A data-driven optimization method for cold spraying of multi-component composite coatings

Abstract

This invention provides a data-driven optimization method for cold spraying processes of multi-component composite coatings, relating to the field of cold spraying. The method includes: establishing a cold spraying process-performance database for metal / ceramic composite coatings; constructing an objective function for optimizing the performance of metal / ceramic composite coatings based on an adaptive neural network; optimizing the cold spraying process of metal / ceramic composite coatings based on a locally enhanced NSGA-II genetic algorithm; and conducting cold spraying experiments to verify the effectiveness of the optimization results based on the optimal combination of process parameters obtained from the optimization. This optimization method, by combining experiments and big data analysis, can quickly and accurately determine the optimal parameter combination for the cold spraying process, effectively improving the quality and performance of the composite coating.

Description

Technical Field

[0001] This invention relates to the field of cold spraying technology, and more specifically to a data-driven method for optimizing the cold spraying process of multi-component composite coatings. Background Technology

[0002] Metal / ceramic composite coatings are composite coating materials formed by combining metal and ceramic materials. They break through the performance limitations of single coatings and have the characteristics of high hardness and wear resistance, high temperature resistance, corrosion resistance, and excellent electrical and thermal conductivity. They provide effective solutions for solving complex material surface protection and performance improvement problems in many fields such as aerospace, marine engineering, and automobile manufacturing. Cold spraying, also known as cold gas dynamic spraying or cold gas kinetic energy spraying, is based on aerodynamics. It uses a high-pressure gas source to propel solid particles to extremely high speeds to impact the substrate, thereby depositing and forming a coating. As a novel material surface coating technology, it has advantages such as minimal thermal impact on the substrate and portable application, showing great potential in the preparation of metal / ceramic composite coatings. However, the cold spraying process is complex and is affected by many factors such as the pressure and temperature of the spraying gas, the particle size of the powder, the distance and angle between the spray gun and the substrate, and the surface condition of the substrate. This can easily lead to problems such as poor interfacial bonding between the constituent materials, high coating porosity, low bonding strength between the coating and the substrate, and uneven thickness, resulting in substandard coating performance and the inability to fully realize the synergistic effect of the metal / ceramic composite coating.

[0003] Currently, the optimization of cold spraying processes for metal / ceramic composite coatings mainly relies on experience or trial-and-error methods. This involves conducting numerous experiments by changing process parameters and adjusting the process based on observed changes in coating quality and performance to achieve relatively satisfactory coating results. This experience-based or trial-and-error-based optimization method not only consumes a lot of labor and time but also makes it difficult to fully consider the interrelationships between various process parameters, thus failing to obtain the truly globally optimal combination of process parameters. Consequently, the performance of cold-sprayed composite coatings is difficult to achieve at its best level. At the same time, experience-based or trial-and-error-based optimization methods also require a high level of experience from operators and are subject to significant human influence, which in turn affects the accuracy of process optimization. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention aims to provide a data-driven optimization method for cold spraying of multi-component composite coatings. This optimization method, by combining experiments and big data analysis, can quickly and accurately determine the optimal parameter combination for the cold spraying process, thereby effectively improving the quality and performance of the composite coating and solving the problems of low efficiency, poor accuracy, and large human influence in traditional cold spraying process optimization methods.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A data-driven method for optimizing the cold spraying process of multi-component composite coatings includes:

[0007] Step S1: Establish a performance database for cold spraying processes of metal / ceramic composite coatings;

[0008] Step S2: Construct an objective function for optimizing the performance of metal / ceramic composite coatings based on an adaptive neural network;

[0009] Step S3: Optimize the cold spraying process of metal / ceramic composite coating based on the local search enhanced NSGA-II genetic algorithm;

[0010] Step S4: Based on the optimized combination of process parameters for cold spraying of metal / ceramic composite coatings, conduct cold spraying tests to verify the effectiveness of the optimization results.

