A p30 high-speed steel band saw blade full life cycle quality prediction method based on digital twinning

Abstract

This invention discloses a method for predicting the quality of P30 high-speed steel band saw blades throughout their entire lifecycle based on digital twins. The method falls within the field of product lifecycle quality prediction and includes: collecting multi-point temperature data and cooling medium flow parameters within the furnace using a sensor array to obtain an initial temperature distribution map and cooling rate curve; adjusting simulation parameters to match the actual alloy element distribution if the initial state of microstructure evolution exceeds a preset threshold, thus obtaining a corrected stress distribution map; acquiring carbide precipitation behavior determined by a neural network model and updating the finite element simulation with real-time monitoring data to obtain a prediction of dynamic hardness uniformity; generating cooling rate adjustment instructions using a process intervention threshold determined by a support vector machine model to obtain an optimized microstructure evolution path; and extracting final stress distribution data from the optimized microstructure evolution path to determine hardness uniformity and defect formation probability, thereby obtaining a stable product quality control scheme.

Description

Technical Field

[0001] This invention belongs to the field of product lifecycle quality prediction, and in particular relates to a method for predicting the full lifecycle quality of P30 high-speed steel band saw blades based on digital twins. Background Technology

[0002] As a critical metal cutting tool, the performance and service life of high-speed steel band saw blades are highly dependent on the quality of the heat treatment process. Currently, the industry's commonly used heat treatment quality control methods mainly rely on offline sampling and testing, and adjustments based on historical experience formulas. These methods assess quality by performing hardness tests or metallographic analyses on samples after production, essentially representing a post-production inspection and reactive remediation approach. Because the heat treatment process involves the complex coupling of temperature fields, microstructure transformations, and stress fields, and because fluctuations in furnace temperature uniformity and cooling medium fluidity occur in actual production, differences in the internal microstructure and properties of products from the same batch can exist. Existing offline methods struggle to achieve real-time sensing and proactive control during the process.

[0003] However, the aforementioned existing technologies have significant limitations. First, offline detection cannot acquire dynamic data on the internal temperature, microstructure, and stress changes of the material during heat treatment, leaving the black box relationship between process parameters and final performance unsolved. Second, empirical adjustments lack precise quantitative basis, making it impossible to accurately predict and intervene in microstructural differences (such as incomplete martensite transformation and abnormal carbide precipitation) caused by uneven distribution of alloying elements in micro-regions and localized abnormal cooling rates, as well as the resulting defects such as internal residual stress concentration, uneven hardness distribution, and even microcrack initiation. Therefore, existing technologies struggle to achieve real-time accurate prediction and closed-loop adaptive optimization of the microstructure and properties of band saw blades during heat treatment, hindering further improvements in product quality stability. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for predicting the entire lifecycle quality of P30 high-speed steel band saw blades based on digital twins, comprising: The initial temperature distribution map and cooling rate curve are obtained by collecting temperature data and cooling medium flow parameters at multiple points inside the furnace through a sensor array. Based on the initial temperature distribution diagram and cooling rate curve, the internal heat conduction process of the saw blade is calculated using the finite element simulation method to determine the initial state of microstructure evolution and the preliminary value of stress distribution. If the initial state of the microstructure evolution exceeds the preset threshold, the simulation parameters are adjusted to match the actual alloy element distribution to obtain a corrected stress distribution map. Key node data are extracted from the modified stress distribution map, and a neural network model is used to predict the degree of martensite transformation and the content of retained austenite based on the key node data in order to determine the carbide precipitation behavior. The carbide precipitation behavior determined by the neural network model is obtained, and the finite element simulation is updated in combination with real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution. If the dynamic hardness uniform distribution prediction indicates an increased risk of internal defects, the quality fluctuation trend is analyzed using a support vector machine model to determine the process intervention threshold. Using the process intervention threshold determined by the support vector machine model, a cooling rate adjustment command is generated to obtain an optimized microstructure evolution path. The final stress distribution data is extracted from the optimized tissue evolution path, and the hardness uniformity and defect formation probability are determined based on the final stress distribution data to obtain a stable product quality control scheme.

[0005] Optionally, the step of acquiring multi-point temperature data and cooling medium flow parameters within the furnace through a sensor array to obtain an initial temperature distribution map and a cooling rate curve includes: The sensor array acquires temperature data at multiple points inside the furnace and the flow parameters of the cooling medium. An initial temperature distribution map is generated based on the temperature data; The temperature sequence of each point is extracted from the initial temperature distribution map, and the cooling rate sequence is calculated based on the temperature difference between adjacent time points; A cooling rate curve is plotted based on the cooling rate sequence.

[0006] Optionally, the step of calculating the internal heat conduction process of the saw blade using the finite element simulation method based on the initial temperature distribution diagram and cooling rate curve, and determining the initial state of microstructure evolution and preliminary stress distribution values, includes: Based on the initial temperature distribution map, a thermal field model of the saw blade is constructed, and the heat conduction process is meshed and numerically calculated using finite element simulation technology to obtain preliminary results of the thermal field distribution. Based on the preliminary results of the thermal field distribution and the cooling rate curve, the variation law of temperature gradient with time is analyzed to determine the key time nodes in the tissue evolution process. For the key time points, the finite element method was used to dynamically simulate the heat conduction process inside the saw blade to obtain the initial state parameters of the tissue evolution. Based on the initial state parameters of the tissue evolution, the stress distribution in different regions inside the saw blade is calculated to obtain preliminary stress distribution values.

