Medium-high carbon steel reverse component design method and system based on conditional generative network

Abstract

The invention relates to a medium-high carbon steel reverse component design method and system based on a conditional generative network. The method comprises the following steps: converting a steel design index into a hardenability curve; inputting the hardenability curve and a preset random latent code into a trained and checked generator; the generator reversely generates steel components and grain size based on the hardenability curve; and the generated grain size and components are adjusted based on the component adjusting module so as to meet the smelting requirement. By adopting the method, a hardenability GAN algorithm can be taken as a core, database input and real hardenability curve input are increased, and a CGAN-based hardenability model is constructed. The model can learn a hardenability data probability distribution dependent on the alloy composition and grain size, and samples extracted from the multi-dimensional distribution will represent the alloy composition and grain size with predictable attributes.

Description

Technical Field

[0001] The present application relates to the technical field of steel design, and in particular to a method and system for reverse composition design of medium and high carbon steel based on a conditional generation network. Background Art

[0002] Hardenability is the ability of steel to obtain the depth of the hardened layer when quenched under certain conditions. After simplification, it can be considered that the hardenability of a steel grade is mainly affected by alloying elements. It can be considered that the hardenability can be obtained only by the alloying element content of the steel, and vice versa. Hardenability is simpler than other performance indicators such as strength, toughness and plasticity. These performance indicators are often strongly correlated with heat treatment processes. There are many performance indicators of steel, which are often related to hardenability, such as hardness, strength, etc. These indicators can be converted into hardenability indicators during research, or after a new steel grade is designed, the corresponding experimental methods can be used to measure some performance indicators that have not been considered. If the requirements cannot be met, adjustments can be made.

[0003] For example, Chinese patent CN110428876B provides a steel material design method based on a physically guided machine learning algorithm, which relates to the design of steel materials and the technical field of machine learning applications. The present invention first collects data, and uses the multiple holdout method to divide the data into a training set and a test set; establishes a support vector machine model based on physical metallurgy guidance according to the training set; uses the support vector machine model based on physical metallurgy guidance with a correlation coefficient greater than 90% as the objective function in the genetic algorithm; obtains the optimized composition, process and material with the best target performance; uses the SVC classifier to classify and screen the large number of design results obtained, and outputs the composition, process and target performance of its typical alloy. This method introduces physical metallurgical mechanisms into machine learning, and combines optimization algorithms to form a complete design platform, and the design results are more in line with the principles of physical metallurgy.

[0004] However, the hardenability design of traditional steel relies on theoretical structural models and expert experience, followed by trial and error experiments. However, the theoretical structural models are not fully mastered by people, and expert experience is also uncertain. Trial and error experiments are expensive and time-consuming. In addition, the hardenability requirements in practical applications are diverse, which makes the development of steel grades with suitable hardenability increasingly challenging. Summary of the invention

[0005] 1. Problem to be solved

[0006] Based on this, it is necessary to design the hardenability of the above-mentioned traditional steels based on theoretical structural models and expert experience, and then conduct trial and error experiments. However, the theoretical structural model is not fully mastered by people, and expert experience is also uncertain. Trial and error experiments are costly and time-consuming, and the requirements for hardenability in practical applications are diverse, which makes the development of steels with suitable hardenability an increasingly challenging technical problem. A method and system for reverse composition design of medium and high carbon steels based on conditional generative networks is provided, which can apply machine learning to study the laws between the hardenability and chemical composition of steel, does not rely on specific physical knowledge, and only needs to establish a nonlinear model between hardenability and chemical composition based on existing data.

[0007] 2. Technical solution

[0008] In a first aspect, the present application provides a method for reverse composition design of medium and high carbon steel based on a conditional generation network. The method comprises:

[0009] Convert steel design parameters into hardenability curves;

[0010] The hardenability curve and the preset random latent code are input into the trained and tested generator;

[0011] The generator generates steel composition and grain size inversely based on the hardenability curve;

[0012] Based on the composition adjustment module, the generated grain size and composition are adjusted to meet the smelting needs.

[0013] In one embodiment, the generator training process further includes:

[0014] Obtaining a hardenability curve and inputting it into a generator to generate a simulation curve corresponding to the hardenability curve;

[0015] Acquire the actual curve and perform difference analysis with the simulated curve input discriminator;

[0016] The discriminator is trained to learn the feature differences between the actual curve and the simulated curve and guide the generator to generate features based on the feature differences;

[0017] Calculate the inconsistency loss between the simulated curve and the actual curve;

[0018] The corresponding weight parameters in the generator and the discriminator are updated by back propagation until the cost function of the generator is minimized and the cost function of the discriminator is maximized.

[0019] In one embodiment, converting the steel design index into a hardenability curve comprises:

[0020] Based on the classification of steel design indicators, there are strong hardenability-related indicators and weak hardenability-related indicators;

[0021] The hardenability-related index is inversely calculated as the hardenability index;

[0022] The hardenability index is input into the generator and inversely calculated into the alloy element ratio of the steel;

[0023] Fine-tuning is performed based on the alloy element ratio to obtain the alloy element ratio for production.

