55SiCrA spring steel performance optimization method based on ISSA-BP neural network

Abstract

The invention provides a 55SiCrA spring steel performance optimization method based on an ISSA-BP neural network, and belongs to the field of metal component optimization and processing. The method comprises the steps that historical production data and performance data are collected and preprocessed to construct a training data set; a 55SiCrA spring steel performance optimization model is constructed based on an ISSA-BP neural network model, the optimization model is trained by adopting a training data set to obtain a mature model, and an optimal solution under a performance optimization target is selected by utilizing an Optuna optimization method to obtain optimal alloy components and rolling process parameters of the 55SiCrA spring steel; according to the optimal alloy components, a steel billet is obtained through converter-refining-continuous casting; and according to the optimal rolling process parameters, rolling to obtain a wire rod, then carrying out cold drawing and heat treatment, cutting out and processing a test sample from the 55SiCrA spring steel subjected to heat treatment according to corresponding standards, and carrying out mechanical property test on the test sample. According to the optimized 55SiCrA spring steel, the tensile strength is larger than or equal to 2000 MPa, and the percentage reduction of area is larger than or equal to 40%.

Description

Technical Field:

[0001] The present invention belongs to the field of optimization and processing of metal components, and particularly relates to a method for optimizing the properties of 55SiCrA spring steel based on an ISSA-BP neural network. Background Art:

[0002] 55SiCrA spring steel is usually applied to railway vehicles, automobiles, etc., and is used to manufacture flat springs that bear large loads or helical springs with a wire diameter below 30 mm, such as automobile suspension springs, which play a role in buffering and shock absorption. With the popularization of automobiles, energy conservation, environmental protection, and safety have attracted extensive attention from people. Reducing the total weight of automobiles through lightweight design of automobiles has become an effective way to reduce energy consumption and exhaust emissions. In order to ensure both the lightweight and high safety of automobiles, it is required that automobile components have more excellent mechanical properties. A large number of studies and practices have shown that by increasing the design stress by 10 MPa, the spring weight can be reduced by about 10%. Therefore, the spring needs to have a higher design stress specification.

[0003] In the prior art, as a typical low-alloy medium-carbon high-silicon steel, 55SiCrA spring steel generally improves its mechanical properties through measures such as alloy composition design, microstructure refinement, effective heat treatment, and inclusion control. However, due to the extremely complex alloy composition and rolling process characteristics of spring steel, which involve the mutual influence of the alloy composition, rolling process, and properties of the material, and the optimization of the alloy composition of spring steel and the design process of rolling process parameters are relatively cumbersome and require a large number of experimental verifications, the R & D cost is high and the R & D cycle is long. Summary of the Invention:

[0004] In order to solve the above problems, the present invention provides a method for optimizing the properties of 55SiCrA spring steel based on an ISSA-BP neural network, which uses the ISSA-BP neural network to optimize the composition, rolling process, and properties of 55SiCrA spring steel, obtains the best alloy composition and rolling process parameters of the 55SiCrA spring steel, and reduces the R & D cost and shortens the R & D cycle on the premise of improving the properties of the 55SiCrA spring steel.

[0005] In order to achieve the above object, the technical solutions adopted in the embodiments of the present invention are as follows:

[0006] A method for optimizing the properties of 55SiCrA spring steel based on an ISSA-BP neural network, the method comprising the following steps:

[0007] Step S1, collecting the historical production data and performance data of 55SiCrA spring steel produced by a steel plant within a predetermined time period, preprocessing the original data, and constructing a training data set;

[0008] Step S2: Based on the ISSA-BP neural network model, construct a performance optimization model for 55SiCrA spring steel. Use the training data set to train the performance optimization model of 55SiCrA spring steel to obtain a mature performance optimization model of 55SiCrA spring steel;

[0009] Step S3: Using the Optuna optimization method, based on the mature performance optimization model of 55SiCrA spring steel, select the optimal solutions of alloying elements and rolling process parameters under the performance optimization goal to obtain the best alloying elements and rolling process parameters of this 55SiCrA spring steel;

[0010] Step S4: According to the obtained best alloying elements, use converter steelmaking - LF furnace refining - VD furnace refining - bloom continuous casting to obtain continuous casting billets;

[0011] Step S5: According to the obtained best rolling process parameters, after heating the obtained continuous casting billets in a heating furnace, then roll them in a continuous rolling mill to obtain hot-rolled wire rods. After the hot-rolled wire rods are spun, they are cooled;

[0012] Step S6: Cold draw and heat-treat the cooled wire rods. Cut and process test specimens from the heat-treated 55SiCrA spring steel according to the corresponding standards, and conduct mechanical property tests on the test specimens.

