Welding metal powder performance prediction optimization method and system

Abstract

The invention discloses a welded metal powder performance prediction optimization method and system, and relates to the technical field of welding powder performance optimizing.The method comprises the steps that historical metal powder performance data are collected, correlation analysis is conducted on the historical metal powder performance data, and welding features associated with the metal powder performance are identified; constructing a performance prediction model based on the welding characteristics, and predicting the melting behaviors of the metal powder under different welding conditions by using the performance prediction model; and according to the prediction result of the metal powder melting behavior, a multi-target optimization model is constructed, and optimal welding parameter configuration is output through the multi-target optimization model, so that performance optimization of the metal powder is achieved. According to the method, correlation analysis is carried out on the historical metal powder performance data, key factors influencing the metal powder welding performance can be recognized, the quality control and prediction capacity of the welding process is improved, and potential modes in the historical data can be fully mined by introducing the deep belief network algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of welding powder performance optimization, and in particular to a method and system for predicting and optimizing welding metal powder performance. Background Art

[0002] Welding metal powder is a fine powder made of metal materials, which is widely used in welding process, especially in powder welding, laser welding, laser melting deposition, additive manufacturing and other technologies. Metal powder is usually composed of high-purity metal or alloy, produced by special powder preparation technology (such as gas atomization, ball milling, etc.), with uniform particle size, high fluidity and good melting characteristics.

[0003] The purpose of optimizing the performance of welding metal powder is to ensure the quality and reliability of the welded joint and avoid a series of defects during the welding process. For example, metal powder exhibits unstable behavior during the melting, diffusion, and alloying processes, resulting in substandard performance of the welded joint. In addition, during the welding process, especially high-energy beam welding such as laser welding and electron beam welding, the temperature change in the welding zone is extremely drastic, and the phase change behavior of the material is affected by complex temperature gradients and stress fields.

[0004] However, existing prediction and optimization methods are usually based on static conditions under certain assumptions and are unable to track and reflect the changes in factors such as temperature field, stress field, and material flow during the welding process in real time and dynamically. Therefore, when faced with a complex and changing welding environment, it is impossible to accurately predict the behavior of metal powder, especially in key processes such as melting, diffusion, and solidification.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a method and system for predicting and optimizing the properties of welding metal powder to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for predicting and optimizing properties of welding metal powder is provided, the method comprising: S1. Collect historical metal powder performance data, perform correlation analysis on the historical metal powder performance data, and identify welding features associated with metal powder performance; S2. Construct a performance prediction model based on welding characteristics, and use the performance prediction model to predict the melting behavior of metal powder under different welding conditions; S3. Construct a multi-objective optimization model based on the prediction results of the melting behavior of the metal powder, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve the performance optimization of the metal powder; Construct a performance prediction model based on welding characteristics, and use the performance prediction model to predict the melting behavior of the metal powder under different welding conditions, including: S21. Construct a thermodynamic model based on the phase diagram analysis algorithm, and use the thermodynamic model to output the phase transformation reaction between the metal powder and the base material under different temperature conditions; S22. Construct a kinetic model based on the reaction diffusion algorithm, and use the kinetic model to output the dynamic diffusion reaction of the metal powder during the welding process; S23. Integrate the phase transformation reaction and the dynamic diffusion reaction to generate a mixed effect prediction model, and use the mixed effect prediction model to predict the melting behavior of the metal powder under different welding conditions.

[0008] Preferably, collect historical metal powder performance data, conduct a correlation analysis on the historical metal powder performance data, and identify the welding characteristics associated with the metal powder performance, including: S11. Extract the historical metal powder performance data of the historical welding records from the database, and identify the metal powder performance characteristics according to the historical metal powder performance data; S12. Use the historical metal powder performance data as the training data set, introduce the deep belief network algorithm to train the training data set to generate a feature tree model, and use the feature tree model to identify the associated feature set; S13. Calculate the contribution degree of the features in the associated feature set, sort the contribution degrees, and use the features within the preset range as the welding characteristics associated with the metal powder performance.

[0009] Preferably, use the historical metal powder performance data as the training data set, introduce the deep belief network algorithm to train the training data set to generate a feature tree model, and use the feature tree model to identify the associated feature set, including: S121. Initialize the first layer of the deep belief network, use the historical metal powder performance data as the visible layer nodes of the deep belief network, and assign weights to the visible layer nodes; S122. Map the weighted visible layer nodes to the input of the visible layer neurons in the deep belief network, and calculate the activation probability of the historical metal powder performance data in the visible layer neurons; S123. Calculate the activation probability of the historical metal powder performance data in the hidden layer neurons based on the activation probability of the historical metal powder performance data in the visible layer; S124. Update the weight matrix between the visible layer neurons and the hidden layer neurons according to the activation probabilities of the visible layer neurons and the hidden layer neurons, and use the weight matrix as the input of the next layer; S125. Iteratively execute S121 - S124 until each layer in the deep belief network is trained, obtaining the final feature mapping relationship between the visible layer neurons and the hidden layer neurons; S126. Construct a feature tree model based on the feature mapping relationship, and identify the key feature set associated with the performance of the metal powder through the feature tree model.

[0010] Preferably, construct a thermodynamic model based on the phase diagram analysis algorithm, and use the thermodynamic model to output the phase change reactions between the metal powder and the substrate under different temperature conditions, including: S211. Obtain the thermodynamic data of the substrate from the thermodynamic database, combine to obtain the component sequences of the substrate and the metal powder, and establish the free energy expression of the metal powder and the substrate based on the component sequences; S212. Solve the free energy expression to obtain the optimized free energy expression, use the free energy expression to calculate the component system formed by the component sequences, and obtain the phase state free energy; S213. Use the Gibbs free energy minimization technique to generate the phase diagram of the phase state free energy under different temperature conditions, and display the phase change reactions between the metal powder and the substrate under different temperature conditions through the phase diagram.

[0011] Preferably, integrate the phase change reaction and the dynamic diffusion reaction to generate a mixed effect prediction model, and use the mixed effect prediction model to predict the melting behavior of the metal powder under different welding conditions, including: S231. Construct a mixed effect prediction model including fixed effects and random effects based on the output results of the thermodynamic model and the kinetic model; S232. Use the mixed effect prediction model and the spatial covariance function to predict the melting behavior of the metal powder at different spatial positions during the melting of the metal powder under different welding conditions; S233. Use the maximum likelihood estimation technique to update the fixed effect parameters and random effect parameters in the mixed effect prediction model.