[0024] In one embodiment, after the steel design index is classified as a weakly related index of hardenability, the method further includes:

[0025] Obtain the chemical composition corresponding to the weakly correlated index of hardenability and convert it into an influencing factor;

[0026] If the coupling of the chemical component with the remaining chemical components exceeds a preset critical value, an inverse operation of the hardenability index is added;

[0027] If the coupling of the chemical component with the other chemical components does not exceed the preset critical value, it is added to the fine-tuning process.

[0028] In one embodiment, the inverse operation of adding the hardenability index includes:

[0029] The approximate expression of the hardenability curve is as follows:

[0030]

[0031] Among them, H0 is the martensite hardness at the water-cooled end, that is, the starting maximum point of the hardenability curve, C f is the curvature coefficient, E s The distance between the body points in the half marathon.

[0032] The formula for the martensite hardness at the water-cooled end is as follows:

[0033] H0=(4+212C 3 ) / (0.1+3.2C 3 )-2C;

[0034] Among them, H0 is the martensite hardness at the water-cooled end;

[0035] Calculates the half marathon point distance based on strength indicators, including:

[0036] The formulas between Rockwell hardness, Brinell hardness, Vickers hardness and tensile strength are obtained respectively. The formulas for ultimate tensile strength and Rockwell hardness are as follows;

[0037] HRC=(0.0015σ b ) 3 +(0.0052σ b ) 2 +0.081σb -26;

[0038] Among them, HRC is Rockwell hardness, σ b is the ultimate tensile strength;

[0039] The formulas for ultimate tensile strength and Brinell and Vickers hardness are as follows;

[0040] HB=(61.8 / d) 2 -10;

[0041] σ b =(32 / d) 3 +235;

[0042] σ b =235+0.14(HB+10) 1.5 ;

[0043] Where HB is the Brinell hardness, d is the indentation diameter of the Brinell hardness test, σ b is the ultimate tensile strength;

[0044] When the quenching process is water quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows:

[0045]

[0046] Where D is the critical diameter of quenching, E s The distance between the body points in the half marathon.

[0047] When the quenching process is oil quenching, the formula between the critical quenching diameter and the half-Martensite point distance is as follows:

[0048]

[0049] Where D is the critical diameter of quenching, E s The distance between the body points in the half marathon.

[0050] When the quenching process is air quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows:

[0051]

[0052] Where D is the critical diameter of quenching, E s It is the body point distance for half marathon.

[0053] In one embodiment, calculating the half marathon body point distance based on impact toughness includes:

[0054] The formula for low temperature tempering toughness and martensite strength is as follows:

[0055] a km=(15356 / σ bm ) 2 ;

[0056] Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength;

[0057] The formulas for quenched and tempered toughness and quenched and tempered strength are as follows:

[0058]

[0059] Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength, σ bq is the quenching strength before quenching and tempering, σ bt It is the quenching strength after quenching and tempering;

[0060] Convert impact toughness to ultimate strength and calculate half marathon point distance.

[0061] In a second aspect, the present application also provides a medium-high carbon steel reverse composition design system based on a conditional generation network. The system includes:

[0062] Hardenability curve conversion module, used to convert steel design indicators into hardenability curves;

[0063] A hardenability curve input module, used for inputting a hardenability curve and a preset random latent code into a trained and tested generator;

[0064] The reverse composition generation module is used for the generator to reversely generate the steel composition and grain size based on the hardenability curve;

[0065] The smelting composition fine-tuning module is used to adjust the generated grain size and composition based on the composition adjustment module to meet the smelting needs.

[0066] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0067] Convert steel design parameters into hardenability curves;

[0068] The hardenability curve and the preset random latent code are input into the trained and tested generator;

[0069] The generator generates steel composition and grain size inversely based on the hardenability curve;

[0070] Based on the composition adjustment module, the generated grain size and composition are adjusted to meet the smelting needs.

[0071] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0072] Convert steel design parameters into hardenability curves;

[0073] The hardenability curve and the preset random latent code are input into the trained and tested generator;

[0074] The generator generates steel composition and grain size inversely based on the hardenability curve;

[0075] Based on the composition adjustment module, the generated grain size and composition are adjusted to meet the smelting needs.

[0076] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0077] Convert steel design parameters into hardenability curves;

[0078] The hardenability curve and the preset random latent code are input into the trained and tested generator;

[0079] The generator generates steel composition and grain size inversely based on the hardenability curve;

[0080] Based on the composition adjustment module, the generated grain size and composition are adjusted to meet the smelting needs.

[0081] The above-mentioned medium-high carbon steel reverse composition design method and system based on conditional generative network converts steel design indicators into hardenability curves; inputs the hardenability curves and preset random latent codes into a trained and tested generator; the generator reversely generates steel composition and grain size based on the hardenability curve; and adjusts the generated grain size and composition based on the composition adjustment module to meet smelting needs.

