Physical metallurgy knowledge graph guided laser cladding wear-resistant coating design method

Abstract

This invention provides a laser cladding wear-resistant coating design method guided by a physical metallurgical knowledge graph, relating to the fields of nickel-based alloy wear-resistant coating design and machine learning applications. First, data is collected, and the data is divided into training and test sets using a multiple hold-out method. A graph neural network model guided by a physical metallurgical knowledge graph is established based on the training set. The graph neural network model with a squared correlation coefficient greater than 90% is used as the objective function in the genetic algorithm. The optimized composition, process, and material with the best target performance are obtained. This method introduces physical metallurgical mechanisms into a statistical graph neural network algorithm in the form of a knowledge graph, further improving the prediction accuracy of the machine learning model. Simultaneously, the combination with optimization algorithms makes the alloy design process more efficient, and the design results more consistent with the principles of physical metallurgy.

Description

Technical Field

[0001] This invention relates to the field of nickel-based alloy wear-resistant coating design and machine learning application technology, and in particular to a laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graphs. Background Technology

[0002] Laser cladding, as an advanced manufacturing technology, can produce dense, metallurgically bonded cladding layers that effectively reduce defects (such as porosity and weak interfacial bonding) while significantly improving wear resistance. It is increasingly being used for surface strengthening of critical components. Among available materials, nickel-based self-fluxing alloys have become a particularly suitable cladding material due to their excellent wettability, inherent self-lubricating properties, and good cost-effectiveness. However, with the rapid development of industrial applications, components are increasingly subjected to harsh service conditions involving sliding, impact, and abrasive wear mechanisms. These demanding environments place higher and more urgent requirements on the performance standards of wear-resistant coatings. Therefore, developing advanced nickel-based self-fluxing alloy wear-resistant coatings with superior performance under extreme conditions has become a key research focus.

[0003] The main solutions currently follow two strategies: (1) optimizing laser cladding process parameters (such as laser power, scanning speed, spot diameter, and powder feeding rate); and (2) adjusting the local chemical composition of nickel-based self-fluxing alloys, or adding high-melting-point carbides, borides, or rare earth oxides to the powder to improve the tribological properties of the coating. However, optimizing process parameters requires extensive orthogonal experiments. In terms of materials, research shows that novel nickel-based self-fluxing alloy coatings (especially those with specific elemental compositions or doped with hard ceramic particles) can significantly improve wear resistance. However, these advances are often accompanied by trade-offs, such as increased cost or increased crack sensitivity, which requires further optimization of laser cladding process parameters. Therefore, a key challenge is how to integrate composition optimization and process optimization based on the inherent characteristics of laser cladding to accelerate the development of high-performance nickel-based self-fluxing alloy coatings with higher wear resistance.

[0004] In recent years, machine learning methods, with their data-driven nature, have demonstrated significant advantages in predicting material properties and accelerating the development of new materials by establishing quantitative relationships between material composition, process parameters, and final performance. However, in the field of laser processing, existing research mainly focuses on improving the printability or mechanical properties of alloys by optimizing process parameters, while attention to tribological properties remains limited. Especially in the sub-field of laser cladding nickel-based self-fluxing alloy coatings, research based on machine learning is even scarcer. For example, some scholars have tried to use random forest models to predict coating wear resistance, but the average prediction accuracy is only 62.7%. This unsatisfactory performance stems from two key limitations: (1) insufficient training samples, making it difficult for traditional data-driven machine learning models to fully capture the complex "composition / process-wear resistance" relationship; and (2) insufficient understanding of the potential physical metallurgical mechanisms in laser cladding (such as molten pool dynamics and solidification behavior), resulting in the model's inability to identify deeper influencing factors. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graphs. By introducing physical metallurgical mechanisms into a statistical graph neural network algorithm in the form of a knowledge graph, the prediction accuracy of the machine learning model is further improved. At the same time, combined with optimization algorithms, the alloy design process is made more efficient and the design results are more in line with the principles of physical metallurgy.

