Multi-field coupling collaborative modeling and parameter autonomous intelligent decision method of roller quenching process

Abstract

This invention discloses a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process, comprising: collecting four key process parameters—nozzle flow rate, water ratio, roller gap, and plate speed—and performing normalization preprocessing, dividing them into training and testing sets; using the four key process parameters from the training set as inputs, and the water-cooling convection heat transfer coefficient and phase transformation plasticity coefficient as outputs, training a BP neural network; predicting the water-cooling convection heat transfer coefficient on the workpiece surface and the phase transformation plasticity coefficient inside the material through the trained BP neural network; solving a multi-field coupled model consisting of the temperature field, microstructure phase transformation field, and stress-strain field to achieve quantitative calculation of the core quality indicators of quenching; constructing a nonlinear optimization model with the goal of minimizing water consumption, under the constraints of multiple core quality indicators such as plate temperature, exit hardness, and plate shape; and using a differential evolution algorithm to perform global optimization of the optimization model and autonomously decide the optimal process parameter settings.

Description

Technical Field

[0001] This invention belongs to the field of steel plate heat treatment technology, and relates to a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process. Background Technology

[0002] Quenching of steel plates is a crucial heat treatment process for enhancing their strength, hardness, and toughness. The precise setting of process parameters directly determines the final microstructure, plate shape, and other core quality indicators of the product. Ensuring that all product quality indicators are met, how to achieve efficient water resource utilization and reduce production costs during the quenching process is a significant challenge in the field of quenching. With the increasing market demand for high-strength and ultra-high-strength steels, more stringent requirements are being placed on the uniformity of their microstructure and the stability of their mechanical properties after quenching. This has made the precise setting of quenching process parameters a research focus for both academia and industry.

[0003] In recent years, with the development of artificial intelligence technology, data-driven models have provided new ideas for optimizing process parameters. However, this method is highly dependent on massive amounts of high-quality field data. In actual industrial environments, acquiring high-quality data that covers all possible operating conditions and has complete labels is costly. In addition, pure data models are like "black boxes," and their predictions lack physical constraints. When operating conditions exceed the data range, they may provide parameter suggestions that violate physical laws and are unreliable, posing production safety risks.

[0004] Therefore, existing technologies either rely too heavily on experience, lacking accuracy and cost-effectiveness, or rely too heavily on data, facing problems such as missing data and unreliable physical properties. Developing a hybrid model that integrates physical mechanisms and data-driven approaches, and an intelligent parameter setting method that takes into account multiple core quenching quality indicators such as steel plate temperature, hardness, and shape while minimizing water consumption as a clear optimization objective, has become a core problem to be solved in the current development of roller quenching machine control technology. Summary of the Invention

[0005] To address the aforementioned technical problems, the purpose of this invention is to provide a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching processes. This method integrates mechanistic models and machine learning models to achieve high-precision prediction of core quality indicators. Based on this, and while ensuring the achievement of cooling process objectives, it performs global optimization with the goal of minimizing cooling water consumption, thereby achieving precise online water-saving and energy-reducing control.

[0006] This invention provides a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching processes, including:

[0007] Step 1: Collect four key process parameters: nozzle flow rate, water ratio, roller gap, and plate speed, perform normalization preprocessing, and then divide them into training set and test set;

[0008] Step 2: Using the four key process parameters of the training set as inputs and the water-cooled convection heat transfer coefficient and phase change plasticity coefficient as outputs, train the BP neural network, and predict the water-cooled convection heat transfer coefficient on the workpiece surface and the phase change plasticity coefficient inside the material through the trained BP neural network.

[0009] Step 3: Solve the multi-field coupled model consisting of the temperature field, microstructure phase transformation field, and stress-strain field to achieve quantitative calculation of the core quality indicators of quenching;

[0010] Step 4: With the goal of minimizing water consumption, a nonlinear optimization model is constructed under the constraints of multiple core quality indicators such as plate temperature, outlet hardness, and plate shape.

