Copper-based superalloy with high mechanical fatigue performance for automobile reed and preparation method of copper-based superalloy

Abstract

The invention relates to the technical field of alloy processing, in particular to a high-mechanical-fatigue-performance copper-based superalloy for an automobile reed and a preparation method thereof.The preparation method comprises the following steps that firstly, raw materials including, by mass, 55%-65% of titanium, 30%-25% of copper, 10%-8% of aluminum, 3%-1% of zinc and 2%-1% of magnesium are weighed; and 2, the raw materials are put into a smelting furnace to be melted, melting parameters in the smelting furnace are collected in the melting process, the melting parameters are input into a covering agent adding model which is constructed in advance and trained, adding parameters of a covering agent are obtained, and the covering agent is added into the smelting furnace based on the adding parameters. According to the method, the adding amount and the optimal adding time of the covering agent are dynamically calculated, so that the air hole forming probability in the casting process is effectively reduced, and the quality of the copper-based superalloy is improved; and then the corrected melting temperature, the corrected melting time and the temperature adjusting rate are calculated based on the added parameters, a dynamic and flexible casting process control system is formed, the casting process is optimized, and the quality and performance of castings are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of alloy processing, and particularly relates to a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs and a preparation method thereof. Background Art

[0002] Automotive leaf springs are important components in the automotive suspension system, used to connect the vehicle body and wheels, playing a role in shock absorption and support. With the development of the automotive industry, the performance requirements for leaf springs are getting higher and higher, especially in terms of mechanical fatigue performance. Traditional leaf spring materials usually use carbon structural steel or alloy structural steel. Although these materials have high strength, they are prone to fatigue fracture during long-term use, affecting the safety and comfort of the vehicle. To improve the mechanical fatigue performance of leaf springs, researchers have started to explore new materials. Copper-based superalloys have become an ideal leaf spring material due to their excellent mechanical properties, corrosion resistance, and processing performance.

[0003] After retrieval, Chinese Patent No. CN201911017399.3 discloses a copper alloy with excellent bending performance, a preparation method thereof, and applications. The preparation method includes the following steps: 1) Melting and casting: Using a conventional copper alloy melting method, melting the copper alloy raw materials at 1100 - 1300 °C, and casting ingots through iron mold casting, horizontal continuous casting, or vertical semi-continuous casting; 2) Hot rolling: Hot rolling the ingots at a temperature of 700 - 980 °C, controlling the cross-sectional area reduction rate of the ingot hot rolling to be not less than 75%, further preferably not less than 90%, to obtain a hot-rolled plate. The above solution controls the area ratio of different orientations of the rolling surface of the alloy strip through the preparation process, so that the crystal orientation of the strip of the copper alloy satisfies within a deviation angle of less than 15°: the area ratio of the Brass orientation {011} is 15.0 - 30.0%, the area ratio of the S orientation {123} is 7.0 - 28.0%, the area ratio of the Copper orientation {112} is 6.5 - 20.0%, and the area ratio of the R orientation {124} is 6.0 - 16.0%, realizing the synchronous improvement of the alloy strength, electrical conductivity, stress relaxation resistance, and bending performance.

[0004] In the above solution, during the processing of existing copper-based alloys, although the relevant parameters of the rolling process are optimized, during the rolling process, the defects such as pores and shrinkage cavities generated during the casting process of the alloy material are not considered. These defects will become the origin points of fatigue cracks, reducing the mechanical fatigue life of the leaf spring; adding covering agents such as charcoal or graphite during the melting process to reduce the dissolution of oxygen and nitrogen, thereby reducing the generation of pores and shrinkage cavities; however, during the addition process of the covering agent, even for the same batch of alloy materials, there will be individual differences, resulting in different addition processes of the covering agent. The existing preparation methods cannot be adaptively adjusted according to the individual differences of the alloy.

[0005] Therefore, a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs and its preparation method are proposed to solve the above-mentioned problems. Summary of the Invention

[0006] Technical Problems to be Solved

[0007] In view of the above-mentioned drawbacks of the prior art, the present invention provides a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs and its preparation method, which can effectively solve the problem in the prior art that the process parameters of the covering agent addition cannot be adjusted adaptively according to the characteristics of the alloy material itself.

[0008] Technical Solution

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0010] The present invention provides a preparation method of a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs, including the following steps:

[0011] Step 1: Weigh the raw materials by mass percentage, including 55%-65% titanium, 30%-25% copper, 10%-8% aluminum, 3%-1% zinc, and 2%-1% magnesium;

[0012] Step 2: Put the raw materials into a furnace for melting. During the melting process, collect the melting parameters in the furnace, input the melting parameters into a pre-constructed and trained covering agent addition model to obtain the addition parameters of the covering agent, and add the covering agent into the furnace based on the addition parameters;

[0013] Step 3: Input the addition parameters into a pre-constructed and trained melting adjustment model to obtain the melting adjustment parameters, and melt the raw materials based on the melting adjustment parameters to obtain a melt;

[0014] Step 4: Pour the melt into a mold for casting to obtain a blank;

[0015] Step 5: Perform heat treatment on the blank, and perform cold processing on the blank after heat treatment to obtain the copper-based superalloy.