[0007] Optionally, if the initial state of microstructure evolution exceeds a preset threshold, adjusting the simulation parameters to match the actual alloy element distribution to obtain a corrected stress distribution map includes: Acquire initial state data of organizational evolution and determine whether the initial state data exceeds a preset threshold; If the deviation exceeds the limit, the simulation parameters are adjusted based on the deviation between the initial state data and the preset threshold to obtain the adjusted parameter configuration. Based on the adjusted parameter configuration and the actual alloy element distribution data, the corresponding simulation results are generated. Based on the comparison and analysis between the simulation results and the actual alloy properties, the stress calculation model is modified, and a modified stress distribution map is generated.

[0008] Optionally, the step of extracting key node data from the corrected stress distribution map and using a neural network model to predict the degree of martensite transformation and the content of retained austenite based on the key node data to determine carbide precipitation behavior includes: Obtain the stress values ​​at key node locations in the corrected stress distribution map; The stress values ​​are input into a pre-trained neural network model to obtain the corresponding values ​​of martensite transformation degree and retained austenite content. Based on whether the value of the degree of martensite transformation is higher than a first preset threshold, the region where significant martensite transformation occurs is determined; The carbide precipitation behavior is determined based on whether the residual austenite content is lower than a second preset threshold and in conjunction with the region where significant martensitic transformation occurs.

[0009] Optionally, the step of obtaining the carbide precipitation behavior determined by the neural network model and updating the finite element simulation in conjunction with real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution includes: Obtain the amount and location distribution of carbide precipitation as determined by the neural network model; Based on the amount and location distribution of carbide precipitation, the phase transition kinetic parameters in the finite element model are corrected to obtain an updated finite element simulation model. Real-time temperature and stress field data are collected by real-time monitoring equipment to obtain real-time temperature and stress sequences. The real-time temperature sequence and stress sequence are input into the updated finite element simulation model, and incremental calculations are performed to obtain the current hardness field distribution and dynamic hardness uniform distribution prediction.

[0010] Optionally, the step of using a support vector machine model to determine the process intervention threshold, generating a cooling rate adjustment command, and obtaining an optimized tissue evolution path includes: Obtain the original cooling process parameter sequence; The classification hyperplane is obtained by training the tissue state samples to classify them using a support vector machine model. The distance from the current tissue state to the intervention boundary is calculated based on the classification hyperplane; If the distance is less than a preset safety margin, a cooling speed adjustment command is generated; The original cooling process parameter sequence is modified according to the cooling rate adjustment command, and the microstructure evolution simulation calculation is performed on the adjusted cooling process parameter sequence to obtain the optimized microstructure evolution path.

[0011] Optionally, the step of extracting final stress distribution data from the optimized microstructure evolution path and determining hardness uniformity and defect formation probability based on the final stress distribution data to obtain a stable product quality control scheme includes: Obtain the final stress distribution data in the optimized tissue evolution path; The final stress distribution data is spatially discretized to obtain a set of discrete stress points, and the hardness value corresponding to each discrete stress point is calculated. The uniformity of hardness distribution is determined based on the statistical characteristics of each hardness value; The discrete stress point set is classified using a support vector machine to obtain high-risk and low-risk defect regions. The probability of defect formation is determined based on the characteristics of stress concentration points within the high-risk defect area. Based on the uniformity of hardness distribution and the probability of defect formation, a stable product quality control scheme is generated.

[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects: The technical solution provided by this invention integrates real-time data acquisition from sensor arrays, finite element numerical simulation, neural network intelligent prediction, and support vector machine decision analysis to construct a multi-source information fusion and dynamic model interaction mechanism for the entire heat treatment process. This method achieves complete closed-loop control, from initial temperature field construction and determination of the initial state of microstructure evolution, to dynamic judgment of carbide precipitation behavior and online prediction of hardness uniformity, and then to defect risk warning and adaptive optimization of process parameters. Its beneficial effects lie in overcoming the lag and blindness of existing offline detection and experience-based adjustments. It can accurately predict the evolution trend of the internal microstructure and properties of materials in real time and actively intervene in the process, effectively controlling key parameters such as cooling rate. This significantly improves the hardness distribution uniformity of band saw blades, greatly reduces the risk of internal defects caused by uncontrolled residual stress and abnormal carbide precipitation, and ultimately forms a stable and reliable product quality control scheme, ensuring the consistency and reliability of heat-treated product performance. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0018] Example 1 This embodiment provides a method for predicting the quality of P30 high-speed steel band saw blades throughout their entire lifecycle based on digital twins, including: The initial temperature distribution map and cooling rate curve are obtained by collecting temperature data and cooling medium flow parameters at multiple points inside the furnace through a sensor array. Based on the initial temperature distribution diagram and cooling rate curve, the finite element simulation method is used to calculate the internal heat conduction process of the saw blade, and to determine the initial state of microstructure evolution and the preliminary value of stress distribution. If the initial state of the microstructure evolution exceeds the preset threshold, the simulation parameters are adjusted to match the actual alloy element distribution to obtain a corrected stress distribution map. Key node data were extracted from the revised stress distribution map, and a neural network model was used to predict the degree of martensite transformation and the content of retained austenite to determine the carbide precipitation behavior. The carbide precipitation behavior determined by the neural network model is obtained, and the finite element simulation is updated by combining real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution. If the dynamic hardness uniform distribution prediction indicates the risk of internal defects, then the quality fluctuation trend is analyzed by using a support vector machine model to determine the process intervention threshold. The process intervention threshold determined by the support vector machine model is used to generate cooling rate adjustment instructions and obtain the optimized microstructure evolution path. The final stress distribution data is extracted from the optimized microstructure evolution path to determine the hardness uniformity and the probability of defect formation, thereby obtaining a stable product quality control scheme.