[0013] As a preferred embodiment of the present invention, the historical production data includes composition data and process data; wherein, the composition data includes the mass percentages of carbon, silicon, chromium, manganese and trace elements phosphorus and sulfur; the process data includes the furnace temperature in the preheating section, the heating time in the preheating section, the furnace temperature in the heating section, the heating time in the heating section, the furnace temperature in the soaking section, the heating time in the soaking section, the temperature entering the rolling mill, the temperature at the entrance of the sizing and reducing mill, the temperature entering the furnace, the time in the furnace, the temperature leaving the furnace, the spinning temperature and the fan frequency; the performance data includes the tensile strength and the reduction of area.

[0014] As a preferred embodiment of the present invention, the preprocessing includes: cleaning, screening and standardizing the collected data.

[0015] As a preferred embodiment of the present invention, step S2 of constructing the performance optimization model of 55SiCrA spring steel specifically includes:

[0016] Step S21: Based on the sparrow search algorithm SSA and the backpropagation BP neural network, construct an SSA-BP neural network model;

[0017] Step S22: Improve the SSA algorithm in the SSA-BP neural network model, and then optimize the hyperparameters of the BP neural network based on the improved sparrow search algorithm ISSA to obtain the ISSA-BP neural network model;

[0018] Step S23: Based on the ISSA - BP neural network model, construct a performance optimization model for 55SiCrA spring steel. The input layer of the model includes chemical compositions and process parameters. The hidden layer realizes non - linear mapping by adjusting the number of neurons and activation functions. The output layer predicts the tensile strength and reduction of area simultaneously.

[0019] Step S24: Train the performance optimization model for 55SiCrA spring steel based on the training data set to obtain a mature performance optimization model for 55SiCrA spring steel.

[0020] As a preferred embodiment of the present invention, when constructing the SSA - BP neural network model in step S21, define the structure, input layer, hidden layer, output layer, activation function and loss function of the BP neural network model; among them, the input layer consists of composition data and process data; the hidden layer adopts a two - layer hidden layer structure, and an appropriate number of neurons are set in each layer; the output layer contains two output nodes, representing the tensile strength and reduction of area respectively; the hidden layer uses the rule activation function, and the output layer uses a linear activation function to adapt to the regression task; the mean square error is used as the loss function.

[0021] As a preferred embodiment of the present invention, when improving the SSA algorithm in step S22, mix the sine - cosine algorithm and the Lévy flight mechanism in the SSA algorithm to obtain the improved sparrow search algorithm ISSA. The process includes:

[0022] Fuse the sine - cosine algorithm in the updation method of the discoverer's position, introduce a non - linear sine learning factor, calculate the discoverer's position according to the learning factor formula, and update the discoverer's position according to the early warning value and safety value; then use the Lévy flight strategy to update the follower's position.

[0023] As a preferred embodiment of the present invention, when optimizing the hyperparameters of the BP neural network, the optimized hyperparameters include: the number of neurons, learning rate and number of iterations.

[0024] As a preferred embodiment of the present invention, in step S3, when the performance optimization target is that the tensile strength of the spring steel ≥ 2000 MPa and the reduction of area ≥ 40%, the mass percentage of the best alloy composition content is: C: 0.51 - 0.59%, Si: 1.2 - 1.6%, Mn: 0.5 - 0.8%, Cr: 0.5 - 0.8%, P: ≤ 0.025%, S: ≤ 0.02%.

[0025] As a preferred embodiment of the present invention, in step S5, a walking - beam type reheating furnace is adopted in the hot - rolling stage, and a three - stage heating system is adopted, namely the pre - heating section, heating section, and soaking section. The rolling line equipment is a continuous rolling mill, and the post - rolling cooling equipment is a Stelmor line equipment.

[0026] As a preferred embodiment of the present invention, in step S6, induction heat treatment is used to heat-treat the test sample of the hot-rolled wire rod. The quenching and holding temperature is 850-900 °C, the cooling method after quenching is water cooling, and the tempering temperature is 420-480 °C.