[0016] Further, the melting parameters in the furnace in Step 2 include: the partial pressure P of the gas and the melting temperature T; the addition parameters of the covering agent include: the addition amount A and the optimal addition time t optimal .

[0017] Further, the method for constructing the covering agent addition model in Step 2 includes:

[0018] Calculate the solubility C of the gas based on the partial pressure P of the gas, and the calculation formula is:

[0019] C(t) = αP; where t is time; C(t) represents the solubility of the gas in the melt at time t; α is the proportionality constant in Henry's law;

[0020] Based on the melting temperature T, the solubility C of the gas is updated and calculated. The calculation formula is:

[0021] C actual (t) = C(t)·(1 + βΔT(t)); where C actual (t) represents the actual solubility of the gas in the melt at time t; β is the melting temperature influence coefficient; ΔT(t) represents the temperature fluctuation at time t, and the calculation formula is: ΔT(t) = T(t) - T expected (t), where T(t) represents the actual melting temperature collected at time t, and T expected (t) represents the expected melting temperature at time t;

[0022] Define the calculation expression for the reaction rate R(T) of the covering agent and the gas:

[0023] R(T) = aT 2 + bT + c; where a, b, and c are constants related to the type of covering agent and the type of gas;

[0024] Define the calculation formula for the addition amount A of the covering agent as:

[0025] A(t) = dC actual (t)R(T); where A(t) represents the addition amount of the covering agent that needs to be added at time t; d is the proportionality constant;

[0026] Set the critical solubility threshold C th (t) and the pore formation rate threshold P defect-th , when C actual (t) > C th (t) and P defect > P defect-th ), update the calculation formula for the addition amount A of the covering agent to obtain; A(t optimal ) = dC actual (t)R(T)·(1 + δ(P defect - P defect-th ));

[0027] Furthermore, the training method of the covering agent addition model in the second step includes:

[0028] Collect the process parameters in the historical melting process, extract the characteristic data in the process parameters, including the partial pressure P of the gas, the melting temperature T, the reaction rate R(T) of the covering agent, the solubility C, the addition amount A, and the optimal addition time toptimal and the pore formation rate P defect ; clean and normalize the feature data; then construct the feature variables in the feature data into an input feature vector X = {P, T, R(T), C, P defect}, and construct the target variable into an output feature vector Y = {A, t optimal}; construct the input feature vector X and the output feature vector Y into training input data, including a training set and a test set;

[0029] Define the random forest regression model SVR as the basic structure of the covering agent addition model, and initialize the network parameters of the basic training structure, including the kernel function K(x i , x j ), the regularization parameter Z, and the insensitive loss function V;

[0030] Input the input feature vector X and the output feature vector Y of the training set into the basic structure; define the form of the loss function of the covering agent addition model as: In the formula, L(w, b, ξ) is the loss function to be minimized; w is the weight vector; b is the bias term; ξ is the slack variable, and the slack variable ξ of each sample i i represents its prediction error;

[0031] Define the constraint conditions for each training sample (x i , y i ), including:

[0032] y i -(w·φ(x i )) + b) ≤ ε + ξ i ;

[0033] (w·φ(x i )) + b) - y i ≤ ε + ξ i ;

[0034] ξ i ≥ 0; where φ is the feature mapping function; ε is a constant;

[0035] Then use the optimization algorithm to update the weight vector w, the bias term b, and the slack variable ξ to minimize the value of the loss function L(w, b, ξ) while satisfying the above constraint conditions;

[0036] Repeat the training on the training set until the covering agent addition model converges or reaches the pre-set iteration times in advance, which means that the melting is completed. Among them, φ is the feature mapping function, which is responsible for mapping the input data to a high-dimensional space. Through feature mapping, SVR can find the best hyperplane in a complex space; adjust the model training to complete.

[0037] Further, the method for constructing the melting adjustment model in step three includes:

[0038] Define the melting adjustment parameters, including the corrected melting temperature T rev (t), the corrected melting time t rev and the temperature adjustment rate T adj (t);

[0039] Define the calculation formula for the corrected melting temperature T rev (t) as:

[0040] T rev (t) = T base + λ·A(t); where T base is the base melting temperature, and λ is the temperature adjustment coefficient;

[0041] Define the calculation formula for the corrected melting time t rev as:

[0042] t rev = t base + μ·A(t); where; t base is the base melting time; μ is the time adjustment coefficient, which is used to describe the influence degree of the covering agent addition amount on the melting time;

[0043] Define the calculation formula for the temperature adjustment rate T adj (t) as:

[0044] where, t rate is the time constant.

