Method and device for high-speed wire mechanical property prediction and model training

Abstract

The application discloses a high-speed wire mechanical property prediction method and device and a model training method and device. The prediction method comprises the following steps: obtaining steelmaking chemical composition, production process parameters and control parameters of air coolers of a wind cooling line of a high-speed wire to be predicted; determining a cooling phase change temperature interval according to the steelmaking chemical composition and screening a matched target air cooler group; extracting the control parameters as phase change region cooling characteristics; calculating the steelmaking chemical composition by using an element interaction operator preset based on a metallurgical strengthening mechanism to generate derived strengthening characteristics; inputting a prediction characteristic set comprising the phase change region cooling characteristics, the derived strengthening characteristics, preset basic element characteristics in the steelmaking chemical composition and the production process parameters into a performance prediction model trained based on historical data to output a mechanical property prediction value. The method avoids temperature field numerical simulation, eliminates interference of air cooler parameters in a non-phase change region and reduces model fitting difficulty, and takes into account prediction efficiency and accuracy.

Description

Technical Field

[0001] This application relates to the field of high-speed wire rod quality control technology, and in particular to a method and apparatus for predicting the mechanical properties of high-speed wire rods and training their models. Background Technology

[0002] In the field of predicting the mechanical properties of high-speed wire rods, existing technologies mainly fall into two categories: one relies on traditional manual tensile testing, which is cumbersome, inefficient, and unable to meet the demands of large-scale, fast-paced testing; the other relies on heat transfer equations or numerical simulations of heat flux density to calculate the wire rod temperature field and then predict performance. This method is computationally cumbersome, sensitive to boundary conditions, and prone to error accumulation. Intermediate calculation steps are easily affected by minor process fluctuations, leading to prediction biases. Furthermore, due to the diverse chemical composition and manufacturing processes of high-speed wire rods, directly inputting all production process data into the model can easily introduce noise signals from non-critical cooling stages, reducing prediction accuracy. Therefore, existing technologies struggle to balance the prediction efficiency and accuracy of high-speed wire rod mechanical properties. Summary of the Invention

[0003] The purpose of this application is to provide a method and apparatus for predicting the mechanical properties of high-speed wire rods and training their models, which takes into account both prediction efficiency and prediction accuracy.

[0004] To achieve the above objectives, one embodiment of this application provides a method for predicting the mechanical properties of high-speed wire rods, comprising:

[0005] Obtain the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted; The cooling phase change temperature range of the high-speed wire rod to be predicted is determined based on the steelmaking chemical composition. Based on the correspondence between the cooling phase change temperature range and the fan position on the air-cooled line, a target fan group that matches the cooling phase change temperature range is selected, and the control parameters of the target fan group are extracted as the cooling characteristics of the phase change zone. The steelmaking chemical composition is calculated using an element interaction operator based on a pre-defined metallurgical strengthening mechanism to generate derived strengthening features; The predicted feature set, which includes the cooling characteristics of the phase transformation zone, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, is input into the performance prediction model, and the predicted mechanical properties of the high-speed wire rod to be predicted are output; wherein, the performance prediction model is trained based on historical data.

[0006] As a further improvement to this application, the step of determining the cooling phase transformation temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition includes: Calculate the phase transformation start temperature and phase transformation end temperature of the high-speed wire rod to be predicted based on the steelmaking chemical composition. Based on the phase transformation start temperature and the phase transformation end temperature, the pearlite phase transformation range or ferrite precipitation range of the high-speed wire to be predicted is determined, and the determined range is used as the cooling phase transformation temperature range.

[0007] As a further improvement of this application, the phase transformation start temperature is the equilibrium heating phase transformation start temperature Ac1, and the phase transformation end temperature is the bainitic phase transformation start temperature Bs. The equilibrium heating phase transformation start temperature and the bainitic phase transformation start temperature Bs are calculated based on the mass percentage of each element in the steelmaking chemical composition using a preset multivariate fitting function.

[0008] As a further improvement to this application, the calculation formula for the multivariate fitting function includes: Ac1 = 224.4 + 992.4C - 465.1C 2 +7.6Mn+16.7Si-6.9Ni+3.7C×Ni-2.7Cr×Ni+0.8Ni 2 +46.7Cr + 19C × Cr - 6.1Cr 2 +10Mo-6.8Cr×Mo; Bs=656-57.7C-35Mn-20Si-15.3Ni-34Cr-41.2Mo; In the formula, C, Si, Mn, Cr, Ni, and Mo represent the mass percentage content of the corresponding chemical elements in the steelmaking chemical composition.

[0009] As a further improvement to this application, the step of selecting target fan units that match the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, and extracting the control parameters of the target fan units as cooling characteristics of the phase change zone, includes: Obtain the wire temperature range corresponding to each fan position on the air-cooled line; The cooling phase change temperature range is matched with the wire temperature range corresponding to each fan to determine a number of target fans that match the cooling phase change temperature range, so as to form the target fan group. The identification information of each target fan in the target fan group is combined with its corresponding gear parameter or air volume parameter to generate the phase change zone cooling characteristics.

[0010] As a further improvement to this application, the step of calculating the steelmaking chemical composition using an element interaction operator preset based on a metallurgical strengthening mechanism to generate derived strengthening features includes: The solid solution strengthening characteristic S1 is calculated using the solid solution strengthening operator. The calculation formula is: S1=∑(yi×Ci); The precipitation strengthening feature S2 is calculated using the precipitation strengthening operator. The calculation formula is: S2 = k1×V + k2×Mo + k3×Nb + k4×Ti. The solid solution strengthening feature S1 and the precipitation strengthening feature S2 are combined to form the derived strengthening feature; Where Ci represents the mass percentage of the i-th solid solution element, yi represents the weighting coefficient of the corresponding solid solution element for solid solution strengthening; V, Mo, Nb, and Ti represent the mass percentage of the microalloying elements vanadium, molybdenum, niobium, and titanium, respectively, and k1 to k4 represent the correction coefficients of the corresponding microalloying elements for precipitation strengthening.

[0011] As a further improvement of this application, the solid solution element Ci is selected from several of carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium, and the weighting coefficients yi corresponding to the solid solution elements carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium are 4570, 3750, 37, 83, 470, -30, 2.9, 80, 80, and 11, respectively; the correction coefficients k1 to k4 range from 0.8 to 1.5.

[0012] As a further improvement to this application, the historical data is mechanical property test data collected from preset sampling positions of wire rod; The preset sampling location is a fixed area on the air-cooled roller conveyor where the wire rod is in the middle cooling zone and is symmetrically positioned. Specifically, it is a continuous sampling area starting from the bottom directly below the wire rod, measured from the clock position of the wire rod cross-section, in the direction from 6 o'clock to 9 o'clock or from 6 o'clock to 3 o'clock. The output predicted mechanical properties correspond to the mechanical properties of the high-speed wire to be predicted at the preset sampling location.

[0013] As a further improvement to this application, the performance prediction model is a gradient boosting decision tree model; The hyperparameters of the gradient boosting decision tree model are fixed hyperparameters that are pre-determined through optimization using a genetic algorithm.

[0014] As a further improvement to this application, the production process parameters include: The temperature of the billet corresponding to the high-speed wire rod to be predicted in the preheating section, the first heating section, the second heating section, and the soaking section in the heating furnace, as well as the furnace time of the preheating section, the first heating section, the second heating section, and the soaking section; The roughing temperature, finishing temperature, and wire drawing temperature of the high-speed wire to be predicted; Production environment temperature.

[0015] To achieve one of the above objectives, one embodiment of this application provides a training method for a high-speed wire mechanical property prediction model, comprising: Acquire historical data, including historical steelmaking chemical composition, historical production process parameters, historical control parameters of each fan in the air-cooled line, and measured values ​​of the mechanical properties of historical high-speed wire rods at preset sampling locations. The cooling phase transformation temperature range of the historical high-speed wire rod is determined based on the historical steelmaking chemical composition. Based on the correspondence between the cooling phase transformation temperature range and the fan position on the air-cooled line, a target fan group that matches the cooling phase transformation temperature range is selected, and the historical control parameters of the target fan group are extracted as the cooling characteristics of the phase transformation zone. The historical steelmaking chemical composition is calculated using an element interaction operator based on a pre-defined metallurgical strengthening mechanism to generate derived strengthening features; The training feature set, which includes the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the historical steelmaking chemical composition, and the historical production process parameters, along with the corresponding measured values ​​of the mechanical properties, are input into the prediction model to be trained. A hyperparameter optimization algorithm is used to determine the hyperparameters of the prediction model to be trained by iteratively optimizing the model with the goal of minimizing the error between the output prediction value of the prediction model to be trained and the measured mechanical properties of the high-speed wire.