[0082] 3. Beneficial effects

[0083] This application adopts the above method, takes the hardenability GAN algorithm as the core, adds database input and real hardenability curve input, and constructs a hardenability model based on CGAN. The model can learn the probability distribution of hardenability data depending on alloy composition and grain size. The samples extracted from the multidimensional distribution will represent alloy composition and grain size with predictable properties;

[0084] Based on the trained generator, a reverse design network for steel composition was constructed. The performance indicators of the steel to be developed are converted into hardenability curves by the C module in the network. The generator G has learned multiple potential spaces of the relationship between performance and characteristic distribution through training, so it can reversely generate the composition and grain size of steel from the hardenability curve converted by the C module; the T module adjusts the composition and grain size generated by the generator in combination with metallurgical principles, and formulates a range that allows fluctuations in each component and a heat treatment system for reference.

[0085] By using the constructed network and combining metallurgical principles, high-strength low-alloy automobile beam steel and gear steel were designed and verified in production. The results show that the tensile strength and impact toughness of the experimental samples meet or partially exceed the design requirements, which is also proved by the metallographic structure experiment. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 A framework diagram of a GAN hardenability curve algorithm in one embodiment;

[0087] Figure 2 A flowchart of a method for reverse composition design of medium-high carbon steel based on a conditional generation network in one embodiment;

[0088] Figure 3 A framework diagram of a hardenability CGNN model in one embodiment;

[0089] Figure 4 A framework diagram for generating a model framework for reverse design of hardenability in one embodiment;

[0090] Figure 5 A framework diagram added for the chemical composition influencing factors of the relevant performance indicators in an embodiment;

[0091] Figure 6 A structural block diagram of a medium-high carbon steel reverse composition design system based on a conditional generation network in one embodiment;

[0092] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0094] The model is based on an end-to-end DNN architecture that has been successfully applied to other material design fields recently. The model relies on implicit feature learning and is able to learn meaningful representations of complex design spaces. Although the original design space may be jagged and discontinuous in many dimensions, the learned space is still low-dimensional and smooth. Among these models, the Generative Adversarial Network (GAN) performs best.

[0095] Reference Figure 1 , the GAN model consists of two DNNs: one is the generator, which is used to learn the mapping between the random space and the target distribution (effectively generating new data); the other is the critic, which is used to learn how to distinguish between observed data and generated data. The so-called "adversarial" means that during the training process, the generator tries to generate data that is closer and closer to the real data, while the discriminator tries to capture the difference between the data generated by the generator and the observed data. Ideally, the generator and the discriminator converge to the Nash equilibrium point through continuous training. When the discriminator cannot distinguish between the two, the data generated by the generator reaches a level sufficient to replace the real data. Compared with ordinary end-to-end networks, the main advantage of adversarial generative networks is that they can learn in a structured way, and the discriminator can control the generator output from a higher perspective, so that the generator output is more in line with the real logic. Starting from the basic idea of ​​GAN, a GAN-based hardenability generation network algorithm block diagram is proposed. Based on the theoretical knowledge, it only needs to start from the existing data to establish a nonlinear model between hardenability and chemical composition.

[0096] Reference Figure 2 , Figure 3 and Figure 4 , the method comprises the following steps:

[0097] Step 202: converting the steel design index into a hardenability curve.

[0098] Among them, with the hardenability GAN algorithm as the core, the database input and the real hardenability curve input are added to build a hardenability model based on CGAN. In the CGAN architecture, the generator is equipped with an additional conditional vector to strengthen the mapping between the latent space and the required space; the generator can learn the probability distribution of hardenability data depending on the alloy composition and grain size, and the samples extracted from the multidimensional distribution will represent the alloy composition and grain size with predictable properties.

[0099] Step 204 , input the hardenability curve and the preset random latent code into the trained and tested generator.

[0100] Among them, the generator G and the discriminator D are both convolutional neural networks (CNN). The conditions are added to the generator and discriminator networks respectively, and the convolutional network is used to extract features and add conditions to assist in error-correction training. The CNN generator generates new data based on the input data, and sends the new and old data together to the discriminator for authenticity judgment.

[0101] In step 206, the generator generates steel composition and grain size in reverse based on the hardenability curve.

[0102] Among them, the function of the C module is to convert the required performance into the hardenability curve, which is composed of a series of calculation functions; the G module is the generator of the generative adversarial network. The generator model is trained and tested using database data. Its main function is to reversely generate the composition and grain size of the input designed hardenability curve; the T module is a composition adjustment module. Its main function is to adjust the composition generated by the generator in combination with the grain size requirements to meet the smelting needs.

[0103] Step 208: Adjust the generated grain size and composition based on the composition adjustment module to meet the smelting requirements.

[0104] In the above-mentioned reverse composition design method for medium and high carbon steel based on conditional generative network, the hardenability GAN algorithm is used as the core, and the database input and real hardenability curve input are added to build a hardenability model based on CGAN. The model can learn the probability distribution of hardenability data depending on alloy composition and grain size, and the samples extracted from the multidimensional distribution will represent alloy composition and grain size with predictable properties.

[0105] It is worth mentioning that the hardenability curve of steel can be described as a random function that varies with the end quenching distance t, as shown in the following formula:

[0106] HRC = F(t);

[0107] Where HRC is the hardenability curve of steel, and F(t) is a random function that varies with the end quenching distance t;

[0108] The measured hardenability values ​​constitute a one-dimensional data set {x^1, x^2, … x^n}, from which m are selected as the training set. The training set can be written as a probability function, as follows:

[0109]

[0110] Through the divergence function KL, the parameter θ that makes L take the maximum value can be obtained * , the formula is as follows:

[0111]

[0112] θ * =argmin θKL(p data (x) / / p model (x;θ));

[0113] The divergence KL between the generated model distribution and the real data distribution is minimized, making the generated model infinitely close to the real data distribution. However, the real distribution of hardenability values ​​is unknown, and only the empirical distribution approximation can be formed using training data.