[0011] Based on further optimization of the above scheme, step S1 specifically includes:

[0012] Step S11: Determine the key process parameters that affect the coating quality and performance based on the coating performance indicators;

[0013] Step S12: Determine the initial value range of the key process parameters for cold spraying, and use super Latin cube sampling to select multiple combinations of key process parameters for cold spraying within the initial value range to ensure the uniformity and representativeness of the parameter space.

[0014] Step S13: Conduct experiments on the preparation of metal / ceramic composite coatings using cold spraying equipment according to the selected key process parameter combinations; during the preparation process, record the parameter sets of each experimental combination in real time, i.e., the... i The set of key process parameters for the group experiment is as follows Q i ; and after the experiment, the properties of the prepared composite coating were tested and recorded in real time to obtain a set of coating properties, the first i The set of coating properties corresponding to the group of tests is as follows R i Finally, the key process parameters and corresponding coating performance data used in each group of experiments were compiled and entered into the database to complete the construction of the cold spraying process-performance database.

[0015] Based on further optimization of the above scheme, the key process parameters include cold spray gas pressure, spray temperature, spray distance, metal / ceramic powder particle size, and the proportion of metal / ceramic powder.

[0016] Based on further optimization of the above scheme, the performance of the composite coating includes coating porosity, coating deposition rate, bonding strength between coating and substrate, coating hardness, etc.

[0017] Based on further optimization of the above scheme, step S2 specifically includes:

[0018] Step S21: Preprocess the cold spraying process performance data to eliminate the dimensionality influence between different data and improve data quality and usability. Specifically:

[0019]

[0020] In the formula: x * Indicates the first preprocessed step i One data point; x i Representing variables x The first in i One set of raw data;

[0021] Step S22: Divide the preprocessed process-performance data into training set, validation set and test set using a hierarchical partitioning method; and determine the number of neurons in the input layer, hidden layer, output layer, as well as the number of hidden layers and activation function according to the process parameters and target performance parameters, and construct a neural network.

[0022] Step S23: First, input the training set into the constructed neural network, calculate the output through forward propagation, and obtain the model output value;

[0023] Then, the mean squared error loss function is used. MSE Calculate the difference between the model output value and the target value:

[0024]

[0025] In the formula: N Indicates the number of samples. M Indicates the number of neurons in the output layer; for the th i One sample, Indicates the first j The true value of the output, i.e., the model output value. Indicates the first j The predicted value of each output, i.e., the target value;

[0026] Then, define the adaptive learning optimizer:

[0027]

[0028] In the formula: w t Indicates the neural network's first... t The parameters for the next iteration; Indicates the learning rate; , "Don't" represents the attenuation coefficient; m t, v t They represent the first t The first and second moments of the gradient are predicted in the next iteration:

[0029]

[0030] In the formula: g t Indicates the first t The gradient of the parameters at the next iteration;

[0031] The weights and biases of the neural network are updated based on backpropagation of the loss.

[0032] Finally, the loss of the validation set is monitored in real time during training. When the loss of the validation set no longer decreases, training is stopped, and the trained neural network model is obtained.

[0033] At the same time, using the mean square error loss function MSE The threshold is used to determine whether the trained neural network model can predict the coating performance based on the process parameters. If not, the number of neurons in the hidden layer is adjusted and the model is retrained.

[0034] Based on further optimization of the above scheme, the mean square error loss function MSE With a threshold of 0.1, the mean squared error loss function MSE Not less than the mean square error loss function MSE At the threshold, a well-trained neural network model is considered capable of predicting coating performance based on process parameters.

[0035] Based on further optimization of the above scheme, step S3 specifically includes:

[0036] Step S31: Using the cold spraying process parameters of the metal / ceramic composite coating as optimization variables, and with the optimization objectives of minimizing coating porosity, maximizing coating deposition rate, and ensuring coating bonding strength meets requirements, establish a cold spraying process optimization model:

[0037]

[0038] Constraints:

[0039] In the formula: F(X) represents the set of optimization objectives; This indicates minimizing the coating porosity. Indicates porosity; This indicates maximizing the coating deposition rate. Indicates the deposition rate; S Indicates the bonding strength. S cThis represents the bonding strength threshold, which is obtained based on the actual optimized composite coating material, substrate material, and experimental data. X This indicates the corresponding cold spraying process parameters. X max , X min These represent the upper and lower limits of the process parameters, respectively.