[0011] Step 5: Use the differential evolution algorithm to perform global optimization on the optimization model and autonomously decide on the optimal process parameter settings.

[0012] The multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process of the present invention has the following beneficial effects:

[0013] 1. By constructing a mechanism- and data-driven intelligent modeling method for the quenching process, a BP neural network model is used to accurately learn and output key physical property parameters that are difficult to measure directly in complex quenching environments, including the water-cooling convection heat transfer coefficient on the workpiece surface and the phase transformation plasticity coefficient inside the material. This effectively compensates for the shortcomings of over-assumption and simplification in traditional mechanism models, and significantly improves the overall prediction accuracy of multi-field coupled models such as temperature field, microstructure field, stress and strain field, providing a reliable foundation for the precise control and optimization of the quenching process.

[0014] 2. This invention aims to minimize water consumption during the quenching process. It constructs a complete constrained nonlinear optimization model by integrating multiple core quality indicators such as plate temperature, exit hardness, and exit plate shape. A differential evolution algorithm is then used to solve the constructed optimization model. This method performs autonomous search and iterative calculations within the allowable range of process parameters, ultimately obtaining the optimal process parameter settings that minimize water consumption while simultaneously satisfying all performance constraints. While ensuring the product's plate temperature, mechanical properties, and plate shape quality, it significantly reduces water consumption, achieving a balance between energy saving, consumption reduction, and quality control, resulting in significant economic and environmental benefits.

[0015] 3. Based on collaborative modeling, high-precision multi-physics coupled calculations are achieved. By inputting key physical property parameters identified by neural networks into a mechanism model based on physical laws, the temperature field, microstructure phase transformation field, and stress-strain field are coupled and solved to achieve quantitative calculation of core quality indicators after quenching. Global optimization of process parameters is then performed based on this. This method combines the reliability of physical mechanisms with the flexibility of data-driven approaches, effectively overcoming the limitations of traditional modeling methods. It achieves rapid calculation and online optimization capabilities for multi-physics parameters, directly serving real-time process control in production, improving the intelligent control level of the quenching process, and providing a complete solution for the precise design and efficient operation of heat treatment processes. Attached Figure Description

[0016] Figure 1 This is a flowchart of a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for a roller quenching process according to the present invention. Detailed Implementation

[0017] This invention provides a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching processes. Taking the process optimization of a medium-thick plate roller quenching machine production line in a steel plant as a specific application scenario, the implementation process of the method described in this invention is explained in detail. This invention aims to construct a high-precision multi-field coupled prediction model for the quenching process by integrating physical mechanisms and data-driven technology. While ensuring the plate temperature, mechanical properties, and plate shape quality after quenching, the key process parameters of the quenching machine are optimized with the goal of minimizing process water consumption.

[0018] This invention employs a hybrid mechanism- and data-driven modeling approach. It establishes a nonlinear mapping relationship between process parameters and key physical property parameters through a BP neural network, and embeds the identification results into a multi-field coupled mechanism model to solve for the temperature field, microstructure phase transformation field, and stress-strain field, thereby constructing a high-precision multi-field coupled prediction model for the quenching process.

[0019] This invention uses a physical model of the temperature field, microstructure field, and stress-strain field of quenched steel plate as its basic framework, and combines a BP neural network with a highly efficient global intelligent optimization algorithm to achieve intelligent optimization control of the quenching process.

[0020] like Figure 1 As shown, the present invention provides a multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for a roller quenching process, comprising:

[0021] Step 1: Collect four key process parameters: nozzle flow rate, water ratio, roller gap, and plate speed, and perform normalization preprocessing before dividing them into training set and test set.

[0022] Step 2: Using the four key process parameters from the training set as input, and the water-cooled convection heat transfer coefficient and phase transformation plasticity coefficient as output, train a BP neural network. Then, use the trained BP neural network to predict the water-cooled convection heat transfer coefficient on the workpiece surface and the phase transformation plasticity coefficient within the material. Specifically:

[0023] Step 2.1: Let the initial number of nodes in the hidden layer be calculated using the following formula:

[0024]

[0025] Where h is the initial number of nodes in the hidden layer, n and m are the number of nodes in the input layer and output layer, respectively, and c is an adjustment constant between 1 and 10.