[0045] Further, the training method of the melting adjustment model includes:

[0046] Define the long short-term memory network model as the basic structure of the melting adjustment model, and the basic structure includes an input layer, a hidden layer, and an output layer;

[0047] Introduce the parameter vector time step d, and combine the vector time step d, the melting adjustment parameters, and the addition parameters to obtain the fusion feature vector where s is the number of vector time steps d;

[0048] Collect the melting adjustment parameters and addition parameters of the target blank in the past fixed time, and correspondingly construct them into a fusion feature vector, denoted as the historical fusion feature vector Unfold the historical fusion feature vector according to the vector time step to form sequence data as the training input of the melting adjustment model, and correspondingly label the sequence data according to the vector time step d, and the labels are the historical addition amount A(t)' and the historical best addition time t' optimal, the historically corrected melting temperature T rev (t)', the historically corrected melting time t r ' ev and the historical temperature adjustment rate T adj (t)';

[0049] Initialize the network parameters of the melting adjustment model; the network parameters include the weight matrix W from the input layer to the hidden layer x 、the recurrent weight matrix W of the hidden layer h 、the bias vector b of the hidden layer h 、the weight matrix W from the hidden layer to the output layer o and the bias vector b of the output layer o ;

[0050] Define the loss function of the melting adjustment model

[0051] In the formula, S is the length of the sequence data; Y true,d is the label annotated at the d-th vector time step, that is, the historical addition amount A(t)', the historical optimal addition time t' optimal , the historically corrected melting temperature T rev (t)', the historically corrected melting time t r ' ev and the historical temperature adjustment rate T adj (t)'; Y pred,d is the prediction result at the d-th vector time step, that is, the predicted addition amount A(t)', the optimal addition time t' optimal , the corrected melting temperature T rev (t)', the corrected melting time t r ' ev and the temperature adjustment rate T adj (t)';

[0052] Transfer the sequence data to the input layer, then pass through the hidden layer and the output layer in turn, output the prediction result, calculate the value of the loss function corresponding to the prediction result, and backpropagate the error gradient from the output layer to the input layer; according to the error gradient, use the optimization algorithm to update the network parameters to reduce the value of the loss function; repeat the training of the sequence data until the melting adjustment model converges or reaches the pre-set number of iterations in advance, which means the training of the melting adjustment model is completed.

[0053] Furthermore, the input layer is used to receive the fused feature vector X d as the input; the hidden layer consists of y state vectors, and the calculation formula of each state vector is: h d = f(W h ·h d-1 + W x ·Xd +b h ); where h d is the hidden state at the current vector time step d, f(...) is the activation function, and W x is the weight matrix from the input layer to the hidden layer, and W h is the recurrent weight matrix of the hidden layer, and b h is the bias vector of the hidden layer;

[0054] Based on the output of the hidden layer, the output layer calculates the output result of the melting adjustment model, and its calculation formula is: Y = η(W O ·h d +b o ); where Y is the output result of the output layer, η(...) is the activation function of the output layer, W o is the weight matrix from the hidden layer to the output layer, and b o is the bias vector of the output layer.

[0055] A copper-based superalloy with high mechanical fatigue performance for automotive leaf springs prepared by the method according to any one of the above.

[0056] Beneficial effects

[0057] The technical solution provided by the present invention has the following beneficial effects compared with the prior art:

[0058] Through the design of the covering agent addition model, the present invention can dynamically calculate the addition amount A and the optimal addition time t of the covering agent optimal , thereby effectively reducing the formation probability of pores during casting and improving the quality of the copper-based superalloy; and by dynamically monitoring the gas solubility C and the reaction rate R(T), it is possible to calculate the addition amount A of the covering agent required at different addition times in real time, effectively reducing the defects caused by the dissolution of gas in the metal and reducing the formation probability of pores; by analyzing historical data and optimizing the addition measurement of the covering agent, based on data analysis and machine learning models, scientific decision-making support is provided, reducing the deviation of human experience and improving the reliability of decision-making;

[0059] Based on the addition parameters, the corrected melting temperature T rev (t), the corrected melting time t rev and the temperature adjustment rate T adj (t) are calculated to form a dynamic and flexible casting process control system, aiming to optimize the casting process and improve the quality and performance of the casting; specifically, by precisely controlling the melting temperature, it can ensure that the reaction between the covering agent and the gas is more sufficient, thereby effectively reducing the gas solubility and reducing the risk of pore formation, and by dynamically adjusting the melting time, it can better adapt to different casting process requirements and material characteristics, ensuring the stability of the production process. Description of the drawings

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying 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 accompanying drawings can be obtained based on these drawings.

[0061] Figure 1 It is a schematic flow chart of the preparation method of the copper-based superalloy in the first embodiment of the present invention. Specific embodiments

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0063] The following further describes the present invention with reference to the embodiments.

