feature map is grouped and normalized to obtain a group normalized feature map, and the group normalized feature map is input into a residual connection module, and a residual connection is performed with the up-sampled feature map, and an element-by-element addition operation is performed to obtain a second-layer convolution feature map.

7. A wear resistance detection device for a surface coating, characterized in that: The device is used to perform the wear resistance detection method of the surface coating according to any one of claims 1 to 6, comprising: A processing module, used for performing surface coating deposition processing on the substrate material based on different combinations of multiple surface coating process parameters to obtain multiple coating samples; A testing module, used to perform multi-parameter testing and 3D scanning on the plurality of coating samples respectively, to obtain wear resistance performance data and 3D surface morphology data of each coating sample; A calculation module, used for performing digital blanket calculations of different scales and directions on the 3D surface topography data to obtain a multi-scale wear resistance index of each coating sample; An analysis module, used for performing fractal analysis on the wear resistance performance data and the multi-scale wear resistance index to obtain a fractal wear resistance feature set of each coating sample; A detection module, used for inputting the fractal wear-resistant feature set into a preset double-layer spatiotemporal graph convolutional network model to perform wear resistance detection, and obtain a wear resistance evaluation index for each coating sample; The optimization module is used to perform particle swarm optimization on multiple groups of surface coating process parameters according to the wear resistance evaluation index of each coating sample to obtain the optimal coating process parameter combination; Wherein, the processing module is specifically used for: measuring the surface roughness of the substrate material to obtain substrate surface roughness data, and according to the substrate surface roughness data, performing plasma cleaning treatment on the substrate material to obtain a surface activated substrate material; performing Latin hypercube sampling on the coating process parameter space to obtain N groups of candidate process parameter combinations, and performing dimensionality reduction analysis on the N groups of candidate process parameter combinations to obtain M target process parameters; constructing a response surface model according to the M target process parameters, and performing simulation analysis on the response surface model to obtain multiple surface coating process parameter combinations; setting parameters for the magnetron sputtering equipment according to the multiple surface coating process parameter combinations to obtain different multiple coating deposition conditions, and performing real-time thickness monitoring on each coating deposition condition to obtain a coating growth curve; according to the coating growth curve, adaptively adjusting the deposition process to obtain a coating sample with uniform thickness, and performing rapid heat treatment on the coating sample to finally obtain multiple coating samples; The calculation module is specifically used to: perform multi-scale decomposition on the 3D surface morphology data to obtain surface morphology components of different scales, and perform directional analysis on the surface morphology components of each scale according to the surface morphology components of different scales to obtain surface morphology features in multiple directions; perform digital blanket coverage calculation on the surface morphology features in multiple directions to obtain blanket area data of each scale and direction, and perform fractal dimension calculation on the surface morphology of each scale and direction according to the blanket area data to obtain a multi-scale fractal dimension matrix; perform singular value decomposition on the multi-scale fractal dimension matrix to obtain The main feature vector of the surface morphology is obtained, and the surface morphology is reconstructed according to the main feature vector to obtain the reconstructed surface morphology data; the reconstructed surface morphology data is subjected to local binary pattern analysis to obtain the surface texture feature descriptor, and the surface morphology is subjected to an anisotropy analysis according to the surface texture feature descriptor to obtain the surface anisotropy index; the surface anisotropy index and the multi-scale fractal dimension matrix are weightedly fused to obtain the preliminary wear resistance index, and the surface morphology is subjected to autocorrelation function analysis according to the preliminary wear resistance index to obtain the multi-scale wear resistance index of each coating sample; The optimization module is specifically used to: encode the multiple groups of surface coating process parameters, convert each group of process parameters into a multidimensional vector, and obtain an initial particle group; construct a fitness function according to the wear resistance evaluation index, evaluate the fitness of each particle, and obtain an initial fitness value; initialize the velocity of the initial particle group, and use Gaussian distribution to randomly generate an initial velocity vector of each particle; according to the initial fitness value, select the particle with the highest fitness as the global optimal solution, and set the initial position of each particle as the individual optimal solution to obtain an initial optimal solution set; iterate the initial particle group Update to obtain an updated particle swarm; according to the updated particle swarm, recalculate the fitness value of each particle, and update the individual optimal solution and the global optimal solution to obtain a new optimal solution set; perform diversity analysis on the new optimal solution set, calculate the aggregation degree of the particle swarm, and obtain a population diversity index; according to the population diversity index, perform dynamic mutation operation on the particle swarm to obtain a mutated particle swarm; perform boundary processing on the mutated particle swarm to ensure that the particle position is within the feasible solution space, obtain a valid particle swarm, and select the particle position corresponding to the global optimal solution, and decode to obtain the optimal coating process parameter combination.

Images