[0026] Step 2.2: The weights and biases are initialized to the range [0,1]. The computation from the input layer to the hidden layer in a neural network is achieved through matrix multiplication and activation functions. For a given input vector... This involves calculating the weighted sum of each hidden node as it passes from the input layer to the hidden layer, and then applying the activation function; for the j-th node in the hidden layer, its weighted sum is calculated. Represented as:

[0027]

[0028] in, These are the weights connecting the input node i and the hidden layer node j. It is the bias term for hidden layer point j. The output of the hidden layer nodes is obtained by using the activation function f:

[0029]

[0030] The computation from hidden layer to output layer in a neural network involves passing the outputs of the hidden layers to the output layer, weighting and summing them, and applying an activation function. For each node k in the output layer, the weighted sum... Represented as:

[0031]

[0032] in, These are the weights connecting hidden layer node j and output node k. It is the output of the hidden layer node. It is the bias term of the output layer node k, which is used to obtain the final output through the activation function:

[0033]

[0034] Step 2.3: Set the loss function to mean squared error loss. When training the neural network using four key process parameters from the historical database, optimize the loss function using the backpropagation algorithm to minimize it.

[0035] Step 2.4: For the error of the output layer, the error of each output node k is... Represented as:

[0036]

[0037] For each node j in the hidden layer, its error The following is calculated through backpropagation of the error from the output layer:

[0038]

[0039] Where m is the number of nodes in the output layer, and gradient descent is used to continuously update each weight and bias to obtain the optimal weights and biases. The update rule for the weights from the hidden layer to the output layer is:

[0040]

[0041] in, This is the learning rate.

[0042] The update rule for the weights from the input layer to the hidden layer is as follows:

[0043]

[0044] Step 2.5: After the model training is completed, the model is validated using a test dataset. The prediction accuracy and generalization ability are evaluated by calculating the error between the predicted and actual values ​​of the water-cooled convection heat transfer coefficient and the phase change plasticity coefficient.

[0045] Step 2.6: Deploy the trained BP neural network into the quenching process optimization system to predict the water-cooling convection heat transfer coefficient on the workpiece surface and the phase transformation plasticity coefficient inside the material.

[0046] Step 3: Solve the multi-field coupled model consisting of the temperature field, microstructure phase transformation field, and stress-strain field to achieve quantitative calculation of the core quality indicators of quenching. Specifically:

[0047] Step 3.1: The governing equations of the temperature field model are based on Fourier's one-dimensional unsteady-state heat conduction law, and consider the latent heat of phase change as an internal heat source. The expression is as follows:

[0048]

[0049] in, For material density, For specific heat capacity, denoted as thermal conductivity, T as steel plate temperature, t as time, z as thickness direction coordinate, and Q as latent heat of phase change.

[0050] Step 3.2: The heat exchange boundary conditions are different in different regions. Among them, the air-cooled zone is the convective heat exchange between the steel plate and the air and the radiative heat exchange between the steel plate and the outside world.

[0051] First, the thickness direction of the steel plate is discretized. Then, a numerical solution method is used to establish the difference equation for each cell. The display format of the internal nodes is as follows:

[0052]

[0053] in, Let represent the temperature of the r-th spatial node at the p-th time step. For time step, For spatial step size, Let r be the internal heat source at position r at time p.

[0054] By introducing Fourier numbers and simplifying, we get:

[0055]

[0056] in: , .

[0057] The discrete equations for the boundary nodes of the air-cooled zone are as follows:

[0058]

[0059] Among them, λ, ε, σ, These represent thermal conductivity, surface emissivity, Stefan-Boltzmann constant, air-cooled convective heat transfer coefficient, air temperature, and ambient temperature, respectively. This represents the temperature of the steel plate surface at time p. This represents the temperature of the steel plate at time p within one unit step.