[0016] As a further improvement to this application, the step of selecting target fan units that match the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, and extracting the historical control parameters of the target fan units as cooling characteristics of the phase change zone, includes: Obtain the wire temperature range corresponding to each fan position on the air-cooled line; The cooling phase change temperature range is matched with the wire temperature range corresponding to each fan to determine a number of target fans that match the cooling phase change temperature range, so as to form the target fan group. The identification information of each target fan in the target fan group is combined with its corresponding gear parameter or air volume parameter to generate the phase change zone cooling characteristics.

[0017] As a further improvement to this application, the step of calculating the historical steelmaking chemical composition using an element interaction operator based on a pre-defined metallurgical strengthening mechanism to generate derived strengthening features includes: The solid solution strengthening characteristic S1 is calculated using the solid solution strengthening operator. The calculation formula is: S1=∑(yi×Ci); The precipitation strengthening feature S2 is calculated using the precipitation strengthening operator. The calculation formula is: S2 = k1×V + k2×Mo + k3×Nb + k4×Ti. The solid solution strengthening feature S1 and the precipitation strengthening feature S2 are combined to form the derived strengthening feature; Where Ci represents the mass percentage of the i-th solid solution element, yi represents the weighting coefficient of the corresponding solid solution element for solid solution strengthening; V, Mo, Nb, and Ti represent the mass percentage of the microalloying elements vanadium, molybdenum, niobium, and titanium, respectively, and k1 to k4 represent the correction coefficients of the corresponding microalloying elements for precipitation strengthening.

[0018] As a further improvement of this application, the prediction model to be trained is a gradient boosting decision tree model, the hyperparameter optimization algorithm is a genetic algorithm, and the error is mean squared error; The search space for the genetic algorithm to optimize the hyperparameters of the gradient boosting decision tree model includes: 150 to 400 decision trees, 5 to 10 maximum tree depth, 0.01 to 0.2 learning rate, 0.2 to 1 subsampling ratio, 0.4 to 1 feature sampling ratio, 50 to 150 leaf nodes, 5 to 100 minimum leaf samples, 0.00001 to 3 L1 regularization coefficient, 0 to 0.5 minimum node splitting gain, and 0 to 6 minimum sum of child node weights. The evolution control parameters of the genetic algorithm are: population size 20 to 40, number of generations 50 to 70, crossover probability 0.5 to 0.7, and mutation probability 0.01 to 0.2.

[0019] As a further improvement of this application, the fitness function of the genetic algorithm in the hyperparameter iterative optimization process is constructed based on the prediction mean square error of the prediction model to be trained; The genetic algorithm uses minimizing the prediction mean square error as the evaluation criterion to perform the selection and evolution of individuals in the population.

[0020] As a further improvement to this application, after obtaining the high-speed wire mechanical property prediction model, the following steps are also included: verifying and iteratively training the high-speed wire mechanical property prediction model: Obtain historical high-speed wire verification samples and input the verification samples into the high-speed wire mechanical property prediction model, and calculate the relative error between the output mechanical property prediction value and the corresponding measured mechanical property value; The number of samples whose relative error is less than or equal to a preset error threshold is counted, and the ratio of the number of samples to the total number of validation samples is determined as the sample hit rate. Determine whether the sample hit rate reaches a preset hit rate threshold; If the sample hit rate does not reach the preset hit rate threshold, the hyperparameters of the prediction model to be trained are adjusted, and the model is retrained. If the sample hit rate reaches the preset hit rate threshold, the high-speed wire mechanical performance prediction model is determined as the final trained model.

[0021] To achieve one of the above objectives, one embodiment of this application provides a high-speed wire mechanical property prediction device, comprising: The data acquisition module is used to acquire the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted. The phase change zone cooling feature extraction module is used to determine the cooling phase change temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition, and to select the target fan group that matches the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan position on the air-cooled line, and to extract the control parameters of the target fan group as the phase change zone cooling feature. The derivative strengthening feature generation module is used to calculate the steelmaking chemical composition using element interaction operators preset based on the metallurgical strengthening mechanism, and generate derivative strengthening features. The mechanical property prediction module is used to input a prediction feature set, including the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, into the performance prediction model, and output the predicted mechanical property value of the high-speed wire rod to be predicted; wherein, the performance prediction model is trained based on historical data.

[0022] To achieve one of the above objectives, one embodiment of this application provides an electronic device, including: Storage module, used to store computer programs; The processing module, when executing the computer program, can implement the steps of the above-mentioned method for predicting the mechanical properties of high-speed wire rods, or the training method for the above-mentioned method for predicting the mechanical properties of high-speed wire rods.

[0023] Compared with commonly used technologies, this application has the following advantages: This method for predicting the mechanical properties of high-speed wire rods determines the cooling phase transformation temperature range through steelmaking chemical composition, matches and screens target fan units, and extracts cooling characteristics of the phase transformation zone. This avoids computationally complex temperature field numerical simulations and eliminates interference from fan parameters in the non-phase transformation zone. Simultaneously, it utilizes element interaction operators based on metallurgical strengthening mechanisms to generate derived strengthening features, directly transforming complex element strengthening effects into explicit feature values, reducing the difficulty of model fitting. The phase transformation zone cooling features, derived strengthening features, basic element features, and production process parameters are combined to form a prediction feature set input to the model, directly outputting predicted mechanical properties. This method eliminates the need for manual tensile testing, reduces prediction deviations caused by process fluctuations, and balances the efficiency and accuracy of high-speed wire rod mechanical property prediction. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for predicting the mechanical properties of high-speed wires according to an embodiment of this application.

[0025] Figure 2 This is a schematic diagram comparing the predicted and measured tensile strength values ​​of steel grade 1 and steel grade 2 using a high-speed wire rod mechanical property prediction method based on an embodiment of this application.

[0026] Figure 3 This is a flowchart of a training method for a high-speed wire mechanical property prediction model according to an embodiment of this application.

[0027] Figure 4 This is a schematic diagram of a module of a high-speed wire mechanical property prediction device according to an embodiment of this application. Detailed Implementation

[0028] The present application will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of this application.

[0029] One embodiment of this application provides a method and apparatus for predicting the mechanical properties of high-speed wire rods and training a model that balances prediction efficiency and accuracy, applicable to the rolling and cooling production of high-speed wire rods. Predicting the mechanical properties of wire rods during high-speed wire rod production is a crucial step in ensuring product quality.

[0030] In existing technologies, traditional performance testing relies on manual tensile testing, which is cumbersome, inefficient, and unsuitable for large-scale, high-speed testing. Another approach relies on heat transfer equations or numerical simulations of heat flux density to calculate the wire temperature field and predict performance. However, this method is computationally intensive, sensitive to boundary conditions, and prone to error accumulation. Intermediate calculation steps are easily affected by minor process fluctuations, leading to prediction bias. Furthermore, due to the diverse chemical composition and manufacturing processes of high-speed wires, directly inputting all production process data into the model can introduce noise signals from non-critical cooling stages, reducing prediction accuracy. Therefore, existing technologies struggle to balance prediction efficiency and accuracy for the mechanical properties of high-speed wires.

[0031] To address the aforementioned issues, this specific embodiment provides a method for predicting the mechanical properties of high-speed wires.

[0032] Example 1 The following is combined with Figures 1-2This application describes a method for predicting the mechanical properties of high-speed wires according to an embodiment. Although this application provides method operation steps as shown in the following embodiments or flowcharts, the execution order of these steps is not limited to the execution order provided in the embodiments of this application, based on conventional or non-creative labor, where there is no necessary causal relationship in the logical process.

[0033] This embodiment provides a method for predicting the mechanical properties of high-speed wires, including: Step S10: Obtain the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted.

[0034] Step S20: Determine the cooling phase transformation temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition, and select the target fan group that matches the cooling phase transformation temperature range based on the correspondence between the cooling phase transformation temperature range and the fan position on the air-cooled line, and extract the control parameters of the target fan group as the cooling characteristics of the phase transformation zone.