[0114] The data is mapped from the input space to the generated space by the generator, x = G(z), which satisfies the distribution or uniform distribution. The generator G can simulate a given type of distribution. The formula of the discriminator's cost function J is as follows:

[0115]

[0116] The generator and the discriminator can be regarded as a zero-sum game, so the cost function of the generator is as follows:

[0117] J (G) +J (D) =0;

[0118] Assume that the cost v function represents J (G) and J (D) , the formula is as follows:

[0119] v(θ (D) ,θ (G) )=E x～Pdata logD(x)+E x～Pz log((G z ));

[0120]

[0121] From the above formula, we can see that we need to find a suitable v function so that J (G) and J (D) As small as possible; for the discriminator, the larger the ν function, the better, and for the generator, the smaller the better, so a game is formed between the two. According to game theory, a Nash equilibrium is formed in the decision combination of the two parties in the game. In this state, neither party can increase its benefits through its own behavior. In the GAN network, the generator and the discriminator need to find a Nash equilibrium point in order to minimize their respective cost functions, that is, to find a v(θ (D) ,θ (G) ), so that the generator requirement is minimized and the discriminator is maximized. It can be defined as a problem of searching for a large minimum value. The formula is as follows:

[0122] arg min G max D v(D,G);

[0123] The ideal discriminator D* and generator G* are formulated as follows:

[0124] D * = arg max D v(D,G);

[0125] G * = arg min G max D ν(D,G)=arg min G v(D * ,G);

[0126] If G is a constant and G(z)=x, we can get:

[0127]

[0128] If x does not change, then:

[0129]

[0130] From the above formula, we can know that D∈(0,1), the ideal discriminator’s mode has a judgment value of 1 for the real data obtained and a judgment value of 0 for the generated data obtained. If the distribution of the generated data is very close to the distribution of the real data, the output result is 1 / 2.

[0131] D * Substituting (x) into the above formula, we can get:

[0132]

[0133]

[0134] Expressing the above formula in KL divergence form, the formula is as follows:

[0135]

[0136] Further converted into Jensen-Shannon (JS) divergence form, the formula is as follows:

[0137] maxv(G,D * )=A+2×JSD(p data ||p g );

[0138] The JS divergence is non-negative, when p data =p g When , the optimal generator G* is obtained.

[0139] In one embodiment, Figure 3 As shown, the generator training process includes:

[0140] The hardenability curve is obtained and input into a generator to generate a simulation curve corresponding to the hardenability curve; the actual curve is obtained and input into a discriminator for difference analysis with the simulation curve; the discriminator is trained to learn the feature difference between the actual curve and the simulation curve and guide the generator to generate features based on the feature difference; the inconsistency loss between the simulation curve and the actual curve is calculated; and the corresponding weight parameters in the generator and the discriminator are updated through back propagation until the cost function of the generator is minimized and the cost function of the discriminator is maximized.

[0141] Among them, the generation network and the adversarial network are independent of each other; the GAN network training mode is interactive and iterative, and the optimized objective function is:

[0142]

[0143] The optimization process is a game of maximum and minimum, and finally the discriminator and generator are combined into a whole; the final objective function of GAN is:

[0144] H=E x～p(x) log D(x)+E z～p(z) log D(x);

[0145] Where x is the true value, p(x) is the statistical distribution of the true value, z is the encoded statistical sampling random number, and p(z) is the statistical distribution of the encoding;

[0146] This application uses the cross entropy loss function. The loss function of the generator G and the function of the discriminator D are as follows:

[0147] L G =-E x～p(x) [log(1-D(G(x)))];

[0148] L D =-E x～p(x) log D(x)-E z～p(z) log(1-D(G(z)));

[0149] In one embodiment, Figure 4 and Figure 5 As shown, the conversion of steel design indicators into hardenability curves includes:

[0150] Based on the classification of steel design indicators, they are divided into indicators with strong correlation to hardenability and indicators with weak correlation to hardenability; the indicators with strong correlation to hardenability are inversely calculated into hardenability indicators; the hardenability indicators are input into a generator and inversely calculated into alloy element ratios of steel; based on the alloy element ratios, fine-tuning is performed to obtain alloy element ratios for production.

[0151] Among them, in the use, production and processing of steel, there are many required indicators, including mechanical performance indicators such as strength, hardness, plasticity, toughness, etc., chemical performance indicators such as corrosion resistance and oxidation resistance, and process performance such as tensile properties, impact properties, fatigue properties, etc. It is difficult to fully consider these performance indicators and it is difficult to achieve. These indicators can be roughly divided into indicators that are strongly related to hardenability and indicators that are weakly related.

