Metal material surface difference analysis method

Abstract

The invention provides a metal material surface difference analysis method which comprises the following steps: calculating Gibbs free energy at different temperatures by adopting a thermodynamic model according to chemical components and an initial microstructure of a material, and drawing a time-temperature-transformation curve of the material through phase diagram analysis software, obtaining a grain growth kinetic parameter and a recrystallization temperature interval; inputting data information of strength, hardness and toughness in different application scenes by utilizing the established multi-objective optimization model, and searching optimal grain size distribution through a genetic algorithm to obtain an optimal microstructure for balancing various performance indexes; according to the optimal microstructure, a heat treatment process route is reversely designed, finite element software is adopted to simulate temperature field and stress field distribution in the heat treatment process, and control parameters of heating, heat preservation and cooling are determined through iterative optimization.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a method for analyzing surface differences of metal materials. Background Art

[0002] In the microstructural analysis of metal materials, grain size (i.e., particle size) is a key parameter. However, in practical applications, there is a thorny technical problem: how to maintain the strength and hardness of the material without sacrificing its toughness too much. This problem stems from the dual effect of particle size on material properties. When the grain size is reduced by methods such as heat treatment or machining, the strength and hardness of the material will indeed increase, which is beneficial for parts that need to withstand high stress. But at the same time, smaller grains will lead to an increase in the grain boundary area, making the material more likely to crack along the grain boundary when impacted, thereby reducing toughness. On the contrary, if the grain size is increased, the toughness of the material will be improved, which is beneficial for parts that need to resist fatigue damage. However, large grains will reduce the resistance of grain boundaries to dislocation movement, making the material more susceptible to plastic deformation, thereby reducing strength and hardness. This contradictory relationship between particle size and performance is particularly prominent in some special applications. For example, in the aerospace field, materials need to have sufficient strength to withstand high temperature and high pressure environments, and maintain a certain toughness to prevent sudden damage. How to find a balance between these two seemingly contradictory requirements has become a major challenge facing material scientists. Summary of the invention

[0003] The present invention provides a method for analyzing surface differences of metal materials, which mainly includes:

[0004] According to the chemical composition and initial microstructure of the material, the thermodynamic model is used to calculate the Gibbs free energy at different temperatures. The time-temperature-transformation curve of the material is drawn by phase diagram analysis software to obtain the grain growth kinetic parameters and recrystallization temperature range.

[0005] According to the grain growth kinetics parameters and recrystallization temperature range, a quantitative relationship model between grain size and heat treatment process parameters is established. The quantitative relationship model is trained by a neural network algorithm, and the target grain size range is input to reversely deduce the corresponding heat treatment temperature, time and cooling rate.

[0006] According to the derived heat treatment process parameters, heat treatment is carried out through the pilot furnace control system, the sample microstructure is observed with a metallographic microscope, the grain size distribution is measured using image analysis software, and statistical analysis is performed to determine whether the actual grain size is within the target range;

[0007] If the actual grain size is within the target range, the mechanical properties of the heat-treated specimens are tested, including tensile, impact and hardness. The data mining algorithm is used to analyze the correlation between grain size and strength, hardness and toughness, and a multi-objective optimization model of material properties and microstructure is established.

[0008] By using the established multi-objective optimization model, inputting data information on strength, hardness and toughness under different application scenarios, searching for the best grain size distribution through genetic algorithms, the optimal microstructure that balances various performance indicators is obtained;

[0009] According to the optimal microstructure, the heat treatment process route is reversely designed, the temperature field and stress field distribution during the heat treatment process are simulated using finite element software, and the control parameters of heating, insulation and cooling are determined through iterative optimization;

[0010] The optimized heat treatment process is implemented on the industrial production line, and an online monitoring system is used to collect the temperature, stress and displacement of the workpiece in real time. The workpiece is analyzed through edge computing equipment. If an abnormality is detected, the process parameters of the heat treatment equipment are adjusted to achieve closed-loop control.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention provides a method for analyzing the surface differences of metal materials. By calculating the Gibbs free energy of the material at different temperatures and using phase diagram analysis, the method can accurately predict and control the grain growth and recrystallization behavior of the material, thereby ensuring that the microstructure meets the performance requirements. By establishing a relationship model between grain size and heat treatment parameters, the optimal heat treatment conditions can be reversely derived based on the target grain size, thereby improving process efficiency and consistency. After heat treatment in a small test furnace, metallographic analysis and statistical methods are used to verify whether the grain size meets expectations, and process parameters are adjusted in time to ensure product quality. The mechanical properties of the material after heat treatment are tested, and data mining technology is used to analyze the relationship between grain size and performance, and a multi-objective optimization model is established to further improve the comprehensive performance of the material. By generating a heat treatment process route corresponding to the optimal microstructure, the number of experiments is reduced and the research and development process is accelerated. The optimized heat treatment process is implemented on the production line, and real-time feedback and automatic adjustment are achieved in combination with online monitoring and edge computing technology to improve production efficiency and quality control level. By implementing a closed-loop control strategy, heat treatment parameters are dynamically adjusted based on real-time data to ensure long-term and stable production of high-quality products, while collecting data for continuous optimization of models and processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flow chart of a method for analyzing surface differences of metal materials.

[0014] Figure 2It is a schematic diagram of a method for analyzing surface differences of metal materials according to the present invention.

[0015] Figure 3 It is another schematic diagram of a method for analyzing surface differences of metal materials according to the present invention. DETAILED DESCRIPTION

[0016] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0017] like Figure 1 -3. In this embodiment, a method for analyzing surface differences of metal materials may specifically include:

[0018] Step S101, according to the chemical composition and initial microstructure of the material, a thermodynamic model is used to calculate the Gibbs free energy at different temperatures, and a time-temperature-transformation curve of the material is drawn by phase diagram analysis software to obtain grain growth kinetic parameters and recrystallization temperature range.