[0006] To achieve the above objectives, the present invention mainly provides the following technical solutions:

[0007] The design method for laser cladding wear-resistant coatings guided by physical metallurgy knowledge graphs includes the following steps:

[0008] Step 1: Establish a dataset of laser cladding nickel-based alloy wear-resistant coatings, and preprocess it to obtain a standard dataset and a knowledge graph dataset;

[0009] Step 1.1: Collect literature on laser cladding nickel-based alloy wear-resistant coatings, and obtain the chemical composition, processing technology, and corresponding wear resistance of g types of nickel-based alloy wear-resistant coatings. Each set of raw data (chemical composition, laser cladding process, and wear resistance) constitutes a set of raw data, and the g sets of raw data form the original dataset. The composition refers to the elements and content of the material, and the process refers to the laser cladding process parameters. The original dataset is used as effective data for predicting the target performance of the material. All data in the original dataset are standardized to form the initial dataset.

[0010] Step 1.2: Based on the physical and metallurgical properties of the nickel-based alloy wear-resistant coating prepared by laser cladding process, the temperature field during single-layer, single-pass laser cladding is simulated for each type of laser-clad nickel-based alloy wear-resistant coating in the original dataset using the finite element method. Subsequently, the corresponding solidification mechanism parameters, including temperature gradient G, solidification growth rate R, and cooling rate, are calculated based on the Cartesian coordinate system. ) and solidification morphology parameters ( (and standardized them, adding each as a one-dimensional variable to the dataset to form a standard dataset);

[0011] The temperature gradient G and solidification growth rate R in the solidification mechanism parameters are calculated as follows:

[0012]

[0013]

[0014] in , , Represents spatial coordinates in the Cartesian coordinate system. For the corresponding temperature, For scanning speed;

[0015] Step 1.3: Based on the standard dataset and the hierarchical relationship between components, process parameters and solidification mechanism parameters, construct a knowledge graph dataset; in this knowledge graph, each data entry consists of nodes and directed edges: nodes represent the values ​​of components, process parameters or solidification mechanism parameters, and directed edges are represented in the form of (1,0) to describe the pointing relationship from component nodes, process parameter nodes to solidification mechanism parameter nodes;

[0016] Step 2: Set the partition ratio to 8:2, and divide the knowledge graph dataset into training and test sets for the i-th time using the multiple hold-out method; where the number of partitions i∈F, and F is the total number of partitions set according to experimental requirements;

[0017] Step 3: Train the graph attention network model (KGAT) based on the training set, i.e., the optimal KGAT model under the i-th partition;

[0018] Step 3.1: Establish a graph attention network model; the graph attention network model consists of graph attention convolutional layers, fully connected layers, and an output layer. The number of fully connected layers is D. The range of H for the graph attention convolutional layers in the model is defined to form a parameter set; the activation function is ReLU, and the approximate derivation of the ReLU formula is as follows:

[0019]

[0020] Where x is the input value;

[0021] Step 3.2: Build a KGAT model based on the training set; use the knowledge graph constructed based on chemical composition, processing technology, and solidification mechanism parameters as the input to the model, and use wear resistance performance as the output; finally, obtain U KGAT models; calculate the squared correlation coefficient R of each of the U models. 2 and mean absolute error (MAE);

[0022]

[0023]

[0024] Where n is the total amount of data in the test set, x i For the i-th data in the test set, f(x) i ) is x i The wear resistance prediction value of the KGAT model, y i For x i The corresponding actual value of wear resistance;

[0025] Step 3.3: Compare the R values ​​of the U models 2 And MAE value, make R 2 The model with the largest value and the smallest MAE value is set as the optimal KGAT model under the i-th partition, and the parameter set corresponding to this model is set as the optimal combination parameter, where i∈50;

[0026] Step 4: Determine the R-value of the KGAT model constructed in Step 3. 2 If the percentage is greater than 90%, delete the model; if it is, proceed to step 5 and use the model as the objective function in the genetic algorithm.