[0019] Furthermore, the step of acquiring multi-point temperature data and cooling medium flow parameters within the furnace through a sensor array to obtain an initial temperature distribution map and a cooling rate curve includes: The sensor array acquires temperature data at multiple points inside the furnace and the flow parameters of the cooling medium. An initial temperature distribution map is generated based on the collected temperature data; Extract the temperature sequence of each point from the initial temperature distribution map; The cooling rate sequence is obtained by calculating the temperature difference between adjacent time points for each temperature sequence. Plot the cooling rate curve based on the cooling rate sequence; Obtain the maximum cooling rate value at each point from the cooling rate curve; If the maximum cooling rate at a certain point exceeds a preset threshold, that point is marked as an abnormal cooling area.

[0020] Furthermore, based on the initial temperature distribution diagram and cooling rate curve, the finite element simulation method is used to calculate the internal heat conduction process of the saw blade, determine the initial state of microstructure evolution and preliminary stress distribution values, including: By collecting initial temperature and distribution data, a thermal field model inside the saw blade is constructed. Finite element simulation technology is used to perform mesh generation and numerical calculation of the heat conduction process, and preliminary results of the thermal field distribution are obtained. Based on the preliminary results of the thermal field distribution, combined with the data of cooling rate and change curve, the variation law of temperature gradient over time is analyzed to determine the key time nodes in the tissue evolution process. For data from key time points, the finite element method is applied to dynamically simulate the heat conduction process inside the saw blade to obtain the initial state parameters of the microstructure evolution. Starting from the initial state parameters of the tissue evolution, the stress distribution in different regions inside the saw blade is calculated to determine the location of stress concentration areas; If the stress concentration value in the region exceeds the preset threshold, the heat conduction process in that region will be locally meshed and the detailed stress distribution data will be recalculated. By analyzing detailed data and integrating the correlation between heat conduction processes and stress distribution, the final preliminary numerical results were determined.

[0021] Furthermore, if the initial state of microstructure evolution exceeds a preset threshold, the simulation parameters are adjusted to match the actual alloy element distribution to obtain a corrected stress distribution map, including: Acquire initial state data of organizational evolution, and determine whether the initial state exceeds a preset threshold by preprocessing the collected data; If the initial state exceeds the preset threshold, the simulation parameters are adjusted based on the deviation between the initial state data and the preset threshold to obtain the adjusted parameter configuration. Based on the adjusted parameter configuration and the element distribution data of the actual alloy, corresponding simulation results are generated to determine the degree of matching of the element distribution. By comparing and analyzing the simulation results with the actual alloy properties, the key differences in element distribution are extracted, and the distribution characteristics of the difference data are obtained. By using the distribution characteristics of the differential data, the stress calculation model is corrected to generate a preliminary stress distribution graph; Based on the preliminary stress distribution diagram and the constraints of the stress analysis, the details of the diagram are optimized to obtain the final corrected stress distribution diagram.

[0022] Furthermore, the step of extracting key node data from the corrected stress distribution map, using a neural network model to predict the degree of martensite transformation and the content of retained austenite, and determining carbide precipitation behavior includes: Obtain the stress values ​​at key node locations in the corrected stress distribution map; The above stress values ​​are received by a pre-trained neural network model to obtain the values ​​of martensite transformation degree and retained austenite content. Based on the fact that the value of the degree of martensitic transformation is higher than a preset threshold, the region where significant martensitic transformation has occurred is identified; If the residual austenite content is lower than the preset threshold, it is determined that the tendency for carbide precipitation in this region is enhanced. Statistical methods were used to count the number of regions that satisfy significant martensitic transformation and have low residual austenite content, thus obtaining a set of high precipitation risk regions. Extract the corresponding location coordinates and stress values ​​from the set of high-extraction-risk areas to construct an extraction behavior distribution dataset; By analyzing the precipitation behavior distribution dataset, we can determine the concentrated areas of carbide precipitation locations and arrive at the final conclusion on the precipitation behavior.

[0023] Furthermore, the step of obtaining the carbide precipitation behavior determined by the neural network model, and updating the finite element simulation with real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution includes: The neural network model is used to judge the carbide precipitation behavior under the current process conditions, and the amount and location distribution of carbide precipitation are obtained. Based on the amount and location distribution of carbide precipitation, the phase transition kinetic parameters in the finite element model are corrected to obtain the updated finite element simulation model. By collecting temperature and stress field data at the current moment through real-time monitoring equipment, the temperature and stress sequences of the actual process can be obtained. Input the temperature and stress sequences into the updated finite element simulation model, perform incremental calculations, and obtain the hardness field distribution at the current moment; Determine the difference between the current hardness field distribution and the previous hardness field distribution. If the difference exceeds a preset threshold, trigger the model parameters to be corrected again; otherwise, maintain the current simulation model. A neural network model is used to make a secondary prediction of the latest hardness field distribution to obtain the hardness change trend in the next time step; The boundary conditions of the finite element simulation for the next time step are adjusted according to the hardness change trend to obtain a continuously updated dynamic hardness distribution prediction sequence.