[0019] The CALPHAD method is used for thermodynamic calculation, and the thermodynamic calculation calculates the Gibbs free energy at different temperatures according to the chemical composition and initial microstructure data of the material, and obtains the volume fraction and chemical composition of each phase. According to the thermodynamic calculation results, the continuous cooling transformation diagram and isothermal transformation diagram of the material are drawn using JMatPro software to obtain the phase change starting temperature and phase change completion temperature. For the continuous cooling transformation diagram and isothermal transformation diagram, the Johnson-Mehl-Avrami-Kolmogorov equation is used to analyze the time-temperature-transformation curve, calculate the grain growth index and rate constant, and determine the grain growth activation energy. If the grain growth kinetic parameters are obtained, the recrystallization starting temperature and completion temperature are determined by calculating the grain growth rate at different temperatures, and the recrystallization temperature range of the material is divided. According to the recrystallization temperature range, the heat treatment process parameters are designed, and the heat treatment process parameters include heating rate and holding time, and the grain size distribution under different heat treatment process conditions is calculated.

[0020] Exemplarily, according to the chemical composition and initial microstructure of the material, the CALPHAD method is used for thermodynamic calculation, the chemical composition and initial microstructure data of the material are input, the Gibbs free energy at different temperatures is calculated, and the volume fraction and chemical composition of each phase are obtained. The above calculation results are imported using JMatPro software, and the continuous cooling transformation diagram and isothermal transformation diagram of the material are drawn to obtain the phase change starting temperature and phase change completion temperature. The Johnson-Mehl-Avrami-Kolmogorov equation is used to analyze the time-temperature-transformation curve, calculate the grain growth index and rate constant, and determine the grain growth activation energy. According to the grain growth kinetic parameters, the recrystallization starting temperature and completion temperature are determined by calculating the grain growth rate at different temperatures, and the recrystallization temperature range of the material is divided. According to the recrystallization temperature range, the heat treatment process parameters, such as heating rate and holding time, are designed, and the grain size distribution under different heat treatment process conditions is calculated. In the CALPHAD method, Fe-0.2C-1.5Mn-0.5Si, that is, the weight percentage, the chemical composition of the alloy and the initial ferrite + pearlite microstructure data are input, and the Gibbs free energy of every 50°C in the temperature range of 700-1200°C is calculated using the thermodynamic database. By minimizing the total Gibbs free energy, the volume fraction and chemical composition of each phase such as ferrite, austenite, and cementite in the equilibrium state are obtained. These data are imported into the JMatPro software to generate a continuous cooling transformation diagram, showing the curve of the cooling rate from 0.1°C / s to 100°C / s. At the same time, an isothermal transformation diagram is drawn with a temperature range of 400-900°C and a time range of 1-10000 seconds. From the figure, the austenitization starting temperature Ac1 is read as 720°C and the completion temperature Ac3 is 850°C. The isothermal transformation curve was analyzed using the Johnson-Mehl-Avrami-Kolmogorov equation ln[-ln(1-f)]=nlnt+lnk, where f is the transformation fraction, t is the time, n is the Avrami index, and k is the rate constant. n=2.5 was obtained by logarithmic fitting, indicating that grain growth is mainly controlled by three-dimensional diffusion. The k value at different temperatures was calculated, and the relationship between lnk and 1 / T was plotted. The slope was the activation energy of grain growth, and Q=280kJ / mol was obtained. According to the grain growth rate equation v=kexp(-Q / RT), the grain growth rate of every 50℃ in the temperature range of 600-1000℃ was calculated. When v reached the preset threshold of 0.1μm / s, the recrystallization start temperature was determined to be 750℃; when v dropped to 0.01μm / s, the recrystallization completion temperature was determined to be 950℃. Based on the recrystallization temperature range of 750-950℃, the heat treatment process parameters are designed as heating rate 5℃ / s, holding temperature 900℃, and holding time 1800 seconds.The extended Burke-Turnbull equation d = (kt) ^ (1 / n) was used to calculate the average grain size after heat treatment, where d is the grain size, k is the temperature-dependent rate constant, t is the time, and n is the grain growth exponent. The average grain size after heat treatment was finally obtained to be 35 μm, and the grain size distribution conformed to the log-normal distribution with a standard deviation of 0.2.

[0021] Step S102, based on the grain growth kinetics parameters and the recrystallization temperature range, a quantitative relationship model between the grain size and the heat treatment process parameters is established, the quantitative relationship model is trained by a neural network algorithm, the target grain size range is input, and the corresponding heat treatment temperature, time and cooling rate are reversely deduced.

[0022] The grain growth kinetic parameters and the recrystallization temperature range, including activation energy and frequency factor, are obtained; the Arrhenius equation is constructed according to the grain growth kinetic parameters and the recrystallization temperature range; the Arrhenius equation is subjected to nonlinear regression using the Levenberg-Marquardt algorithm to obtain a quantitative relationship function between the grain size and the heat treatment temperature, time and cooling rate. Experimental data are collected, the experimental data including the grain size measurement results under different alloy compositions, heat treatment temperatures, times and cooling rates; the experimental data are divided into a training set and a validation set in proportion. A multilayer perceptron neural network is set, the multilayer perceptron neural network including an input layer, a hidden layer and an output layer; the multilayer perceptron neural network is trained using an Adam optimizer and a mean square error loss function to obtain a trained quantitative relationship model. A target grain size range is preset; if the target grain size range is a preset value, a genetic algorithm is used for reverse calculation; the fitness function of the genetic algorithm is to minimize the error between the predicted grain size and the target size. A heat treatment temperature, time and cooling rate parameter combination is obtained by the genetic algorithm; the parameter combination is used to generate grains within the target grain size range.