[0027] Step 5: Within the scope of the original dataset, combine the KGAT model with the genetic algorithm to design the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating; use the KGAT model as the objective function of the genetic algorithm to optimize the design of the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating, and obtain the laser cladding nickel-based alloy wear-resistant coating with the best wear resistance performance.

[0028] Step 5.1: Based on the standard dataset, establish a prediction model for solidification mechanism parameters: use chemical composition and processing technology as input to the CNN model, and use solidification mechanism parameters as the output of the model; finally, obtain the prediction model for solidification mechanism parameters.

[0029] Step 5.2: Initialize the population. First, randomly generate W chromosomes from the original dataset. Each chromosome represents a combination of chemical composition and processing technology of a laser cladding nickel-based alloy coating. The W chromosomes form a chromosome set Q.

[0030] Step 5.3: Calculate the chemical composition and coagulation mechanism parameters corresponding to the processing technology in each chromosome using the coagulation mechanism parameter prediction model, forming the generation o dataset, where o represents the current iteration number; standardize the generation o dataset and input it into the optimal KGAT model to calculate the wear resistance performance;

[0031] Step 5.4: Calculate the fitness of each chromosome according to the fitness function and sort them in descending order; determine if the current iteration is the first iteration. If yes, retain the top 90% of individuals by fitness, delete the remaining 10%, and proceed to step 5.5; if no, retain the top 90% of individuals by fitness, and perform selection, crossover, and mutation on the remaining 10% of individuals to generate new chromosomes, resulting in a new chromosome set M. o Proceed to step 5.5; the chromosome set M o Including chromosome M o-1 The 90% of individuals with high fitness and newly generated chromosomes;

[0032] The fitness function FitnV is:

[0033]

[0034] Where Nind is the number of individuals in the population, sp is the selection pressure difference, and the value of sp ranges from [1,2]; Pos is the position of the individual in the ordination population;

[0035] Step 5.5: Determine whether the termination condition has been met. If yes, output the alloy composition, process and target performance, and put the result into the design result set M. If no, execute step 5.3. The termination condition is that all chromosome individuals in the current chromosome group converge to the same result, and the results of ten consecutive iterations are consistent.

[0036] The beneficial effects of adopting the above technical solution are as follows: This invention provides a laser cladding wear-resistant coating design method guided by a physical metallurgical knowledge graph. This method uses a graph neural network model guided by a physical metallurgical knowledge graph to establish the relationship between composition, process, solidification mechanism parameters and target performance. A genetic algorithm is then applied to quickly and accurately optimize the wear resistance design within the original dataset, achieving highly reliable design results and forming a complete advanced materials rational design platform. This invention is the first to integrate a physical metallurgical knowledge graph into machine learning, giving statistically based machine learning a physical metallurgical meaning. Furthermore, this method can improve the model's generalization ability, making the design more efficient and the design results more consistent with the principles of physical metallurgy. Attached Figure Description

[0037] Figure 1 Design flowchart provided for embodiments of the present invention;

[0038] Figure 2 This is a diagram showing the optimal prediction result of the friction coefficient under multiple random divisions provided in an embodiment of the present invention.

[0039] Figure 3 The diagram shows the optimal prediction results of solidification mechanism parameters provided in the embodiments of the present invention.

[0040] Figure 4 A comparison chart of calculation verification based on precipitation strengthening and crack sensitivity provided for embodiments of the present invention; Detailed Implementation

[0041] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0042] In this invention, solidification mechanism parameters highly correlated with target performance are added to the dataset to participate in the construction of the knowledge graph, model training, and design process, thus imbuing the machine learning process with physical meaning. Simultaneously, a genetic algorithm is applied to optimize the composition process design, the design flow of which is as follows: Figure 1 As shown. Compared to traditional machine learning, the graph neural network model guided by knowledge graphs has higher prediction accuracy and the design results are more in line with the principles of physical metallurgy. Based on this rational design method, the wear resistance of the laser-clad nickel-based alloy coating was successfully optimized. The method of this embodiment is described below.