[0027] The solution of the embodiment of the present invention has the following beneficial effects:

[0028] (1) By combining the optimized sparrow search algorithm and the BP neural network, an ISSA-BP neural network model is constructed, which not only realizes the organic coupling of different chemical compositions and properties, but also realizes the organic coupling of rolling process parameters and properties. While improving the mechanical properties of the material, it reduces the material cost of spring steel, optimizes the production process of spring steel, and saves time cost and economic cost to the greatest extent;

[0029] (2) By integrating the sine-cosine (SCA) algorithm idea into the SSA sparrow search algorithm and introducing a non-linear sine learning factor and Lévy flight strategy, the search ability and convergence speed of the traditional sparrow algorithm are improved, so that the optimized sparrow algorithm can find the global optimal solution faster when optimizing the weights of the BP neural network, avoiding the problem of falling into local optima, thereby improving the prediction accuracy and generalization ability of the ISSA-BP neural network model;

[0030] (3) Using the ISSA sparrow search algorithm to optimize the BP neural network model, the accuracy of the algorithm is controlled between 90%-95%.

[0031] (4) The 55SiCrA spring steel designed by the ISSA-BP model has achieved good mechanical properties, with a tensile strength ≥ 2000 MPa and an area reduction ≥ 40%.

[0032] Of course, it is not necessary for any product or method implementing the present invention to achieve all the above advantages simultaneously. Description of the Drawings:

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 It is the flowchart of the method for optimizing the properties of 55SiCrA spring steel based on the ISSA-BP neural network described in the embodiments of the present invention;

[0035] Figure 2It is the training flow chart of the ISSP-BP neural network obtained by improving the SSA algorithm in the embodiment of the present invention;

[0036] Figure 3 It is the comparison chart of the true value and the predicted value of the ISSA-BP neural network model in the embodiment of the present invention;

[0037] Figure 4 It is the scatter plot of the comparison between the true value and the predicted value of the ISSA-BP neural network model in the embodiment of the present invention. Specific implementation manner:

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Usually, the components of the embodiments of the present invention described and shown in the accompanying drawings here can be arranged and designed in various different configurations. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can also be combined with each other.

[0039] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. In the description of the present invention, the terms "first", "second", "third", "fourth", etc. are only used for distinguishing descriptions, and cannot be understood as indicating or implying relative importance.

[0040] Aiming at the problems of high R & D cost and long R & D cycle required for optimizing the properties of 55SiCrA spring steel by experimental means in the laboratory, the embodiment of the present invention provides a method and system for optimizing the properties of 55SiCrA spring steel based on the ISSA-BP neural network, applying the machine learning algorithm to the R & D process of 55SiCrA spring steel, using the ISSA-BP neural network to optimize the composition and rolling process of 55SiCrA spring steel and predict its mechanical properties, and obtaining the optimal alloy composition and rolling process parameters of the 55SiCrA spring steel; then based on the process parameters and alloy composition obtained by the algorithm, continuous casting billets are obtained by converter steelmaking - LF furnace refining - VD furnace refining - bloom continuous casting, the obtained continuous casting billets are homogenized and then rolled on a continuous rolling mill to obtain hot-rolled wire rods, and after cooling through the Stelmor line, cold drawing and heat treatment are carried out, and the mechanical properties of the tested steel after heat treatment are tested. The technical solution of the present invention formulates an optimized alloy composition and rolling process design plan based on the machine learning algorithm, reasonably reduces the content of added elements in the steel and optimizes the rolling process, further reduces the cost, shortens the R & D cycle, and successfully develops a high-performance spring steel with a tensile strength greater than 2000 MPa and a reduction of area greater than 40%.

[0041] Such asFigure 1 As shown in Figure 1 , the method for optimizing the properties of 55SiCrA spring steel based on the ISSA-BP neural network in this embodiment includes the following steps:

[0042] Step S1: Collect the historical production data and property data of 55SiCrA spring steel produced by the steel mill within a predetermined time period, preprocess the original data, and construct a training data set.