[0023] For example, according to the grain growth kinetic parameters and recrystallization temperature range, including activation energy and frequency factor, the Arrhenius equation is constructed, and the Levenberg-Marquardt algorithm is used for nonlinear regression to fit the quantitative relationship function between grain size and heat treatment temperature, time and cooling rate, and obtain a preliminary mathematical model. 1000 sets of experimental data are collected, covering grain size measurement results under different alloy compositions, heat treatment temperatures such as 600-1200℃, time such as 1-10000 seconds and cooling rates such as 0.1-100℃ / s, and divided into training set and validation set in a ratio of 8:2. A multi-layer perceptron neural network is used, and the input layer such as 4 nodes are set for alloy composition, temperature, time, cooling rate, hidden layer such as 2 layers, 20 nodes per layer, and output layer such as 1 node for grain size. The Adam optimizer and mean square error loss function are used to adjust the learning rate and batch size through cross-validation to train the quantitative relationship model. The target grain size range was set to 10-50μm, and the genetic algorithm was used for reverse calculation. The population size was set to 100, the maximum number of iterations was 1000, the crossover probability was 0.8, and the mutation probability was 0.1. The fitness function was set to minimize the error between the predicted grain size and the target size, and the corresponding heat treatment temperature, time and cooling rate parameter combination was obtained. In the implementation process, the kinetic parameters obtained, such as activation energy Q = 280kJ / mol and frequency factor A = 1.5×10^15s^-1, were first used to construct the Arrhenius equation d = A×t^n×exp(-Q / RT), where d is the grain size, t is the time, T is the temperature, and R is the gas constant. The Levenberg-Marquardt algorithm was used for nonlinear regression to fit 100 sets of initial experimental data and obtain a preliminary mathematical model. Subsequently, 1000 sets of experimental data were collected, including the grain size measurement results of Fe-0.10.5C-12Mn-0.30.7Si alloy in the temperature range of 6001200℃, 110000 seconds heat treatment time, and 0.1100℃ / s cooling rate. The data set was randomly divided into training set and validation set in a ratio of 8:2. A multi-layer perceptron neural network was constructed, with 4 nodes in the input layer corresponding to alloy composition, temperature, time and cooling rate, 2 hidden layers with 20 nodes each using ReLU activation function, and 1 node in the output layer corresponding to grain size. The Adam optimizer was used with an initial learning rate of 0.001 and a batch size of 32. The hyperparameters were optimized by 5-fold cross validation. After 2000 rounds of training, the mean square error reached 0.05μm^2 on the validation set. Finally, the target grain size range was set to 10-50μm, and the genetic algorithm was used for reverse calculation. The population size is set to 100, the maximum number of iterations is 1000, the crossover probability is 0.8, and the mutation probability is 0.1.The fitness function is defined as |dpredicted - dtarget| / dtarget, where dpredicted is the grain size predicted by the neural network and dtarget is the target grain size. After 950 iterations, the optimal solution is a heat treatment temperature of 920°C, a time of 3600 seconds, a cooling rate of 5°C / s, and a predicted grain size of 32μm, which meets the target range requirements.

[0024] Step S103, according to the derived heat treatment process parameters, heat treatment is performed in a small test furnace control system, the sample microstructure is observed using a metallographic microscope, the grain size distribution is measured using image analysis software, and statistical analysis is performed to determine whether the actual grain size is within the target range.

[0025] According to the heat treatment process parameters, the heat treatment temperature, holding time and cooling rate of the pilot furnace control system are set by the PID control algorithm; the sample is heat treated by the pilot furnace control system to obtain the heat-treated sample. The heat-treated sample is cut, inlaid, ground and polished by an automatic grinding and polishing machine; if the polishing accuracy reaches a preset threshold, a sample for metallographic observation is obtained. The surface of the sample for metallographic observation is scanned by the metallographic microscope automatic scanning system; the scanning result is processed by the scale-invariant feature transformation algorithm to generate a microstructure panorama. The image analysis software is used to perform image segmentation and grain identification on the microstructure panorama; the watershed algorithm is used for image segmentation, and the roundness and area threshold method is used to identify the grains, and the grain size distribution is measured. The average grain size and standard deviation of the grain size distribution are calculated by statistical analysis; if the average grain size is within the target range and the standard deviation is less than the preset percentage of the target range, the sample for metallographic observation is judged to be qualified.

[0026] For example, according to the derived heat treatment process parameters, the heat treatment temperature, holding time and cooling rate are set by the pilot furnace control system using the PID control algorithm, with an accuracy of ±5°C, ±1 minute and ±0.1°C / s, respectively, and the sample is heat treated, and a temperature sensor is used to monitor the temperature change curve during the heat treatment process in real time. The heat-treated sample is cut, inlaid, ground and polished by an automatic grinding and polishing machine, and finally polished to an accuracy of 0.05μm to obtain a sample for metallographic observation, and etched with a 3% nitric acid alcohol solution through an electrolytic corrosion device for 10-15 seconds. The sample surface is scanned by the metallographic microscope automatic scanning system, and microstructure images of multiple fields of view are collected, and a microstructure panorama with a resolution of 4096×3072 pixels is generated through a scale-invariant feature transformation algorithm. Image analysis software is used to perform image segmentation and grain identification on the microstructure panorama. The watershed algorithm is used for image segmentation, and the roundness and area threshold methods are used to identify grains. The grain size distribution is measured, and the average grain size and standard deviation are calculated through statistical analysis. If the average grain size is within the target range and the standard deviation is less than 10% of the target range, it is judged to be qualified.