[0064] Embodiment 1:

[0065] Referring to the attached Figure 1 , this case proposes a preparation method of a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs, including the following steps:

[0066] Step 1: Weigh the raw materials by mass percentage, including 55%-65% titanium, 30%-25% copper, 10%-8% aluminum, 3%-1% zinc, and 2%-1% magnesium;

[0067] Step 2: Put the raw materials into a furnace for melting. During the melting process, collect the melting parameters in the furnace, input the melting parameters into a pre-constructed and trained covering agent addition model to obtain the addition parameters of the covering agent, and add the covering agent into the furnace based on the addition parameters;

[0068] Step 3: Input the addition parameters into a pre-constructed and trained melting adjustment model to obtain the melting adjustment parameters, and melt the raw materials based on the melting adjustment parameters to obtain a melt;

[0069] Step 4: Pour the melt into a mold for casting to obtain a blank;

[0070] Step 5: Heat-treat the blank, and perform cold working on the heat-treated blank to obtain a copper-based superalloy.

[0071] Specifically, in this case, the method of constructing the covering agent addition model includes:

[0072] Define the melting parameters in the furnace, including: the partial pressure P of the gas and the melting temperature T;

[0073] Define the addition parameters of the covering agent, including: the addition amount A and the optimal addition time t optimal ;

[0074] Calculate the solubility C of the gas based on the partial pressure P of the gas. The calculation formula is:

[0075] C(t) = αP; where t is the time; C(t) represents the solubility of the gas in the molten metal (i.e., the melt) at time t, that is, the number of moles of gas that can be dissolved in the melt per unit volume. The solubility C of the gas is one of the very important parameters in the casting process. The higher the solubility C of the gas, the higher the concentration of the gas in the metal, and the greater the risk of forming pores; α is the proportionality constant in Henry's law, indicating the relationship between the solubility of the gas and the partial pressure of the gas in a specific metal liquid; P is the partial pressure of the gas, representing the partial pressure of the gas in the mixed gas or gas phase;

[0076] Update and calculate the solubility C of the gas based on the melting temperature T. The calculation formula is:

[0077] C actual (t) = C(t)·(1 + βΔT(t)); where C actual (t) represents the actual solubility of the gas in the molten metal at time t, taking into account the influence of the melting temperature fluctuation, and is more accurate than the solubility C calculated according to Henry's law; β is the melting temperature influence coefficient, indicating the degree of influence of the melting temperature fluctuation on the gas solubility; ΔT(t) represents the temperature fluctuation at time t, that is, the difference between the actual melting temperature and the expected melting temperature. The calculation formula is: ΔT(t) = T(t) - T expected (t), where T(t) represents the actual melting temperature collected at time t, and T expected (t) represents the expected melting temperature at time t;

[0078] Define the calculation expression of the reaction rate R(T) between the covering agent and the gas:

[0079] R(T) = aT 2 + bT + c; where the reaction rate R(T) represents the amount of the covering agent consumed by the reaction or the amount of the product generated by the reaction per unit time per unit volume of the molten metal; a, b, c are constants related to the specific covering agent type and gas type, determining the specific influence of the melting temperature on the reaction rate, and are usually calibrated based on experimental data;

[0080] Calculate the change in the solubility of a gas based on the reaction rate R(T), and the calculation formula is:

[0081] In the formula, ΔC(t) represents the change in the solubility of the gas at time t; R(T)(t′) represents the reaction rate at time t′, where t′ is an integration variable representing different time points considered during the integration process and is used to describe all time points from 0 to t;

[0082] Define the calculation formula for the addition amount A of the covering agent as:

[0083] A(t) = dC actual (t)R(T); in the formula, A(t) represents the addition amount of the covering agent to be added at time t; d is a proportionality constant;

[0084] Set the critical solubility threshold C th (t) and the pore formation rate threshold P defect-th ; Real-time monitor the actual solubility C actual (t) and the pore forming rate P defect (monitored in real time based on image processing or other detection methods); when C actual (t) > C th (t) and P defect > P defect-th At this time, trigger the addition of the covering agent and update the calculation formula for the addition amount A of the covering agent to obtain; A(t optimal ) = dC actual (t)R(T)·(1 + δ(P defect - P defect-th )); in the formula, t optimal is the optimal addition time of the covering agent;

[0085] More specifically, the training method of the covering agent addition model includes:

[0086] Collect the process parameters in the historical smelting process, extract the characteristic data in the process parameters, including the partial pressure P of the gas, the smelting temperature T, the reaction rate R(T) of the covering agent, the solubility C, the addition amount A, the optimal addition time t optimal and the pore forming rate P defect ; Clean and normalize the characteristic data; then construct the characteristic variables in the characteristic data into the input feature vector X = {P, T, R(T), C, P defect}, and construct the target variable into the output feature vector Y = {A, t optimal}; Construct the input feature vector X and the output feature vector Y into the training input data, including the training set and the test set. The training set accounts for 80% of the training input data, and the test set accounts for 20% of the training input data;