[0043] In this step, the historical production data includes composition data and process data. Among them, the composition data is the chemical composition content of 55SiCrA spring steel, including but not limited to the mass percentages of carbon (C), silicon (Si), chromium (Cr), manganese (Mn), and trace elements phosphorus (P) and sulfur (S); the process data are the key parameters of the steel mill during the rolling process such as rough rolling, medium rolling, finish rolling, and wire laying, including but not limited to: the furnace temperature in the preheating section, the heating time in the preheating section, the furnace temperature in the heating section, the heating time in the heating section, the furnace temperature in the soaking section, the heating time in the soaking section, the temperature entering the rolling mill, the temperature at the entrance of the reducing and sizing mill, the furnace inlet temperature, the time in the furnace, the furnace outlet temperature, the wire laying temperature, the frequencies of fan 1, fan 2, fan 3, fan 4, and fan 5.

[0044] The property data are the mechanical property parameters of 55SiCrA spring steel obtained through mechanical property tests, including the tensile strength and reduction of area, etc.

[0045] When collecting the above data, each production batch is uniquely identified to form a one-to-one corresponding closed-loop data set for the composition, process parameters, and performance test results.

[0046] The preprocessing includes: cleaning, screening, and standardizing the collected data to ensure the accuracy and consistency of the data. Cleaning includes removing outliers and missing values and eliminating data with obvious experimental errors or incomplete records. Screening selects variables that have a significant impact on the tensile strength and reduction of area, such as C, Si, and the furnace outlet temperature, etc. Standardizing performs normalization processing on the input variables, mapping all values to the [0, 1] interval to avoid the influence of too large differences in eigenvalue on model training. The cleaned data are randomly divided into a training set, a validation set, and a test set, and are allocated according to a ratio of 7:2:1 to ensure the scientific nature of model training and testing.

[0047] Step S2: Build a 55SiCrA spring steel property optimization model based on the ISSA-BP neural network model, and use the training data set to train the 55SiCrA spring steel property optimization model.

[0048] This step specifically includes:

[0049] Step S21: Construct an SSA-BP neural network model based on the Sparrow Search Algorithm (SSA) and the BackPropagation (BP) neural network.

[0050] In this step, when constructing the SSA-BP neural network model, first establish a BP neural network model and define its structure, input layer, hidden layer, output layer, activation function, and loss function. Among them, the input layer consists of composition data (such as C, Si, Cr, etc.) and process data (such as soaking furnace temperature in the soaking section, soaking time in the soaking section, tapping time, etc.); Hidden layer: Adopt a two-layer hidden layer structure, and set an appropriate number of neurons in each layer; The output layer contains two output nodes, representing the tensile strength and the reduction of area respectively. The rule activation function is used in the hidden layer, and the linear activation function is used in the output layer to adapt to the regression task. The mean square error (MSE) is used as the loss function to measure the error between the predicted value and the true value.

[0051] Step S22: Improve the SSA algorithm in the SSA-BP neural network model, and then optimize the hyperparameters of the BP neural network based on the Improved Sparrow Search Algorithm (ISSA) to obtain the ISSA-BP neural network model.

[0052] In this step, when improving the SSA algorithm, the sine-cosine algorithm and the Lévy flight mechanism are mixed in the traditional sparrow search algorithm to obtain the improved sparrow search algorithm ISSA. The process includes: integrating the sine-cosine algorithm in the updation method of the discoverer's position, introducing a non-linear sine learning factor, calculating the discoverer's position according to the learning factor formula, and updating the discoverer's position according to the warning value and the safety value; Then use the Lévy flight strategy to update the followers' positions. The improved sparrow search algorithm (ISSA) integrates the sine-cosine algorithm (SCA) and the Lévy flight mechanism, significantly enhancing the global search ability, balancing exploration and exploitation through the non-linear learning factor, and effectively escaping from the local optimum by combining the long-jump characteristics of Lévy flight, while improving the convergence speed and ensuring the stability of the algorithm.

[0053] Specifically, the discoverer's position is updated according to the warning value and the safety value using the following formula:

[0054]

[0055] ω = ω min +(ω max -ω min )·sin(tΠ / iter max )

[0056] In formula (1), ω is the learning factor, ω max is the maximum value of the learning factor, ω min is the minimum value of the learning factor; r1 is a random number within [0, 2π]; r2 is a random number within [0, 2]; t represents the current iteration number, and X i,j t represents the position information of the i-th sparrow at the j-th dimension when the iteration number is t, iter max represents the maximum iteration number, and X best represents the global optimal position, and R2 and ST represent the warning value and the safety value respectively.