[0027] In actual operation, the pilot furnace control system uses PID control algorithm, setting the heat treatment temperature to 920℃, the holding time to 60 minutes, and the cooling rate to 5℃ / s. The PID parameters are optimized, with the proportional coefficient Kp=10, the integral time Ti=120s, and the differential time Td=30s, and the control accuracy reaches ±3℃,

[0028] ±30 seconds and ±0.05℃ / s. The sampling frequency of the temperature sensor is 1Hz, and the temperature curve of the whole process is recorded. After heat treatment, the automatic grinding and polishing machine uses 240#, 400#, 600#, 800#, and 1200# sandpaper for coarse grinding in sequence, and then fine grinding with 3μm and 1μm diamond suspension, and finally polishing to a mirror surface with 0.05μm alumina suspension. The electrolytic corrosion device uses 3% nitric acid alcohol solution, voltage 3V, current 0.2A, and corrosion for 12 seconds. The automatic scanning system of the metallographic microscope is set to a 50x objective lens, a scanning step of 0.1mm, and a total of 100×100 field of view images are collected. The scale-invariant feature transformation algorithm sets the feature point matching threshold to 0.7 and the minimum number of matching points to 4, generating a panoramic image of 4096×3072 pixels. The image analysis software uses the watershed algorithm for segmentation, sets the roundness threshold to 0.6, and the area threshold to 50-5000 pixels to identify grains. The average grain size calculated by statistical analysis is 35μm, the standard deviation is 3μm, the target range is 30-40μm, and the standard deviation is less than 1μm, which is considered qualified. The whole process is highly automated and does not require manual intervention. It takes about 120 minutes from heat treatment to result determination.

[0029] Step S104: If the actual grain size is within the target range, mechanical properties tests are performed on the heat-treated samples, including tensile, impact and hardness tests. A data mining algorithm is used to analyze the correlation between grain size and strength, hardness and toughness, and a multi-objective optimization model of material properties and microstructure is established.

[0030] The heat-treated specimens were obtained, and tensile, impact and hardness test specimens were prepared according to ASTM standards. The specimens were tested to obtain mechanical properties data such as yield strength, tensile strength, elongation, impact toughness and Vickers hardness. The metallographic structure images of the specimens were processed by image analysis software to extract microstructural characteristic parameters such as average grain size, grain size distribution standard deviation and grain shape factor. For the mechanical properties data and microstructural characteristic parameters, the correlation between grain size and strength, hardness and toughness was analyzed by random forest algorithm. If the number of trees is a preset value and the minimum number of leaf node samples is a preset value, cross validation is used to select the best parameters. According to the correlation and influence weight, a multi-objective optimization model of material properties and microstructure is constructed. The objective function is set to maximize strength and toughness while minimizing hardness deviation. A genetic algorithm is used to find the optimal combination of heat treatment process parameters to obtain optimized microstructural parameters. For the optimized microstructural parameters, the corresponding heat treatment temperature, time and cooling rate are obtained by reverse calculation.

[0031] Exemplarily, 30 tensile specimens were prepared according to ASTM E8 standard, 30 impact specimens were prepared according to ASTM E23 standard, and 30 hardness test specimens were prepared by automated testing equipment. Tensile, impact and hardness tests were performed on the heat-treated specimens to obtain mechanical properties data such as yield strength, tensile strength, elongation, impact toughness and Vickers hardness. The metallographic structure images of the specimens were processed using image analysis software to extract microstructural characteristic parameters such as average grain size, standard deviation of grain size distribution and grain shape factor, such as roundness defined as 4π×area / perimeter2. The correlation between grain size and strength, hardness and toughness was analyzed by random forest algorithm, the number of trees was set to 500, the minimum number of leaf node samples was set to 5, cross-validation was used to select the best parameters, root mean square error and determination coefficient were used to evaluate the model performance, and the influence weight of each microstructural feature on mechanical properties was calculated. Based on the obtained correlation and influence weight, a multi-objective optimization model of material properties and microstructure is constructed. The objective function is set to maximize strength and toughness while minimizing hardness deviation. The genetic algorithm is used to find the optimal combination of heat treatment process parameters. The population size is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the number of iterations is 1000. The optimized microstructure parameters are reversely calculated to obtain the corresponding heat treatment temperature, time and cooling rate. In actual operation, the automated testing equipment prepares 30 tensile specimens with a gauge length of 10mm according to the ASTME8 standard, 30 V-notch impact specimens with a gauge length of 10×10×55mm according to the ASTME23 standard, and 30 hardness test specimens with a gauge length of 15×15×10mm according to the ASTME23 standard. The tensile test uses a loading rate of 1mm / min to obtain a yield strength of 480MPa, a tensile strength of 620MPa, and an elongation of 18%. The impact test is carried out at room temperature, and the average impact toughness is 85J / cm2. The hardness test uses a load of 10kg to obtain an average Vickers hardness of 205HV10. The image analysis software processed the metallographic images of 1000×1000 pixels, and extracted the average grain size of 35μm, the standard deviation of grain size distribution of 5μm, and the average grain roundness of 0.85. The random forest algorithm set 500 decision trees, the minimum number of leaf node samples was 5, and the optimal parameters were selected by 5-fold cross validation. The model performance evaluation obtained a root mean square error of 15MPa, which is strength, 5J / cm2, which is toughness, and 8HV, which is hardness. The determination coefficients R2 were 0.92, 0.88, and 0.90, respectively. The influence weight analysis showed that the grain size had the greatest influence on strength, such as 60%, followed by toughness, such as 30%, and had a smaller influence on hardness, such as 10%. The multi-objective optimization model set the objective function to maximize 0.6×strength+0.3×toughness-0.1×|hardness-target hardness|. The genetic algorithm used a population of 100 individuals, a crossover probability of 0.8, a mutation probability of 0.1, and 1000 iterations.The final optimization results show that the best microstructure parameters are an average grain size of 30 μm, a standard deviation of 3 μm, and a roundness of 0.9. The corresponding heat treatment process parameters obtained by reverse calculation are a heating temperature of 900 °C, a holding time of 45 minutes, and a cooling rate of 10 °C / s.