[0040] Step S32: Using real number encoding, represent each cold spraying process parameter with a real number, and randomly generate a certain scale (let's call it a real number). N The initial population of 0) G 0, the parameter values ​​for each individual are randomly generated within their corresponding value range;

[0041] Step S33: For each individual in the population, substitute its decoded process parameter value into the trained neural network model to replace the actual experimental process and obtain the objective function value of each individual;

[0042] Step S34: First, the initial population... G All individuals in the population are sorted and their crowding is calculated according to non-dominance relationships. For any two individuals in the population... i and j The degree of crowding between them CI for:

[0043]

[0044] in, N 0 represents the population size; x ik Represents an individual i The eigenvector of the first k One portion, x jk Represents an individual j The eigenvector of the first k One component;

[0045] Select, crossover, and mutation operations are performed on individuals with non-dominated frontiers or high crowding.

[0046] Then, the newly generated combination of process parameters is returned to the neural network model to perform coating performance prediction, and a progeny population is generated. g 0, and simultaneously merge with the parent population to generate a capacity of 2. N Population of 0 G 0 g 0;

[0047] Step S35, then analyze the population. G 0 g 0. Perform non-dominated sorting and crowding calculation, and select based on elite strategy.N The new parent population consists of 0 individuals. G i ; for the new paternal population G i Perform selection, crossover, mutation operations and objective function evaluation to select the next generation population. G i+1 ;

[0048] Step S36: Monitor the changes in the objective function values ​​of individuals in the population. If the distance between the current non-dominated solution set and the known optimal frontier is less than... d When this occurs, a local search is initiated; the neighborhood search range is set ± for the cold spraying process parameters of interest. h Starting from the selected initial individual, the values ​​of the decision variables are changed sequentially within the given neighborhood search range; if the objective function value corresponding to the new individual is better than that of the current individual, the new individual is accepted as the current individual, until no better solution can be found in the local region;

[0049] Step S37: When the optimal solution of the population does not change significantly for several consecutive generations or when the number of iterations reaches a preset value, stop the iteration and output the final population, that is, obtain the optimal combination of process parameters within the range of process parameter values.

[0050] The following are the effects achieved by the technical means of this invention:

[0051] This invention combines experimental design and big data analysis, employing data preprocessing, mean square error loss function, and adaptive learning optimizer. Based on a neural network model, it establishes a mapping relationship between cold spraying process parameters and coating performance. By inputting process parameters, it predicts coating performance, effectively solving the problems of numerous process parameters, complex processes, and nonlinearity in metal / ceramic composite coating spraying. Simultaneously, it utilizes the NSGA-II genetic algorithm to rapidly search for the optimal combination of process parameters. This not only overcomes the limitations of traditional process optimization methods but also effectively avoids the insufficient local search capability of the traditional NSGA-II genetic algorithm, enabling precise searching for the optimal combination. This improves the optimization efficiency and accuracy of the cold spraying process for metal / ceramic composite coatings, thus producing high-quality, high-performance composite coatings to meet the stringent surface performance requirements of various fields. Attached Figure Description

[0052] Figure 1 This is a flowchart of a data-driven method for optimizing the cold spraying process of multi-component composite coatings in an embodiment of the present invention.

[0053] Figure 2 The image shows the morphology of the Ni / Al2O3 composite coating prepared using a traditional empirical optimization method.

[0054] Figure 3The image shows the morphology of the Ni / Al2O3 composite coating prepared using the data-driven, multi-component composite coating cold spraying process optimization method of this invention. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below. In the following description, specific details such as specific system structures and technologies are presented for illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention.