[0087] Define the random forest regression model SVR as the basic structure of the covering agent addition model, and initialize the network parameters of the basic training structure, including the kernel function K(x i ,x j ), the regularization parameter Z, and the insensitive loss function V; among them, the kernel function K(x i ,x j ) selects the RBF kernel. The RBF (radial basis function) is the most commonly used kernel function in SVR and is suitable for dealing with nonlinear problems. Its form is is a function for calculating the similarity between sample points x i and x j . The kernel function can effectively calculate the relationship between data points in the high-dimensional space without explicitly performing high-dimensional mapping, thus avoiding complex calculations; the sample points x i and x j represent two sample points in the input data, which may contain multiple feature data, and i and j are the sample point numbers; ||x i -x j || 2 is the Euclidean distance, and the expression is: where m is the number of feature data, x ik and x jk are the values of the k-th feature data of the sample points x i and x j ; σ is a scalar parameter used to control the width of the RBF kernel; exp(...) is the exponential function; the regularization parameter Z is used to control the complexity and error tolerance of the model. A larger Z value will make the model more complex and minimize the training error, but may lead to overfitting. A smaller Z value will make the model simpler and allow a certain training error. Usually, Z = 1 is set as the initial value; the insensitive loss function V defines the tolerance of the model. This value determines the error range between the predicted value and the actual value. The allowed error within this range will not be regarded as a loss. Usually, V = 1 is also set as the initial value;

[0088] Input the input feature vector X and the output feature vector Y of the training set into the basic structure; define the form of the loss function of the covering agent addition model as: In the formula, L(w, b, ξ) is the loss function to be minimized, representing the error and complexity of the model; w is the weight vector, which determines the importance of features in prediction. A larger weight value indicates a greater impact of the feature on the prediction result; b is the bias term, a constant term used to adjust the position of the hyperplane to make the model more flexible, and its role is similar to the intercept in a linear equation; ξ is the slack variable, representing the error between the predicted value and the actual value. When the model cannot accurately predict, this variable is used to relax the constraint conditions. The slack variable ξ of each sample i i represents its prediction error, which allows the model to tolerate a certain prediction error in some cases; is a measure of the model complexity. By introducing this term, the goal of the model is to keep the weights small to avoid overfitting;

[0089] Define the constraint conditions for each training sample (x i , y i ), including:

[0090] y i -(w·φ(x i )) + b) ≤ ε + ξ i ; This constraint ensures that the gap between the true value y i and the predicted value (w·φ(x i )) + b) is less than or equal to ε + ξ i , that is, allowing an error within a certain range ε;

[0091] (w·φ(x i )) + b) - y i ≤ ε + ξ i ; This constraint ensures that the gap between the predicted value (w·φ(x i )) + b) and the true value y i is less than or equal to ε + ξ i , ensuring that the positive error between the predicted value and the actual value is also within the limit;

[0092] ξ i ≥ 0; This means that each slack variable ξ must be non - negative, ensuring that the error cannot be negative, that is, the error cannot be rejected. Among them, φ is the feature mapping function, responsible for mapping the input data to a high - dimensional space. Through feature mapping, SVR can find the optimal hyperplane in a complex space; ε is a constant, defining the insensitive region between the predicted value and the true value. The error within this region will not be considered in the loss calculation.

[0093] Then use an optimization algorithm (such as sequential minimal optimization algorithm, gradient descent or other optimal algorithms) to update the weight vector w, the bias term b, and the slack variable ξ to minimize the value of the loss function L(w, b, ξ) while satisfying the above - mentioned constraint conditions;

[0094] The training set is repeatedly trained until the covering agent addition model converges (i.e., the value of the damage function no longer changes) or reaches a pre-set number of iterations in advance, indicating that the molten adjustment model training is completed.

[0095] It should be noted that in this solution,

[0096] Furthermore, in this solution, through the design of the covering agent addition model, the addition amount A and the optimal addition time t of the covering agent can be dynamically calculated optimal , thereby effectively reducing the formation probability of pores during the casting process and improving the quality of the copper-based superalloy; and by dynamically monitoring the gas solubility C and the reaction rate R(T), it is possible to calculate in real time the addition amount A of the covering agent required at different addition times, effectively reducing the defects caused by the dissolution of gas in the metal and reducing the formation probability of pores; by analyzing historical data and optimizing the addition measurement of the covering agent, based on data analysis and machine learning models, scientific decision-making support is provided, reducing the deviation of human experience and improving the reliability of decision-making.