[0032] Step S105, using the established multi-objective optimization model, inputting data information on strength, hardness and toughness in different application scenarios, searching for the best grain size distribution through a genetic algorithm, and obtaining the optimal microstructure that balances various performance indicators.

[0033] According to the target value and weight set in the application scenario, the comprehensive performance evaluation function F is constructed, wherein F=w1*(σ / σ_target)+w2*(H / H_target)+w3*(K / K_target), the σ, H, K represent strength, hardness and toughness respectively, the w1, w2, w3 are corresponding weights, and the target is the target value; the grain size distribution parameter population is initialized by a genetic algorithm, including the average grain size, standard deviation, roundness average value and roundness standard deviation; the fitness value of each individual is calculated by the multi-objective optimization model, and a new generation of population is generated by roulette wheel selection, single point crossover and Gaussian mutation operations; if the change amplitude of the optimal solution for several consecutive generations is less than a preset threshold or reaches the maximum number of iterations, the individual with the highest fitness is selected from the final population as the optimal solution; the optimal grain size distribution parameters that balance various performance indicators are obtained to obtain the optimal microstructure; according to the pre-established multivariate regression mapping relationship between the heat treatment process parameters and the grain size distribution parameters, the heat treatment temperature, time and cooling rate are obtained by reverse calculation.

[0034] For example, target values ​​and weights of strength, hardness and toughness are set according to different application scenarios to construct a comprehensive performance evaluation function.

[0035] F = w1 * (σ / σ_target) + w2 * (H / H_target) + w3 * (K / K_target), where σ, H, K represent strength, hardness and toughness respectively, w1, w2, w3 are the corresponding weights, target is the target value, and the target value and weight are input into the multi-objective optimization model. The genetic algorithm is used to initialize the population of grain size distribution parameters, including the average grain size, standard deviation, roundness average and roundness standard deviation, and the population size is set to 100, the crossover probability is 0.8, and the mutation probability is 0.1. The fitness value of each individual is calculated by the multi-objective optimization model, and roulette selection, single-point crossover and Gaussian mutation operations are performed to generate a new generation of populations. The iterative process is repeated until the change in the optimal solution for 50 consecutive generations is less than 1% or the maximum number of iterations is 1000 generations. The individual with the highest fitness is selected from the final population as the optimal solution to obtain the optimal grain size distribution parameters that balance various performance indicators and obtain the optimal microstructure. The mapping relationship between the heat treatment process parameters and the grain size distribution parameters is established through multivariate regression, and the corresponding heat treatment temperature, time and cooling rate are obtained by reverse calculation. In practical applications, for the application scenario of automobile suspension springs, the strength target value is set to 1200MPa, the hardness target value is set to 400HV, and the toughness target value is set to 50J / cm2, with weights of 0.5, 0.3 and 0.2 respectively. Construct a comprehensive performance evaluation function

[0036] F = 0.5 * (σ / 1200) + 0.3 * (H / 400) + 0.2 * (K / 50). The genetic algorithm initializes a population of 100 individuals, each of which contains an average grain size of 10-50 μm, a standard deviation of 1-10 μm, a roundness average of 0.6-1.0, and a roundness standard deviation of 0.05-0.2. The crossover probability is set to 0.8 and the mutation probability is set to 0.1. The multi-objective optimization model uses the previously trained random forest algorithm to predict performance indicators based on grain parameters and calculate fitness values. Individuals are selected using the roulette wheel selection method, and single-point crossover and Gaussian mutation are performed, such as a mean of 0 and a standard deviation of 5% of the parameter range. During the iteration process, the optimal solution gradually improves from the initial F = 0.85 to F = 0.98 at the 782nd generation, and the change amplitude is less than 1% for 50 consecutive generations, meeting the convergence conditions. Finally, the optimal grain parameters were obtained as average size 28μm, standard deviation 4μm, roundness average value 0.85, and roundness standard deviation 0.08. The multivariate regression model y=β0+β1x1+β2x2+β3x3 was used for reverse calculation, where y is the grain parameter and x is the heat treatment parameter. The optimal heat treatment process was obtained as temperature 940℃, time 40 minutes, and cooling rate 15℃ / s.

[0037] Step S106, reverse design the heat treatment process route according to the optimal microstructure, use finite element software to simulate the temperature field and stress field distribution during the heat treatment process, and determine the control parameters of heating, insulation and cooling through iterative optimization.

[0038] According to the optimal microstructure parameters, an initial heat treatment process route is obtained from the heat treatment process database, and the initial heat treatment process route includes a heating rate, a holding temperature, a holding time and a cooling rate; the geometric model and mesh division of the heat treatment process are established by using the initial heat treatment process route to obtain the temperature field and the stress field distribution; for the temperature field and the stress field distribution, the root mean square error with the target microstructure is calculated by the gradient descent method; if the root mean square error is greater than a preset threshold, the heat treatment process parameters are adjusted, and the finite element simulation is repeated until the root mean square error is less than the preset threshold; a heating curve, a holding curve and a cooling curve are generated according to the optimized heat treatment process parameters to obtain a heat treatment process control scheme; a small batch test is carried out using the heat treatment process control scheme to obtain an actual microstructure; it is determined whether the deviation between the actual microstructure and the target microstructure exceeds the preset threshold, and if so, the heat treatment process parameters are returned to be adjusted.