[0152] The strength and toughness of steel (including tensile properties, yield strength, plasticity, impact toughness, hardness, etc.) indicators mainly depend on the organizational state of steel, which is the transformation product of high-temperature austenite cooling down. This transformation product depends on the supercooling ability of austenite. The important indicator for measuring the supercooling ability of austenite is hardenability. In other words, the process of obtaining some of the indicators of steel, such as hardness, strength, and toughness, is to formulate the corresponding heat treatment process based on hardenability, and obtain the corresponding steel organization through the heat treatment process. The organization of steel determines these properties. Based on the above two points, the idea of ​​reverse development is to inversely calculate the indicators that are strongly related to hardenability into hardenability indicators, and then inversely calculate the hardenability indicators into the alloy element ratio of steel through the generator in the generative adversarial network, and finally obtain the alloy element ratio that can actually be used in production through fine-tuning.

[0153] There are some performance indicators: such as the corrosion resistance of steel, which mainly depends on elements such as Cr, Ni, and Ti. The magnetic permeability of steel mainly depends on the Si element. Other performance indicators such as fatigue resistance, high temperature creep resistance, corrosion resistance, weldability, etc. are also affected by different chemical compositions and hot working processes. These performance indicators can be roughly considered to be determined by the organization and chemical composition. These performance indicators are strongly correlated with individual chemical compositions and the organization of steel, and have a relatively weak relationship with hardenability. The current research idea is to convert the chemical compositions that affect these performance indicators into influencing factors. If these chemical compositions have a strong coupling with the other chemical compositions, they are added to the hardenability calculation process. If they have a weak coupling with the other chemical compositions, they are added to the composition fine-tuning calculation process.

[0154] Simplifying the unimportant factors that affect hardenability, it can be roughly considered that it is mainly affected by alloying elements. It can be considered that hardenability can be obtained only by the alloying element content of steel, and vice versa. Hardenability is simpler than other performance indicators such as strength, toughness and plasticity, and performance indicators such as hardness and strength have a strong correlation with hardenability. These indicators can be converted into hardenability indicators during research, or after a new steel grade is designed, the corresponding experimental methods can be used to measure some performance indicators that are not taken into account. If the requirements cannot be met, fine-tuning can be performed. This application is designed with hardness, strength and impact toughness as the main required indicators.

[0155] In actual production, it is difficult to coordinate the different requirements of these indicators for composition by relying solely on expert experience combined with simulation and physical experiments. In order to coordinate the different requirements of different indicators for composition, it is a better choice to convert these indicators into hardness requirements and then design the composition. Converting other indicators into hardness indicators and then further converting them into the hardenability of steel is Figure 5 The C module in is completed, and the calculation process of the C module is as follows:

[0156] The hardenability curve expresses the nonlinear relationship between the end quenching hardness and the end quenching distance. The approximate expression is as follows:

[0157]

[0158] Among them, H0 is the martensite hardness at the water-cooled end, the starting maximum point of the hardenability curve, and C f is the curvature coefficient, E s The distance between the body points in the half marathon.

[0159] Among them, the hardenability curve is a hardness function with the end quenching distance as a variable. The maximum value H0 of the function is the martensite hardness, and the function decreases monotonically. If the quenched martensite hardness H0 can be determined, the curvature coefficient C f The end quenching distance E from quenching to obtain half of the martensite s , the hardenability curve of the steel can be approximately depicted.

[0160] H0 is the martensite hardness at the water-cooled end, which is also the starting point of the function. It can be approximately considered to be only related to the carbon element. It is the maximum hardness that this type of steel can take. It can be directly given during design. The formula for the martensite hardness at the water-cooled end is as follows:

[0161] H0=(4+212C 3 ) / (0.1+3.2C 3 )-2C;

[0162] Among them, H0 is the martensite hardness of the water-cooled end; when designing most steel grades, C f It can take values ​​between 0.5 and 2.5. f When the value is 0.5, the inflection point of the function and E s Point coincidence, C f decreases, the inflection point of the function moves toward the direction of increasing end-quenching distance, C f increases, the inflection point moves toward the direction of decreasing end quenching distance. In addition, C f The value directly affects the minimum value of the hardenability curve and is modified according to the tissue farthest from the water spray end;

[0163] Currently, C f There is no reliable formula yet, and most people make choices based on experience; E sIt is a critical point on the hardenability curve, which directly determines the position of the inflection point of the hardenability curve. The structure at this point has higher hardness and better toughness. s The value of can be obtained by calculating the strength index and toughness index required during design.

[0164] Calculates the half marathon point distance based on strength indicators, including:

[0165] The formulas between Rockwell hardness, Brinell hardness, Vickers hardness and tensile strength are obtained respectively. The formulas for ultimate tensile strength and Rockwell hardness are as follows;

[0166] HRC=(0.0015σ b ) 3 +(0.0052σ b ) 2 +0.081σ b -26;

[0167] Among them, HRC is Rockwell hardness, σ b is the ultimate tensile strength;

[0168] The formulas for ultimate tensile strength and Brinell and Vickers hardness are as follows;

[0169] HB=(61.8 / d) 2 -10;

[0170] σ b =(32 / d) 3 +235;

[0171] σ b =235+0.14(HB+10) 1.5 ;

[0172] Where HB is the Brinell hardness, d is the indentation diameter of the Brinell hardness test, σ b is the ultimate tensile strength;

[0173] When the quenching process is water quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows:

[0174]

[0175] Where D is the critical diameter of quenching, E s The distance between the body points in the half marathon.