[0012] Preferably, construct a mixed effect prediction model including fixed effects and random effects based on the phase change reaction and the dynamic diffusion reaction, including: S2311. Use the multivariate joint analysis technique to obtain the joint evolution relationship between the phase change reaction and the dynamic diffusion reaction under different welding conditions; S2312. Based on the predefined state variables, establish the partial differential equation systems of the phase change reaction and the dynamic diffusion respectively, and determine the boundary conditions and initial conditions of the partial differential equation systems. S2313. Based on the combined evolution relationship between the phase change reaction and the dynamic diffusion reaction, combine the partial differential equations of the phase change reaction and the dynamic diffusion reaction to obtain a mixed effect prediction model including fixed effects and random effects, and perform iterative solution on the mixed effect prediction model.

[0013] Preferably, construct a multi-objective optimization model according to the prediction results of the melting behavior of the metal powder, and output the optimal welding parameter configuration through the multi-objective optimization model, including: S31. Construct a multi-objective optimization model with the phase change reaction and the dynamic diffusion reaction of the metal powder as the optimization objectives; S32. Calculate the crowding degree selection operator of the optimization objective, and combine the non-dominated sorting genetic algorithm to solve the multi-objective optimization model; S33. In the solution of the multi-objective optimization model, use the welding characteristics as the input to simulate the prediction results of the melting behavior of the metal powder to obtain the distribution of the important phase change regions; S34. Based on the distribution of the important phase change regions obtained by the simulation, calculate the optimization objective values, select the optimal solution according to the evaluation of the optimization objective values, and use the optimal solution as the optimal welding parameter configuration.

[0014] Preferably, the free energy expression of the metal powder and the substrate is: ; In the formula, G represents the free energy of the metal powder and the substrate; N i represents the amount of substance of component i in the substrate; represents the chemical potential of component i under the standard state; R represents a constant; T represents the temperature; N j represents the deviation degree of component j in the metal powder at the preset temperature; V ij represents the interaction energy between component i and component j ; a j represents the activity of component j in the metal powder.

[0015] Preferably, the calculation formula for the crowding degree is: ; In the formula, represents the crowding degree of the optimization objective; represents the phase change reaction e +1 of the kAn optimized target value; Indicates a dynamic diffusion reaction e The k An optimized target value; Indicates the k Maximum value of an optimization target; Indicates the k Minimum value of an optimization target.

[0016] According to another aspect of the present invention, there is also provided a welding metal powder performance prediction and optimization system, which includes: A welding feature recognition module, which is used to collect historical metal powder performance data, perform correlation analysis on the historical metal powder performance data, and identify welding features associated with the metal powder performance; A melting behavior prediction module, which is used to construct a performance prediction model based on welding features and use the performance prediction model to predict the melting behavior of metal powder under different welding conditions; A welding parameter optimization module, which is used to construct a multi-objective optimization model according to the prediction results of the metal powder melting behavior, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve the performance optimization of the metal powder.

[0017] The beneficial effects of the present invention are as follows: 1. The present invention performs correlation analysis on historical metal powder performance data, which helps to identify key factors affecting the welding performance of metal powder, improves the quality control and prediction ability of the welding process. By introducing the deep belief network algorithm, it can fully explore the potential patterns in historical data, use deep feature learning to extract complex non-linear relationships, improve the prediction accuracy of metal powder performance, and by training the deep belief network layer by layer, calculating the activation probabilities of visible layer and hidden layer neurons, and continuously optimizing the weight matrix, a robust feature mapping relationship can be formed. The feature tree model based on the deep belief network can effectively identify key welding features closely related to the metal powder performance, and thus provide data support for welding process optimization, material design, and intelligent welding systems.

[0018] 2. By introducing the thermodynamic model and kinetic model, the present invention can deeply analyze the behavior of metal powder from the perspectives of phase change and diffusion respectively, providing a theoretical basis for predicting its performance during the welding process. The thermodynamic model helps to understand the phase change reactions between metal powder and base material under different temperature conditions, while the kinetic model reveals how the metal powder evolves over time during the welding process by describing the reaction diffusion process. By integrating these two models, the formed hybrid effect prediction model can take into account the complex multi-factor interactions during the welding process, thus not only improving the prediction ability of the welding process, but also being able to make a quick response under different welding conditions, and further improving the welding quality.

[0019] 3. Based on the prediction results of the melting behavior of metal powder materials, the present invention constructs a multi-objective optimization model and outputs the optimal welding parameter configuration through this model, which can significantly improve the performance optimization of metal powder materials. By introducing a crowding degree selection operator and combining with a non-dominated sorting genetic algorithm, it can efficiently solve the multi-objective optimization problem and find an optimal welding parameter configuration. At the same time, by calculating the optimized objective values and selecting the optimal solution according to the evaluation criteria, the optimal welding parameter configuration can be obtained, thereby optimizing the melting behavior of metal powder materials and avoiding performance losses caused by inappropriate welding parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 is a flowchart of a method for predicting and optimizing the performance of welded metal powder materials according to an embodiment of the present invention; Figure 2 is a schematic block diagram of a system for predicting and optimizing the performance of welded metal powder materials according to an embodiment of the present invention.

[0022] In the figure: 1. Welding feature recognition module; 2. Melting behavior prediction module; 3. Welding parameter optimization module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To further illustrate the embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, mainly used to illustrate the embodiments, and can be combined with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention.

[0024] According to an embodiment of the present invention, a method and system for predicting and optimizing the performance of welded metal powder materials are provided.

[0025] Now, the present invention will be further described in combination with the drawings and specific implementation manners. As Figure 1 shown, according to an embodiment of the present invention, a method for predicting and optimizing the performance of welded metal powder materials is provided, and the method includes: S1. Collect historical metal powder material performance data, perform correlation analysis on the historical metal powder material performance data, and identify welding features associated with the metal powder material performance.

[0026] Among them, collecting historical metal powder material performance data, conducting a correlation analysis on the historical metal powder material performance data, and identifying welding characteristics associated with the metal powder material performance include: S11. Extract the historical metal powder material performance data of historical welding records from the database, and identify the metal powder material performance characteristics based on the historical metal powder material performance data; S12. Use the historical metal powder material performance data as the training data set, introduce the deep belief network algorithm to train the training data set to generate a feature tree model, and use the feature tree model to identify the associated feature set.

[0027] Among them, using the historical metal powder material performance data as the training data set, introducing the deep belief network algorithm to train the training data set to generate a feature tree model, and using the feature tree model to identify the associated feature set includes: S121. Initialize the first layer of the deep belief network, use the historical metal powder material performance data as the visible layer nodes of the deep belief network, and assign weights to the visible layer nodes; S122. Map the weighted visible layer nodes to the inputs of the visible layer neurons in the deep belief network, and calculate the activation probability of the historical metal powder material performance data in the visible layer neurons; S123. Calculate the activation probability of the historical metal powder material performance data in the hidden layer neurons based on the activation probability of the historical metal powder material performance data in the visible layer; S124. Update the weight matrix between the visible layer neurons and the hidden layer neurons according to the activation probabilities of the visible layer neurons and the hidden layer neurons, and use the weight matrix as the input of the next layer; S125. Iteratively execute S121 - S124 until each layer in the deep belief network is trained, and obtain the final feature mapping relationship between the visible layer neurons and the hidden layer neurons; S126. Construct a feature tree model based on the feature mapping relationship, and identify the key feature set associated with the metal powder material performance through the feature tree model.