[0060]

[0061] The discrete equations for the boundary nodes of the water-cooled zone are as follows:

[0062]

[0063] in, For cooling water temperature, The coefficient of heat transfer during water cooling convection is given.

[0064]

[0065] Step 3.3: The latent heat of phase change Q is calculated by the following formula:

[0066]

[0067] in, For the first Phase transition enthalpy of seed phase For the first The volume fraction of the seed phase is calculated in real time by coupling with a tissue field model. This refers to all the phases that the material undergoes during the quenching process.

[0068] Step 3.4: The steel plate organization process specifically includes:

[0069] Diffusion-type phase transition models: These are generally described using isothermal transformations. However, quenching is a non-isothermal continuous cooling process and cannot be directly described using isothermal kinetic equations. For non-isothermal transformation problems, the continuous cooling process is discretized into multiple small time steps. Each time step is considered an isothermal transformation process. Then, the Scheil superposition rule is used to determine the start time of the phase transition. The incubation period at the current temperature is calculated by superimposing the values ​​within each time step. The phase transition is considered to have started when the following equation is satisfied:

[0070]

[0071] in, For the l-th isothermal time step, This represents the incubation period of the phase transition at the l-th isothermal time step. This represents the total number of discrete time steps elapsed from the start of the phase transition to the present moment.

[0072] The diffusion-type phase transformations of ferrite, pearlite, and bainite are described by the Avrami equation, whose basic form is:

[0073]

[0074] in: The volume fraction of phase transitioned is t; t is the time after the phase transition begins. and It is the phase transformation correlation coefficient; it can be extracted from the TTT curve and changes with the chemical composition of the steel and the austenitizing temperature. and It can be obtained using the following formula:

[0075]

[0076]

[0077] Where t1 and t2 represent two different isothermal times at a given isothermal temperature T, and V1 and V2 are the variables corresponding to these isothermal times; by changing the isothermal temperature T, we can obtain the isothermal temperatures at different isothermal temperatures. and The value of .

[0078] When using the superposition rule to calculate the microstructure transformation in steel, the concept of virtual time needs to be introduced: when calculating the microstructure transformation in the l-th time step, it is necessary to first determine the microstructure volume fraction at the end of the previous time step. At the current temperature The time required for the transition under the given conditions is called virtual time, and its specific calculation equation is as follows:

[0079]

[0080] In the formula, , The phase transition kinetic parameters at this time step are then calculated, followed by the calculation of the phase transition kinetic parameters at the current time step at temperature. Isothermal Δt i Phase volume fraction after time :

[0081]

[0082] Non-diffusional phase transformation model: For martensitic phase transformation, the Koistinen-Marburger equation is used, and its transformation variable is only related to temperature.

[0083]

[0084] in: It represents the volume fraction of martensite generated at temperature T, and Ms is the temperature at which the martensitic transformation begins. It is a constant for the martensitic transformation rate, taken as 0.011.

[0085] Step 3.5: Steel plate hardness calculation includes:

[0086] At the end of quenching, the hardness is calculated using the mixed hardness law, and the formula is:

[0087]

[0088] in: , and These are the Vickers hardness values ​​for austenite-ferrite-pearlite mixtures, bainite, and martensite, respectively. These are the volume fractions of austenite, ferrite, pearlite, bainite, and martensite, respectively.

[0089] Step 3.6: Calculation of the stress-strain field of the steel plate includes:

[0090] Based on the thermo-elastic-plastic constitutive theory, the total strain increment Decomposed into elastic strain increments Thermal strain increment Plastic strain increment Phase transition strain increment and phase transformation plastic strain increment The constitutive relation of the sum of the five components is expressed as:

[0091]

[0092]

[0093] Where E is Young's modulus; Poisson's ratio; denoted by , where u represents the direction of stress, and v represents the normal direction of stress.

[0094]

[0095] in, It is the coefficient of thermal expansion of the material, which is a function of temperature.

[0096]

[0097] Where H is the work hardening index, It represents the equivalent stress.