[0039] Exemplarily, according to the optimal microstructure parameters, the heat treatment process database constructed based on experimental data and empirical models is used to reversely query the initial heat treatment process route, including heating rate, holding temperature, holding time and cooling rate, as the initial input parameters of finite element simulation. ANSYS software is used to establish the geometric model and meshing of the heat treatment process, set material physical parameters such as thermal conductivity, specific heat capacity, density, boundary conditions and initial conditions, and simulate the temperature field and stress field distribution in the heating, holding and cooling stages. The difference between the simulation results and the target microstructure is compared by the gradient descent method, the heat treatment process parameters are adjusted, and the finite element simulation is repeated until the root mean square error of the temperature field and stress field distribution and the target microstructure is less than 5%. According to the optimized heat treatment process parameters, a piecewise function is used to generate accurate heating curves, holding curves and cooling curves to form a complete heat treatment process control scheme. Through small batch tests and microstructure analysis, the actual effect of the optimized heat treatment process is verified, and the actual microstructure obtained is compared with the target structure. If the deviation exceeds the preset threshold, the optimization step is returned to iterate again. In practical applications, for a high-strength low-alloy steel, the optimal microstructural parameters are an average grain size of 30 μm and a standard deviation of 5 μm. Using a heat treatment process database containing 1000 sets of experimental data, the initial heat treatment process route is obtained by reverse query through interpolation algorithm, which is a heating rate of 5 ° C / s, a holding temperature of 900 ° C, a holding time of 30 minutes, and a cooling rate of 10 ° C / s. A 100 × 100 × 20 mm sample geometry model is established in ANSYS, and hexahedral mesh division is used with 50000 units. The material physical parameters are set as thermal conductivity 45 W / (m·K), specific heat capacity 480 J / (kg·K), and density 7850 kg / m2. The boundary conditions include convective heat transfer coefficient 20 W / (m2·K) and radiation emissivity 0.8. The transient thermal analysis module is used to simulate the heat treatment process, with a time step of 1 s and a total simulation time of 3600 s. The simulation results show that the maximum temperature reaches 898℃ and the maximum thermal stress is 250MPa. The parameters are optimized by gradient descent method, the learning rate is set to 0.01, and after 50 iterations, the root mean square error of the temperature field and stress field distribution and the target microstructure is reduced to 4.8%. The optimized process parameters are heating rate 5.5℃ / s, holding temperature 910℃, holding time 35 minutes, and cooling rate 9.5℃ / s. Cubic spline interpolation is used to generate an accurate control curve of 1800 discrete points. In a small batch test, 10 specimens were heat treated, and the metallographic analysis results showed that the average grain size was 29.5μm, the standard deviation was 5.2μm, and the deviation from the target structure was within 3%, which verified the effectiveness of the optimization process.

[0040] Step S107, implement the optimized heat treatment process on the industrial production line, use the online monitoring system to collect the temperature, stress and displacement of the workpiece in real time, and analyze it through the edge computing device. If an abnormality is detected, adjust the process parameters of the heat treatment equipment to achieve closed-loop control.

[0041] Acquire real-time data of the workpiece, the real-time data is collected by temperature sensors, stress sensors and displacement sensors installed on the workpiece on the industrial production line; use the isolation forest algorithm to calculate the deviation between the current process parameter and the set value based on the real-time data; determine whether the deviation exceeds a preset threshold, if the deviation exceeds the preset threshold, it is determined to be abnormal information; for the abnormal information, generate a process parameter adjustment instruction according to a pre-established adjustment rule library; send the process parameter adjustment instruction to the control system of the heat treatment equipment, after receiving the process parameter adjustment instruction, the control system of the heat treatment equipment updates at least one process parameter of the heating power, the holding time or the cooling rate within the set adjustment amplitude limit and the continuous adjustment number limit; acquire the updated process parameters of the heat treatment equipment, calculate the control accuracy and response time; and perform performance evaluation on the closed-loop control system according to the control accuracy and response time.

[0042] For example, three temperature sensors, two stress sensors and one displacement sensor are installed on each workpiece on an industrial production line to build a real-time data acquisition network, and the collected workpiece temperature, stress and deformation data are transmitted to the edge computing device through a high-speed industrial bus. The edge computing device uses the isolation forest algorithm to quickly analyze the collected real-time data, calculate the deviation between the current process parameters and the set values, and determine it as an abnormal situation when the deviation exceeds the preset threshold. If an abnormality is detected, the edge computing device automatically generates a process parameter adjustment instruction based on the adjustment rule library based on expert experience and historical data analysis, and sends the instruction to the control system of the heat treatment equipment through the control interface. After receiving the adjustment instruction, the heat treatment equipment updates the process parameters such as heating power, holding time or cooling rate in real time within the set adjustment range limit and continuous adjustment number limit, realizes closed-loop control of the production process, and feeds the adjusted parameters back to the monitoring system for recording and analysis. By calculating indicators such as control accuracy and response time, the performance of the entire closed-loop control system is evaluated, and the abnormality detection algorithm and adjustment rule library are optimized according to the evaluation results to continuously improve the control effect of the system. In practical applications, three K-type thermocouples are installed on each workpiece surface with a measuring range of -200 to 1300°C and an accuracy of ±1.5°C, and two strain gauges are installed.

[0043] ±3000με, accuracy 0.1%, 1 laser displacement sensor such as range 0-100mm, resolution 0.01mm. The EtherCAT bus is used to transmit data to the edge computing device with a transmission rate of 100Mbps. The edge computing device uses the isolation forest algorithm such as the number of trees is 100, the sub-sampling size is 256, and it is judged as abnormal when the abnormal score exceeds 0.6. The adjustment rule base contains 50 rules, such as "If the temperature deviation is > +10℃, reduce the heating power by 5%." The PID controller of the heat treatment equipment receives the adjustment instruction and adjusts it continuously for up to 3 times within the adjustment range of ±10%. The adjustment accuracy of the heating power is 0.1kW, the adjustment accuracy of the holding time is 1 second, and the adjustment accuracy of the cooling rate is 0.1℃ / s. The monitoring system records the adjustment parameters every 10 seconds, calculates the average control deviation if the target is <±5℃ and the response time if the target is <30 seconds. If the average control deviation of 100 consecutive batches is less than ±3°C and the response time is less than 20 seconds, the anomaly detection threshold is increased by 0.05 and the parameter value in the adjustment rule base is updated.