[0035] Step S30: Calculate the steelmaking chemical composition using an element interaction operator preset based on the metallurgical strengthening mechanism to generate derived strengthening features.

[0036] Step S40: Input the predicted feature set, which includes the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, into the performance prediction model, and output the predicted mechanical properties of the high-speed wire rod to be predicted; wherein, the performance prediction model is trained based on historical data.

[0037] In step S10, the steelmaking chemical composition refers to the mass percentage content of each chemical element measured during the steelmaking or continuous casting process. The production process parameters refer to the temperature and time parameters during the heating, rolling, and wire drawing processes. The control parameters for each fan in the air-cooling line refer to the operating speed, air volume, or air velocity of each fan arranged sequentially along the air-cooling line. These data can be retrieved through the production line's data acquisition system.

[0038] In step S20, the phase transformation process of the high-speed wire rod during cooling plays a decisive role in its final mechanical properties, and the temperature range in which the phase transformation occurs is determined by its steelmaking chemical composition. The cooling phase transformation temperature range for the wire rod is calculated using the steelmaking chemical composition. This temperature range is then matched with the wire rod temperature range covered by each fan on the air-cooled line, accurately identifying the fan corresponding to the phase transformation and designating it as the target fan group. The control parameters of the target fan group are extracted as cooling characteristics of the phase transformation zone, eliminating or reducing interference from other fan parameters outside the phase transformation zone, avoiding the influence of irrelevant data noise on the prediction, and improving the correlation between cooling characteristics and the phase transformation process.

[0039] In step S30, the element interaction operator refers to a pre-constructed calculation formula based on metallurgical principles, used to reflect the comprehensive effects between different chemical elements. The derived strengthening feature refers to the specific numerical value calculated by the element interaction operator, used to represent the strengthening effect under the combined action of multiple elements. By pre-calculating the chemical composition using the element interaction operator, the complex interactions between elements are directly transformed into explicit feature values, avoiding the model directly fitting complex element interactions, reducing the model's fitting difficulty, and enabling the features to more clearly reflect the influence of chemical composition on performance.

[0040] In step S40, the preset basic element characteristics refer to the content of key elements in the steelmaking chemical composition that play a dominant role in the basic properties of steel, such as the content of elements that have a significant impact on phase transformation, such as C, Si, and Mn. The performance prediction model is a model trained using historical production data and corresponding measured performance values, used to establish the correspondence between input features and mechanical properties. By combining the phase transformation zone cooling characteristics, derived strengthening characteristics, preset basic element characteristics, and production process parameters to form a prediction feature set and inputting it into the performance prediction model, the predicted mechanical properties of the wire rod to be predicted can be directly calculated and output.

[0041] This method for predicting the mechanical properties of high-speed wire rods avoids the computationally complex numerical simulation of temperature fields, eliminates the need for manual tensile testing, removes the interference of fan parameters in non-critical cooling stages, reduces the difficulty of model fitting, and minimizes prediction errors caused by process fluctuations, thus balancing the efficiency and accuracy of predicting the mechanical properties of high-speed wire rods.

[0042] In one embodiment, the production process parameters include the preheating section temperature, first heating section temperature, second heating section temperature, and soaking section temperature of the billet corresponding to the high-speed wire rod to be predicted in the heating furnace, as well as the furnace time of the preheating section, first heating section, second heating section, and soaking section, and also include the roughing temperature, finishing temperature, wire drawing temperature, and production environment temperature of the high-speed wire rod to be predicted.

[0043] The heating process of steel billets in the heating furnace is divided into four stages: preheating, primary heating, secondary heating, and soaking. Recording the heating temperature and the time the billet spends in the furnace at each stage can fully reflect the heating thermal history of the billet. The heating temperature and furnace time determine the uniformity of the internal temperature distribution of the billet, as well as the initial microstructure and the dissolution state of compounds, which have a fundamental impact on the deformation resistance and microstructure evolution during wire rod rolling.

[0044] Roughing temperature, finishing temperature, and wire drawing temperature reflect the thermal state of wire rod in the initial stages of deformation and cooling. Roughing temperature and finishing temperature determine the deformation and recrystallization state of austenite during rolling. Wire drawing temperature is the initial temperature at which wire rod enters the air-cooling line and begins forced cooling, directly determining the initial thermal boundary of the subsequent phase transformation process.

[0045] The production environment temperature refers to the actual temperature of the workshop or environment where the air-cooled line is located. The level of the ambient temperature will affect the natural heat dissipation rate of the wire on the air-cooled roller conveyor and the initial temperature of the cooling medium of the fan, thus affecting the actual cooling speed of the wire on the air-cooled line.

[0046] By inputting the temperatures of the four heating furnaces, the corresponding furnace time, roughing temperature, finishing temperature, wire drawing temperature, and ambient temperature into the performance prediction model as production process parameters, the model comprehensively covers the entire thermal and deformation process from billet heating and plastic deformation to the initial state of air cooling. This completes the process variables that affect microstructure evolution and improves the completeness of the input data and the accuracy of the prediction results.

[0047] The microstructure transformation of high-speed wire rod on the air-cooled roller conveyor depends on the actual temperature range during the cooling process of the wire rod. The temperature boundaries for phase transformation of steel with different chemical compositions are different.

[0048] In one embodiment, step S20, which determines the cooling phase transformation temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition, includes: Step S21: Calculate the phase transformation start temperature and phase transformation end temperature of the high-speed wire rod to be predicted based on the steelmaking chemical composition.

[0049] Step S22: Based on the phase transformation start temperature and the phase transformation end temperature, determine the pearlite phase transformation range or ferrite precipitation range of the high-speed wire to be predicted, and use the determined range as the cooling phase transformation temperature range.

[0050] In step S21, the phase transformation start temperature and phase transformation end temperature of the wire rod are first calculated based on the steelmaking chemical composition, thereby clarifying the upper and lower temperature boundaries of the wire rod when it undergoes phase transformation.

[0051] In step S22, the temperature range between the phase transformation start temperature and the phase transformation end temperature is determined as the pearlite phase transformation range or ferrite precipitation range of the wire, and is used as the cooling phase transformation temperature range for subsequent matching fans. By accurately defining the core temperature range where the wire undergoes phase transformation, a temperature benchmark is provided for the subsequent selection of fans corresponding to the phase transformation zone.

[0052] In one embodiment, the phase transformation start temperature is the equilibrium heating phase transformation start temperature Ac1, and the phase transformation end temperature is the bainitic phase transformation start temperature Bs; the equilibrium heating phase transformation start temperature and the bainitic phase transformation start temperature Bs are calculated based on the mass percentage of each element in the steelmaking chemical composition using a preset multivariate fitting function.

[0053] The equilibrium heating phase transformation onset temperature Ac1 reflects the critical temperature reference point for the occurrence of phase transformation in steel, while the bainitic phase transformation onset temperature Bs represents the lower temperature limit at which the pearlite or ferrite transformation terminates and transforms into bainite. Using the equilibrium heating phase transformation onset temperature Ac1 as the phase transformation onset temperature and the bainitic phase transformation onset temperature Bs as the phase transformation end temperature can accurately cover the main temperature range in which pearlite phase transformation or ferrite precipitation occurs in high-speed wire rods on air-cooled lines.

[0054] A multivariate fitting function is a pre-established calculation formula that comprehensively considers the mass percentage content of multiple chemical elements. By substituting the mass percentage content of each element in the steelmaking chemical composition into the multivariate fitting function, the equilibrium heating phase transformation initiation temperature Ac1 and the bainitic phase transformation initiation temperature Bs corresponding to that chemical composition can be directly calculated. Through multivariate fitting function calculations, there is no need for tedious phase transformation experiments on each batch of wire rod. The phase transformation temperature range can be quickly calculated based on the chemical composition, improving the efficiency and accuracy of phase transformation range determination.

[0055] In one embodiment, the formula for calculating the multivariate fitting function includes: Ac1 = 224.4 + 992.4C - 465.1C 2 +7.6Mn+16.7Si-6.9Ni+3.7C×Ni-2.7Cr×Ni+0.8Ni 2 +46.7Cr + 19C × Cr - 6.1Cr 2 +10Mo-6.8Cr×Mo.

[0056] Bs=656-57.7C-35Mn-20Si-15.3Ni-34Cr-41.2Mo.