[0043] This invention provides a method for designing laser cladding wear-resistant coatings guided by a physical metallurgical knowledge graph, such as... Figure 1 Specifically, the process includes the following steps:

[0044] Step 1: Establish a dataset of laser cladding nickel-based alloy wear-resistant coatings, and preprocess it to obtain a standard dataset and a knowledge graph dataset;

[0045] Step 1.1: Collect literature on laser cladding nickel-based alloy wear-resistant coatings, obtaining the chemical composition, processing technology, and corresponding friction coefficients of 53 nickel-based alloy wear-resistant coatings (g=53). Each nickel-based alloy wear-resistant coating's chemical composition, laser cladding process, and wear resistance performance constitutes a set of raw data. These 53 sets of raw data form the original dataset. The composition refers to the elements and content of the material, and the process refers to the laser cladding process parameters. The original dataset serves as effective data for predicting the target performance of the material. All data in the original dataset are standardized to form the initial dataset. The standardization formula is: , where z is the standardized data, x is the original data, μ is the mean of each dimension of the original data, and σ is the standard deviation of each dimension of the original data;

[0046] Step 1.2: Based on the physical and metallurgical properties of the nickel-based alloy wear-resistant coating prepared by laser cladding process, the temperature field during single-layer, single-pass laser cladding is simulated for each type of laser-clad nickel-based alloy wear-resistant coating in the original dataset using the finite element method. Subsequently, the corresponding solidification mechanism parameters, including temperature gradient G, solidification growth rate R, and cooling rate, are calculated based on the Cartesian coordinate system. ) and solidification morphology parameters ( (and standardized them, adding each as a one-dimensional variable to the dataset to form a standard dataset);

[0047] The temperature gradient G and solidification growth rate R in the solidification mechanism parameters are calculated as follows:

[0048]

[0049]

[0050] in , , Represents spatial coordinates in the Cartesian coordinate system. For the corresponding temperature, For scanning speed;

[0051] Step 1.3: Based on the standard dataset and the hierarchical relationship between components, process parameters and solidification mechanism parameters, construct a knowledge graph dataset; in this knowledge graph, each data entry consists of nodes and directed edges: nodes represent the values ​​of components, process parameters or solidification mechanism parameters, and directed edges are represented in the form of (1,0) to describe the pointing relationship from component nodes, process parameter nodes to solidification mechanism parameter nodes;

[0052] The data distribution within the knowledge graph dataset in this embodiment is shown in Table 1.

[0053] Table 1. Data Distribution in the Knowledge Graph Dataset

[0054] Step 2: Set the partition ratio to 8:2, and divide the knowledge graph dataset into training and test sets for the i-th time using the multiple hold-out method; where the number of partitions i∈F, and F is the total number of partitions set according to experimental requirements;

[0055] In this embodiment, F = 50;

[0056] Step 3: Train the graph attention network model, i.e., the KGAT model, based on the training set;

[0057] Step 3.1: Establish a graph attention network model; the graph attention network model consists of graph attention convolutional layers, fully connected layers, and an output layer. The number of fully connected layers is 4, and the number of graph attention convolutional layers in the model is set to H=6; the activation function is ReLU, and the ReLU formula is approximately derived as follows:

[0058]

[0059] Where x is the input value;

[0060] Step 3.2: Build a KGAT model based on the training set; use the knowledge graph constructed based on chemical composition, processing technology and solidification mechanism parameters as the input of the model, and use the friction coefficient as the output of the model;