[0056] In this embodiment, a data-driven optimization method for cold spraying of multi-component composite coatings is described in the flowchart below. Figure 1 As shown, taking the optimization of the cold spraying process for a composite coating of metallic Ni and ceramic Al2O3 as an example, it includes:

[0057] Step S1: Establish a performance database for cold spraying processes of metal / ceramic composite coatings.

[0058] Step S11: Based on the coating performance indicators, determine the key process parameters that affect the coating quality and performance. Key process parameters include cold spray gas pressure, spray temperature, spray distance, metal / ceramic powder particle size, and the proportion of metal / ceramic powder. Specifically, in this embodiment, key process parameters include gas pressure P, spray temperature T, particle size D1 of metal powder Ni, particle size D2 of ceramic powder Al2O3, and the proportion φ of ceramic powder Al2O3 in the composite powder.

[0059] Step S12: Determine the initial value range of the key process parameters for cold spraying (for example, in this embodiment, the spraying pressure range is: 1MPa≤P≤3MPa, the spraying temperature range is: 200℃≤T≤500℃, the particle size range of metal powder Ni is: 5μm≤D1≤50μm, the particle size range of ceramic powder Al2O3 is: 20μm≤D2≤100μm, and the proportion of ceramic powder Al2O3 in the composite powder is: 0<φ≤0.5). Then, use super Latin cubic sampling to select multiple combinations of key process parameters for cold spraying within the initial value range (50 combinations are selected in this embodiment) to ensure the uniformity and representativeness of the parameter space.

[0060] Step S13: Conduct experiments on the preparation of metal / ceramic composite coatings using cold spraying equipment according to the selected key process parameter combinations; during the preparation process, record the parameter sets of each experimental combination in real time, i.e., the... i The set of key process parameters for the group experiment is as follows Q i After the experiment, the properties of the prepared composite coating were tested and recorded in real time. The composite coating properties included coating porosity, coating deposition rate, bonding strength between the coating and the substrate, and coating hardness, etc., to obtain a set of coating properties.i The set of coating properties corresponding to the group of tests is as follows R i Finally, the key process parameters and corresponding coating performance data used in each group of experiments were compiled and entered into the database to complete the construction of the cold spraying process-performance database.

[0061] Step S2: Construct an objective function for optimizing the performance of metal / ceramic composite coatings based on an adaptive neural network.

[0062] Step S21: Preprocess the cold spraying process performance data to eliminate the dimensionality influence between different data and improve data quality and usability. Specifically:

[0063]

[0064] In the formula: x * Indicates the first preprocessed step i One data point; x i Representing variables x The first in i One set of raw data;

[0065] Step S22: Divide the preprocessed process-performance data into training set, validation set, and test set using a hierarchical partitioning method (wherein the ratio of training set, validation set, and test set is 7:2:1); and determine the number of neurons in the input layer, hidden layer, and output layer (in this embodiment, the number of neurons in the input layer, hidden layer, and output layer are 5, 10, and 4, respectively), the number of hidden layers (in this embodiment, the number of hidden layers is 1), and the activation function (using the sigmoid activation function) according to the process parameters and target performance parameters, and construct the neural network;

[0066] Step S23: First, input the training set into the constructed neural network, calculate the output through forward propagation, and obtain the model output value;

[0067] Then, the mean squared error loss function is used. MSE Calculate the difference between the model output value and the target value:

[0068]

[0069] In the formula: N Indicates the number of samples. M Indicates the number of neurons in the output layer; for the th i One sample, Indicates the first j The true value of the output, i.e., the model output value. Indicates the first j The predicted value of each output, i.e., the target value;

[0070] Then, define the adaptive learning optimizer:

[0071]

[0072] In the formula: w t Indicates the neural network's first... t The parameters for the next iteration; Indicates the learning rate; , "Don't represent the attenuation coefficient (in this embodiment, , ); m t , v t They represent the first t The first and second moments of the gradient are predicted in the next iteration:

[0073]

[0074] In the formula: g t Indicates the first t The gradient of the parameters at the next iteration;

[0075] The weights and biases of the neural network are updated based on backpropagation of the loss.