[0098]

[0099] in: For the transformation volume of the organization; Here are the expansion coefficients of each phase. The calculation formula is as follows:

[0100]

[0101] in: It is the density of austenite; It is the density of ferrite, pearlite, bainite and martensite after phase transformation.

[0102]

[0103] in: The phase transformation plasticity coefficient, For tissue volume fraction, It represents the change in tissue volume fraction over a time step.

[0104] Step 4: With the goal of minimizing water consumption, and under the constraints of multiple core quality indicators such as plate temperature, outlet hardness, and plate shape, a nonlinear optimization model is constructed, specifically as follows:

[0105] Step 4.1: The core optimization objective is to minimize the total water consumption during the quenching process. Its mathematical expression is as follows:

[0106]

[0107] Where F represents the objective function, Indicates the number of upper nozzles; This represents the cooling water flow rate of the w-th upper nozzle; This represents the water ratio, which is the ratio of the flow rate of the lower nozzle to the flow rate of the upper nozzle at the same position.

[0108] Step 4.2: To ensure that the optimization results simultaneously guarantee stable product quality, the core quality indicators of the quenched workpiece are used as hard constraints. In the optimization model, for any set of process parameters, including the flow rate of each nozzle, water ratio, roller gap, plate speed, and the oscillation time of the steel plate in the low-pressure zone, the corresponding core quenching quality indicators, including the center temperature of the quenched steel plate, the hardness of the quenched steel plate, and the surface shape, need to be calculated using the multi-field coupling model described in Step 3. These calculation results are used to determine whether the constraints are met.

[0109] Final temperature constraint: The core temperature of the workpiece after quenching must reach the target range required by the process.

[0110]

[0111] in, This refers to the center temperature of the steel plate after quenching. This is the target temperature for the steel plate after quenching.

[0112] Hardness constraint: the hardness of the workpiece after quenching must be within a specified range.

[0113]

[0114] in, This represents the hardness value of the steel plate after quenching. To ensure that the product meets the minimum hardness requirements for mechanical properties, The maximum hardness limit allowed to ensure subsequent processing performance.

[0115] Plate shape quality constraints: The surface shape of the workpiece after quenching must meet quality requirements.

[0116]

[0117] in, This refers to the surface shape of the steel plate after quenching. This refers to the target shape of the steel plate after quenching.

[0118] Step 5: Use the differential evolution algorithm to perform global optimization on the optimization model and autonomously decide on the optimal process parameter settings.

[0119] The optimization problem solved in step 5 uses the penalty function method to handle constraints. The specific steps are as follows: if a solution violates the constraint, a large penalty term is added to the objective function value to make its fitness worse, thus eliminating it in the evolution.

[0120] Therefore, the mathematical expression for the objective function of optimizing the quenching cooling water flow rate is:

[0121]

[0122] in, This is a penalty for the center temperature of the steel plate after quenching. This is a penalty for the hardness of the steel plate after quenching. For the surface shape penalty item of the steel plate after quenching:

[0123]

[0124]

[0125]

[0126] in, This represents the corresponding penalty coefficient. The corresponding threshold is given.

[0127] Step 5 specifically involves:

[0128] Step 5.1: Initialize the population by assigning a value to each dimension of each individual. Each individual is a complete set of parameters, including the flow rate of each nozzle, water ratio, roll gap, plate speed, and the oscillation time of the steel plate in the low-pressure zone. The initial population is generated randomly.

[0129]

[0130] Where e represents the e-th individual and d represents the d-th component; This represents the lower bound of the d-th component. This represents the upper bound of the d-th component.

[0131] Set the limits of the parameter variables as ;

[0132] Step 5.2: After generating the initial population, perform mutation operations. The mutated individuals in generation G are as follows:

[0133]

[0134] in, , , Three distinct individuals are randomly selected from the Gth generation population, and are different from the target individual. J is the variation factor.