[0044] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for analyzing surface differences of metal materials, characterized in that: The method comprises: According to the chemical composition and initial microstructure of the material, the thermodynamic model is used to calculate the Gibbs free energy at different temperatures. The time-temperature-transformation curve of the material is drawn by phase diagram analysis software to obtain the grain growth kinetic parameters and recrystallization temperature range. According to the grain growth kinetics parameters and recrystallization temperature range, a quantitative relationship model between grain size and heat treatment process parameters is established. The quantitative relationship model is trained by a neural network algorithm, and the target grain size range is input to reversely deduce the corresponding heat treatment temperature, time and cooling rate. According to the derived heat treatment process parameters, heat treatment is carried out through the pilot furnace control system, the sample microstructure is observed with a metallographic microscope, the grain size distribution is measured using image analysis software, and statistical analysis is performed to determine whether the actual grain size is within the target range; If the actual grain size is within the target range, the mechanical properties of the heat-treated specimens are tested, including tensile, impact and hardness. The data mining algorithm is used to analyze the correlation between grain size and strength, hardness and toughness, and a multi-objective optimization model of material properties and microstructure is established. By using the established multi-objective optimization model, inputting data information on strength, hardness and toughness under different application scenarios, searching for the best grain size distribution through genetic algorithms, the optimal microstructure that balances various performance indicators is obtained; According to the optimal microstructure, the heat treatment process route is reversely designed, the temperature field and stress field distribution during the heat treatment process are simulated using finite element software, and the control parameters of heating, insulation and cooling are determined through iterative optimization; The optimized heat treatment process is implemented on the industrial production line, and an online monitoring system is used to collect the temperature, stress and displacement of the workpiece in real time. The workpiece is analyzed through edge computing equipment. If an abnormality is detected, the process parameters of the heat treatment equipment are adjusted to achieve closed-loop control.

2. The method according to claim 1, wherein: The method uses a thermodynamic model to calculate the Gibbs free energy at different temperatures based on the chemical composition and initial microstructure of the material, and uses a phase diagram analysis software to draw the time, temperature, and transformation curves of the material to obtain the grain growth kinetic parameters and the recrystallization temperature range, including: The CALPHAD method is used to perform thermodynamic calculations, wherein the thermodynamic calculations calculate the Gibbs free energy at different temperatures based on the chemical composition and initial microstructure data of the material to obtain the volume fraction and chemical composition of each phase; According to the thermodynamic calculation results, the continuous cooling transformation diagram and isothermal transformation diagram of the material are drawn using JMatPro software to obtain the phase change starting temperature and phase change completion temperature; For the continuous cooling transformation diagram and isothermal transformation diagram, Johnson-Mehl-Avrami- Kolmogorov equation is used to analyze the time-temperature-transformation curve, calculate the grain growth index and rate constant, and determine the activation energy of grain growth; If the grain growth kinetic parameters are obtained, the recrystallization starting temperature and completion temperature are determined by calculating the grain growth rate at different temperatures, and the recrystallization temperature range of the material is divided; According to the recrystallization temperature range, heat treatment process parameters are designed, wherein the heat treatment process parameters include heating rate and holding time, and the grain size distribution under different heat treatment process conditions is calculated.

3. The method according to claim 1, wherein: According to the grain growth kinetics parameters and the recrystallization temperature range, a quantitative relationship model between the grain size and the heat treatment process parameters is established, the quantitative relationship model is trained by a neural network algorithm, the target grain size range is input, and the corresponding heat treatment temperature, time and cooling rate are reversely derived, including: Obtain grain growth kinetic parameters and recrystallization temperature range, including activation energy and frequency factor; Constructing an Arrhenius equation according to the grain growth kinetics parameters and the recrystallization temperature range; The Levenberg-Marquardt algorithm is used to perform nonlinear regression on the Arrhenius equation to obtain a quantitative relationship function between grain size and heat treatment temperature, time and cooling rate; Collecting experimental data, the experimental data including grain size measurement results under different alloy compositions, heat treatment temperatures, times and cooling rates; Dividing the experimental data into a training set and a validation set in proportion; Setting a multi-layer perceptron neural network, wherein the multi-layer perceptron neural network comprises an input layer, a hidden layer and an output layer; The multi-layer perceptron neural network is trained using an Adam optimizer and a mean square error loss function to obtain a trained quantitative relationship model; Preset target grain size range; If the target grain size range is a preset value, a genetic algorithm is used for reverse calculation; The fitness function of the genetic algorithm is to minimize the error between the predicted grain size and the target size; Obtaining a heat treatment temperature, time, and cooling rate parameter combination through the genetic algorithm; The parameter combination is used to generate grains within the target grain size range.