[0176] When the quenching process is oil quenching, the formula between the critical quenching diameter and the half-Martensite point distance is as follows:

[0177]

[0178] Where D is the critical diameter of quenching, Es The distance between the body points in the half marathon.

[0179] When the quenching process is air quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows:

[0180]

[0181] Where D is the critical diameter of quenching, E s It is the body point distance for half marathon.

[0182] In one embodiment, the toughness of steel reflects the steel's resistance to impact loads, and the toughness is closely related to the organizational conditions and metallurgical quality; under the general metallurgical quality level and the fully martensitic quenching state, the toughness at the same strength level is higher than the toughness of the low-temperature quenched martensite, that is, the toughness decreases with the increase of strength, the degradation rate is slower, and the square of the toughness is inversely proportional to the strength. The toughness of the quenched organization at medium and high temperatures increases with the increase of resistance, and the decrease of toughness is inversely proportional to the cube of resistance; in addition, the degree of hardening before medium or high temperature tempering also has a great influence on the tempering toughness, and after tempering, it is relative to water at the same strength. The lower the hardness, the lower the toughness after tempering.

[0183] Calculate the half marathon body point distance based on impact toughness, including:

[0184] The formula for low temperature tempering toughness and martensite strength is as follows:

[0185] a km =(15356 / σ bm ) 2 ;

[0186] Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength;

[0187] The formulas for quenched and tempered toughness and quenched and tempered strength are as follows:

[0188]

[0189] Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength, σ bq is the quenching strength before quenching and tempering, σ bt It is the quenching strength after quenching and tempering;

[0190] Convert impact toughness to ultimate strength and calculate half marathon point distance.

[0191] Input the designed hardenability curve into the generator. The generator has extracted the relationship between composition, grain size and hardenability curve during training, so it can output the composition and grain size of the input hardenability curve through reverse calculation.

[0192] It is worth mentioning that the composition adjustment is completed by the T module in the network model, and the composition percentage content and grain size generated by the generator are both single values, which are difficult to be directly used for smelting control and process formulation.

[0193] The solution of this application is based on the components generated by the generator, which are converted by the T module in the network to formulate a range within which each component is allowed to fluctuate; the T module takes the percentage content of the components generated by the generator as the intermediate value, and combines the principles of metallurgy to formulate a width function for each element that allows fluctuations; the width function has the property of decreasing with increasing content, that is, the higher the content of the component, the stricter the control and the smaller the allowable fluctuation range; the C element has the most obvious effect on hardness and is also the largest in the percentage component content, so its fluctuation range should be more strictly controlled than other elements; other elements can refer to the correlation analysis results and adopt a relatively wide fluctuation range. The width function formula of the C element is as follows:

[0194]

[0195] The width function formulas of other elements are as follows:

[0196]

[0197] It should be pointed out that the influence of each element is complex, and the calculation in the above formula is not accurate enough. It is only suitable for the regression results of the research and is not universally representative.

[0198] The effects of chemical elements on performance are complex and they are coupled to each other. The difficulty and cost of adding different elements are also inconsistent, especially rare elements, which are expensive. Therefore, from an economic point of view, the amount of cheap elements should be increased and the amount of expensive elements should be reduced as much as possible without affecting performance. This is also an important part of ingredient adjustment. Currently, the adjustment of chemical composition mainly relies on expert experience and lacks strict theoretical guidance.

[0199] In hot working, the deformation termination temperature of steel structure has an important influence on grain size; the higher the deformation termination temperature, the stronger the tendency of grain aggregation and the coarser the austenite grains. Therefore, the deformation termination temperature should be reduced as much as possible, but generally not lower than the Ar3 line. The grain size generated by the generator can provide guidance for the formulation of slow cooling system and final rolling temperature.

[0200] In actual production, according to the composition ratio generated by CGAN and combined with the actual steelmaking practice, the original solution content and target content of each element are calculated to determine the final amount of each alloy material and form an alloy material usage table.

[0201] During the design phase, appropriate indicators are generally selected based on application requirements and usage environment. Generally speaking, high-strength steel can be selected when high strength is required and weight is not an issue, while ductile steel and plastic steel can be selected when weight is the primary consideration or there is a risk of local damage and deformation.

[0202] In addition to tensile strength and yield strength indicators, some other strength indicators need to be considered in the design. For example, in some specific cases, impact resistance must be considered, that is, the ability to absorb energy when subjected to impact loads. In addition, fatigue strength must also be considered when steel works under cyclic loads. The strength index of steel is one of the most important parameters in design and engineering applications. The appropriate steel strength index should be selected according to the specific application needs and structural requirements to ensure the safety and reliability of the structure, but other performance indicators must also be considered, such as corrosion resistance and weldability, etc., to meet the requirements of special engineering environments. In order to reduce the difficulty of research, tensile strength and impact toughness are used as the main indicators for design in this application.