[0028] S13. Calculate the contribution degrees of the features in the associated feature set, sort the contribution degrees, and use the features within the preset range as the welding characteristics associated with the metal powder material performance.

[0029] Among them, the welding characteristics associated with the metal powder material performance include welding temperature, welding speed, laser power, and welding atmosphere.

[0030] The following further illustrates collecting historical metal powder material performance data, conducting a correlation analysis on the historical metal powder material performance data, and identifying the welding characteristics associated with the metal powder material performance in combination with specific embodiments: Step 1: Extract the historical metal powder property data from the database and identify the metal powder property characteristics based on the historical metal powder property data, including: Data extraction: Extract the historical metal powder property data from the welding experiment database. These data usually include the type of metal powder, chemical composition, particle size, melting point, thermal conductivity, etc. At the same time, relevant parameters of the welding process, such as welding current, welding speed, heat input, etc., also need to be extracted.

[0031] Performance characteristic identification: Identify the characteristics related to the metal powder properties through data analysis. These characteristics include melting temperature, thermal stability, diffusion rate, welding joint strength, etc.

[0032] For example, the following data is extracted from the database: First data: Metal powder type: stainless steel powder, particle size: 50 µm, melting temperature: 1400 °C, welding current: 180 A, welding speed: 6 mm / s, molten pool stable.

[0033] Second data: Metal powder type: aluminum alloy powder, particle size: 70 µm, melting temperature: 660 °C, welding current: 220 A, welding speed: 8 mm / s, pores appear.

[0034] These data can be used to identify the melting behavior, stability, and reaction characteristics during the welding process of the metal powder.

[0035] Step 2: Use the historical metal powder property data as the training data set, introduce the Deep Belief Network (DBN) algorithm to train the training data set to generate a feature tree model, and use the feature tree model to identify the associated feature set, including: Data preprocessing: Organize the historical metal powder property data into a data set suitable for training the Deep Belief Network (DBN). Each row of the data set represents an experiment, and each column represents different metal powder property characteristics (such as particle size, melting temperature, etc.) and welding parameters (such as current, speed, etc.).

[0036] Train the DBN model: Use these data as input and train through the Deep Belief Network (DBN) algorithm to generate a feature tree model. The DBN will extract the features of the data layer by layer and establish the relationship between the input features and the output results.

[0037] For example, use the following data as the training set: Sample 1: Metal powder type: stainless steel powder, particle size: 50 µm, melting temperature: 1400 °C, welding current: 180 A, welding speed: 6 mm / s, molten pool stable.

[0038] Sample 2: Metal powder type: Aluminum alloy powder, particle size: 70 µm, melting temperature: 660 °C, welding current: 220 A, welding speed: 8 mm / s, pores appeared.

[0039] Use these data to train the DBN and construct a feature tree model to identify the relationship between the metal powder and the key features of the welding process.

[0040] Step 3: Initialize the first layer of the deep belief network, use the historical metal powder performance data as the visible layer nodes of the deep belief network, and assign weights to the visible layer nodes, including: Initialize the input layer: In the DBN, the historical metal powder performance data (such as particle size, melting temperature, etc.) are used as the input of the visible layer nodes, and each input data feature (such as welding current, powder particle size) will correspond to a neuron in the visible layer.

[0041] Assign weights: Assign an initial weight to each visible layer node. Usually, these weights can be randomly initialized or set according to prior knowledge.

[0042] Input data sample 1: Features such as metal powder particle size (50 µm), welding current (180 A), welding speed (6 mm / s), etc., and assign these data to the visible layer nodes of the DBN respectively.

[0043] Map the weighted visible layer nodes to the input of the visible layer neurons in the deep belief network, and calculate the activation probability of the historical metal powder performance data in the visible layer neurons, including: Map the input: Pass the visible layer nodes after weight assignment to the visible layer neurons in the deep belief network. Each visible layer neuron receives the corresponding input data (such as welding current, powder particle size) and performs calculations.

[0044] Activation function: Use an activation function (such as Sigmoid or ReLU) to calculate the activation probability of the visible layer neurons, indicating the activation degree of the input data to the neurons.

[0045] For the welding current (180 A) in sample 1, after weighted processing, calculate its activation probability for the visible layer neurons through the activation function.

[0046] Calculate the activation probability of the historical metal powder performance data in the hidden layer neurons based on the activation probability of the historical metal powder performance data in the visible layer, including: Calculate the hidden layer activation probability: According to the activation results of the visible layer neurons, use them as the input to the hidden layer neurons and calculate the activation probability of the hidden layer neurons. The activation probability of the hidden layer neurons is determined by the input and weights of the visible layer.

[0047] Assume that the activation probability of the visible layer is 0.75, and the activation probability of the hidden layer can be obtained by weighted calculation based on the result of the visible layer, for example, 0.6.

[0048] Update the weight matrix between the visible layer neurons and the hidden layer neurons according to the activation probabilities of the visible layer neurons and the hidden layer neurons, and use the weight matrix as the input of the next layer, including: Weight update: Through the backpropagation algorithm, update the weight matrix between neurons according to the activation probabilities and errors of the visible layer and hidden layer neurons. The updated weight matrix will be used for the input calculation of the next layer.

[0049] Iteratively train each layer of the deep belief network until the training of each layer in the deep belief network is completed, and obtain the final feature mapping relationship between the visible layer neurons and the hidden layer neurons, including: Iterative training: Through multiple iterations, the deep belief network will gradually adjust the weights until the training of each layer is completed, and the network will converge to obtain the final feature mapping relationship between the visible layer and the hidden layer; through multiple trainings, finally obtain the feature mapping relationship between the welding current, powder particle size and metal powder properties.

[0050] Construct a feature tree model based on the feature mapping relationship, and identify the key feature set associated with the metal powder properties through the feature tree model, including: Construct a feature tree model: Based on the trained deep belief network, construct a feature tree model. The feature tree model will show the hierarchical relationship between the input features and the metal powder properties, and identify the key feature set.

[0051] For example, the feature tree model may show that the welding current and powder particle size have a greater impact on the melting temperature, while the welding speed has a smaller impact on the melting behavior.

[0052] Step 4: Calculate the contribution degree of the features in the associated feature set, sort the contribution degrees, and use the features within the preset range as the welding features associated with the metal powder properties, including: Calculate the contribution degree: Use methods such as information gain and Gini index to calculate the contribution degree of each feature. The higher the contribution degree of the feature, the greater the impact on the metal powder properties.

[0053] Sorting and selection: Sort the features and select the features with larger contribution degrees as the key welding features.