[0076] Finally, the loss of the validation set is monitored in real time during training. When the loss of the validation set no longer decreases, training is stopped, and the trained neural network model is obtained.

[0077] At the same time, using the mean square error loss function MSE The threshold, i.e., 0.1, is used to determine whether the trained neural network model can predict coating performance based on process parameters. When the mean square error loss function... MSE When the value is ≥0.1, the trained neural network model is considered to be able to predict coating performance based on process parameters; otherwise, it cannot, and the number of neurons in the hidden layer should be adjusted and the model should be retrained.

[0078] Step S3: Optimize the cold spraying process of metal / ceramic composite coatings based on the locally enhanced NSGA-II genetic algorithm:

[0079] Step S31: Using the cold spraying process parameters of the metal / ceramic composite coating as optimization variables, and with the optimization objectives of minimizing coating porosity, maximizing coating deposition rate, and ensuring coating bonding strength meets requirements, establish a cold spraying process optimization model:

[0080]

[0081] Constraints:

[0082] In the formula: F(X) represents the set of optimization objectives; This indicates minimizing the coating porosity. Indicates porosity; This indicates maximizing the coating deposition rate. Indicates the deposition rate; S Indicates the bonding strength. S c Indicates the binding strength threshold (in this embodiment, S c =40MPa), combined with the strength threshold obtained based on the actual optimized composite coating material, substrate material and test data; X This indicates the corresponding cold spraying process parameters. X max , X min These represent the upper and lower limits of the process parameters, respectively.

[0083] Step S32: Using real number encoding, represent each cold spraying process parameter with a real number, and randomly generate a certain scale (let's call it a real number). N 0, for example: N The initial population was 0 (or 20). G 0, the parameter values ​​for each individual are randomly generated within their corresponding value range; the... i individual X i It can be represented as ;

[0084] Step S33: For each individual in the population, substitute its decoded process parameter value into the trained neural network model to replace the actual experimental process and obtain the objective function value of each individual, that is, obtain the coating porosity, deposition rate and coating bonding strength.

[0085] Step S34: First, the initial population... G All individuals in the population are sorted and their crowding is calculated according to non-dominance relationships. For any two individuals in the population... i and j The degree of crowding between them CI for:

[0086]

[0087] in, N 0 represents the population size; x ik Represents an individual i The eigenvector of the first k One portion, x jk Represents an individualj The eigenvector of the first k One component;

[0088] For individuals with non-dominated frontiers or high crowding, selection, crossover, and mutation operations are performed (in this embodiment, the crossover probability is 0.6 and the mutation probability is 0.1).

[0089] Then, the newly generated combination of process parameters is returned to the neural network model to perform coating performance prediction, and a progeny population is generated. g 0, and simultaneously merge with the parent population to generate a capacity of 2. N A population of 0 (i.e., a capacity of 40) G 0 g 0;

[0090] Step S35, then analyze the population. G 0 g 0. Perform non-dominated sorting and crowding calculation, and select based on elite strategy. N A new paternal population consists of 0 or 20 individuals. G i ; for the new paternal population G i Perform selection, crossover, mutation operations and objective function evaluation to select the next generation population. G i+1 ;

[0091] Step S36: Monitor the changes in the objective function values ​​of individuals in the population. If the distance between the current non-dominated solution set and the known optimal frontier is less than... d (In this embodiment, d When the value is 0.01, a local search is initiated; the neighborhood search range is set to ± for the cold spraying process parameters of interest. h % (In this embodiment, h =5), starting from the selected initial individual, the values ​​of the decision variables are changed sequentially within the given neighborhood search range; if the objective function value corresponding to the new individual is better than that of the current individual, the new individual is accepted as the current individual, until no better solution can be found in the local region;