[0135] Step 5.3: Cross-reference each individual with its offspring mutation vectors. The purpose is to increase the diversity of the interference parameter vectors. Specifically, for each component, offspring mutation vectors are selected according to a certain probability to generate experimental individuals. The cross-reference operation is as follows:

[0136]

[0137] Where CR is the crossover operator. This represents randomly generated integers between [0, D], where D is the dimension of the problem to be solved; This represents the d-th component of the variant individual in generation G. This represents the d-th component of the target individual. This represents the d-th component of the experimental individual.

[0138] Step 5.4: The differential evolution algorithm uses a greedy algorithm. Based on the value of the objective function, it selects the better individual from the target individual and the trial individual as the next generation. The population evolves to the next generation and repeats the cycle until the number of algorithm iterations reaches the predetermined maximum number or the optimal solution of the population reaches the predetermined error accuracy.

[0139] This invention constructs a high-precision multi-field coupled prediction model for the quenching process based on mechanistic and data-driven collaborative modeling. By collecting operational data from the quenching production line, a one-dimensional unsteady-state temperature field model based on Fourier's law is established, coupled with a microstructure phase transformation field model and a stress-strain field model, enabling accurate prediction of the workpiece's temperature, microstructure composition, and stress distribution after quenching. Furthermore, an efficient global optimization algorithm is employed, with the goal of minimizing water consumption during the quenching process. Under the premise of satisfying multiple process constraints such as plate temperature, exit hardness, and exit surface shape, the algorithm globally optimizes key process parameters such as nozzle flow rate, water ratio, roller gap, plate speed, and oscillation time. This method provides a reliable computational basis and decision support for achieving precise optimization of the quenching process, focusing on water conservation and energy saving while ensuring product performance.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the ideas of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for a roller quenching process, characterized in that, include: Step 1: Collect four key process parameters: nozzle flow rate, water ratio, roller gap, and plate speed, perform normalization preprocessing, and then divide them into training set and test set; Step 2: Using the four key process parameters of the training set as inputs and the water-cooled convection heat transfer coefficient and phase change plasticity coefficient as outputs, train the BP neural network, and predict the water-cooled convection heat transfer coefficient on the workpiece surface and the phase change plasticity coefficient inside the material through the trained BP neural network. Step 3: Solve the multi-field coupled model consisting of the temperature field, microstructure phase transformation field, and stress-strain field to achieve quantitative calculation of the core quality indicators of quenching; Step 4: With the goal of minimizing water consumption, a nonlinear optimization model is constructed under the constraints of multiple core quality indicators such as plate temperature, outlet hardness, and plate shape. Step 5: Use the differential evolution algorithm to perform global optimization on the optimization model and autonomously decide on the optimal process parameter settings; Step 4 specifically involves: Step 4.1: The core optimization objective is to minimize the total water consumption during the quenching process. Its mathematical expression is as follows: Where F represents the objective function, Indicates the number of upper nozzles; This represents the cooling water flow rate of the w-th upper nozzle; This represents the water ratio, which is the ratio of the flow rate of the lower nozzle to the flow rate of the upper nozzle at the same position. Step 4.2: To ensure that the optimization results simultaneously guarantee stable product quality, the core quality indicators of the quenched workpiece are used as hard constraints. In the optimization model, for any set of process parameters, including the flow rate of each nozzle, water ratio, roller gap, plate speed, and the oscillation time of the steel plate in the low-pressure zone, the corresponding core quenching quality indicators, including the center temperature of the quenched steel plate, the hardness of the quenched steel plate, and the surface shape, need to be calculated using the multi-field coupling model described in Step 3. These calculation results are used to determine whether the constraint conditions are met. Final temperature constraint: The core temperature of the workpiece after quenching must reach the target range required by the process. in, This refers to the center temperature of the steel plate after quenching. The target temperature for the steel plate after quenching; Hardness constraint: the hardness of the workpiece after quenching must be within a specified range. in, This represents the hardness value of the steel plate after quenching. To ensure the minimum hardness requirement that must be met to guarantee the mechanical properties of the product, The maximum hardness limit allowed to ensure subsequent processing performance; Plate shape quality constraints: The surface shape of the workpiece after quenching must meet quality requirements. in, This refers to the surface shape of the steel plate after quenching. This refers to the target shape of the steel plate after quenching.