[0054] Through calculation, it may be found that the contribution degree of the welding current is 0.85, the contribution degree of the welding speed is 0.6, and the contribution degree of the powder particle size is 0.3. According to the sorting, select the welding current and welding speed as the key features.

[0055] It should be noted that the present invention extracts key features from historical metal powder property data, and identifies welding features closely related to the metal powder properties through a deep belief network and a feature tree model. These features can help optimize the welding process and improve the accuracy of metal powder property prediction.

[0056] S2. Construct a performance prediction model based on the welding features, and use the performance prediction model to predict the melting behavior of the metal powder under different welding conditions.

[0057] Among them, constructing a performance prediction model based on the welding features and using the performance prediction model to predict the melting behavior of the metal powder under different welding conditions includes: S21. Construct a thermodynamic model based on the phase diagram analysis algorithm, and use the thermodynamic model to output the phase change reactions between the metal powder and the base material under different temperature conditions.

[0058] Among them, constructing a thermodynamic model based on the phase diagram analysis algorithm and using the thermodynamic model to output the phase change reactions between the metal powder and the base material under different temperature conditions includes: S211. Obtain the thermodynamic data of the base material from the thermodynamic database, combine to obtain the component sequences of the base material and the metal powder, and establish a free energy expression of the metal powder and the base material based on the component sequences.

[0059] Among them, the free energy expression of the metal powder and the base material is: ; In the formula, G represents the free energy of the metal powder and the base material; N i represents the amount of substance of component i in the base material; represents the chemical potential of component i under the standard state; R represents a constant; T represents the temperature; N j represents the deviation degree of component j in the metal powder at the preset temperature; V ij represents the interaction energy between component i and component j ; a j represents the activity of component j in the metal powder.

[0060] S212. Solve the free energy expression to obtain an optimized free energy expression, and use the free energy expression to calculate the component system formed by the component sequences to obtain the phase state free energy; S213. Use the Gibbs free energy minimization technique to generate a phase diagram of the phase free energy under different temperature conditions, and display the phase transformation reaction between the metal powder and the substrate through the phase diagram.

[0061] S22. Construct a kinetic model based on the reaction-diffusion algorithm, and use the kinetic model to output the dynamic diffusion reaction of the metal powder during the welding process.

[0062] It should be noted that constructing a kinetic model based on the reaction-diffusion algorithm and using the kinetic model to output the dynamic diffusion reaction of the metal powder during the welding process includes: Step 1. Define the model parameters and boundary conditions: The key parameters include the temperature field (T): The change in temperature during the welding process is the main factor driving the diffusion reaction, and the temperature field determines the diffusion coefficient (D) and the reaction rate.

[0063] Diffusion coefficient (D): The diffusion coefficient is related to temperature, pressure, the properties of the material itself, etc. According to the change in temperature, the diffusion coefficient can be calculated according to the Arrhenius equation; Among them, the expression of the Arrhenius equation is: ; In the formula, k represents the reaction rate constant (indicating the magnitude of the reaction rate); A represents the frequency factor or pre-exponential factor, which is related to the frequency and geometric structure of the collision of reactant molecules; E a represents the activation energy, indicating the energy barrier that needs to be overcome for the reaction to occur; R represents the gas constant; T represents the absolute temperature.

[0064] Reaction rate (k): Describes the reaction rate between the powder and the substrate, and is usually also affected by temperature and other environmental conditions.

[0065] Boundary conditions: Initial concentration: The initial distribution of the metal powder at the start of welding.

[0066] Boundary conditions during the welding process: The changes in the temperature field and stress field, and the supply method of the metal powder during welding (such as the distribution of laser beam, electron beam, welding current, etc.).

[0067] Step 2. Establish a reaction-diffusion kinetic model: Based on the diffusion behavior of the metal powder during the welding process, construct a kinetic model based on Fick's Law.

[0068] Among them, Fick's law describes the basic law of mass diffusion and is particularly applicable to the case of diffusion in a uniform temperature and concentration field. During the welding process, Fick's law can be used as a kinetic model to predict the diffusion process.

[0069] The basic form of Fick's law is J =− D ∇ C , where J represents the diffusion flux, D represents the diffusion coefficient, and ∇ C represents the concentration gradient.

[0070] Step 3. Implement numerical simulation: After establishing the reaction-diffusion model, the next step is to establish a numerical simulation framework, which includes the following steps: Mesh generation: To solve the diffusion equation, it is necessary to generate a mesh for the welding area, usually using the finite difference method (FDM) or the finite element method (FEM) for discretization.

[0071] Solve the diffusion equation: Solve the above diffusion equation through numerical methods (such as explicit or implicit difference methods) to obtain the concentration distribution of the metal powder under different time and space conditions.

[0072] Reaction rate calculation: Based on the temperature field and diffusion field, calculate the reaction rate and feedback it into the diffusion equation. The reaction rate can be calculated through a thermodynamic model or fitted according to experimental data.

[0073] Step 4. Consider the influence of temperature and stress on diffusion: During the welding process, the temperature gradient and stress field are two key factors affecting the diffusion of metal powder, so the model needs to consider the influence of these factors.

[0074] Influence of temperature field: The change of the temperature field will directly affect the diffusion coefficient, so it is necessary to input the temperature field data into the diffusion equation.

[0075] Influence of stress field: During the welding process, due to thermal expansion and contraction, a strong stress field may be generated, affecting the flow and diffusion of the powder. The stress field can also be simulated through numerical methods and coupled with the diffusion model.

[0076] Once the model is established and the numerical simulation is completed, the dynamic diffusion behavior of the metal powder during the welding process can be output, and these output results include: Concentration distribution: The concentration distribution of the metal powder under different time and space conditions.

[0077] Diffusion rate: The diffusion rate of the metal powder in different regions.

[0078] Reaction product: The result of the reaction between the metal powder and the substrate, such as the extent of the alloying region.

[0079] S23. Integrate the phase transformation reaction and the dynamic diffusion reaction to generate a hybrid effect prediction model, and use the hybrid effect prediction model to predict the melting behavior of the metal powder under different welding conditions.

[0080] Among them, integrating the phase transformation reaction and the dynamic diffusion reaction to generate a hybrid effect prediction model and using the hybrid effect prediction model to predict the melting behavior of the metal powder under different welding conditions includes: S231. Construct a hybrid effect prediction model including fixed effects and random effects based on the phase transformation reaction and the dynamic diffusion reaction.

[0081] Among them, constructing a hybrid effect prediction model including fixed effects and random effects based on the phase transformation reaction and the dynamic diffusion reaction includes: S2311. Use multivariate joint analysis technology to obtain the joint evolution relationship between the phase transformation reaction and the dynamic diffusion reaction under different welding conditions.