[0097] Further, in this case, the addition of the covering agent may require adjusting the process parameters of the melting process, including the melting temperature, the melting time, and the melting temperature adjustment rate. For example, if the covering agent causes a decrease in the gas content in the melt, it may be necessary to lower the melting temperature or shorten the heating time to avoid performance deterioration caused by overheating;

[0098] The ways to construct the molten adjustment model include:

[0099] Define the molten adjustment parameters, including the corrected molten temperature T rev (t), the corrected molten time t rev and the temperature adjustment rate T adj (t);

[0100] Define the calculation formula for the corrected molten temperature T rev (t) as:

[0101] T rev (t) = T base + λ·A(t); where the corrected molten temperature T rev (t) represents the actual molten temperature adjusted according to the addition amount A(t) of the covering agent at time t; T base is the basic molten temperature, which is the target temperature set according to the material properties and the casting process without the influence of the covering agent, usually the theoretical temperature set during the optimal heat treatment or melting process; λ is the temperature adjustment coefficient, indicating the influence intensity of the covering agent type and the gas properties on the melting temperature. The magnitude of this coefficient directly affects the adjustment degree of the covering agent addition amount on the actual melting temperature, and is usually determined through experiments or theoretical analysis;

[0102] Define the corrected melting time t rev The calculation formula is as follows:

[0103] t rev = t base + μ·A(t); where, the corrected melting time t rev represents the actual required melting time adjusted according to the addition amount A(t) of the covering agent at time t; t base is the basic melting time, which refers to the melting time required under standard conditions, without a covering agent or under normal circumstances. This time is usually determined based on experiments and experience in actual production; μ is the time adjustment coefficient, which is used to describe the influence degree of the addition amount of the covering agent on the melting time;

[0104] Define the temperature adjustment rate T adj (t) The calculation formula is as follows:

[0105] In the formula, the temperature adjustment rate T adj (t) represents the speed of temperature change, which depends on the control situation of the current melting process; t rate is the time constant, which represents the response time in the temperature control process. This value reflects the thermal propulsion effect in the melting process and affects the time required to reach the target temperature. The smaller it is, the more sensitive the temperature adjustment is and the faster the change is;

[0106] The melting adjustment model calculates the corrected melting temperature T rev (t), the corrected melting time t rev and the temperature adjustment rate T adj (t) to form a dynamic and flexible casting process control system, aiming to optimize the casting process and improve the quality and performance of castings; specifically, by precisely controlling the melting temperature, it can ensure that the reaction between the covering agent and gas is more sufficient, thereby effectively reducing the gas solubility and reducing the risk of pore formation. By dynamically adjusting the melting time, it can better adapt to different casting process requirements and material characteristics, ensuring the stability of the production process.

[0107] Furthermore, in this case, the training method of the melting adjustment model includes:

[0108] Define the long short-term memory network model as the basic structure of the melting adjustment model, and the basic structure includes an input layer, a hidden layer, and an output layer;

[0109] Introduce the parameter vector time step d, and combine the vector time step d, the melting adjustment parameter, and the addition parameter to obtain the fused feature vector where, s is the number of vector time steps d;

[0110] Collect the melting adjustment parameters and addition parameters of the target blank within a fixed past time, and correspondingly construct a fused feature vector, denoted as the historical fused feature vector. Unfold the historical fused feature vector According to the vector time step to form sequence data as the training input of the melting adjustment model. Correspondingly, label the sequence data according to the vector time step d, and the labels are the historical addition amount A(t)′, the historical optimal addition time t o ′ ptimal , the historical adjusted melting temperature T rev (t)′, the historical adjusted melting time t r ′ ev and the historical temperature adjustment rate T adj (t)′;

[0111] The input layer is used to receive the fused feature vector X d as the input; the hidden layer consists of y state vectors, and the calculation formula for each state vector is: h d = f(W h ·h d-1 + W x ·X d + b h ); In the formula, h d is the hidden state of the current vector time step d, f(...) is the activation function, W x is the weight matrix from the input layer to the hidden layer, W h is the recurrent weight matrix of the hidden layer, b h is the bias vector of the hidden layer;

[0112] The output layer calculates the output result of the melting adjustment model based on the output of the hidden layer, and its calculation formula is: Y = η(W O ·h d + b o ); where, Y is the output result of the output layer, η(...) is the activation function of the output layer, W o is the weight matrix from the hidden layer to the output layer, b o is the bias vector of the output layer;

[0113] Initialize the network parameters of the melting adjustment model; the network parameters include the weight matrix W x from the input layer to the hidden layer, the recurrent weight matrix W h of the hidden layer, the bias vector b h of the hidden layer, the weight matrix W o from the hidden layer to the output layer, and the bias vector b o of the output layer;