[0057] Furthermore, the position of the followers is updated. To avoid the algorithm falling into local optimality, the Lévy flight strategy is introduced into the follower update formula to improve the global search ability; the improved formula is as follows:

[0058]

[0059] In formulas (2)-(4), X p t+1 represents the optimal discoverer position when the iteration number is t + 1, X worst t represents the worst sparrow position when the iteration number is t, n represents the number of joiners, Q represents a random number obeying the normal distribution, r3 and r4 are both random numbers within the range of [0, 1], the value of ε can be taken as 1.5, Γ(x) = (x - 1)!; Lévy(d) represents the flight mechanism.

[0060] In this step, when optimizing the hyperparameters of the BP neural network, the optimized hyperparameters include: the number of neurons, the learning rate, the iteration number, etc., to improve the global search ability and performance of the model. Specifically, as Figure 2 shown, first, the sparrow population is initialized, and the search range is defined, including the value ranges of hyperparameters such as the learning rate, the number of hidden layer neurons, and the training batch; the prediction error (MSE) of the BP neural network on the training set is used as the fitness function to evaluate the quality of each hyperparameter combination; the "discoverer" individuals perform global search to find the learning rate and neuron combination suitable for model training; the "followers" individuals perform local search near the global solution to further optimize the hyperparameters; the dynamic escape mechanism: when the search falls into local optimality, the population is redirected to explore a new solution space; according to the ISSA optimization results, the optimal learning rate, the number of hidden layer neurons, and the iteration number of the BP neural network are determined.

[0061] Step S23: Based on the ISSA-BP neural network model, construct a performance optimization model for 55SiCrA spring steel. The input layer of the model includes chemical compositions and process parameters. The hidden layer realizes non-linear mapping by adjusting the number of neurons and activation functions. The output layer predicts the tensile strength and reduction of area simultaneously.

[0062] Step S24: Train the 55SiCrA spring steel performance optimization model based on the training dataset to obtain a mature 55SiCrA spring steel performance optimization model.

[0063] In this step, when training the model, input the composition data and process data into the input layer of the network. The network outputs the tensile strength and reduction of area, which are compared with the experimental test values. When updating the gradient, based on the error backpropagation mechanism of the BP algorithm, adjust the weights and thresholds through the optimized learning rate. Monitor the change of the training loss value (MSE), and set an early stopping mechanism to avoid overfitting. As Figure 3 and Figure 4 shown, the predicted values output by the model highly overlap with the collected real values and show consistent performance in the test data outside the training set, and its generalization ability is improved. The optimized ISSA-BP neural network model has good accuracy and stability in optimizing the composition and process of 55SiCr spring steel and predicting its performance.

[0064] Step S3: Using the Optuna optimization method, select the optimal solutions of alloy compositions and rolling process parameters under the performance optimization objective based on the mature 55SiCrA spring steel performance optimization model to obtain the best alloy composition and rolling process parameters of the 55SiCrA spring steel.

[0065] In this step, in a specific application example, when the performance optimization objective is set as the tensile strength of the spring steel ≥ 2000 MPa and the reduction of area ≥ 40%, the mass percentages of the obtained best alloy composition are: C: 0.51 - 0.59%, Si: 1.2 - 1.6%, Mn: 0.5 - 0.8%, Cr: 0.5 - 0.8%, P: ≤ 0.025%, S: ≤ 0.02%.

[0066] Step S4: According to the obtained best alloy composition, use converter steelmaking - LF furnace refining - VD furnace refining - bloom continuous casting to obtain continuous casting billets.

[0067] Step S5: According to the obtained rolling process parameters, heat the obtained continuous casting billets in a heating furnace and then roll them in a continuous rolling mill to obtain hot-rolled wire rods, and then carry out cooling after the hot-rolled wire rods are spun.

[0068] In this step, a walking beam type reheating furnace is adopted in the hot rolling stage, and a three-stage heating system is adopted, namely a preheating section, a heating section, and a soaking section. The rolling line equipment is a continuous rolling mill, and the post-rolling cooling equipment is a Stelmor line equipment.

[0069] Step S6: Cold draw and heat-treat the cooled wire rods, intercept and process test specimens from the heat-treated 55SiCrA spring steel according to corresponding standards, and conduct mechanical property tests on the test specimens.

[0070] In this step, the test specimens of the hot-rolled wire rods are heat-treated by induction heat treatment. The quenching and holding temperature is 850 - 900 °C, the cooling method after quenching is water cooling, and the tempering temperature is 420 - 480 °C. The tensile strength of the obtained 55SiCrA spring steel is 1985 MPa - 2200 MPa, and the reduction of area is 42% - 46%.