[0024] Furthermore, if the dynamic hardness uniform distribution prediction indicates an internal defect risk, then the quality fluctuation trend is analyzed using a support vector machine model to determine the process intervention threshold, including: Obtain dynamic hardness prediction results; If the dynamic hardness prediction results indicate an increased risk of internal defects, then extract the quality data for the current batch. The extracted quality data is trained and analyzed using a support vector machine to obtain the trend of quality changes. Calculate the sensitivity sequence of each process parameter to defect risk based on the quality change trend; For process parameters that rank high in the sensitivity sequence, the corresponding intervention threshold is determined using preset rules; If the current process parameters exceed the intervention threshold, an adjustment command is generated and output. The adjustment instructions are transmitted to the execution system to complete the correction of process parameters.

[0025] Furthermore, the process intervention threshold determined using a support vector machine model is used to generate a cooling rate adjustment command to obtain an optimized tissue evolution path, including: Obtain the original cooling process parameter sequence; The classification hyperplane is obtained by training the tissue state samples to classify them using a support vector machine model. The distance from the current tissue state to the intervention boundary is calculated using a classification hyperplane; If the distance from the current organizational state to the intervention boundary is less than the preset safety margin, a cooling rate adjustment command is generated. Modify the original cooling process parameter sequence according to the cooling rate adjustment command to obtain the adjusted cooling process parameter sequence; A microstructure evolution simulation calculation was performed on the adjusted cooling process parameter sequence to obtain a new microstructure state sequence. The final organizational morphology parameters are extracted from the new organizational state sequence to obtain the optimized organizational evolution path.

[0026] Furthermore, the step of extracting the final stress distribution data from the optimized microstructure evolution path, determining hardness uniformity and defect formation probability, and obtaining a stable product quality control scheme includes: Obtain the final stress distribution data in the optimized microstructure evolution path; The stress distribution data is spatially discretized using a mesh generation method to obtain a set of discrete stress points. For a set of discrete stress points, calculate the hardness value at each point to obtain the set of hardness value points; The uniformity of hardness distribution is determined by the standard deviation of the hardness values ​​at each point in the set of hardness values. If the standard deviation is less than a preset threshold, the hardness distribution is considered uniform. If the standard deviation is greater than or equal to the preset threshold, the hardness distribution is considered non-uniform. Support vector machines are used to classify the discrete stress point set to obtain high-risk and low-risk defect regions; The probability of defect formation is determined based on the number of stress concentration points and the rate of change of stress gradient within the high-risk defect area. If the probability of defect formation is lower than the preset probability threshold, output the stable quality control scheme under the current process parameters; if the probability of defect formation is higher than or equal to the preset probability threshold, adjust the parameters of the key temperature control nodes in the microstructure evolution path. The adjusted organizational evolution path is re-acquired, and the process of extracting stress distribution data to determine the probability of defect formation is repeated to obtain an updated stable quality control scheme.

[0027] Example 2 like Figure 1 As shown, this embodiment provides a method for predicting the quality of P30 high-speed steel band saw blades throughout their entire lifecycle based on digital twins, including: S101. Collect temperature data and cooling medium flow parameters at multiple points inside the furnace through a sensor array to obtain an initial temperature distribution map and a cooling rate curve.

[0028] Temperature data and cooling medium flow parameters at multiple points within the furnace are acquired using a sensor array. An initial temperature distribution map is generated based on the collected temperature data. Temperature sequences for each point are extracted from the initial temperature distribution map. The cooling rate sequence is obtained by calculating the temperature difference between adjacent time points for each temperature sequence. A cooling rate curve is plotted based on the cooling rate sequence. The maximum cooling rate value for each point is obtained from the cooling rate curve. If the maximum cooling rate value at a point exceeds a preset threshold, that point is marked as an abnormal cooling area.

[0029] A sensor array was used to collect temperature data and cooling medium flow parameters at multiple points within the furnace, constructing an initial temperature distribution map and cooling rate curve. First, 100 high-precision temperature sensors, positioned at key locations within the furnace, collected temperature data every 5 seconds. The sensors had an accuracy of ±0.1℃, covering the temperature range from the center to the edge of the furnace; for example, the temperature at the center was 1200℃, and at the edge it was 800℃. These temperature values ​​were transmitted in real-time to the central processing unit via a data acquisition system. Interpolation algorithms, such as Kriging, were used to estimate the temperature between the collected points, generating a two-dimensional temperature distribution map with a resolution of 1cm × 1cm. The temperature gradient in the map decreased from the center to the edge. Analysis showed that the temperature change rate in the central region was less than 1℃ / min, while it reached 3℃ / min in the edge region. Simultaneously, flow and velocity sensors installed on the cooling medium pipeline measured the flow rate of the cooling medium (e.g., water) at 50L / min and the velocity at 2m / s. Combined with the pipeline cross-sectional area of ​​0.05m², the calculated medium volumetric flow rate deviated from the theoretical value by less than 2%, ensuring data reliability. Next, the temperature data was combined with the cooling medium parameters, and the finite element method was used to simulate the heat conduction process inside the furnace. Input boundary conditions, such as an ambient temperature of 25℃ and a convective heat transfer coefficient of 10 W / (m²·K), yielded a cooling rate curve. The curve showed an initial cooling rate of 5℃ / min, which decreased to 2℃ / min after 30 minutes. Analysis indicated a positive correlation between the cooling rate and the medium flow rate; for every 0.5 m / s increase in flow rate, the cooling rate increased by approximately 0.8℃ / min. Through the above data processing and algorithm analysis, a complete temperature distribution and cooling dynamic model was formed, providing a basis for subsequent optimization of cooling strategies. For example, when localized excessively high temperatures are detected, the algorithm can automatically adjust the medium flow rate to 60 L / min to accelerate cooling, ensuring that the temperature uniformity inside the furnace is controlled within ±5℃, thereby improving process stability.