[0082] It should be noted that using multivariate joint analysis technology to obtain the joint evolution relationship between the phase transformation reaction and the dynamic diffusion reaction under different welding conditions includes: Collect various parameter data during the welding process, such as temperature field, stress field, concentration distribution of the metal powder, diffusion rate, etc. Establish a multi-dimensional dataset containing these variables, and use correlation analysis algorithms (such as Pearson correlation coefficient, Spearman rank correlation, mutual information, etc.) to analyze the correlation between these variables. These algorithms can reveal the mutual dependence relationship between variables such as temperature and stress and the phase transformation reaction and diffusion reaction by calculating the correlation degree between variables. Through this method, the key factors affecting the phase transformation and diffusion processes during welding can be identified, providing a basis for establishing a multivariate joint model. Based on the analysis results, further construct a joint model that can simultaneously consider factors such as temperature, stress, and concentration of the metal powder, dynamically simulate the joint evolution of the phase transformation reaction and the diffusion reaction, and can predict the behavior of the metal powder under different welding conditions.

[0083] S2312. Based on predefined state variables, establish partial differential equations for the phase transformation reaction and the dynamic diffusion reaction respectively, and determine the boundary conditions and initial conditions of the partial differential equations; S2313. Based on the joint evolution relationship between the phase transformation reaction and the dynamic diffusion reaction, combine the partial differential equations of the phase transformation reaction and the dynamic diffusion reaction to obtain a hybrid effect prediction model including fixed effects and random effects, and perform iterative solution on the hybrid effect prediction model.

[0084] S232. Use the mixed - effect prediction model and the spatial covariance function to predict the melting behavior of metal powder at different spatial positions under different welding conditions. S233. Use the maximum - likelihood estimation technique to update the fixed - effect parameters and random - effect parameters in the mixed - effect prediction model.

[0085] The following further illustrates the construction of a performance prediction model based on welding characteristics and the use of the performance prediction model to predict the melting behavior of metal powder under different welding conditions with specific embodiments: Step 1. Construct a thermodynamic model based on the phase - diagram analysis algorithm, and use the thermodynamic model to output the phase - change reactions between the metal powder and the substrate under different temperature conditions. The core objective of this step is to use the thermodynamic model and combine phase - diagram analysis to predict the phase - change behavior of the metal powder and the substrate under different temperature conditions, which includes: Obtain the thermodynamic data of the substrate from the thermodynamic database, combine to obtain the component sequences of the substrate and the metal powder, and establish the free energy of the metal powder and the substrate based on the component sequences.

[0086] Among them, obtaining the thermodynamic data of the substrate includes obtaining the thermodynamic data of the substrate from the thermodynamic database (such as FactSage, Thermo - Calc), and the data content includes the chemical potential, melting point, specific heat, phase - change data, etc. of each component under the standard state.

[0087] For example, assume the substrate is stainless steel (ferroalloy), and its composition is: Iron (Fe): 65%; Chromium (Cr): 20%; Nickel (Ni): 15%; Metal powder: The metal powder is an alloy containing aluminum and molybdenum, and its composition is as follows: Aluminum (Al): 80%; Molybdenum (Mo): 20%; Combine the component sequences of the substrate and the metal powder: Combine the component information of the metal powder and the substrate to form a mixed component sequence. Assume that the free - energy expression to be established includes the components of the substrate (iron, chromium, nickel) and the metal powder (aluminum, molybdenum). Based on the thermodynamic relationship (such as the Gibbs free - energy formula), construct the free - energy expression of the metal powder and the substrate.

[0088] Solve the free - energy expression to obtain the optimized free - energy expression, and use the free - energy expression to calculate the component system formed by the component sequences to obtain the phase - state free energy, including: Free - energy optimization: Through numerical methods (such as the least - squares method or the quasi - Newton method), optimize the free - energy expression to obtain the optimized free energy of the substrate and the metal powder under different temperature conditions.

[0089] Calculate the phase free energy: Using the optimized free energy expression, calculate the phase free energy of the substrate and the metal powder, especially at different temperatures, calculate the thermodynamic stability of the metal powder and the substrate, and form a relationship diagram between the formation temperature and the free energy.

[0090] Use the Gibbs free energy minimization technique to generate a phase diagram of the phase free energy under different temperature conditions. The phase diagram shows the phase change reactions between the metal powder and the substrate under different temperature conditions, including: Gibbs free energy minimization: By minimizing the Gibbs free energy, solve the phase state of the metal powder and the substrate at different temperatures. Minimizing the free energy helps predict the stable phase of substances at different temperatures.

[0091] Generate a phase diagram: Based on the calculated free energy, draw a phase diagram under different temperature conditions to show the phase change behavior of the metal powder and the substrate at different temperatures, such as solid-liquid phase change, solid-solid phase change, etc.

[0092] For example, by minimizing the free energy, the generated phase diagram shows the phase change of aluminum and molybdenum near the melting point in the temperature range of 1000 °C to 1600 °C. At 1200 °C, aluminum begins to melt, and at the interface in contact with the substrate, aluminum reacts with iron.

[0093] Step 2: Construct a kinetic model based on the reaction-diffusion algorithm, and use the kinetic model to output the dynamic diffusion reaction of the metal powder during welding, including: Reaction-diffusion algorithm: Establish a kinetic model to describe the diffusion behavior of the metal powder during welding, considering the influence of the temperature gradient during welding on the diffusion of the metal powder. The diffusion rate of the metal powder is closely related to the temperature and can be described by the diffusion equation.

[0094] Dynamic diffusion reaction: Use the model to calculate the diffusion rate of the metal powder on the substrate surface and the mass transfer during the diffusion process.

[0095] For example, at 1200 °C, the diffusion coefficient of the aluminum powder is 2.5×10 −9 m 2 / s, and the diffusion rate of aluminum is calculated through the kinetic model.

[0096] Step 3: Integrate the phase change reaction and the dynamic diffusion reaction to generate a mixed effect prediction model, and use the mixed effect prediction model to predict the melting behavior of the metal powder under different welding conditions, including: Define state variables: Temperature T ( x , y , z , t );Concentration C (x , y , z , t ); welding speed v ; cooling rate r ; Establish a differential equation system for the phase transformation reaction, and its expression is: ; In the formula, D represents the thermal diffusion coefficient; represents the heat source term of the phase transformation reaction; represents the heat source term of the phase transformation reaction.

[0097] Establish a differential equation system for the dynamic diffusion reaction, and its expression is: ; In the formula, R ( C , T ) represents the reaction rate, which can be a linear or non-linear function.

[0098] Couple the diffusion and reaction equations to obtain a mixed effect prediction model, and its expression is: ; Determine the boundary conditions and initial conditions: Initial conditions: C ( x , y , z , 0) = C 0 ( x , y , z ) ϕ ( x , y , z , 0); Boundary conditions: T ( x , y , z , 0) = T 0 ( x , y , z ) ϕ ( x , y , z , 0); Define the mixed effect model: Fixed effect: Represents the deterministic and predictable part in the system, such as the thermal diffusion coefficient D, Heat source term of phase change reaction Q ( T , ϕ ) and so on.