[0071] In several specific application examples of the present invention, 55SiCrA springs are prepared using the alloying components and process parameters obtained by the above 55SiCrA spring steel performance optimization model.

[0072] The components and process parameters adopted in Example 1 are shown in Table 1 and Table 2; steel billets are smelted according to the alloying components shown in Table 1, hot-rolled wire rods are obtained by rolling according to the process parameters shown in Table 2, and the wire rods are processed in the order of cold drawing and heat treatment to obtain 55SiCrA spring steel with a φ16 specification. The performance test results of the obtained 55SiCrA spring steel are shown in Table 3.

[0073] Table 1 Composition of steel billets (wt.%)

[0074]

[0075] Table 2 Rolling process parameters

[0076]

[0077]

[0078] Table 3 Mechanical properties

[0079]

[0080] The components and process parameters adopted in Example 2 are shown in Table 4 and Table 5; steel billets are smelted according to the alloying components shown in Table 4, hot-rolled wire rods are obtained by rolling according to the process parameters shown in Table 5, and the wire rods are processed in the order of cold drawing and heat treatment to obtain 55SiCrA spring steel with a φ16 specification. The performance test results of the obtained 55SiCrA spring steel are shown in Table 6.

[0081] Table 4 Composition of steel billet (wt.%)

[0082]

[0083] Table 5 Rolling process parameters

[0084]

[0085] Table 6 Mechanical properties

[0086]

[0087] In Example 3, the composition and process parameters used are shown in Table 7 and Table 8; a steel billet is smelted according to the alloy composition shown in Table 7, hot-rolled wire rods are obtained by rolling according to the process parameters shown in Table 8, and the wire rods are processed in the order of cold drawing and heat treatment to obtain 55SiCrA spring steel with a φ16 specification. The performance test results of the prepared 55SiCrA spring steel are shown in Table 9.

[0088] Table 7 Composition of steel billet (wt.%)

[0089]

[0090] Table 8 Rolling process parameters

[0091]

[0092] Table 9 Mechanical properties

[0093]

[0094] Using the alloy composition and process parameters commonly used in the prior art and the same process flow as in the preparation process of this example, 55SiCrA spring steel of the same specification is prepared as Comparative Example 1 and Comparative Example 2 of this example. In Comparative Example 1, the alloy composition and process parameters are shown in Table 10 and Table 11, and the test performance of the prepared 55SiCrA spring steel is shown in Table 12; in Comparative Example 2, the alloy composition and process parameters are shown in Table 13 and Table 14, and the test performance of the prepared 55SiCrA spring steel is shown in Table 15.

[0095] Table 10 Composition of steel billet (wt.%)

[0096]

[0097] Table 11 Rolling process parameters of steel billet

[0098]

[0099] Table 12 Mechanical properties

[0100]

[0101] Table 13 Composition of steel billet (wt.%)

[0102]

[0103] Table 14 Rolling process parameters of steel billet

[0104]

[0105] Table 15 Mechanical properties

[0106]

[0107] By analyzing the results in Tables 1-9 and Tables 10-15, it can be seen that before the optimization of the composition and rolling process, the mechanical properties were relatively poor and could not meet the requirements of actual production applications. After the optimization of the composition and rolling process, within the composition range described in this patent and under the rolling process described, the 55SiCrA spring steel achieved relatively excellent mechanical properties and could meet the requirements of actual production applications.

[0108] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles applied, and is not intended to limit the scope of the present invention claimed, but only represents the preferred embodiments of the present invention. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the properties of 55SiCrA spring steel based on ISSA-BP neural network, characterized in that, The method includes the following steps: Step S1, collect the historical production data and performance data of 55SiCrA spring steel produced by the steel mill within a predetermined time period, preprocess the original data, and construct a training data set; Step S2, construct a 55SiCrA spring steel performance optimization model based on the ISSA-BP neural network model, and use the training data set to train the 55SiCrA spring steel performance optimization model to obtain a mature 55SiCrA spring steel performance optimization model; Step S3, use the Optuna optimization method to select the optimal solutions of alloying elements and rolling process parameters under the performance optimization goal based on the mature 55SiCrA spring steel performance optimization model, and obtain the best alloying elements and rolling process parameters of the 55SiCrA spring steel; Step S4, according to the obtained best alloying elements, use converter steelmaking - LF furnace refining - VD furnace refining - bloom continuous casting to obtain continuous casting billets; Step S5, according to the obtained best rolling process parameters, heat the obtained continuous casting billets in a heating furnace and then roll them in a continuous rolling mill to obtain hot-rolled wire rods, and cool the hot-rolled wire rods after spinning; Step S6, perform cold drawing and heat treatment on the cooled wire rods, cut and process test specimens from the heat-treated 55SiCrA spring steel according to the corresponding standards, and perform mechanical property tests on the test specimens.