[0114] Define the loss function of the melting adjustment model

[0115] Wherein, S is the length of the sequence data; Y true,d is the label marked at the d-th vector time step, that is, the historical addition amount A(t)', the historical best addition time t' optimal , the historical corrected melting temperature T rev (t)', the historical corrected melting time t r ' ev and the historical temperature adjustment rate T adj (t)'; Y pred,d is the prediction result at the d-th vector time step, that is, the predicted addition amount A(t)', the best addition time t' optimal , the corrected melting temperature T rev (t)', the corrected melting time t' rev and the temperature adjustment rate T adj (t)';

[0116] The sequence data is passed to the input layer, then successively passes through the hidden layer and the output layer, the prediction result is output, the value of the loss function corresponding to the prediction result is calculated, and the error gradient is propagated backward from the output layer to the input layer; according to the error gradient, the network parameters are updated using an optimization algorithm to reduce the value of the loss function; the training of the sequence data is repeated until the melting adjustment model converges (that is, the value of the damage function no longer changes) or reaches the pre-set number of iterations in advance, which indicates that the training of the melting adjustment model is completed.

[0117] It is worth mentioning that in this case, the melting adjustment model can continuously learn the latest data during the production process. By regularly retraining the model, it can adapt to different casting environments and process conditions. This adaptive ability enables the model to continuously update and optimize the prediction results. With the continuous accumulation of data, the model can optimize decisions based on big data analysis, making the decision-making process of operators more scientific and data-driven, and reducing the uncertainty brought by relying on personal experience.

[0118] Embodiment 2:

[0119] On the basis of Embodiment 1, this case proposes a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs, which is prepared by using the preparation method of a copper-based superalloy with high mechanical fatigue performance for automotive leaf springs in Embodiment 1.

[0120] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs, characterized in that: The following steps are involved: Step 1: weigh the raw materials by mass percentage, including 55%-65% titanium, 30%-25% copper, 10%-8% aluminum, 3%-1% zinc, and 2%-1% magnesium; Step 2: Put the raw materials into the furnace for melting, collect the melting parameters in the furnace during the melting process, input the melting parameters into the pre-built and trained covering agent addition model, obtain the covering agent addition parameters, and add the covering agent into the furnace based on the addition parameters; Step 3: inputting the added parameters into a pre-built and trained melting adjustment model to obtain melting adjustment parameters, and melting the raw materials based on the melting adjustment parameters to obtain a melt; Step 4: pouring the melt into a casting mold for casting to obtain a blank; Step 5: heat-treating the blank, and cold-working the blank after the heat treatment to obtain a copper-based superalloy.

2. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 1, characterized in that: The melting parameters in the furnace in step 2 include: gas partial pressure P and melting temperature T; the adding parameters of the covering agent include: addition amount A and optimal adding time t optimal .

3. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 2, characterized in that: The method of constructing the covering agent addition model in step 2 includes: The solubility C of the gas is calculated based on the partial pressure P of the gas. The calculation formula is: C(t) = αP; where t is time; C(t) represents the solubility of the gas in the melt at time t; α is the proportionality constant in Henry's law; The solubility C of the gas is updated and calculated based on the melting temperature T. The calculation formula is: C actual (t) = C (t) (1 + βΔT (t)); where C actual (t) represents the actual solubility of the gas in the melt at time t; β is the melting temperature influence coefficient; ΔT(t) represents the temperature fluctuation at time t, and the calculation formula is: ΔT(t) = T(t)-T expected (t), where T(t) represents the actual melting temperature collected at time t, T expected (t) represents the expected melting temperature at time t; Define the calculation expression of the reaction rate R(T) between the covering agent and the gas: R(T)=aT 2 +bT+c; where a, b, c are constants related to the type of covering agent and the type of gas; The calculation formula for defining the amount of covering agent added A is: A(t)=dC actual (t)R(T); where A(t) represents the amount of covering agent to be added at time t; d is the proportional constant; Set the critical solubility threshold C for determining the optimal time to add the covering agent th (t) and the pore formation rate threshold P defect-th , when C actual (t)>C th (t) and P defect >P defect-th When the amount of covering agent added A is updated, the calculation formula is obtained: A(t optimal )=dC actual (t)R(T)·(1+δ(P defect -P defect-th )).

4. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 3, characterized in that: The training method of the covering agent addition model in step 2 includes: Collect the process parameters in the historical smelting process and extract the characteristic data in the process parameters, including gas partial pressure P, smelting temperature T, capping agent reaction rate R (T), solubility C, addition amount A, and optimal addition time t optimal and pore forming rate P defect ; Clean and normalize the feature data; then construct the feature variables in the feature data as the input feature vector X = {P, T, R (T), C, P defect }, construct the target variable as the output feature vector Y = {A, t optimal }; Construct the input feature vector X and the output feature vector Y as training input data, including a training set and a test set; Define the random forest regression model SVR as the cover to add the basic structure of the model and initialize the network parameters of the basic training structure, including the kernel function K(x i ,x j ), regularization parameter Z and insensitive loss function V; The input feature vector X and output feature vector Y of the training set are input into the basic structure; the loss function of the covering agent addition model is defined as: Where L(w,b,ξ) is the loss function to be minimized; w is the weight vector; b is the bias term; ξ is the slack variable, and the slack variable ξ for each sample i is i represents its prediction error; Define each training sample (x i ,y i ) include: y i -(w·φ(x i )+b)≤ε+ξ i ; (w·φ(x i )+b)-y i ≤e+ξ i ; ξ i ≥0; where φ is the characteristic mapping function; ε is a constant; Then use the optimization algorithm to update the weight vector w, bias term b and slack variable ξ to minimize the value of the loss function L(w,b,ξ) while satisfying the above constraints; Repeat the training set until the cover agent addition model converges or reaches the preset number of iterations, which indicates melting. Where φ is the feature mapping function, which is responsible for mapping the input data to a high-dimensional space. Through feature mapping, SVR can find the best hyperplane in a complex space; the adjustment model training is completed.

5. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 4, characterized in that: The method of constructing the melting adjustment model in step 3 includes: Define melting adjustment parameters, including the corrected melting temperature T rev (t), corrected melting time t rev and temperature adjustment rate T adj (t); Define the corrected melting temperature T rev The calculation formula of (t) is: T rev (t) = T base +λ·A(t); where T base is the basic melting temperature, λ is the temperature adjustment coefficient; Define the corrected melting time t rev The calculation formula is: t rev =t base +μ·A(t); where; t base is the basic melting time; μ is the time adjustment coefficient, which is used to describe the influence of the amount of covering agent added on the melting time; Define the temperature adjustment rate T adj The calculation formula of (t) is: Where, t rate is the time constant.

6. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 5, characterized in that: The training method of the melting adjustment model includes: Define the long short-term memory network model as the basic structure of the melting adjustment model, the basic structure includes an input layer, a hidden layer, and an output layer; Introduce the parameter vector time step d, combine the vector time step d, the melting adjustment parameter, and the addition parameter to obtain the fused feature vector Where s is the number of vector time steps d; Collect the melting adjustment parameters and added parameter phases of the target billet in the past fixed time, and construct the corresponding fusion feature vector, which is recorded as the historical fusion feature vector Fusion of history into feature vector Expand according to the vector time step to form sequence data as the training input of the melting adjustment model. Correspondingly, the sequence data is labeled according to the vector time step d. The label is the historical addition amount A(t)′ and the historical optimal addition time t′ optimal , historically corrected melting temperature T rev (t)′, historically corrected melting time t′ rev and the historical temperature adjustment rate T adj (t)′; Initialize the network parameters of the melting adjustment model; the network parameters include the weight matrix W from the input layer to the hidden layer x , the recurrent weight matrix W of the hidden layer h , the bias vector b of the hidden layer h , the weight matrix W from the hidden layer to the output layer o and the bias vector b of the output layer o ; Define the loss function for the melt adjustment model In the formula, S is the length of the sequence data; Y true,d The label of the dth vector time step, that is, the historical addition amount A(t)′, the historical best addition time t′ optimal , historically corrected melting temperature T rev (t)′, historically corrected melting time t′ rev and the historical temperature adjustment rate T adj (t)′;Y pred,d is the prediction result of the dth vector time step, that is, the predicted addition amount A(t)′, the optimal addition time t′ optimal , corrected melting temperature T rev (t)′, corrected melting time t′ rev and temperature adjustment rate T adj (t)′; The sequence data is passed to the input layer, and then passes through the hidden layer and the output layer in turn, the prediction result is output, and the value of the loss function of the corresponding prediction result is calculated, and the error gradient is back-propagated from the output layer to the input layer; according to the error gradient, the network parameters are updated using the optimization algorithm to reduce the value of the loss function; the sequence data is repeatedly trained until the melting adjustment model converges or reaches the preset number of iterations, which means that the melting adjustment model training is completed.

7. The method for preparing a copper-based superalloy with high mechanical fatigue performance for automobile springs according to claim 6, characterized in that: The input layer is used to receive the fused feature vector X d As input; the hidden layer consists of y state vectors, and the calculation formula for each state vector is: d =f(W h ·h d-1 +W x ·X d +b h );where h d is the hidden state of the current vector at time step d, f(...) is the activation function, W x is the weight matrix from the input layer to the hidden layer, W h is the recurrent weight matrix of the hidden layer, b h is the bias vector of the hidden layer; The output layer calculates the output result of the melting adjustment model based on the output of the hidden layer. The calculation formula is: Y = η (W O ·h d +b o ), where Y is the output result of the output layer, η(...) is the activation function of the output layer, and W o is the weight matrix from the hidden layer to the output layer, b o is the bias vector of the output layer.

8. A copper-based superalloy with high mechanical fatigue performance for automobile springs prepared according to the method of any one of claims 1 to 9.

Images