[0092] Step S37: When the optimal solution of the population does not change significantly for several consecutive generations (e.g., 10 generations) or when the number of iterations reaches a preset value (e.g., the preset iteration value is 100 times), stop the iteration and output the final population. That is, obtain the optimal combination of process parameters within the range of process parameter values. For example: the optimal combination of process parameters that satisfies the requirements of minimizing coating porosity, maximizing coating deposition rate, and coating bonding strength not less than 40 MPa is determined as follows: cold spraying pressure P is 1.3 MPa, cold spraying temperature T is 345℃, normal distribution particle size of Ni metal powder is 18 μm, normal distribution particle size of Al2O3 ceramic powder is 45 μm, and the proportion of Al2O3 in the composite powder is 18.5%.

[0093] Step S4: Based on the optimized combination of process parameters for cold spraying of the metal / ceramic composite coating, conduct cold spraying experiments to verify the effectiveness of the optimization results; the final coating structure performance characterization is as follows: Figure 3 As shown, the prepared coating has a porosity of 1.3%, a deposition efficiency of 35.2%, and a bonding strength of 42.3 MPa. Compared with the Ni / Al2O3 composite coating prepared by the traditional empirical optimization method, the coating prepared in this embodiment has lower porosity, higher deposition efficiency, and higher bonding strength (it can bond...). Figure 2 and Figure 3 (As shown in the morphological diagram).

Claims

1. A data-driven method for optimizing the cold spraying process of multi-component composite coatings, characterized in that: include: Step S1: Establish a performance database for cold spraying processes of metal / ceramic composite coatings; Step S2: Construct an objective function for optimizing the performance of metal / ceramic composite coatings based on an adaptive neural network; Step S3: Optimize the cold spraying process of metal / ceramic composite coating based on the local search enhanced NSGA-II genetic algorithm; Specifically: Step S31: Using the cold spraying process parameters of the metal / ceramic composite coating as optimization variables, and with the optimization objectives of minimizing coating porosity, maximizing coating deposition rate, and ensuring coating bonding strength meets requirements, establish a cold spraying process optimization model: Constraints: In the formula: F(X) represents the set of optimization objectives; Indicates minimizing coating porosity, Indicates porosity; This indicates maximizing the coating deposition rate. Indicates the deposition rate; S Indicates the bonding strength. S c This represents the bonding strength threshold, which is obtained based on the actual optimized composite coating material, substrate material, and experimental data. X This indicates the corresponding cold spraying process parameters. X max , X min These represent the upper and lower limits of the process parameters, respectively. Step S32: Using real number encoding, each cold spraying process parameter is represented by a real number, and an initial population of a certain size is randomly generated. G 0, the parameter values ​​for each individual are randomly generated within their corresponding value range; Step S33: For each individual in the population, substitute its decoded process parameter value into the trained neural network model to replace the actual experimental process and obtain the objective function value of each individual; Step S34: First, the initial population... G All individuals in the population are sorted and their crowding is calculated according to non-dominance relationships. For any two individuals in the population... i and j The degree of crowding between them CI for: in, N 0 represents the population size; x ik Represents an individual i The eigenvector of the first k One portion, x jk Represents an individual j The eigenvector of the first k One component; Select, crossover, and mutation operations are performed on individuals with non-dominated frontiers or high crowding. Then, the newly generated combination of process parameters is returned to the neural network model to perform coating performance prediction, and a progeny population is generated. g 0, and simultaneously merge with the parent population to generate a capacity of 2. N Population of 0 G 0 g 0; Step S35, then analyze the population. G 0 g 0. Perform non-dominated sorting and crowding calculation, and select based on elite strategy. N The new parent population consists of 0 individuals. G i ; for the new paternal population G i Perform selection, crossover, mutation operations and objective function evaluation to select the next generation population. G i+1 ; Step S36: Monitor the changes in the objective function values ​​of individuals in the population. If the distance between the current non-dominated solution set and the known optimal frontier is less than... d When this occurs, a local search is initiated; the neighborhood search range is set ± for the cold spraying process parameters of interest. h Starting from the selected initial individual, the values ​​of the decision variables are changed sequentially within the given neighborhood search range; if the objective function value corresponding to the new individual is better than that of the current individual, the new individual is accepted as the current individual, until no better solution can be found in the local region; Step S37: When the optimal solution of the population does not change significantly for several consecutive generations or when the number of iterations reaches a preset value, stop the iteration and output the final population, that is, obtain the optimal combination of process parameters within the range of process parameter values. Step S4: Based on the optimized combination of process parameters for cold spraying of metal / ceramic composite coatings, conduct cold spraying tests to verify the effectiveness of the optimization results.