[0030] S102. Based on the initial temperature distribution diagram and cooling rate curve, the internal heat conduction process of the saw blade is calculated using the finite element simulation method to determine the initial state of microstructure evolution and the preliminary value of stress distribution.

[0031] By collecting initial temperature and distribution data, a thermal field model of the saw blade's interior is constructed. Finite element method (FEM) simulation is used to mesh and numerically calculate the heat conduction process, yielding preliminary results of the thermal field distribution. Based on these preliminary results, combined with data on cooling rate and variation curves, the temperature gradient over time is analyzed to determine key time points in the microstructure evolution process. For the data at these key time points, the FEM is applied to dynamically simulate the heat conduction process within the saw blade, obtaining initial state parameters for microstructure evolution. Starting from these initial state parameters, the stress distribution in different regions within the saw blade is calculated, identifying the locations of stress concentration areas. If the stress concentration values ​​exceed a preset threshold, the heat conduction process in that region is locally meshed, and detailed stress distribution data is recalculated. Through analysis of the detailed data, the correlation between the heat conduction process and stress distribution is integrated to determine the final preliminary numerical results.

[0032] Based on the initial temperature distribution map and cooling rate curve, the finite element method is used to calculate the internal heat conduction process of the saw blade. The technical implementation of determining the initial state of microstructure evolution and preliminary stress distribution values ​​can be carried out through the following integration method. First, using the initial temperature distribution map, assuming the saw blade surface temperature is 800℃ and the center temperature is 600℃, a three-dimensional temperature field model is constructed. Finite element software such as ANSYS is used for mesh generation, dividing the saw blade into 10,000 tetrahedral elements. The mesh size is controlled within 0.5mm to ensure accuracy. The heat conduction equation is then used... (Where α is the thermal diffusivity, taken as 0.02 m² / s) Iterative calculations were performed with a time step of 0.1 s to simulate temperature changes over 10 s, obtaining temperature change curves for each node over time. It was found that when the surface temperature dropped to 500℃, the center temperature remained at 550℃, indicating a significant temperature difference. Secondly, combining the cooling rate curves, assuming an initial surface cooling rate of 50℃ / s and a center rate of 20℃ / s, the cooling rate distribution in each region was calculated using a finite element model interpolation algorithm. The influence of the cooling rate on the microstructure transformation was analyzed, and the JMAK equation (transformation fraction) was applied. The onset time of the austenite-to-martensite transformation was predicted using a formula (k=0.01, n=2). The calculated onset time was 2.5 s for the surface transformation and 4.0 s for the center, forming a spatiotemporal distribution map of the initial state of the microstructure evolution. Next, based on the temperature field and microstructure transformation data, thermal stress and phase transformation stress were calculated using thermoelastic theory. (E is the elastic modulus, taken as 200 GPa, α is the coefficient of thermal expansion, and ΔT is the temperature difference). Based on finite element simulation, the peak surface thermal stress is 300 MPa, and the center is 150 MPa. Considering the additional stress caused by phase transformation volume change (assumed to be 50 MPa), the preliminary stress distribution is determined to be 350 MPa on the surface and 200 MPa at the center. Through the above analysis, the distribution of temperature, microstructure, and stress forms a logical closed loop, providing data support for subsequent process optimization and ensuring that the calculation results match the actual saw blade heat treatment process.

[0033] S103. If the initial state of the microstructure evolution exceeds the preset threshold, adjust the simulation parameters to match the actual alloy element distribution and obtain the corrected stress distribution map.

[0034] Initial state data of microstructure evolution is acquired. The collected data is preprocessed to determine if the initial state exceeds a preset threshold. If it does, the simulation parameters are adjusted to account for the deviation between the initial state data and the preset threshold, resulting in an adjusted parameter configuration. Based on the adjusted parameter configuration and the elemental distribution data of the actual alloy, corresponding simulation results are generated to determine the degree of elemental distribution matching. By comparing and analyzing the simulation results with the actual alloy properties, key differences in elemental distribution are extracted, and the distribution characteristics of the difference data are obtained. Using these distribution characteristics, the stress calculation model is corrected, generating a preliminary stress distribution diagram. Based on the preliminary stress distribution diagram and the constraints of stress analysis, the diagram details are optimized to obtain the final corrected stress distribution diagram.