4. The method according to claim 1, wherein: According to the derived heat treatment process parameters, heat treatment is performed in a small test furnace control system, the sample microstructure is observed using a metallographic microscope, the grain size distribution is measured using image analysis software, and statistical analysis is performed to determine whether the actual grain size is within the target range, including: According to the heat treatment process parameters, the PID control algorithm is used to set the heat treatment temperature, holding time and cooling rate of the pilot furnace control system; Performing heat treatment on the sample by the small test furnace control system to obtain the heat-treated sample; Using an automatic grinding and polishing machine to cut, inlay, grind and polish the heat-treated sample; If the polishing accuracy reaches a preset threshold, a sample for metallographic observation is obtained; Scanning the surface of the sample for metallographic observation by using an automatic scanning system of a metallographic microscope; Processing the scan results by a scale-invariant feature transformation algorithm to generate a microscopic tissue panorama; Using image analysis software to perform image segmentation and grain identification on the microstructure panorama; The watershed algorithm was used for image segmentation, and the roundness and area threshold methods were used to identify grains and measure the grain size distribution; calculating the average grain size and standard deviation of the grain size distribution by statistical analysis; If the average grain size is within the target range and the standard deviation is less than a preset percentage of the target range, the sample for metallographic observation is determined to be qualified.

5. The method according to claim 1, wherein: If the actual grain size is within the target range, the mechanical properties of the heat-treated specimens are tested, including tensile, impact and hardness. The data mining algorithm is used to analyze the correlation between grain size and strength, hardness and toughness, and a multi-objective optimization model of material properties and microstructure is established, including: Obtain the heat-treated specimens and prepare tensile, impact and hardness test specimens according to ASTM standards; Testing the sample to obtain mechanical property data such as yield strength, tensile strength, elongation, impact toughness and Vickers hardness; The metallographic structure image of the sample is processed by image analysis software to extract microstructural characteristic parameters such as average grain size, grain size distribution standard deviation and grain shape factor; Based on the mechanical property data and microstructural characteristic parameters, the correlation between grain size and strength, hardness and toughness is analyzed by random forest algorithm; If the number of trees is the preset value and the minimum number of leaf node samples is the preset value, cross validation is used to select the best parameters; According to the correlation and influence weights, a multi-objective optimization model of material properties and microstructure is constructed; The objective function is set to maximize strength and toughness while minimizing hardness deviation; Genetic algorithm is used to find the best combination of heat treatment process parameters and obtain optimized microstructure parameters; According to the optimized microstructure parameters, the corresponding heat treatment temperature, time and cooling rate are obtained by reverse calculation.

6. The method according to claim 1, wherein: The established multi-objective optimization model is used to input data information on strength, hardness and toughness in different application scenarios, and the optimal grain size distribution is searched through a genetic algorithm to obtain the optimal microstructure that balances various performance indicators, including: According to the target value and weight set in the application scenario, the comprehensive performance evaluation function F is constructed, wherein F=w1*(σ / σ_target)+w2*(H / H_target)+w3*(K / K_target), σ, H, K represent strength, hardness and toughness respectively, w1, w2 and w3 are corresponding weights, and target is the target value; Initializing the population of grain size distribution parameters using a genetic algorithm, including average grain size, standard deviation, roundness average value, and roundness standard deviation; The fitness value of each individual is calculated by the multi-objective optimization model, and a new generation of population is generated by roulette wheel selection, single point crossover and Gaussian mutation operations; If the change range of the optimal solution for several consecutive generations is less than the preset threshold or reaches the maximum number of iterations, the individual with the highest fitness is selected from the final population as the optimal solution; Obtain the best grain size distribution parameters to balance various performance indicators and obtain the optimal microstructure; The heat treatment temperature, time and cooling rate are obtained by reverse calculation according to the pre-established multivariate regression mapping relationship between the heat treatment process parameters and the grain size distribution parameters.

7. The method according to claim 1, wherein: According to the optimal microstructure, the heat treatment process route is reversely designed, the temperature field and stress field distribution during the heat treatment process are simulated by finite element software, and the control parameters of heating, insulation and cooling are determined by iterative optimization, including: Acquire an initial heat treatment process route from the heat treatment process database according to the optimal microstructure parameters, wherein the initial heat treatment process route includes a heating rate, a holding temperature, a holding time, and a cooling rate; The initial heat treatment process route is used to establish a geometric model and mesh division of the heat treatment process to obtain the temperature field and stress field distribution; For the temperature field and the stress field distribution, calculating the root mean square error with the target microstructure by gradient descent method; If the root mean square error is greater than a preset threshold, adjusting the heat treatment process parameters, and repeating the finite element simulation until the root mean square error is less than the preset threshold; Generate a heating curve, a heat preservation curve and a cooling curve according to the optimized heat treatment process parameters to obtain a heat treatment process control scheme; The heat treatment process control scheme is used to conduct small batch tests to obtain the actual microstructure; It is determined whether the deviation between the actual microstructure and the target microstructure exceeds a preset threshold, and if so, the heat treatment process parameters are adjusted.

8. The method according to claim 1, wherein: The optimized heat treatment process is implemented on the industrial production line, and the temperature, stress and displacement of the workpiece are collected in real time by an online monitoring system, and analyzed by edge computing equipment. If an abnormality is detected, the process parameters of the heat treatment equipment are adjusted to achieve closed-loop control, including: Acquire real-time data of the workpiece, which is collected by temperature sensors, stress sensors and displacement sensors installed on the workpiece on the industrial production line; Based on the real-time data, an isolation forest algorithm is used to calculate the deviation between the current process parameter and the set value; Determine whether the deviation exceeds a preset threshold, and if the deviation exceeds the preset threshold, determine it as abnormal information; In response to the abnormal information, a process parameter adjustment instruction is generated according to a pre-established adjustment rule library; The process parameter adjustment instruction is sent to the control system of the heat treatment equipment, and after receiving the process parameter adjustment instruction, the control system of the heat treatment equipment updates at least one process parameter of the heating power and the holding time or the cooling rate within the set adjustment amplitude limit and the continuous adjustment number limit; Obtaining updated process parameters of the heat treatment equipment, and calculating control accuracy and response time; The performance of the closed-loop control system is evaluated based on the control accuracy and response time.

Images