[0099] Random effect: Represents the random and unpredictable part in the system, such as variations in initial conditions, fluctuations in environmental conditions, etc.

[0100] The expression of the mixed effect model is: y ij = β 0 + β 1 T ij + β 2 C ij + β 3 ϕ ij + b i + ϵ ij ; In the formula, β 0, β 1, β 2, β 3 all represent fixed effect parameters; T ij , C ij , ϕ ij represent covariates of fixed effects; b i represents the random effect of the i th observation unit; ϵ ij represents random error.

[0101] Calculate the mean square error (MSE) and root mean square error (RMSE) between the predicted value and the actual value.

[0102] Using the mixed effect prediction model and the spatial covariance function, predict the melting behavior of metal powder at different spatial positions under different welding conditions, including: Spatial covariance function: Use the spatial covariance function to describe the spatial distribution of metal powder during welding and predict the melting behavior of metal powder at different positions (such as the welding joint area, heat affected zone).

[0103] It should be noted that the spatial covariance function is used to quantify the correlation between different positions in space. During the welding process, due to the spatial variations in the temperature field and stress field, the melting behavior of the metal powder also exhibits spatial correlation. Therefore, the spatial covariance function can help describe the dependence relationship of the melting behavior of the metal powder between different spatial positions. For example, the spatial covariance function includes the exponential covariance function, and its expression is: ; In the formula, h represents the distance between spatial positions, σ 2 represents the scale of variation, θ represents the scale parameter, that is, the correlation range in space.

[0104] The spatial covariance function can help describe the spatial correlation between different positions. In the modeling of the melting behavior of the metal powder, considering the spatial distribution of factors such as temperature and stress and their influence on the diffusion of the metal powder, the melting behavior of the metal powder at different spatial positions can be better modeled and predicted through the covariance function.

[0105] Melting behavior prediction: According to the spatial covariance function, predict the melting process of the metal powder on the substrate surface and in the joint area under different welding conditions. Specifically, a spatial autoregressive model can be used. The spatial autoregressive model (Spatial Autoregressive Model) is a model that combines the spatial covariance function and random effects. This model allows the melting behavior of the metal powder to be affected not only by the welding process parameters but also by the interaction between spatial positions. Through this model, the spatial distribution of the melting behavior can be captured more precisely.

[0106] In this model, the prediction of the melting behavior depends not only on the fixed effects and random effects but also on the melting states of its neighboring positions. Therefore, using the spatial autoregressive model can take into account the spatial interaction and improve the prediction of the melting behavior of the metal powder.

[0107] Using the maximum likelihood estimation technique to update the fixed effect parameters and random effect parameters in the mixed effect prediction model for estimation includes: Maximum likelihood estimation: Use the maximum likelihood estimation (MLE) technique to optimize the parameters in the mixed effect model. Adjust the fixed effect and random effect parameters of the model by minimizing the error between the model prediction and the actual experimental data.

[0108] Parameter optimization: Estimate parameters such as welding current, welding speed, and the microstructure of the substrate to optimize the accuracy of the model.

[0109] It should be noted that the present invention constructs a complete welding process model, combines a thermodynamic model, a kinetic model, and a hybrid effect model to accurately predict the melting behavior of metal powder under different welding conditions. Through phase diagram analysis, diffusion algorithms, and maximum likelihood estimation techniques, the influence of various factors during the welding process (such as temperature, welding parameters, substrate differences) can be considered, thereby providing an accurate prediction of the melting behavior of metal powder.

[0110] S3. Construct a multi-objective optimization model based on the prediction results of the melting behavior of metal powder, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve the performance optimization of metal powder.

[0111] Among them, constructing a multi-objective optimization model based on the prediction results of the melting behavior of metal powder and outputting the optimal welding parameter configuration through the multi-objective optimization model includes: S31. Construct a multi-objective optimization model with the phase change reaction and dynamic diffusion reaction of metal powder as the optimization objectives; S32. Calculate the crowding degree selection operator of the optimization objective, and combine the non-dominated sorting genetic algorithm to solve the multi-objective optimization model.

[0112] Among them, the calculation formula for the crowding degree is: ; In the formula, represents the crowding degree of the optimization objective; represents the phase change reaction e +1 of the k th optimization objective value; represents the dynamic diffusion reaction e -1 of the k th optimization objective value; represents the k th maximum value of the optimization objective; represents the k th minimum value of the optimization objective.

[0113] S33. In the solution of the multi-objective optimization model, use the welding characteristics as the input to simulate the prediction results of the melting behavior of metal powder to obtain the distribution of important phase change regions; S34. Based on the distribution of important phase change regions obtained from the simulation, calculate the optimization objective values, select the optimal solution according to the evaluation of the optimization objective values, and use the optimal solution as the optimal welding parameter configuration.

[0114] The following further illustrates the construction of a multi-objective optimization model based on the prediction results of the melting behavior of metal powder and the output of the optimal welding parameter configuration through the multi-objective optimization model to achieve the performance optimization of metal powder in combination with specific embodiments: Step 1: Taking the phase change reaction and dynamic diffusion reaction of metal powder as the optimization objectives, construct a multi-objective optimization model, including: Phase change reaction: The phase change behavior between metal powder and substrate during the welding process, mainly considering solid-liquid phase change, eutectic reaction, etc.

[0115] Dynamic diffusion reaction: The diffusion behavior of metal powder during the welding process, which affects the distribution and melting degree of metal powder in the substrate.

[0116] Mathematical expression of the optimization objective: The goal is to minimize / maximize the relevant indicators of the phase change reaction and diffusion reaction between metal powder and substrate, such as minimizing the depth of the melting zone or minimizing the diffusion time.

[0117] Establish a multi-objective optimization model: Taking the phase change reaction and dynamic diffusion reaction as the optimization objective functions, a multi-objective optimization problem is formed. The model can be expressed as: f 1 (phase change reaction) and f 2 (dynamic diffusion reaction), where f 1 represents the objective function (optimization objective) of the phase change reaction, f 2 represents the objective function of the diffusion reaction.

[0118] For example: Phase change reaction data: Assume that at 1200 °C, the phase change point between the substrate and metal powder occurs at a depth of 0.5 mm.

[0119] Dynamic diffusion data: At 1200 °C, the diffusion coefficient of aluminum is 2.5×10 −9 m 2 / s. Assume that it is necessary to optimize the diffusion rate to keep it within a certain range in the heat affected zone.

[0120] Step 2: Calculate the crowding degree selection operator of the optimization objective, and solve the multi-objective optimization model by combining with the non-dominated sorting genetic algorithm: Crowding degree calculation formula: Crowding degree refers to the sparsity of each solution relative to other solutions in the objective space in multi-objective optimization.