[0061] Step 3.3: Use the multiple hold-out method (50 random datasets divided into groups, squared correlation coefficient R) 2 The performance of the KGAT models was evaluated using the mean and maximum values, resulting in 50 KGAT models. The squared correlation coefficient R0 of each of the 50 models was then calculated. 2 and Mean Absolute Error (MAE), where R on the training and test sets 2 More than 90% of the model prediction results are displayed in Figure 2 Specifically, such as Figure 2 As shown in (a), when i=1, R on the training and test sets... 2 They were 92.81% and 93.5% respectively; Figure 2 As shown in (b), when i=30, R on the training and test sets... 2 They were 91.62% and 97.45% respectively; Figure 2 As shown in (c), when i=34, R on the training and test sets... 2 The percentages were 94.54% and 97.87% respectively; Figure 2 As shown in (d), when i=39, R on the training and test sets... 2 They were 94.29% and 92.54% respectively; Figure 2 As shown in (e), when i=49, R on the training and test sets... 2 The percentages were 94.43% and 91.92%, respectively. The vast majority of data points in both the training and test sets were very close to the diagonal line in the graph, which fully reflects that most predicted values ​​are very consistent with experimental values, indicating that the KGAT model has high prediction accuracy.

[0062]

[0063]

[0064] Where n is the total amount of data in the test set, x iFor the i-th data in the test set, f(x) i ) is x i The wear resistance prediction value of the KGAT model, y i For x i The corresponding actual value of wear resistance;

[0065] Step 4: Determine the R-value of the KGAT model constructed in Step 3. 2 If the percentage is greater than 90%, delete the model; if it is, proceed to step 5 and use the model as the objective function in the genetic algorithm.

[0066] Step 5: Within the scope of the original dataset, combine the KGAT model with a genetic algorithm to design the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating; use the KGAT model as the objective function of the genetic algorithm to optimize the design of the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating, and obtain the laser cladding nickel-based alloy wear-resistant coating with the best wear resistance under specific test conditions (test load 30 N, sliding speed 10 mm / s);

[0067] Step 5.1: Based on the standard dataset, establish a solidification mechanism parameter prediction model: use chemical composition and processing technology as input to the CNN model, and use the solidification mechanism parameters as the output of the model; finally, obtain the solidification mechanism parameter prediction model; the model's prediction results for the solidification mechanism parameters are displayed on... Figure 3 Specifically, the model's prediction of the solidification growth rate is as follows: Figure 3 As shown in (a), R on the training and test sets 2 The accuracy rates were 96.58% and 92.97% respectively; while the prediction of temperature gradients was as follows: Figure 3 As shown in (b), R on the training and test sets 2 The accuracy rates were 97.39% and 94.04% respectively, indicating that the model has high prediction accuracy and the vast majority of data points are concentrated near the diagonal.

[0068] Step 5.2: Initialize the population. First, randomly generate W chromosomes from the original dataset. Each chromosome represents a combination of chemical composition and processing technology of a laser cladding nickel-based alloy coating. The W chromosomes form a chromosome set Q.

[0069] Step 5.3: Calculate the chemical composition and coagulation mechanism parameters corresponding to the processing technology in each chromosome using the coagulation mechanism parameter prediction model, forming the generation o dataset, where o represents the current iteration number; standardize the generation o dataset and input it into the optimal KGAT model to calculate the wear resistance performance;

[0070] Step 5.4: Calculate the fitness of each chromosome according to the fitness function and sort them in descending order; determine if the current iteration is the first iteration. If yes, retain the top 90% of individuals by fitness, delete the remaining 10%, and proceed to step 5.5; if no, retain the top 90% of individuals by fitness, and perform selection, crossover, and mutation on the remaining 10% of individuals to generate new chromosomes, resulting in a new chromosome set M. o Proceed to step 5.5; the chromosome set M o Including chromosome M o-1 The 90% of individuals with high fitness and newly generated chromosomes;

[0071] The fitness function FitnV is:

[0072]

[0073] Where Nind is the number of individuals in the population, sp is the selection pressure difference, and the value of sp ranges from [1,2]; Pos is the position of the individual in the ordination population;

[0074] Step 5.5: Determine whether the termination condition has been met. If yes, output the alloy composition, process and target performance, and put the result into the design result set M. If no, execute step 5.3. The termination condition is that all chromosome individuals in the current chromosome group converge to the same result, and the results of ten consecutive iterations are consistent.