[0057] In the formula, C, Si, Mn, Cr, Ni, and Mo represent the mass percentage content of the corresponding chemical elements in the steelmaking chemical composition.

[0058] Specifically, the formula for calculating the equilibrium heating phase transformation initiation temperature Ac1 comprehensively considers the individual effects of elements such as carbon, silicon, manganese, chromium, nickel, and molybdenum on the austenite formation temperature. It introduces quadratic terms for carbon, chromium, and nickel, as well as cross-product terms between carbon and nickel, chromium and nickel, carbon and chromium, and chromium and molybdenum. Among these, the main elements such as carbon, chromium, and nickel have a nonlinear influence on the phase transformation temperature. The inclusion of quadratic terms and multi-element cross-product terms accurately expresses the interaction and nonlinear synergistic law of different alloying elements on the critical temperature of austenite transformation when they coexist.

[0059] The formula for calculating the bainitic transformation initiation temperature Bs reflects the controlling effects of elements such as carbon, silicon, manganese, chromium, nickel, and molybdenum on the stability of supercooled austenite and the lower limit temperature of bainitic transformation. The content of each chemical element is adjusted for the Bs value through corresponding coefficients, quantitatively reflecting the cumulative influence of the main alloying elements on the phase transformation termination limit.

[0060] The phase transformation temperature ranges differ significantly among different steel grades; for example, the pearlite phase transformation window is completely different between low-carbon steel and medium-carbon alloy steel. For most steel grades, the pearlite phase transformation mainly occurs within the Ar1 to Bs temperature range, while the ferrite precipitation region occurs near Ar3 to Ar1. In practical calculations and applications, Ac1 calculated using a multivariate fitting formula can approximately represent the Ar1 temperature under equilibrium conditions. After calculating Ac1 and Bs, and combining the CCT curve characteristics of the wire rod with on-site commissioning data, the cooling phase transformation temperature range that most needs to be controlled on the air-cooled line for this steel grade can be accurately determined; for example, Ac1 to Bs can be defined as the pearlite phase transformation range.

[0061] By employing the aforementioned multivariate fitting formula, the equilibrium heating phase transformation initiation temperature Ac1 and the bainitic phase transformation initiation temperature Bs are directly calculated using the mass percentage of each element in the steelmaking chemical composition. This quantifies the influence of major chemical elements and their interactions on the critical phase transformation temperature. This method avoids the bias caused by using fixed empirical constants, adapts to the significant differences in phase transformation ranges among different steel grades, and improves the accuracy of phase transformation temperature range determination. It provides a reliable temperature basis for accurately defining the phase transformation region corresponding to the high-speed wire rod to be predicted and for selecting target wind turbine units.

[0062] In one embodiment, step S20, based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, filters out target fan units that match the cooling phase change temperature range, and extracts the control parameters of the target fan units as cooling characteristics of the phase change zone, including: Step S23: Obtain the wire temperature range corresponding to each fan position on the air-cooled line.

[0063] Step S24: Match the cooling phase change temperature range with the wire temperature range corresponding to each fan, and determine a number of target fans that match the cooling phase change temperature range to form the target fan group.

[0064] Step S25: Combine the identification information of each target fan in the target fan group with its corresponding gear parameters or air volume parameters to generate the phase change zone cooling characteristics.

[0065] As the wire is transported on the air-cooled roller conveyor, it continuously dissipates heat and cools down. Each fan, arranged sequentially along the transport direction, corresponds to a specific temperature range as the wire passes through that location. In step S23, the temperature range of the wire corresponding to each fan's installation location is obtained, thus establishing the mapping relationship between the fan's spatial location and the wire's real-time temperature.

[0066] In step S24, if the temperature range of the wire covered by one or more fans falls into or overlaps with the cooling phase change temperature range, it indicates that the wire is undergoing a phase change when passing through the fan. The fan or the fans are then identified as the target fans, and all the fans that meet the conditions together constitute the target fan group.

[0067] For example, the phase transformation temperature ranges of some steel grades and their corresponding target wind turbine units are shown in Table 1 below:

[0068] Table 1 In step S25, the identification information (such as fan number or position number) of each fan in the target fan group is extracted and combined sequentially with the control data such as the gear parameters, air volume parameters or wind speed parameters during operation to construct a feature vector representing the cooling state of the phase change zone.

[0069] This method achieves precise spatial binding between fan control parameters and wire phase change process. For other fans that are weakly correlated with the cooling phase change temperature range, they can be discarded or have their weight reduced, thereby eliminating or reducing the interference of fan parameters in non-phase change regions on the model. This allows the extracted cooling features to directly and centrally reflect the thermal history of the key phase change stage, improving the fit between features and phase change process.

[0070] In addition, step S25 can also calculate simple derived features such as average air volume and air volume difference for the selected target wind turbine units to enhance feature expression.

[0071] In one embodiment, step S30 utilizes an element interaction operator preset based on a metallurgical strengthening mechanism to calculate the steelmaking chemical composition and generate derived strengthening features, including: Step S31: Calculate the solid solution strengthening characteristic S1 using the solid solution strengthening operator. The calculation formula is: S1=∑(yi×Ci).

[0072] Step S32: Calculate the precipitation strengthening feature S2 using the precipitation strengthening operator. The calculation formula is: S2=k1×V+k2×Mo+k3×Nb+k4×Ti.

[0073] Step S33: Combine the solid solution strengthening feature S1 with the precipitation strengthening feature S2 to form the derived strengthening feature.

[0074] Where Ci represents the mass percentage of the i-th solid solution element, yi represents the weighting coefficient of the corresponding solid solution element for solid solution strengthening; V, Mo, Nb, and Ti represent the mass percentage of the microalloying elements vanadium, molybdenum, niobium, and titanium, respectively, and k1 to k4 represent the correction coefficients of the corresponding microalloying elements for precipitation strengthening.

[0075] Solid solution strengthening refers to the phenomenon where alloying elements dissolve into the matrix solid solution, causing lattice distortion and thus increasing the strength of steel. In step S31, the solid solution strengthening operator multiplies the mass percentage content Ci of various solid solution elements with their corresponding weighting coefficients yi and then sums them to obtain the characteristic value S1 that quantitatively characterizes the solid solution strengthening effect.

[0076] Precipitation strengthening refers to the phenomenon where microalloying elements precipitate second-phase particles such as carbonitrides during cooling, hindering dislocation movement and thus increasing strength. In step S32, the precipitation strengthening operator multiplies the mass percentage content of microalloying elements vanadium, molybdenum, niobium, and titanium by their corresponding correction coefficients k1 to k4 and then sums them to obtain the characteristic value S2 that quantitatively characterizes the precipitation strengthening effect.

[0077] In step S33, the calculated solid solution strengthening feature S1 and precipitation strengthening feature S2 are combined as derived strengthening features.

[0078] Steps S31-S33 quantify the two main metallurgical strengthening mechanisms of solid solution strengthening and precipitation strengthening into specific features through solid solution strengthening operator and precipitation strengthening operator, transforming the complex nonlinear strengthening law among multiple elements into directly usable feature values, reducing the burden of fitting complex physical metallurgical laws on the performance prediction model.

[0079] In one embodiment, the solid solution element Ci is selected from several of carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium, and the weighting coefficients yi corresponding to the solid solution elements carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium are 4570, 3750, 37, 83, 470, -30, 2.9, 80, 80, and 11, respectively; the correction coefficients k1 to k4 range from 0.8 to 1.5.

[0080] Carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium are common strengthening elements in steel. Each element has different solid solution capabilities in the matrix and causes varying degrees of lattice distortion. The set of weighting coefficients yi (e.g., 4570 for carbon, 3750 for silicon, 37 for manganese, etc.) is predetermined based on metallurgical principles and empirical data, used to quantitatively reflect the contribution of each element's unit mass percentage to solid solution strengthening. During the actual cooling process of microalloying elements vanadium, molybdenum, niobium, and titanium, the size, quantity, and sufficiency of their precipitated phases are affected by the actual cooling rate and temperature changes. Correction coefficients k1 to k4, ranging from 0.8 to 1.5, are used to correct the precipitation strengthening effect of each microalloying element according to the actual cooling conditions of the specific production line.