[0121] Non-dominated sorting genetic algorithm (NSGA-II): Non-dominated sorting: Determine the priority of solutions through non-dominated sorting, and select multiple solutions for reproduction. Non-dominated sorting considers multiple objective functions simultaneously and selects the optimal solution set.

[0122] Genetic algorithm selection: Find the optimal welding parameter configuration through genetic operations such as selection, crossover, and mutation.

[0123] For example: Phase change reaction: The goal is to minimize the melting depth (assuming the target value is 0.6 mm and the previous generation was 0.5 mm).

[0124] Dynamic diffusion reaction: The goal is to maximize the diffusion rate (assuming the target is 2.5×10 −9 m 2 / s and the previous generation was 2.0×10 −9 m 2 / s).

[0125] Through the above crowding degree calculation, the priority of each objective function can be measured, and the welding conditions can be optimized by the NSGA-II algorithm.

[0126] Step 3: In the solution of the multi-objective optimization model, use the welding characteristics as the input to simulate the prediction results of the melting behavior of the metal powder, and obtain the distribution of important phase change regions, including: Use the welding characteristics as the input: The input welding characteristics include welding current, welding speed, laser power, etc.

[0127] For example, assume the input is: welding current 200 A, welding speed 5 mm / s, laser power 1500 W.

[0128] Simulation of the melting behavior of the metal powder: Use thermodynamic models and kinetic models (such as phase change analysis and diffusion simulation) to predict the melting behavior of the metal powder during welding. Through simulation, calculate the melting depth and molten pool width of the metal powder under different welding conditions, and obtain the phase change regions of the metal powder and the base material under specific welding conditions. These regions are usually the molten pool region, the heat affected zone, and the cooling zone.

[0129] For example: Welding current: 200 A; Welding speed: 5 mm / s; Welding temperature: 1200 °C; The simulation results show that the maximum depth of the molten pool is 0.7 mm and the width of the heat affected zone is 2 mm.

[0130] Step 4: Based on the distribution of important phase change regions obtained from the simulation, calculate the optimization target values, select the optimal solution according to the evaluation of the optimization target values, and use the optimal solution as the optimal welding parameter configuration, including: Calculate the optimization target values: Through the distribution of important phase change regions obtained from the simulation, calculate the optimization target values under different welding conditions. The calculation indicators include: Objective of the phase change reaction: Minimize the melting depth, and the target value is 0.6 mm; Dynamic diffusion reaction objective: Maximize the diffusion rate, with the target value being 2.5×10 −9 m 2 / s.

[0131] Select the optimal solution according to the evaluation of the optimized target value: Calculate the objective function values under different welding conditions, and select the optimal solution based on non-dominated sorting and crowding degree in the multi-objective optimization algorithm.

[0132] The optimal solution is used as the optimal welding parameter configuration: Use the obtained optimal solution as the final welding parameter configuration. The optimized welding parameter configuration can achieve higher welding quality, such as more uniform melting behavior, more precise molten pool control, etc.

[0133] For example, through non-dominated sorting and crowding degree calculation, the finally obtained optimal welding parameters are: Welding current: 200A; Welding speed: 6mm / s; Laser power: 1600W; The optimal solution results in a melting depth of 0.6mm and a diffusion rate reaching 2.5×10 −9 m 2 / s.

[0134] According to another embodiment of the present invention, as Figure 2 shown, there is also provided a welding metal powder performance prediction and optimization system, which includes: Welding feature recognition module 1, used to collect historical metal powder performance data, perform correlation analysis on the historical metal powder performance data, and identify welding features associated with the metal powder performance; Melting behavior prediction module 2, used to construct a performance prediction model based on welding features and use the performance prediction model to predict the melting behavior of metal powder under different welding conditions; Welding parameter optimization module 3, used to construct a multi-objective optimization model according to the prediction results of the metal powder melting behavior, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve the performance optimization of the metal powder.

[0135] In summary, by means of the above technical solutions of the present invention, the present invention conducts a correlation analysis on the performance data of historical metal powder materials, which helps to identify the key factors affecting the welding performance of metal powder materials, improve the quality control and prediction ability of the welding process. By introducing the deep belief network algorithm, the potential patterns in historical data can be fully mined, complex non-linear relationships can be extracted by using deep feature learning, and the prediction accuracy of the performance of metal powder materials can be improved. By training the deep belief network layer by layer, calculating the activation probabilities of the neurons in the visible layer and the hidden layer, and continuously optimizing the weight matrix, a robust feature mapping relationship can be formed. The feature tree model constructed based on the deep belief network can effectively identify the key welding features closely related to the performance of metal powder materials, and further provide data support for welding process optimization, material design, and intelligent welding systems. By introducing the thermodynamic model and the kinetic model, the present invention can deeply analyze the behavior of metal powder materials from the perspectives of phase change and diffusion respectively, providing a theoretical basis for predicting their performance in the welding process. The thermodynamic model helps to understand the phase change reactions between metal powder materials and the base material under different temperature conditions, while the kinetic model reveals how the metal powder materials evolve over time during the welding process by describing the reaction diffusion process. By integrating these two models, the formed hybrid effect prediction model can take into account the complex multi-factor interactions in the welding process, thereby not only improving the prediction ability of the welding process but also being able to make a rapid response under different welding conditions, and further improving the welding quality. The present invention constructs a multi-objective optimization model based on the prediction results of the melting behavior of metal powder materials, and outputs the optimal welding parameter configuration through this model, which can significantly improve the performance optimization of metal powder materials. By introducing the crowding degree selection operator and combining it with the non-dominated sorting genetic algorithm, the multi-objective optimization problem can be efficiently solved to find an optimal welding parameter configuration. At the same time, by calculating the optimized objective values and selecting the optimal solution according to the evaluation criteria, the optimal welding parameter configuration can be obtained, thereby optimizing the melting behavior of metal powder materials and avoiding performance losses caused by inappropriate welding parameters.

[0136] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting and optimizing the properties of welding metal powder, characterized in that: The method includes: S1. Collect historical metal powder performance data, perform correlation analysis on the historical metal powder performance data, and identify welding features associated with metal powder performance; S2. Construct a performance prediction model based on welding characteristics, and use the performance prediction model to predict the melting behavior of metal powder under different welding conditions; S3. Construct a multi-objective optimization model based on the prediction results of the melting behavior of the metal powder, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve performance optimization of the metal powder; The method of constructing a performance prediction model based on welding characteristics and using the performance prediction model to predict the melting behavior of metal powder under different welding conditions includes: S21. Construct a thermodynamic model based on a phase diagram analysis algorithm, and use the thermodynamic model to output the phase change reaction between the metal powder and the substrate under different temperature conditions; S22. constructing a kinetic model based on a reaction-diffusion algorithm, and using the kinetic model to output a dynamic diffusion reaction of metal powder during welding; S23. Integrate the phase change reaction and the dynamic diffusion reaction to generate a mixed effect prediction model, and use the mixed effect prediction model to predict the melting behavior of metal powder under different welding conditions.