[0035] If the initial state of the tissue evolution exceeds a preset threshold of 0.15, the two-dimensional distribution data of Cr, Ni, and Mo in the actual alloy are first acquired using an electron probe microanalyzer to obtain a concentration field image with a grid resolution of 512×512. Then, the root mean square error (RMSE) between the elemental concentration field at the simulation's initial moment and the actual measured concentration field is calculated. When the RMSE is greater than 0.15, a parameter correction process is triggered. Next, the particle swarm optimization algorithm (PSO) is used to simulate the diffusion coefficient. , , Using the interface energy parameter σ as the optimization variable, the objective function is set to minimize the sum of squares of the differences between the concentration field RMSE and the actual value. The population size is set to 60, the number of iterations to 120, the inertia weight is linearly decreased from 0.9 to 0.4, and the learning factors c1=c2=2.0. After the 45th iteration, the RMSE converges to below 0.042. At this point, the corrected diffusion coefficients are obtained as follows: , , Then, these corrected parameters were re-input into the phase-field model, and the same heat treatment process simulation path was run (heated to 1050℃, held for 2 hours, and then cooled to room temperature at a rate of 5℃ / s), outputting the full-field von Mises stress distribution. Finally, the corrected stress distribution map was post-processed, with Gaussian filtering (3 grids in the core radius) used to smooth the noise, and the stress peak curve along the grain boundary direction of the maximum principal stress path was extracted. Compared with the uncorrected simulation results, it was found that the peak stress decreased from the original 628MPa to the corrected 541MPa, and the deviation decreased from the initial 18.7% to 6.3%, thus verifying that the parameter correction effectively improved the physical accuracy of the stress distribution prediction.

[0036] S104. Extract key node data from the corrected stress distribution map, use a neural network model to predict the degree of martensite transformation and the content of retained austenite, and determine the carbide precipitation behavior.

[0037] The stress values ​​at key node locations in the corrected stress distribution map are obtained. These stress values ​​are received by a pre-trained neural network model to obtain the martensitic transformation degree and retained austenite content. Regions with significant martensitic transformation are identified if the martensitic transformation degree is higher than a preset threshold. If the retained austenite content is lower than a preset threshold, the region is considered to have an increased tendency for carbide precipitation. Statistical methods are used to count the number of regions that satisfy significant martensitic transformation and have low retained austenite content, resulting in a set of high-precipitation-risk regions. The corresponding location coordinates and stress values ​​are extracted from this set of high-precipitation-risk regions to construct a precipitation behavior distribution dataset.

[0038] From the corrected stress distribution map, image processing algorithms were first used to divide the cloud map into grids and locate the pixels at the pixel level. Data from 25 key nodes, spaced 0.5 mm apart along the direction of maximum principal stress, were extracted. Each node recorded an equivalent stress value ranging from 420 MPa to 1280 MPa. Subsequently, the stress values, temperature (set as actual temperature curve data after quenching and cooling to room temperature, e.g., an average cooling rate of 35 K / s), and local carbon content (assuming a matrix carbon content of 0.42%) of these nodes were used as input feature vectors to a pre-trained BP neural network model. This model contains three hidden layers with 64, 32, and 16 neurons per layer, respectively. The ReLU activation function is used. The output layer has two nodes corresponding to the percentage of martensite transformation and the volume fraction of retained austenite, respectively. During training, the mean squared error loss function and the Adam optimizer were used. With a learning rate of 0.001 and a batch size of 32, after 120 epochs of training, the model's root mean square error (RMSE) on the validation set was less than 1.8%. After batch prediction of the extracted 25 node data, the average martensitic transformation degree was 92.4% ± 3.1%. The highest retained austenite content was 7.6% (located in the low stress region of the core) and the lowest was 1.2% (in the high stress region of the surface). Then, based on the distribution of retained austenite content and the degree of local carbon enrichment, a carbide precipitation tendency criterion was constructed. When the retained austenite content was greater than 5.5% and the local carbon supersaturation exceeded 0.15%, it was judged as a high precipitation risk region. Combined with the region with an equivalent stress exceeding 950 MPa, it was defined as a carbide easy precipitation sensitive region. Statistical analysis showed that this sensitive region was mainly concentrated in the annular region 1.2 mm to 3.8 mm from the surface, accounting for about 18.7% of the total volume fraction.

[0039] S105. Obtain the carbide precipitation behavior judged by the neural network model, and update the finite element simulation by combining real-time monitoring data to obtain the prediction of dynamic hardness uniform distribution.

[0040] A neural network model is used to assess carbide precipitation behavior under current process conditions, yielding the amount and location distribution of carbide precipitation. Based on this, the phase transformation kinetic parameters in the finite element model are corrected, resulting in an updated finite element simulation model. Real-time monitoring equipment is used to collect temperature and stress field data, obtaining the actual process temperature and stress sequences. These sequences are then input into the updated finite element simulation model for incremental calculations to obtain the current hardness field distribution. The difference between the current and previous hardness field distributions is assessed; if the difference exceeds a preset threshold, the model parameters are re-corrected; otherwise, the current simulation model is maintained. A neural network model is then used to perform a secondary prediction of the latest hardness field distribution, yielding the hardness trend for the next time step. Based on this hardness trend, the finite element simulation boundary conditions for the next time step are adjusted, resulting in a continuously updated dynamic hardness distribution prediction sequence.