[0081] High-speed wire rods are conveyed and cooled on air-cooled roller conveyors in an overlapping, loose-wound configuration. The heat dissipation conditions differ between the overlapping edge areas and the sparser areas within the coils, resulting in inconsistent cooling rates at different locations within the same coil. In other words, there are temperature differences within the same coil during cooling, meaning performance variations may exist at different locations within the same coil, potentially causing predictions to diverge and fail to converge. This performance variation within the same coil presents a significant challenge to wire rod performance prediction.

[0082] To eliminate data fluctuations caused by inconsistent sampling locations, in this embodiment, the historical data is mechanical performance test data collected from preset sampling locations on the wire rod.

[0083] Specifically, the preset sampling position is a fixed area symmetrically positioned in the middle cooling zone of the wire rod on the air-cooled roller conveyor. Specifically, it is a continuous sampling area starting from the bottom directly below the wire rod, measured in clockwise direction from 6 o'clock to 9 o'clock or from 6 o'clock to 3 o'clock on the wire rod's cross-section. This position is less affected by the overlap of the two side edges and has relatively stable heat dissipation conditions, enabling stable characterization of the typical mechanical properties of the wire rod.

[0084] The output predicted mechanical properties correspond to the mechanical properties of the high-speed wire to be predicted at the preset sampling location.

[0085] In this way, historical data is limited to detection data collected from the preset sampling location, and the predicted value output by the model directly corresponds to the mechanical performance of the preset sampling location. This ensures the consistency of input features, training labels, and final prediction targets at the sampling points in physical space, avoids data noise introduced by sampling location deviations, and improves the training quality and prediction correspondence of the mechanical performance prediction model.

[0086] In one embodiment, the performance prediction model is a gradient boosting decision tree model; the model hyperparameters of the gradient boosting decision tree model are fixed hyperparameters that are pre-determined through optimization using a genetic algorithm.

[0087] Gradient boosting decision tree models, as an ensemble learning model, possess strong nonlinear fitting capabilities and anti-overfitting abilities, effectively handling complex nonlinear relationships in tabular data such as chemical composition and process parameters. Gradient boosting decision tree models can be implemented using lightweight gradient boosting machine algorithms, such as LightGBM. The performance of gradient boosting decision tree models is significantly influenced by their internal hyperparameters (e.g., the number of trees, maximum tree depth, learning rate, etc.). Before the model is used for prediction, historical data is used, and a genetic algorithm is employed to perform global iterative optimization within a defined hyperparameter search space to find the optimal set of hyperparameters that minimizes the model's prediction error, which is then used as the model's fixed parameters.

[0088] By adopting a gradient boosting decision tree model as the performance prediction model, and using fixed hyperparameters pre-determined by using a genetic algorithm, the model can maintain stable and high-precision output during the prediction and inference stage. This fully leverages the fitting advantage of gradient boosting decision trees for nonlinear relationships in metallurgy and processes, while avoiding model prediction bias caused by untuned parameters, thus ensuring the reliability of mechanical performance prediction results.

[0089] To verify the prediction accuracy and reliability of the high-speed wire mechanical property prediction method provided in this application, a test sample set was constructed using historical production data for verification. The prediction results are shown in Table 2 below. The comparison scheme uses traditional prediction methods; the scheme in this application refers to the scheme that extracts the cooling characteristics of the phase transition region, generates derived strengthening characteristics, and then constructs a feature set input model.

[0090]

[0091] Table 2 As can be seen from the test data in Table 2, in different test sample groups, regardless of whether the preset error threshold is set to a relative error of less than or equal to 3% or 4%, the sample hit rate of the proposed solution is significantly higher than that of the comparative solution without feature optimization. For example, in test sample group 1, when the preset error threshold is 3%, the sample hit rate of the proposed solution reaches 94.8%, which is 6.1 percentage points higher than that of the comparative solution; when the preset error threshold is 4%, the sample hit rate increases to 98.2%. This result shows that by determining the cooling phase transformation temperature range based on the steelmaking chemical composition and screening the target fan group, the noise interference of the fan parameters in the non-phase transformation region on the model can be effectively eliminated. At the same time, by using the element interaction operator based on the metallurgical strengthening mechanism to generate derived strengthening features, the fitting difficulty of the model is reduced, and the prediction accuracy and stability of the model under different production conditions are significantly improved, which can better meet the needs of high-precision control of the mechanical properties of high-speed wire rods.

[0092] Additionally, refer to Figure 2 The figure further illustrates the continuous sample comparison curves of the predicted and measured tensile strength values ​​of steel grade 1 and steel grade 2 provided by the prediction model of this application. The horizontal axis represents the continuously collected sample number, the vertical axis represents the tensile strength of the wire, the red line represents the measured mechanical property value measured at a fixed preset sampling location, and the blue line represents the predicted mechanical property value output by the model of this application.

[0093] from Figure 2 It can be seen that, regardless of whether it is steel grade 1 with a lower strength base or steel grade 2 with a higher strength base, the tensile strength prediction value output by the model in this application can closely match the fluctuation trend of the actual measured value. Even at sample nodes where slight fluctuations in the production process cause fluctuations in the actual tensile strength of the wire, the prediction curve can still reflect the changing trend of the actual value well, without obvious lag or large deviation. This proves that by screening the cooling characteristics of the phase change region and constructing derived strengthening characteristics, the model in this application not only eliminates the noise interference of the fan parameters in the non-phase change region, but also effectively reflects the inherent metallurgical strengthening and cooling control mechanism of different steel grades, possesses good dynamic anti-interference ability and multi-steel grade generalization performance, and improves the high accuracy and high reliability of the prediction.

[0094] Compared with commonly used technologies, this embodiment has the following advantages: This method for predicting the mechanical properties of high-speed wire rods determines the cooling phase transformation temperature range through steelmaking chemical composition and matches and screens target fan units to extract cooling characteristics in the phase transformation zone. This avoids computationally complex temperature field numerical simulations and eliminates interference from fan parameters in the non-phase transformation zone. Furthermore, it utilizes element interaction operators based on metallurgical strengthening mechanisms to generate derived strengthening features, directly transforming complex element strengthening effects into explicit feature values, thus reducing the difficulty of model fitting. The phase transformation zone cooling features, derived strengthening features, basic element features, and production process parameters are combined to form a prediction feature set, which is then input into the model to directly output predicted mechanical properties. This method eliminates the need for manual tensile testing, reduces prediction bias caused by process fluctuations, and balances efficiency and accuracy in predicting the mechanical properties of high-speed wire rods.

[0095] Example 2 One embodiment of this application provides a training method for a prediction model of the mechanical properties of high-speed wires, such as... Figure 3 As shown, it includes: Step S51: Obtain historical data, wherein the historical data includes historical steelmaking chemical composition, historical production process parameters, historical control parameters of each fan in the air-cooled line, and historical measured values ​​of mechanical properties of high-speed wire rod at preset sampling locations.

[0096] Step S52: Determine the cooling phase transformation temperature range of the historical high-speed wire rod based on the historical steelmaking chemical composition, and select target fan groups that match the cooling phase transformation temperature range based on the correspondence between the cooling phase transformation temperature range and the fan positions on the air-cooled line, and extract the historical control parameters of the target fan groups as cooling characteristics of the phase transformation zone.

[0097] Step S53: Calculate the historical steelmaking chemical composition using the element interaction operator preset based on the metallurgical strengthening mechanism to generate derived strengthening features.

[0098] Step S54: Input the training feature set, which includes the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the historical steelmaking chemical composition, and the historical production process parameters, along with the corresponding measured values ​​of the mechanical properties, into the prediction model to be trained.

[0099] Step S55: Using a hyperparameter optimization algorithm, with the goal of minimizing the error between the output prediction value of the prediction model to be trained and the measured mechanical properties, the hyperparameters of the prediction model to be trained are determined through iterative optimization, and the prediction model of the mechanical properties of high-speed wire is trained.

[0100] In step S51, the historical steelmaking chemical composition refers to the mass percentage content of each chemical element measured in previous production batches; the historical production process parameters refer to the temperatures and times of each section of the heating furnace, roughing rolling temperature, finishing rolling temperature, wire drawing temperature, and ambient temperature during previous production processes. The historical control parameters of each fan in the air-cooling line refer to the actual control parameters such as the operating speed, air volume, or air velocity of each fan during previous production.