2. The multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process according to claim 1, characterized in that, Step 2 specifically involves: Step 2.1: Set the initial number of nodes in the hidden layer; Step 2.2: Initialize the weights and biases to [0,1]; Step 2.3: Set the loss function to mean squared error loss. When training the neural network using four key process parameters from the historical database, optimize the loss function using the backpropagation algorithm to minimize it. Step 2.4: When training the neural network, reduce the output error through iterative training and continuously update each weight and bias using gradient descent to obtain the optimal weight and bias; Step 2.5: After the model training is completed, the model is validated using a test dataset. The prediction accuracy and generalization ability are evaluated by calculating the error between the predicted and actual values ​​of the water-cooled convection heat transfer coefficient and the phase change plasticity coefficient. Step 2.6: Deploy the trained BP neural network into the quenching process optimization system to predict the water-cooling convection heat transfer coefficient on the workpiece surface and the phase transformation plasticity coefficient inside the material.

3. The multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process according to claim 1, characterized in that, Step 3 specifically involves: Step 3.1: The governing equations of the temperature field model are based on Fourier's one-dimensional unsteady-state heat conduction law, and consider the latent heat of phase change as an internal heat source. The expression is as follows: in, For material density, For specific heat capacity, Where T is the thermal conductivity, t is the temperature of the steel plate, z represents the thickness direction coordinate, and Q is the latent heat of phase change. Step 3.2: The heat transfer boundary conditions differ in different regions. In the air-cooled zone, there is convective heat transfer between the steel plate and the air, and radiative heat transfer between the steel plate and the outside environment. First, the thickness direction of the steel plate is discretized. Then, a numerical solution method is used to establish the difference equation for each cell. The display format of the internal nodes is as follows: in, Let represent the temperature of the r-th spatial node at the p-th time step. For time step, For spatial step size, Let r be the internal heat source at position r at time p; By introducing Fourier numbers and simplifying, we get: in: , ; The discrete equations for the boundary nodes of the air-cooled zone are as follows: Among them, λ, ε, σ, These represent thermal conductivity, surface emissivity, Stefan-Boltzmann constant, air-cooled convective heat transfer coefficient, air temperature, and ambient temperature, respectively. This represents the temperature of the steel plate surface at time p. This represents the temperature of the steel plate at time p within the next unit step. The discrete equations for the boundary nodes of the water-cooled zone are as follows: in, For cooling water temperature, The coefficient of heat transfer during water cooling convection; Step 3.3: The latent heat of phase change Q is calculated by the following formula: in, For the first Phase transition enthalpy of seed phases For the first The volume fraction of the seed phase is calculated in real time by coupling with a tissue field model. All phases that the material undergoes transformations during the quenching process; Step 3.4: The steel plate organization process specifically includes: Diffusion-type phase transition model: For non-isothermal transition problems, the continuous cooling process is discretized into multiple small time steps. Each time step is considered an isothermal transition process. Then, the Scheil superposition rule is used to determine the start time of the phase transition. The incubation period at the current temperature is calculated by superimposing the values ​​within each time step. The phase transition is considered to have started when the following equation is satisfied: in, For the l-th isothermal time step, This represents the incubation period of the phase transition at the l-th isothermal time step. This represents the total number of discrete time steps elapsed from the start of the phase transition to the present moment. The diffusion-type phase transformations of ferrite, pearlite, and bainite are described by the Avrami equation, whose basic form is: in: The volume fraction of phase transitioned is t; t is the time after the phase transition begins. and This is the phase transition correlation coefficient; and It can be obtained using the following formula: Where t1 and t2 represent two different isothermal times at a given isothermal temperature T, and V1 and V2 are the variables corresponding to these isothermal times; by changing the isothermal temperature T, we can obtain the isothermal temperatures at different isothermal temperatures. and The value of ; When using the superposition rule to calculate the microstructure transformation in steel, the concept of virtual time needs to be introduced: when calculating the microstructure transformation in the l-th time step, it is necessary to first determine the microstructure volume fraction at the end of the previous time step. At the current temperature The time required for the transition under the given conditions is called virtual time, and its specific calculation equation is as follows: In the formula, , The phase transition kinetic parameters at this time step are then calculated, followed by the calculation of the phase transition kinetic parameters at the current time step at temperature. Isothermal Δt i Phase volume fraction after time : Non-diffusional phase transformation model: For martensitic phase transformation, the Koistinen-Marburger equation is used, and its transformation variable is only related to temperature. in: It represents the volume fraction of martensite generated at temperature T, and Ms is the temperature at which the martensitic transformation begins. It is a constant representing the rate of martensitic transformation; Step 3.5: Steel plate hardness calculation includes: At the end of quenching, the hardness is calculated using the mixed hardness law, and the formula is: in: , and These are the Vickers hardness values ​​for austenite-ferrite-pearlite mixtures, bainite, and martensite, respectively. These are the volume fractions of austenite, ferrite, pearlite, bainite, and martensite, respectively. Step 3.6: Calculation of the stress-strain field of the steel plate includes: Based on the thermo-elastic-plastic constitutive theory, the total strain increment Decomposed into elastic strain increments Thermal strain increment Plastic strain increment Phase transition strain increment and phase transformation plastic strain increment The constitutive relation of the sum of the five components is expressed as: Where E is Young's modulus; Poisson's ratio; For stress components; u represents the direction of stress, and v represents the normal direction of stress. in, It is the coefficient of thermal expansion of the material, a function of temperature; Where H is the work hardening index, Indicates equivalent stress; in: For the transformation volume of the organization; The expansion coefficients of each phase are calculated using the following formulas: in: It is the density of austenite; It is the density of ferrite, pearlite, bainite, and martensite after phase transformation; in: The phase transformation plasticity coefficient, For tissue volume fraction, It represents the change in tissue volume fraction over a time step.