2. The method according to claim 1, characterized in that, The historical production data includes composition data and process data; among them, the composition data includes the mass percentages of carbon, silicon, chromium, manganese and trace elements phosphorus and sulfur; the process data includes the furnace temperature in the preheating section, the heating time in the preheating section, the furnace temperature in the heating section, the heating time in the heating section, the furnace temperature in the soaking section, the heating time in the soaking section, the temperature entering the rolling mill, the temperature at the entrance of the sizing mill, the temperature entering the furnace, the time in the furnace, the temperature leaving the furnace, the spinning temperature and the fan frequency; The performance data includes tensile strength and reduction of area.

3. The method according to claim 1, characterized in that, The preprocessing includes: cleaning, screening and standardizing the collected data.

4. The method according to claim 1, wherein Step S2 of constructing the 55SiCrA spring steel performance optimization model specifically includes: Step S21, construct an SSA-BP neural network model based on the sparrow search algorithm SSA and the backpropagation BP neural network; Step S22, improve the SSA algorithm in the SSA-BP neural network model, and then optimize the hyperparameters of the BP neural network based on the improved sparrow search algorithm ISSA to obtain the ISSA-BP neural network model; Step S23, based on the ISSA-BP neural network model, construct a 55SiCrA spring steel performance optimization model. The input layer of the model includes chemical compositions and process parameters, the hidden layer realizes non-linear mapping by adjusting the number of neurons and activation functions, and the output layer predicts tensile strength and reduction of area simultaneously; Step S24, train the 55SiCrA spring steel performance optimization model based on the training data set to obtain a mature 55SiCrA spring steel performance optimization model.

5. The method according to claim 4, wherein When constructing the SSA-BP neural network model in step S21, define the structure, input layer, hidden layer, output layer, activation function and loss function of the BP neural network model; among them, the input layer consists of composition data and process data; the hidden layer adopts a two-layer hidden layer structure, and an appropriate number of neurons are set in each layer; the output layer contains two output nodes, representing the tensile strength and the reduction of area respectively; the rule activation function is used in the hidden layer, and the linear activation function is used in the output layer to adapt to the regression task; the mean square error is used as the loss function.

6. The method according to claim 4, characterized in that, When improving the SSA algorithm in step S22, mix the sine-cosine algorithm and the Lévy flight mechanism in the SSA algorithm to obtain the improved sparrow search algorithm ISSA. The process includes: Fuse the sine-cosine algorithm in the updation method of the discoverer's position, introduce a non-linear sine learning factor, calculate the discoverer's position according to the learning factor formula, and update the discoverer's position according to the warning value and the safety value; then use the Lévy flight strategy to update the follower's position.

7. The method according to claim 4, characterized in that, When optimizing the hyperparameters of the BP neural network, the hyperparameters to be optimized include: the number of neurons, the learning rate, and the number of iterations.

8. The method according to claim 1, wherein In step S3, when the performance optimization target is that the tensile strength of the spring steel is ≥2000 MPa and the reduction of area is ≥40%, the mass percentage of the best alloy composition content is: C: 0.51-0.59%, Si: 1.2-1.6%, Mn: 0.5-0.8%, Cr: 0.5-0.8%, P: ≤0.025%, S: ≤0.02%.

9. The method according to claim 1, wherein In step S5, a walking beam type reheating furnace is adopted in the hot rolling stage, and a three-stage heating system is adopted, namely a preheating section, a heating section, and a soaking section. The rolling line equipment is a continuous rolling mill, and the post-rolling cooling equipment is a Stelmor line equipment.

10. The method according to claim 4, wherein In step S6, the test sample of the hot-rolled wire rod is heat-treated by induction heat treatment. The quenching and holding temperature is 850-900 °C, the cooling method after quenching is water cooling, and the tempering temperature is 420-480 °C.

Images