[0101] The measured mechanical properties refer to the actual mechanical property data obtained from tensile tests and other examinations conducted on previously produced wires at fixed preset sampling locations. Using the measured values ​​at these fixed sampling locations as training labels ensures the consistency of the training data across the sampling locations and eliminates data noise introduced by random variations in sampling locations. The fixed preset sampling location can be, as described above, a continuous sampling area starting directly below the bottom of the wire rod, and extending from the clock position of the wire rod's cross-section towards the 6 o'clock to 9 o'clock direction or from the 6 o'clock to 3 o'clock direction.

[0102] In step S52, the historical steelmaking chemical composition of each historical data point is used to calculate the corresponding historical wire rod cooling phase transformation temperature range. This cooling phase transformation temperature range is then matched with the wire rod temperature range covered by each fan during production. The fans corresponding to the historical wire rod phase transformation are selected to form the target fan group. Control parameters such as the fan speed or airflow of the target fan group during production are extracted and combined to generate the cooling characteristics of the phase transformation zone. This process eliminates interference from fan control parameters in non-phase transformation areas of the historical data, ensuring that the historical cooling characteristics reflect the key stages that determine the microstructure evolution.

[0103] In step S53, the pre-constructed solid solution strengthening operator and precipitation strengthening operator are used to calculate the historical steelmaking chemical composition, obtaining quantitative characterizations of the derived strengthening features of solid solution strengthening and precipitation strengthening. By calculating the nonlinear strengthening effects between multiple elements in the historical chemical composition as explicit values ​​in advance, the difficulty of blindly fitting complex metallurgical laws during model training is reduced, which helps to improve the model's convergence speed and fitting effect.

[0104] In step S54, the preset basic element features refer to the content of key elements in the historical steelmaking chemical composition that play a major role in the basic properties of steel. The phase transformation zone cooling features, derived strengthening features, preset basic element features, and historical production process parameters extracted from each historical sample are combined to form a training feature set, which is then correlated and aligned with the measured mechanical properties corresponding to the sample, and input together into the prediction model to be trained, providing the model with high-quality training sample pairs with clear mapping relationships.

[0105] In step S55, the prediction model to be trained needs to determine a set of optimal hyperparameters during the training process. The hyperparameter optimization algorithm iteratively searches within the set hyperparameter search space. In each iteration, the model uses the current hyperparameters to predict the training samples, calculates the error between the output predicted value and the actual measured mechanical properties, and adjusts the hyperparameter direction according to the error. When the error reaches its minimum or a preset condition is met, the optimal hyperparameters are determined, thereby training a high-speed wire mechanical property prediction model with high prediction accuracy and stability.

[0106] This training method eliminates interference noise in historical data by performing targeted preprocessing and screening of cooling characteristics and chemical composition, providing a high-quality dataset that conforms to physical laws for model training. Combined with hyperparameter optimization algorithms to determine the optimal parameters, it ensures that the high-speed wire mechanical property prediction model obtained in the final training has accurate and stable prediction capabilities in practical applications.

[0107] It should be noted that for details not disclosed in the training method of the high-speed wire mechanical property prediction model in the embodiments of this application, please refer to the details disclosed in the high-speed wire mechanical property prediction method in the embodiments of this application.

[0108] In one embodiment, the prediction model to be trained is a gradient boosting decision tree model, the hyperparameter optimization algorithm is a genetic algorithm, and the error is mean squared error. The gradient boosting decision tree model has strong nonlinear fitting and generalization ability for industrial tabular data. Using mean squared error as an indicator to evaluate the deviation between predicted and measured values ​​can quantitatively drive the model to converge towards the direction of minimum global deviation.

[0109] The search space for hyperparameter optimization of the gradient boosting decision tree model by the genetic algorithm includes: 150 to 400 decision trees, maximum tree depth 5 to 10, learning rate 0.01 to 0.2, subsampling ratio 0.2 to 1, feature sampling ratio 0.4 to 1, number of leaf nodes 50 to 150, minimum number of leaf samples 5 to 100, L1 regularization coefficient 0.00001 to 3, minimum node splitting gain 0 to 0.5, and minimum sum of child node weights 0 to 6. The evolutionary control parameters of the genetic algorithm are: population size 20 to 40, number of generations 50 to 70, crossover probability 0.5 to 0.7, and mutation probability 0.01 to 0.2.

[0110] The predictive performance of gradient boosting decision tree models is highly dependent on the configuration of their structural hyperparameters and regularization parameters. Setting the number of decision trees to 150 to 400, the maximum tree depth to 5 to 10, the number of leaf nodes to 50 to 150, and the minimum number of leaf samples to 5 to 100 limits the range of structural complexity for individual trees and the overall tree forest. This ensures the model's ability to express complex multivariate effects while limiting overfitting caused by an excessively large tree structure. Setting the learning rate to 0.01 to 0.2, the subsampling ratio to 0.2 to 1, and the feature sampling ratio to 0.4 to 1 introduces reasonable randomness when building weak learners in each iteration, enhancing the model's robustness to data fluctuations. Setting the L1 regularization coefficient to 0.00001 to 3, the minimum gain for node splitting to 0 to 0.5, and the minimum sum of child node weights to 0 to 6 further eliminates redundant branches caused by non-critical features by constraining node splitting and weight updates.

[0111] Regarding the evolution control parameters of the genetic algorithm, setting the population size to 20 to 40 can control the computational load per generation while ensuring the diversity of the optimized individuals; setting the number of generations to 50 to 70 can ensure that the algorithm completes sufficient global iterations and tends to converge; setting the crossover probability to 0.5 to 0.7 can promote the recombination and combination of superior chromosome genes; and setting the mutation probability to 0.01 to 0.2 can prevent the algorithm from getting trapped in local optima too early.

[0112] By using a genetic algorithm with the aforementioned evolutionary control parameters to optimize the hyperparameters of the gradient boosting decision tree model within the defined search space, a fast search for model parameter configurations was achieved with reasonable computational resource consumption, resulting in the acquisition of the optimal model hyperparameters adapted to this specific dataset.

[0113] In one embodiment, the fitness function and its evaluation criteria in the hyperparameter iterative optimization process of the genetic algorithm are further defined. The fitness function of the genetic algorithm in the hyperparameter iterative optimization process is constructed based on the prediction mean square error of the prediction model to be trained; the genetic algorithm uses minimizing the prediction mean square error as the evaluation criterion to perform the selection and evolution of individuals in the population.

[0114] Specifically, the fitness function is a quantitative standard used by genetic algorithms to evaluate the quality of various hyperparameter combinations in the population. Mean squared error (MSE) quantitatively reflects the overall deviation between the model's predicted output and the actual measured value. In this embodiment, the MSE of the prediction model on the data samples is used as the basis for constructing the fitness function. The smaller the MSE of the model, the higher the fitting accuracy of the model under that set of hyperparameters, and the higher the fitness score of the corresponding hyperparameter individual.

[0115] During the evolutionary process, the genetic algorithm uses the fitness function to calculate the fitness score of each hyperparameter individual in the population. Following the principle of prioritizing the retention of individuals with higher fitness, it performs selection, crossover, and mutation operations to drive the hyperparameter population to iteratively evolve towards a direction with smaller prediction mean square error. By using the minimization of prediction mean square error as the evaluation criterion, the optimization orientation of the genetic algorithm in the hyperparameter optimization process is clarified, ensuring that the optimized hyperparameter combination can improve the accuracy of the prediction model in global numerical fitting.

[0116] In one embodiment, after obtaining the high-speed wire mechanical property prediction model, the method further includes the following steps: verifying and iteratively training the high-speed wire mechanical property prediction model. Step S61: Obtain verification samples of historical high-speed wires and input the verification samples into the high-speed wire mechanical property prediction model, and calculate the relative error between the output mechanical property prediction value and the corresponding measured mechanical property value.

[0117] Step S62: Count the number of samples whose relative error is less than or equal to a preset error threshold, and determine the ratio of the number of samples to the total number of verification samples as the sample hit rate.

[0118] Step S63: Determine whether the sample hit rate reaches the preset hit rate threshold.

[0119] Step S64: If the sample hit rate does not reach the preset hit rate threshold, adjust the hyperparameters of the prediction model to be trained and return to retrain the prediction model to be trained.

[0120] Step S65: If the sample hit rate reaches the preset hit rate threshold, then the high-speed wire mechanical property prediction model is determined as the finally trained model.