[0041] In the process of implementing a neural network model to determine carbide precipitation behavior, updating finite element simulations with real-time monitoring data, and predicting the uniform distribution of dynamic hardness, the following specific methods can be used for technical implementation. First, the neural network model is used to predict carbide precipitation behavior. A convolutional neural network (CNN) algorithm is adopted, with input material composition data such as carbon content of 0.45%, chromium content of 1.2%, and heat treatment temperature of 850℃. The model learns from a training dataset (containing 5000 sets of historical experimental data) and outputs the carbide precipitation probability. Assuming the predicted result is a precipitation probability of 0.78, the model is classified using a sigmoid activation function. If the probability is greater than 0.5, it is judged as significant precipitation. The analysis shows that the precipitation tendency is obvious in a high-carbon, high-chromium environment, providing a basis for subsequent simulations. Next, the finite element simulation was updated based on real-time monitoring data. Temperature field data collected by sensors (e.g., a real-time temperature distribution matrix with a center temperature of 820℃ and an edge temperature of 780℃) was input into the ANSYS finite element software via a data interface. Using a thermal-structural coupling algorithm, with a material thermal conductivity of 15 W / m·K and a specific heat capacity of 460 J / kg·K, a temperature gradient of 5℃ / cm was calculated, thus updating the stress field distribution. Analysis showed that the thermal stress caused by the temperature gradient concentrated in the central region, with a maximum value of 120 MPa, providing dynamic boundary conditions for hardness prediction. Finally, combining the above results, a dynamic uniform hardness distribution was predicted. The stress field output from the finite element simulation and the carbide distribution data predicted by the neural network were used to calculate the hardness using the hardness calculation formula. (σ is local stress, unit MPa). The hardness of the central area is calculated to be 450 HB and that of the edge is 400 HB. The uniformity of hardness distribution is analyzed (standard deviation is 15 HB). If the uniformity is lower than the set threshold of 20 HB, it is optimized by adjusting the heat treatment parameters (e.g., reducing the temperature to 830℃).

[0042] S106. If the prediction of uniform distribution of dynamic hardness indicates the risk of internal defects, the quality fluctuation trend is analyzed by using a support vector machine model to determine the threshold for process intervention.

[0043] Obtain dynamic hardness prediction results. If the dynamic hardness prediction results indicate an increased risk of internal defects, extract the current batch quality data. Train and analyze the extracted quality data using a support vector machine to obtain the quality change trend. Calculate the sensitivity sequence of each process parameter to defect risk based on the quality change trend. For process parameters ranking high in the sensitivity sequence, determine the corresponding intervention threshold using preset rules. If the current process parameter exceeds the intervention threshold, generate and output an adjustment command. Transmit the adjustment command to the execution system to complete the process parameter correction.

[0044] When the dynamic hardness uniform distribution prediction model outputs an internal defect risk index exceeding 0.75, the system automatically triggers a support vector machine (SVM) model to analyze quality fluctuation trends. First, dynamic hardness data sequences from the most recent 30 batches of products are collected (with at least 120 sampling points per batch). Feature vectors are constructed, including 8 dimensions: hardness mean, standard deviation, skewness, kurtosis, and hardness difference between adjacent batches. These feature vectors are then combined with labeled defect occurrence tags (0 for normal, 1 for defect) to form a training set. A support vector machine (SVM) with a radial basis function (RBF) kernel is used, with a penalty parameter C=10.0 and kernel parameters... The parameters are optimized through grid search cross-validation. After model training, the system predicts the hardness fluctuation data of the current 10 consecutive batches to obtain a quality fluctuation trend score (range 0 to 1). If the predicted trend score rises continuously and exceeds 0.68, it is determined that there is a risk of quality deterioration. The system automatically calculates the process intervention threshold: using the output value of the SVM decision function as the confidence level, when the confidence level is greater than 0.85, the intervention threshold is determined to be the current mean hardness minus 1.2 times the standard deviation (for example, if the current mean is HV650 and the standard deviation is HV12, then the intervention threshold is HV635.6). This threshold is then pushed to the process control system to trigger automatic adjustment of the heating temperature by 2.5℃ or the cooling rate by 0.8℃ / s, etc., as compensation actions.

[0045] S107. Using the process intervention threshold determined by the support vector machine model, generate cooling rate adjustment instructions to obtain the optimized microstructure evolution path.

[0046] Obtain the original cooling process parameter sequence. Classify and train the tissue state samples using a support vector machine model to obtain a classification hyperplane. Calculate the distance from the current tissue state to the intervention boundary using the classification hyperplane. If the distance from the current tissue state to the intervention boundary is less than a pre-set safety margin, generate a cooling rate adjustment command. Modify the original cooling process parameter sequence according to the cooling rate adjustment command to obtain the adjusted cooling process parameter sequence. Perform tissue evolution simulation calculations on the adjusted cooling process parameter sequence to obtain a new tissue state sequence. Extract the final tissue morphology parameters from the new tissue state sequence to obtain the optimized tissue evolution path.