[0075] This embodiment uses 50 different KGAT models, and selects 5 of them in the design section. 2 >90% of the models were designed by combining the selected models with GA. The solidification mechanism prediction model was used to calculate the solidification mechanism parameters of each individual in the genetic algorithm. Finally, under specific test conditions (test load 30 N, sliding speed 10 mm / s), five design results were obtained, as shown in Table 2. The friction coefficient of the five design results was better than that of the original dataset.

[0076] Table 2 Design Results of Laser Cladding Nickel-Based Alloy Coatings

[0077] Considering the common defects such as cracks and porosity in the laser cladding process of nickel-based self-fluxing alloy coatings, a comprehensive evaluation of the printability of the designed knowledge graph-optimized alloy is crucial. Meanwhile, the wear resistance of laser-clad nickel-based self-fluxing alloy coatings is mainly dominated by the strengthening effect of hard precipitates. To collaboratively evaluate the printability and wear resistance of the designed alloy, two key parameters were calculated based on ThermoCalc software and the Scheil-Gulliver model: (1) the crack sensitivity factor in the final solidification stage (solid fraction range: 0.9–0.99), which is directly related to the hot cracking tendency; and (2) the phase composition in the near-final solidification stage (solid fraction = 0.99), which determines the generation potential of the hard phase. The calculation results are as follows: Figure 4 As shown, the results demonstrate that the alloy coating designed using the knowledge graph exhibits an optimal combination of printability and wear resistance, located in the upper right quadrant (marked with a green star). Specifically, the alloy displays a significantly lower crack sensitivity factor (100-500 K), indicating a lower tendency for hot cracking during solidification. Simultaneously, the precipitated phase fraction reaches approximately 50% at a solid fraction of 0.99, ensuring sufficient hard phase to enhance wear resistance. This synergistic improvement is attributed to the optimized composition design, highlighting the importance of incorporating physical metallurgy knowledge graphs to guide knowledge graph attention network models for accurate performance prediction and reliable alloy design.

Claims

1. A method for designing laser cladding wear-resistant coatings guided by a physical metallurgical knowledge graph, characterized in that, Includes the following steps: Step 1: Establish a dataset of laser cladding nickel-based alloy wear-resistant coatings, and preprocess it to obtain a standard dataset and a knowledge graph dataset; Step 2: Set the partition ratio to 8:2, and use the multiple hold-out method to partition the knowledge graph dataset into training and test sets for the i-th time; Where the number of partitions i∈F, and F is the total number of partitions set according to the experimental requirements; Step 3: Train the graph attention network model based on the training set, i.e., the optimal KGAT model under the i-th partition; Step 4: Determine the R-value of the KGAT model constructed in Step 3. 2 If the percentage is greater than 90%, delete the model; if it is, proceed to step 5 and use the model as the objective function in the genetic algorithm. Step 5: Within the scope of the original dataset, combine the KGAT model with a genetic algorithm to design the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating; use the KGAT model as the objective function of the genetic algorithm to optimize the design of the chemical composition and processing technology of the laser cladding nickel-based alloy wear-resistant coating, and obtain the laser cladding nickel-based alloy wear-resistant coating with the best wear resistance performance; the termination condition is that all chromosome individuals in the current chromosome group converge to the same result, and the results of ten consecutive iterations are consistent.

2. The laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graph according to claim 1, characterized in that, Step 1 is as follows: Step 1.1: Collect literature on laser cladding nickel-based alloy wear-resistant coatings, obtain the chemical composition, processing technology and corresponding wear resistance of g kinds of nickel-based alloy wear-resistant coatings. The chemical composition, laser cladding process and wear resistance of each nickel-based alloy wear-resistant coating are a set of raw data, and the g sets of raw data constitute the original dataset. Step 1.2: Based on the physical and metallurgical properties of the nickel-based alloy wear-resistant coating prepared by laser cladding process, the temperature field during single-layer, single-pass laser cladding is simulated for each type of laser-clad nickel-based alloy wear-resistant coating in the original dataset using the finite element method. Subsequently, the corresponding solidification mechanism parameters, including temperature gradient G, solidification growth rate R, and cooling rate, are calculated based on the Cartesian coordinate system. ) and solidification morphology parameters ( (and standardized them, adding each as a one-dimensional variable to the dataset to form a standard dataset); The temperature gradient G and solidification growth rate R in the solidification mechanism parameters are calculated as follows: in , , Represents spatial coordinates in the Cartesian coordinate system. For the corresponding temperature, For scanning speed; Step 1.3: Based on the standard dataset and the hierarchical relationship between components, process parameters and solidification mechanism parameters, construct a knowledge graph dataset; In this knowledge graph, each data entry consists of nodes and directed edges: nodes represent the values ​​of components, process parameters, or solidification mechanism parameters, and directed edges are represented in the form of (1,0) to describe the pointing relationship from component nodes, process parameter nodes to solidification mechanism parameter nodes.