[0121] Specifically, validation samples refer to historical production data with known measured mechanical properties, independent of the model training process.

[0122] In step S61, the verification samples are input into the pre-trained prediction model, and the relative error between the predicted mechanical properties and the corresponding measured mechanical properties is calculated. The formula for calculating the relative error is: E=|Y pred -Y act | / Y act ×100% In the formula, E represents the relative error, and Y... pred Y represents the predicted mechanical properties. act This represents the corresponding measured mechanical property value. By calculating the absolute value of the difference between the predicted value and the measured value and then dividing it by the measured value, the influence of the absolute value of different mechanical property indicators can be eliminated, thus accurately reflecting the degree of deviation of the model's prediction on unseen samples.

[0123] In step S62, the preset error threshold is a deviation limit set according to the accuracy standards required for actual production quality control. The relative error E calculated for each verification sample is compared with the preset error threshold. The number of qualified samples whose relative error E is less than or equal to the preset error threshold is counted, and the ratio of this number to the total number of verification samples is calculated, which is determined as the sample hit rate. The formula for calculating the sample hit rate is: Ratio = m / n × 100% In the formula, Ratio represents the sample hit rate, m represents the number of validation samples whose relative error E is less than or equal to the preset error threshold, and n represents the total number of validation samples. The sample hit rate Ratio quantitatively reflects the actual prediction success rate of the model under the acceptable accuracy requirements in industrial settings.

[0124] In steps S63 to S65, the preset hit rate threshold is the quality standard for determining whether the model can be deployed or put into use online. It is determined whether the sample hit rate Ratio reaches the preset hit rate threshold. In step S64, if the calculated sample hit rate Ratio does not reach the preset hit rate threshold, it indicates that the model's current prediction accuracy does not meet the requirements of on-site control. At this time, the hyperparameters of the prediction model to be trained need to be readjusted, and retraining is performed to form an iterative closed loop of dynamic optimization. In step S65, when the sample hit rate Ratio reaches or exceeds the preset hit rate threshold, it indicates that the model exhibits the required prediction accuracy and stability performance on independent validation data. At this time, it is determined as the finally trained model and used for subsequent actual predictions.

[0125] By introducing a validation and iterative training process based on sample hit rate after model training, and clarifying the specific quantitative calculation formulas for relative error and sample hit rate, a model quality control process guided by the actual field pass rate index was established. This process ensures that the finally trained model not only fits well on the training samples, but also stably reaches the preset prediction accuracy standard on independent validation data. This avoids the use of unqualified models due to overfitting or prediction bias, and improves the reliability of the mechanical performance prediction model in practical field applications.

[0126] Example 3 In Example 3, a high-speed wire mechanical property prediction device is provided, such as... Figure 4 As shown. The high-speed wire mechanical property prediction device includes the following modules, and the specific functions of each module are as follows: The data acquisition module is used to acquire the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted.

[0127] The phase change zone cooling feature extraction module is used to determine the cooling phase change temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition, and to select target fan groups that match the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, and to extract the control parameters of the target fan groups as phase change zone cooling features.

[0128] The derivative strengthening feature generation module is used to calculate the steelmaking chemical composition using element interaction operators preset based on the metallurgical strengthening mechanism, and generate derivative strengthening features.

[0129] The mechanical property prediction module is used to input a prediction feature set, including the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, into the performance prediction model, and output the predicted mechanical property value of the high-speed wire rod to be predicted; wherein, the performance prediction model is trained based on historical data.

[0130] It should be noted that for details not disclosed in the high-speed wire mechanical property prediction device of this application embodiment, please refer to the details disclosed in the high-speed wire mechanical property prediction method of this application embodiment.

[0131] In addition, this application also proposes an electronic device, which includes a storage module and a processing module. When the processing module executes the computer program, it can implement the steps in the above-mentioned high-speed wire mechanical property prediction method or the above-mentioned high-speed wire mechanical property prediction model training method. That is, it can implement the steps in any one of the above-mentioned high-speed wire mechanical property prediction method or the above-mentioned high-speed wire mechanical property prediction model training method.

[0132] The electronic device may be integrated into a high-speed wire mechanical property prediction device, a local terminal device, or part of a cloud server.

[0133] The processing module can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processing module is the control center of the high-speed wire mechanical property prediction device, connecting all parts of the device via various interfaces and lines.

[0134] The storage module can be used to store the computer programs and / or modules. The processing module realizes various functions of the high-speed wire mechanical property prediction device by running or executing the computer programs and / or modules stored in the storage module and calling the data stored in the storage module. The storage module may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the storage module may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0135] For example, the computer program may be divided into one or more modules / units, which are stored in a storage module and executed by a processing module to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in a high-speed wire mechanical property prediction device.

[0136] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0137] The detailed descriptions listed above are merely specific descriptions of feasible implementation methods of this application, and are not intended to limit the scope of protection of this application. All equivalent implementation methods or modifications made without departing from the specific spirit of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the mechanical properties of high-speed wire rods, characterized in that, include: Obtain the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted; The cooling phase change temperature range of the high-speed wire rod to be predicted is determined based on the steelmaking chemical composition. Based on the correspondence between the cooling phase change temperature range and the fan position on the air-cooled line, a target fan group that matches the cooling phase change temperature range is selected, and the control parameters of the target fan group are extracted as the cooling characteristics of the phase change zone. The steelmaking chemical composition is calculated using an element interaction operator based on a pre-defined metallurgical strengthening mechanism to generate derived strengthening features; The predicted feature set, which includes the cooling characteristics of the phase transformation zone, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, is input into the performance prediction model, and the predicted mechanical properties of the high-speed wire rod to be predicted are output; wherein, the performance prediction model is trained based on historical data.

2. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The step of determining the cooling phase transformation temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition includes: Calculate the phase transformation start temperature and phase transformation end temperature of the high-speed wire rod to be predicted based on the steelmaking chemical composition. Based on the phase transformation start temperature and the phase transformation end temperature, the pearlite phase transformation range or ferrite precipitation range of the high-speed wire to be predicted is determined, and the determined range is used as the cooling phase transformation temperature range.

3. The method for predicting the mechanical properties of high-speed wire rods according to claim 2, characterized in that, The phase transformation start temperature is the equilibrium heating phase transformation start temperature Ac1, and the phase transformation end temperature is the bainitic phase transformation start temperature Bs. The equilibrium heating phase transformation start temperature and the bainitic phase transformation start temperature Bs are calculated based on the mass percentage of each element in the steelmaking chemical composition using a preset multivariate fitting function.

4. The method for predicting the mechanical properties of high-speed wire rods according to claim 3, characterized in that, The calculation formula for the multivariate fitting function includes: Ac1=224.4+992.4C-465.1C 2 +7.6Mn+16.7Si-6.9Ni+3.7C×Ni-2.7Cr×Ni+0.8Ni 2 +46.7Cr+19C×Cr-6.1Cr 2 +10Mo-6.8Cr×Mo; Bs=656-57.7C-35Mn-20Si-15.3Ni-34Cr-41.2Mo; In the formula, C, Si, Mn, Cr, Ni, and Mo represent the mass percentage content of the corresponding chemical elements in the steelmaking chemical composition.

5. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The step of selecting target fan units that match the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, and extracting the control parameters of the target fan units as cooling characteristics of the phase change zone, includes: Obtain the wire temperature range corresponding to each fan position on the air-cooled line; The cooling phase change temperature range is matched with the wire temperature range corresponding to each fan to determine a number of target fans that match the cooling phase change temperature range, so as to form the target fan group. The identification information of each target fan in the target fan group is combined with its corresponding gear parameter or air volume parameter to generate the phase change zone cooling characteristics.

6. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The calculation of the steelmaking chemical composition using a pre-defined element interaction operator based on a metallurgical strengthening mechanism to generate derived strengthening features includes: The solid solution strengthening characteristic S1 is calculated using the solid solution strengthening operator. The calculation formula is: S1=∑(yi×Ci); The precipitation strengthening feature S2 is calculated using the precipitation strengthening operator. The calculation formula is: S2 = k1×V + k2×Mo + k3×Nb + k4×Ti. The solid solution strengthening feature S1 and the precipitation strengthening feature S2 are combined to form the derived strengthening feature; Where Ci represents the mass percentage of the i-th solid solution element, yi represents the weighting coefficient of the corresponding solid solution element for solid solution strengthening; V, Mo, Nb, and Ti represent the mass percentage of the microalloying elements vanadium, molybdenum, niobium, and titanium, respectively, and k1 to k4 represent the correction coefficients of the corresponding microalloying elements for precipitation strengthening.