4. The multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process according to claim 1, characterized in that, The optimization problem solved in step 5 uses the penalty function method to handle constraints. The specific steps are as follows: if a solution violates the constraint, a large penalty term is added to the objective function value to make its fitness worse, thus eliminating it in the evolution. Therefore, the mathematical expression for the objective function of optimizing the quenching cooling water flow rate is: in, This is a penalty for the center temperature of the steel plate after quenching. This is a penalty for the hardness of the steel plate after quenching. For the surface shape penalty item of the steel plate after quenching: in, This represents the corresponding penalty coefficient. The corresponding threshold is given.

5. The multi-field coupled collaborative modeling and parameter autonomous intelligent decision-making method for roller quenching process according to claim 1, wherein step 5 specifically comprises: Step 5.1: Initialize the population by assigning a value to each dimension of each individual in the population. Each individual is a complete combination of parameter settings, including the flow rate of each nozzle, water ratio, roll gap, plate speed, and the oscillation time of the steel plate in the low-pressure zone. The initial population is generated randomly. in, e represents the e-th individual, and d represents the d-th component; This represents the lower bound of the d-th component. This represents the upper bound of the d-th component; Set the limits of the parameter variables as ; Step 5.2: After generating the initial population, perform mutation operations. The mutated individuals in generation G are as follows: in, , , Three distinct individuals are randomly selected from the Gth generation population, and are different from the target individual. J is the variation factor; Step 5.3: Cross over each individual and its generated offspring mutation vectors. For each component, select an offspring mutation vector with a certain probability to generate experimental individuals. The crossover operation is as follows: Where CR is the crossover operator. This represents randomly generated integers between [0, D], where D is the dimension of the problem to be solved; This represents the d-th component of the variant individual in generation G. This represents the d-th component of the target individual. This represents the d-th component of the experimental individual; Step 5.4: The differential evolution algorithm uses a greedy algorithm. Based on the value of the objective function, it selects the better individual from the target individual and the trial individual as the next generation. The population evolves to the next generation and repeats the cycle until the number of algorithm iterations reaches the predetermined maximum number or the optimal solution of the population reaches the predetermined error accuracy.

Images