[0203] There are a lot of literature on steel research, including various simulation experiments, physical experiments, and theoretical research. How to add such a large amount of data to the database, mine it using machine learning methods, and add the mined information as a condition to the CGAN network to achieve comprehensive prediction and reverse engineering development of multiple performance indicators is a relatively difficult and huge research topic, which still requires a long exploration process.

[0204] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0205] Based on the same inventive concept, the embodiment of the present application also provides a medium-high carbon steel reverse composition design device based on a conditional generation network for implementing the medium-high carbon steel reverse composition design method based on a conditional generation network involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the medium-high carbon steel reverse composition design device based on a conditional generation network provided below can refer to the limitations of the medium-high carbon steel reverse composition design method based on a conditional generation network above, and will not be repeated here.

[0206] In one embodiment, Figure 6 As shown, a medium-high carbon steel reverse composition design device based on a conditional generation network is provided, comprising: a hardening curve conversion module, a hardening curve input module, a reverse composition generation module and a smelting composition fine-tuning module, wherein:

[0207] Hardenability curve conversion module, used to convert steel design indicators into hardenability curves;

[0208] A hardenability curve input module, used for inputting a hardenability curve and a preset random latent code into a trained and tested generator;

[0209] The reverse composition generation module is used for the generator to reversely generate the steel composition and grain size based on the hardenability curve;

[0210] The smelting composition fine-tuning module is used to adjust the generated grain size and composition based on the composition adjustment module to meet the smelting needs.

[0211] In one embodiment, the hardenability curve input module is also used to: obtain the hardenability curve and input it into the generator to generate a simulation curve corresponding to the hardenability curve; obtain the actual curve and input it into the discriminator for difference analysis with the simulation curve; train the discriminator to learn the feature differences between the actual curve and the simulation curve and guide the generator to generate features based on the feature differences; calculate the inconsistency loss between the simulation curve and the actual curve; and update the corresponding weight parameters in the generator and the discriminator through back propagation until the cost function of the generator is minimized and the cost function of the discriminator is maximized.

[0212] In one embodiment, the hardenability curve conversion module is also used to: classify steel design indicators into strong hardenability related indicators and weak hardenability related indicators; inversely calculate the hardenability related indicators to obtain the hardenability indicators; input the hardenability indicators into the generator to inversely calculate the alloy element ratio of the steel; and fine-tune the alloy element ratio based on the alloy element ratio to obtain the alloy element ratio for production.

[0213] In one embodiment, the hardenability curve conversion module is also used to: obtain the chemical composition corresponding to the weakly correlated index of hardenability and convert it into an influencing factor; if the coupling between the chemical composition and the other chemical compositions exceeds a preset critical value, then add the inverse operation of the hardenability index; if the coupling between the chemical composition and the other chemical compositions does not exceed the preset critical value, then add it to the fine-tuning process.

[0214] In one embodiment, the hardenability curve conversion module is further used for: the approximate expression of the hardenability curve is as follows:

[0215] Among them, H0 is the martensite hardness at the water-cooled end, that is, the starting maximum point of the hardenability curve, C f is the curvature coefficient, E s is the distance between the half-martensite points; the formula for the hardness of the water-cooled end martensite is as follows: H0 = (4 + 212C 3 ) / (0.1+3.2H 3 )-2H; where H0 is the martensite hardness at the water-cooled end; the semi-martensite point distance is calculated based on the strength index, including: obtaining the formulas between Rockwell hardness, Brinell hardness, Vickers hardness and tensile strength respectively, and the formulas for tensile strength limit and Rockwell hardness are as follows; HRC=(0.0015σ b ) 3 +(0.0052σ b ) 2 +0.081σ b -26; HRC is Rockwell hardness, σ b is the tensile strength limit; the formula for tensile strength limit and Brinell hardness and Vickers hardness is as follows; HB = (61.8 / d) 2 -10; σ b =(32 / d) 3 +235; σ b =235+0.14(HB+10) 1.5 ; Among them, HB is the Brinell hardness, d is the indentation diameter of the Brinell hardness test, σ b is the tensile strength limit; when the quenching process is water quenching, the formula between the quenching critical diameter and the semi-Martensite point distance is as follows: Where D is the critical diameter of quenching, E s is the semi-Marathon body point distance; when the quenching process is oil quenching, the formula between the quenching critical diameter and the semi-Marathon body point distance is as follows: Where D is the critical diameter of quenching, E s is the semi-Marathon body point distance; when the quenching process is air quenching, the formula between the quenching critical diameter and the semi-Marathon body point distance is as follows: Where D is the critical diameter of quenching, E sIt is the body point distance for half marathon.

[0216] In one embodiment, the hardening curve conversion module is further used to: calculate the half-martensite point distance based on the impact toughness, including: the formula of low temperature tempering toughness and martensite strength is as follows: a km =(15356 / σ bm ) 2 ; Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength; the formulas for quenched and tempered toughness and quenched and tempered strength are as follows: Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength, σ bq is the quenching strength before quenching and tempering, σ bt is the quenching strength after quenching and tempering; convert the impact toughness into the ultimate strength and calculate the half-Martensite point distance.

[0217] Each module in the above-mentioned medium-high carbon steel reverse composition design system based on conditional generative network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0218] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for reverse composition design of medium and high carbon steel based on a conditional generation network is implemented.