2. The data-driven optimization method for cold spraying of multi-component composite coatings according to claim 1, characterized in that: Step S1 specifically involves: Step S11: Determine the key process parameters that affect the coating quality and performance based on the coating performance indicators; Step S12: Determine the initial value range of the key process parameters for cold spraying, and use super Latin cube sampling to select multiple combinations of key process parameters for cold spraying within the initial value range to ensure the uniformity and representativeness of the parameter space. Step S13: Conduct an experiment to prepare a metal / ceramic composite coating using a cold spraying device according to the selected combination of key process parameters; During the preparation process, the parameter sets of each experimental combination are recorded in real time, i.e., the first set of parameters. i The set of key process parameters for the group experiment is as follows G i ; and after the experiment, the properties of the prepared composite coating were tested and recorded in real time to obtain a set of coating properties, the first i The set of coating properties corresponding to the group of tests is as follows R i Finally, the key process parameters and corresponding coating performance data used in each group of experiments were compiled and entered into the database to complete the construction of the cold spraying process-performance database.

3. The data-driven optimization method for cold spraying of multi-component composite coatings according to claim 2, characterized in that: The key process parameters include cold spray gas pressure, spray temperature, spray distance, metal / ceramic powder particle size, and the proportion of metal / ceramic powder.

4. The data-driven optimization method for cold spraying of multi-component composite coatings according to claim 1 or 2, characterized in that: The properties of the composite coating include coating porosity, coating deposition rate, and bonding strength between the coating and the substrate.

5. The data-driven optimization method for cold spraying of multi-component composite coatings according to claim 1, characterized in that: Step S2 specifically involves: Step S21: Preprocess the cold spraying process performance data to eliminate the dimensionality influence between different data and improve data quality and usability. Specifically: In the formula: x * Indicates the first preprocessed... i One data point; x i Representing variables x The first in i One set of raw data; Step S22: Divide the preprocessed process-performance data into training set, validation set and test set using a hierarchical partitioning method; and determine the number of neurons in the input layer, hidden layer, output layer, as well as the number of hidden layers and activation function according to the process parameters and target performance parameters, and construct a neural network. Step S23: First, input the training set into the constructed neural network, calculate the output through forward propagation, and obtain the model output value; Then, the mean squared error loss function is used. MSE Calculate the difference between the model output value and the target value: In the formula: N Indicates the number of samples. M Indicates the number of neurons in the output layer; For the i One sample, Indicates the first j The true value of the output, i.e., the model output value. Indicates the first j The predicted value of each output, i.e., the target value; Then, define the adaptive learning optimizer: In the formula: w t Indicates the neural network's first... t The parameters for the next iteration; Indicates the learning rate; , "Don't" represents the attenuation coefficient; m t , v t They represent the first t The first and second moments of the gradient are predicted in the next iteration: In the formula: g t Indicates the first t The gradient of the parameters at the next iteration; The weights and biases of the neural network are updated based on backpropagation of the loss. Finally, the loss of the validation set is monitored in real time during training. When the loss of the validation set no longer decreases, training is stopped, and the trained neural network model is obtained. At the same time, using the mean square error loss function MSE The threshold is used to determine whether the trained neural network model can predict the coating performance based on the process parameters. If not, the number of neurons in the hidden layer is adjusted and the model is retrained.

Images