2. A welding metal powder performance prediction and optimization method according to claim 1, characterized in that: The collecting of historical metal powder material performance data, performing correlation analysis on the historical metal powder material performance data, and identifying welding features associated with the metal powder material performance include: S11, extracting historical metal powder performance data of historical welding records from a database, and identifying metal powder performance characteristics based on the historical metal powder performance data; S12, using historical metal powder performance data as a training data set, introducing a deep belief network algorithm to train the training data set to generate a feature tree model, and using the feature tree model to identify associated feature sets; S13, calculating the contribution of the features in the associated feature set, sorting the contribution, and taking the features within a preset range as welding features associated with the metal powder performance.

3. A welding metal powder performance prediction and optimization method according to claim 2, characterized in that: The method of using the historical metal powder performance data as a training data set, introducing a deep belief network algorithm to train the training data set to generate a feature tree model, and using the feature tree model to identify the associated feature set includes: S121, initializing the first layer of the deep belief network, using the historical metal powder performance data as the explicit layer nodes of the deep belief network, and weighting the explicit layer nodes; S122, mapping the weighted display layer nodes to the input of the display layer neurons in the deep belief network, and calculating the activation probability of the historical metal powder performance data in the display layer neurons; S123, calculating the activation probability of the historical metal powder material performance data in the hidden layer neurons based on the activation probability of the historical metal powder material performance data in the visible layer; S124, updating the weight matrix between the visible layer neurons and the hidden layer neurons according to the activation probabilities of the visible layer neurons and the hidden layer neurons, and using the weight matrix as the input of the next layer; S125, iteratively executing S121-S124 until the training of each layer in the deep belief network is completed, and the final feature mapping relationship between the neurons in the visible layer and the neurons in the hidden layer is obtained; S126. Construct a feature tree model based on the feature mapping relationship, and identify a key feature set associated with the performance of the metal powder through the feature tree model.

4. A welding metal powder performance prediction and optimization method according to claim 1, characterized in that: The method of constructing a thermodynamic model based on a phase diagram analysis algorithm and using the thermodynamic model to output the phase change reaction between the metal powder and the substrate under different temperature conditions includes: S211, obtaining thermodynamic data of the substrate from a thermodynamic database, combining to obtain a component sequence of the substrate and the metal powder, and establishing a free energy expression of the metal powder and the substrate based on the component sequence; S212, solving the free energy expression to obtain an optimized free energy expression, and using the free energy expression to calculate the component system formed by the component sequence to obtain the phase free energy; S213. Use Gibbs free energy minimization technology to generate a phase diagram of phase free energy under different temperature conditions, and use the phase diagram to display the phase change reaction between metal powder and substrate under different temperature conditions.

5. A welding metal powder performance prediction and optimization method according to claim 4, characterized in that: The phase change reaction and the dynamic diffusion reaction are integrated to generate a mixed effect prediction model, and the mixed effect prediction model is used to predict the melting behavior of metal powder under different welding conditions, including: S231. Construct a mixed effect prediction model including fixed effects and random effects based on phase change reaction and dynamic diffusion reaction; S232. Predicting the melting behavior of metal powder at different spatial positions under different welding conditions using a mixed effect prediction model and a spatial covariance function; S233. Use maximum likelihood estimation technology to update the fixed effect parameters and random effect parameters in the mixed effect prediction model.

6. A welding metal powder performance prediction and optimization method according to claim 5, characterized in that: The mixed effect prediction model including fixed effect and random effect based on phase change reaction and dynamic diffusion reaction is constructed as follows: S2311. Use multivariate joint analysis technology to obtain the joint evolution relationship between phase change reaction and dynamic diffusion reaction under different welding conditions; S2312, based on the predefined state variables, respectively establish partial differential equations for phase change reaction and dynamic diffusion, and determine boundary conditions and initial conditions of the partial differential equations; S2313. Based on the joint evolution relationship between phase change reaction and dynamic diffusion reaction, the partial differential equations of phase change reaction and dynamic diffusion are combined to obtain a mixed effect prediction model including fixed effects and random effects, and the mixed effect prediction model is iteratively solved.

7. A welding metal powder performance prediction and optimization method according to claim 1, characterized in that: The multi-objective optimization model is constructed according to the prediction results of the melting behavior of the metal powder, and the optimal welding parameter configuration is output through the multi-objective optimization model, including: S31. Taking the phase change reaction and dynamic diffusion reaction of metal powder as optimization targets, a multi-objective optimization model is constructed; S32, calculating the crowding degree selection operator of the optimization target, and solving the multi-objective optimization model in combination with the non-dominated sorting genetic algorithm; S33. In solving the multi-objective optimization model, the welding characteristics are used as input to simulate the predicted results of the melting behavior of the metal powder to obtain the distribution of important phase change areas; S34. Based on the distribution of important phase change areas obtained by simulation, the optimization target value is calculated, the optimal solution is selected according to the evaluation of the optimization target value, and the optimal solution is used as the optimal welding parameter configuration.

8. A welding metal powder performance prediction and optimization method according to claim 5, characterized in that: The free energy expression of the metal powder and the substrate is: ； In the formula, G It represents the free energy between metal powder and substrate; N i Indicates the components in the substrate i The amount of substance; Indicates components i Chemical potential at standard conditions; R represents a constant; T Indicates temperature; N j Indicates the components in metal powder j The degree of deviation from the preset temperature; V ij Indicates components i With components j The interaction energy; a j Indicates the components in metal powder j activity.

9. A welding metal powder performance prediction and optimization method according to claim 7, characterized in that: The calculation formula of the congestion degree is: ； In the formula, Indicates the congestion degree of the optimization objective; Phase change reaction e +1 k An optimization target value; Represents a dynamic diffusion reaction e -1st k An optimization target value; Indicates k The maximum value of the optimization objectives; Indicates k The minimum value of the optimization objective.

10. A welding metal powder performance prediction and optimization system, used to implement the welding metal powder performance prediction and optimization method according to any one of claims 1 to 9, characterized in that: The system includes: A welding feature recognition module is used to collect historical metal powder performance data, perform correlation analysis on the historical metal powder performance data, and identify welding features associated with the metal powder performance; A melting behavior prediction module is used to build a performance prediction model based on welding characteristics, and use the performance prediction model to predict the melting behavior of metal powder under different welding conditions; The welding parameter optimization module is used to build a multi-objective optimization model based on the prediction results of the melting behavior of metal powder, and output the optimal welding parameter configuration through the multi-objective optimization model to achieve performance optimization of metal powder.

Images