[0219] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a reverse composition design method for medium and high carbon steel based on a conditional generation network is implemented.

[0220] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0221] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0222] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0223] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0224] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0225] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0226] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0227] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for reverse composition design of medium and high carbon steel based on conditional generation network, characterized in that: The method comprises: Convert steel design parameters into hardenability curves; The hardenability curve and the preset random latent code are input into the trained and tested generator; The generator generates steel composition and grain size inversely based on the hardenability curve; Based on the composition adjustment module, the generated grain size and composition are adjusted to meet the smelting needs.

2. The method for reverse composition design of medium and high carbon steel based on conditional generative network according to claim 1 is characterized in that: The generator training process also includes: Obtaining a hardenability curve and inputting it into a generator to generate a simulation curve corresponding to the hardenability curve; Acquire the actual curve and perform difference analysis with the simulated curve input discriminator; The discriminator is trained to learn the feature differences between the actual curve and the simulated curve and guide the generator to generate features based on the feature differences; Calculate the inconsistency loss between the simulated curve and the actual curve; The corresponding weight parameters in the generator and the discriminator are updated by back propagation until the cost function of the generator is minimized and the cost function of the discriminator is maximized.

3. The method for reverse composition design of medium and high carbon steel based on conditional generative network according to claim 2 is characterized in that: The method of converting the steel design index into a hardenability curve comprises: Based on the classification of steel design indicators, there are strong hardenability-related indicators and weak hardenability-related indicators; The hardenability-related index is inversely calculated as the hardenability index; The hardenability index is input into the generator and inversely calculated into the alloy element ratio of the steel; Fine-tuning is performed based on the alloy element ratio to obtain the alloy element ratio for production.

4. The method for reverse composition design of medium and high carbon steel based on conditional generative network according to claim 3 is characterized in that: After the steel material design index is classified as a weakly related index of hardenability, it also includes: Obtain the chemical composition corresponding to the weakly correlated index of hardenability and convert it into an influencing factor; If the coupling of the chemical component with the remaining chemical components exceeds a preset critical value, an inverse operation of the hardenability index is added; If the coupling of the chemical component with the other chemical components does not exceed the preset critical value, it is added to the fine-tuning process.

5. The method for reverse composition design of medium and high carbon steel based on conditional generative network according to claim 4, characterized in that: The inverse operation of adding the hardenability index includes: The approximate expression of the hardenability curve is as follows: Among them, H0 is the martensite hardness at the water-cooled end, that is, the starting maximum point of the hardenability curve, C f is the curvature coefficient, E s The distance between the body points in the half marathon. The formula for the martensite hardness at the water-cooled end is as follows: H0=(4+212C 3 ) / (0.1+3.2C 3 )-2C; Among them, H0 is the martensite hardness at the water-cooled end; Calculates the half marathon point distance based on strength indicators, including: The formulas between Rockwell hardness, Brinell hardness, Vickers hardness and tensile strength are obtained respectively. The formulas for ultimate tensile strength and Rockwell hardness are as follows; HRC=(0.0015σ b ) 3 +(0.0052σ b ) 2 +0.081σ b -26; Among them, HRC is Rockwell hardness, σ b is the ultimate tensile strength; The formulas for ultimate tensile strength and Brinell and Vickers hardness are as follows; HB=(61.8 / d) 2 -10; s b =(32 / d) 3 +235; σ b =235+0.14(HB+10) 1.5 ; Where HB is the Brinell hardness, d is the indentation diameter of the Brinell hardness test, σ b is the ultimate tensile strength; When the quenching process is water quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows: Where D is the critical diameter of quenching, E s The distance between the body points in the half marathon. When the quenching process is oil quenching, the formula between the critical quenching diameter and the half-Martensite point distance is as follows: Where D is the critical diameter of quenching, E s The distance between the body points in the half marathon. When the quenching process is air quenching, the formula between the quenching critical diameter and the half-Marne body point distance is as follows: Where D is the critical diameter of quenching, E s It is the body point distance for half marathon.

6. The method for reverse composition design of medium and high carbon steel based on conditional generative network according to claim 5, characterized in that: The method further comprises: Calculate the half marathon body point distance based on impact toughness, including: The formula for low temperature tempering toughness and martensite strength is as follows: a km =(15356 / s bm ) 2 ; Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength; The formulas for quenched and tempered toughness and quenched and tempered strength are as follows: Among them, a km is the quenched and tempered toughness, σ bm is the martensite strength, σ bq is the quenching strength before quenching and tempering, σ bt It is the quenching strength after quenching and tempering; Convert impact toughness to ultimate strength and calculate half marathon point distance.

7. A medium-high carbon steel reverse composition design device based on conditional generation network, characterized in that: The device comprises: Hardenability curve conversion module, used to convert steel design indicators into hardenability curves; A hardenability curve input module, used for inputting a hardenability curve and a preset random latent code into a trained and tested generator; The reverse composition generation module is used for the generator to reversely generate the steel composition and grain size based on the hardenability curve; The smelting composition fine-tuning module is used to adjust the generated grain size and composition based on the composition adjustment module to meet the smelting needs.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images