3. The laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graph according to claim 2, characterized in that, The components refer to the elements and content of the material, and the process refers to the laser cladding process parameters.

4. The laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graph according to claim 1, characterized in that, Step 3 specifically involves: Step 3.1: Establish a graph attention network model; the graph attention network model consists of graph attention convolutional layers, fully connected layers, and an output layer. The number of fully connected layers is D. The range of H for the graph attention convolutional layers in the model is defined to form a parameter set; the activation function is ReLU, and the approximate derivation of the ReLU formula is as follows: Where x is the input value; Step 3.2: Build a KGAT model based on the training set; use the knowledge graph constructed based on chemical composition, processing technology, and solidification mechanism parameters as the input to the model, and use wear resistance performance as the output; finally, obtain U KGAT models; calculate the squared correlation coefficient R of each of the U models. 2 and mean absolute error (MAE); Where n is the total amount of data in the test set, x i For the i-th data in the test set, f(x) i ) is x i The wear resistance prediction value of the KGAT model, y i For x i The corresponding actual value of wear resistance; Step 3.3: Compare the R values ​​of the U models 2 And MAE value, make R 2 The model with the largest value and the smallest MAE value is set as the optimal KGAT model under the i-th partition, and the parameter set corresponding to this model is set as the optimal combination parameter, where i∈50.

5. The laser cladding wear-resistant coating design method guided by physical metallurgical knowledge graph according to claim 1, characterized in that, Step 5 specifically involves: Step 5.1: Based on the standard dataset, establish a prediction model for solidification mechanism parameters: use chemical composition and processing technology as input to the CNN model, and use solidification mechanism parameters as the output of the model; finally, obtain the prediction model for solidification mechanism parameters. Step 5.2: Initialize the population. First, randomly generate W chromosomes from the original dataset. Each chromosome represents a combination of chemical composition and processing technology of a laser cladding nickel-based alloy coating. The W chromosomes form a chromosome set Q. Step 5.3: Using the coagulation mechanism parameter prediction model, calculate the coagulation mechanism parameters corresponding to the chemical composition and processing technology of each chromosome to form the generation o dataset, where o represents the current iteration number; The o-th generation dataset is standardized and input into the optimal KGAT model to calculate the wear resistance performance. Step 5.4: Calculate the fitness of each chromosome according to the fitness function and sort them in descending order; determine if the current iteration is the first iteration. If yes, retain the top 90% of individuals by fitness, delete the remaining 10%, and proceed to step 5.5; if no, retain the top 90% of individuals by fitness, and perform selection, crossover, and mutation on the remaining 10% of individuals to generate new chromosomes, resulting in a new chromosome set M. o Proceed to step 5.5; the chromosome set M o Including chromosome M o-1 The 90% of individuals with high fitness and newly generated chromosomes; The fitness function FitnV is: Where Nind is the number of individuals in the population, sp is the selection pressure difference, and the value of sp ranges from [1,2]; Pos is the position of the individual in the ordination population; Step 5.5: Determine whether the termination condition has been met. If yes, output the alloy composition, process and target performance, and put the result into the design result set M; if no, execute step 5.

3. The termination condition is that all chromosome individuals within the current chromosome set converge to the same result, and the results of ten consecutive iterations are consistent.

Images