7. The method for predicting the mechanical properties of high-speed wire rods according to claim 6, characterized in that, The solid solution element Ci is selected from several of carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium. The weighting coefficients yi corresponding to the solid solution elements carbon, silicon, manganese, chromium, nickel, copper, molybdenum, vanadium, titanium, and niobium are 4570, 3750, 37, 83, 470, -30, 2.9, 80, 80, and 11, respectively. The correction coefficients k1 to k4 range from 0.8 to 1.

5.

8. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The historical data is mechanical property test data collected from preset sampling locations of the wire rod; The preset sampling location is a fixed area on the air-cooled roller conveyor where the wire rod is in the middle cooling zone and is symmetrically positioned. Specifically, it is a continuous sampling area starting from the bottom directly below the wire rod, measured from the clock position of the wire rod cross-section, in the direction from 6 o'clock to 9 o'clock or from 6 o'clock to 3 o'clock. The output predicted mechanical properties correspond to the mechanical properties of the high-speed wire to be predicted at the preset sampling location.

9. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The performance prediction model is a gradient boosting decision tree model; The hyperparameters of the gradient boosting decision tree model are fixed hyperparameters that are pre-determined through optimization using a genetic algorithm.

10. The method for predicting the mechanical properties of high-speed wire rods according to claim 1, characterized in that, The production process parameters include: The temperature of the billet corresponding to the high-speed wire rod to be predicted in the preheating section, the first heating section, the second heating section, and the soaking section in the heating furnace, as well as the furnace time of the preheating section, the first heating section, the second heating section, and the soaking section; The roughing temperature, finishing temperature, and wire drawing temperature of the high-speed wire to be predicted; Production environment temperature.

11. A training method for a high-speed wire rod mechanical property prediction model, characterized in that, include: Acquire historical data, including historical steelmaking chemical composition, historical production process parameters, historical control parameters of each fan in the air-cooled line, and measured values ​​of the mechanical properties of historical high-speed wire rods at preset sampling locations. The cooling phase transformation temperature range of the historical high-speed wire rod is determined based on the historical steelmaking chemical composition. Based on the correspondence between the cooling phase transformation temperature range and the fan position on the air-cooled line, a target fan group that matches the cooling phase transformation temperature range is selected, and the historical control parameters of the target fan group are extracted as the cooling characteristics of the phase transformation zone. The historical steelmaking chemical composition is calculated using an element interaction operator based on a pre-defined metallurgical strengthening mechanism to generate derived strengthening features; The training feature set, which includes the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the historical steelmaking chemical composition, and the historical production process parameters, along with the corresponding measured values ​​of the mechanical properties, are input into the prediction model to be trained. A hyperparameter optimization algorithm is used to determine the hyperparameters of the prediction model to be trained by iteratively optimizing the model with the goal of minimizing the error between the output prediction value of the prediction model to be trained and the measured mechanical properties of the high-speed wire.

12. The training method for the high-speed wire mechanical property prediction model according to claim 11, characterized in that, The step of selecting target fan units that match the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan positions on the air-cooled line, and extracting the historical control parameters of the target fan units as cooling characteristics of the phase change zone, includes: Obtain the wire temperature range corresponding to each fan position on the air-cooled line; The cooling phase change temperature range is matched with the wire temperature range corresponding to each fan to determine a number of target fans that match the cooling phase change temperature range, so as to form the target fan group. The identification information of each target fan in the target fan group is combined with its corresponding gear parameter or air volume parameter to generate the phase change zone cooling characteristics.

13. The training method for the high-speed wire mechanical property prediction model according to claim 11, characterized in that, The calculation of the historical steelmaking chemical composition using a pre-defined element interaction operator based on a metallurgical strengthening mechanism to generate derived strengthening features includes: The solid solution strengthening characteristic S1 is calculated using the solid solution strengthening operator. The calculation formula is: S1=∑(yi×Ci); The precipitation strengthening feature S2 is calculated using the precipitation strengthening operator. The calculation formula is: S2 = k1×V + k2×Mo + k3×Nb + k4×Ti. The solid solution strengthening feature S1 and the precipitation strengthening feature S2 are combined to form the derived strengthening feature; Where Ci represents the mass percentage of the i-th solid solution element, yi represents the weighting coefficient of the corresponding solid solution element for solid solution strengthening; V, Mo, Nb, and Ti represent the mass percentage of the microalloying elements vanadium, molybdenum, niobium, and titanium, respectively, and k1 to k4 represent the correction coefficients of the corresponding microalloying elements for precipitation strengthening.

14. The training method for the high-speed wire mechanical property prediction model according to claim 11, characterized in that, The prediction model to be trained is a gradient boosting decision tree model, the hyperparameter optimization algorithm is a genetic algorithm, and the error is the mean squared error. The search space for the genetic algorithm to optimize the hyperparameters of the gradient boosting decision tree model includes: 150 to 400 decision trees, 5 to 10 maximum tree depth, 0.01 to 0.2 learning rate, 0.2 to 1 subsampling ratio, 0.4 to 1 feature sampling ratio, 50 to 150 leaf nodes, 5 to 100 minimum leaf samples, 0.00001 to 3 L1 regularization coefficient, 0 to 0.5 minimum node splitting gain, and 0 to 6 minimum sum of child node weights. The evolution control parameters of the genetic algorithm are: population size 20 to 40, number of generations 50 to 70, crossover probability 0.5 to 0.7, and mutation probability 0.01 to 0.

2.

15. The training method for the high-speed wire mechanical property prediction model according to claim 14, characterized in that, The fitness function of the genetic algorithm in the hyperparameter iterative optimization process is constructed based on the prediction mean square error of the prediction model to be trained. The genetic algorithm uses minimizing the prediction mean square error as the evaluation criterion to perform the selection and evolution of individuals in the population.

16. The training method for the high-speed wire mechanical property prediction model according to claim 11, characterized in that, After obtaining the high-speed wire mechanical property prediction model, the following steps are also included to verify and iteratively train the high-speed wire mechanical property prediction model: Obtain historical high-speed wire verification samples and input the verification samples into the high-speed wire mechanical property prediction model, and calculate the relative error between the output mechanical property prediction value and the corresponding measured mechanical property value; The number of samples whose relative error is less than or equal to a preset error threshold is counted, and the ratio of the number of samples to the total number of validation samples is determined as the sample hit rate. Determine whether the sample hit rate reaches a preset hit rate threshold; If the sample hit rate does not reach the preset hit rate threshold, the hyperparameters of the prediction model to be trained are adjusted, and the model is retrained. If the sample hit rate reaches the preset hit rate threshold, the high-speed wire mechanical performance prediction model is determined as the final trained model.

17. A device for predicting the mechanical properties of high-speed wire rods, characterized in that, include: The data acquisition module is used to acquire the steelmaking chemical composition, production process parameters, and control parameters of each fan in the air-cooled line for the high-speed wire rod to be predicted. The phase change zone cooling feature extraction module is used to determine the cooling phase change temperature range of the high-speed wire rod to be predicted based on the steelmaking chemical composition, and to select the target fan group that matches the cooling phase change temperature range based on the correspondence between the cooling phase change temperature range and the fan position on the air-cooled line, and to extract the control parameters of the target fan group as the phase change zone cooling feature. The derivative strengthening feature generation module is used to calculate the steelmaking chemical composition using element interaction operators preset based on the metallurgical strengthening mechanism, and generate derivative strengthening features. The mechanical property prediction module is used to input a prediction feature set, including the phase transformation zone cooling characteristics, the derived strengthening characteristics, the preset basic element characteristics in the steelmaking chemical composition, and the production process parameters, into the performance prediction model, and output the predicted mechanical property value of the high-speed wire rod to be predicted; wherein, the performance prediction model is trained based on historical data.

18. An electronic device, characterized in that, include: Storage module, used to store computer programs; The processing module, when executing the computer program, can implement the steps of the high-speed wire mechanical property prediction method according to any one of claims 1 to 10, or the training method of the high-speed wire mechanical property prediction model according to any one of claims 11 to 16.

Images