[0047] The process intervention threshold was determined using a support vector machine (SVM) model. First, a linear kernel SVM classifier was trained using historical data to divide the cooling rate into normal and intervention-required intervals. The classification label was whether the grain size was less than 15 μm and the bainite proportion was greater than 65%. The training set contained 800 sets of data corresponding to process parameters and the final microstructure. A grid search was used to optimize the penalty parameter C to 12.5 and gamma to 0.008. The final model achieved an accuracy of 92.3% on the validation set. The resulting decision function was... Where the threshold corresponds to The cooling rate threshold is 8.7℃ / s, meaning that intervention is triggered when the real-time monitored cooling rate deviates from the target path by more than this threshold; subsequently, the current cooling rate is calculated based on the deviation between the current temperature-time state and the ideal TTT curve. With target cooling rate The difference ,like Then a deceleration command is generated, the command format of which is "reduce the fan power from the current value P to..." ",in Taking the historical maximum deviation of 28.4℃ / s, for example, the current... The calculated fan power adjustment ratio is as follows: This means a power reduction of approximately 7.3%; if Δv < -8.7℃ / s, an acceleration command is generated, using similar proportional control to increase the fan power or activate the auxiliary spray system. The command is "Fan power increased to..." This allows the actual cooling curve to gradually approach the optimized path. After multiple closed-loop adjustments, the microstructure evolution path is gradually optimized from the initial pearlite + a small amount of ferrite to a microstructure dominated by fine acicular bainite. Finally, the average grain size is reduced from the original 21.6 μm to 13.8 μm, and the bainite volume fraction is increased from 52% to 71%.

[0048] S108. Extract the final stress distribution data from the optimized microstructure evolution path, determine the hardness uniformity and defect formation probability, and obtain a stable product quality control scheme.

[0049] Obtain the final stress distribution data in the optimized microstructure evolution path. Spatial discretization of the stress distribution data is performed using a mesh generation method to obtain a set of discrete stress points. The hardness value of each point in the set of discrete stress points is calculated to obtain a set of hardness value points. The uniformity of hardness distribution is determined based on the standard deviation of the hardness values ​​in the set of hardness value points. If the standard deviation is less than a preset threshold, the hardness distribution is considered uniform; if the standard deviation is greater than or equal to the preset threshold, the hardness distribution is considered non-uniform. A support vector machine is used to classify the set of discrete stress points, identifying high-risk and low-risk defect regions. The probability of defect formation is determined based on the number of stress concentration points and the rate of change of the stress gradient within the high-risk defect region. If the probability of defect formation is lower than a preset probability threshold, a stable quality control scheme under the current process parameters is output; if the probability of defect formation is higher than or equal to the preset probability threshold, the parameters of the key temperature control nodes in the microstructure evolution path are adjusted. The adjusted microstructure evolution path is re-obtained, and the process from extracting stress distribution data to determining the probability of defect formation is repeated to obtain an updated stable quality control scheme.

[0050] When extracting the final stress distribution data from the optimized microstructure evolution path obtained from finite element simulation, the odb file output by Abaqus or ANSYS is first read. A Python script combined with the odbAccess module is used to extract the Mises equivalent stress values ​​at all integration points in the final increment step. For example, the Mises stress ranges from 320.5 MPa to 478.2 MPa, with an average of 389.7 MPa and a standard deviation of 42.1 MPa. Then, a three-dimensional stress distribution matrix is ​​constructed based on the extracted stress field data. The stress gradient is stored and calculated using a NumPy array. After calculating the gradients in the x, y, and z directions using a Sobel filter, the maximum stress gradient value is obtained as 18.6 MPa / mm. Finally, when determining hardness uniformity, the Mises stress is compared with a pre-calibrated hardness-stress relationship curve (hardness...). Mapping using R²=0.978, the calculated hardness range is 168.2 HV to 217.5 HV, with a standard deviation of 13.1 HV. The coefficient of variation (CV) = standard deviation / mean = 0.069 is used as the uniformity index; a CV less than 0.08 indicates acceptable hardness uniformity. Next, when assessing the probability of defect formation, a Weibull distribution model based on stress and critical fracture stress is introduced, setting the material's Weibull modulus m=12.4 and the characteristic stress... Calculate the failure probability at each integration point. The cumulative volume-weighted average failure probability was 0.0047. When this value is below 0.008, the risk of defect formation is considered controllable. Finally, based on the above analysis results, a stable product quality control scheme was formulated, setting the control thresholds as follows: peak stress not exceeding 460 MPa, hardness coefficient of variation (CV) ≤ 0.075, and defect probability ≤ 0.006. When the simulation results exceed any of the thresholds, process parameters are automatically adjusted, such as reducing the quenching cooling rate by 5% or increasing the tempering temperature by 8°C. The adjusted parameters are then re-input into the microstructure evolution model for iterative optimization until all indicators meet the stable range. On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0051] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0052] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the entire lifecycle quality of P30 high-speed steel band saw blades based on digital twins, characterized in that, include: The initial temperature distribution map and cooling rate curve are obtained by collecting temperature data and cooling medium flow parameters at multiple points inside the furnace through a sensor array. Based on the initial temperature distribution diagram and cooling rate curve, the internal heat conduction process of the saw blade is calculated using the finite element simulation method to determine the initial state of microstructure evolution and the preliminary value of stress distribution. If the initial state of the microstructure evolution exceeds the preset threshold, the simulation parameters are adjusted to match the actual alloy element distribution to obtain a corrected stress distribution map. Key node data are extracted from the modified stress distribution map, and a neural network model is used to predict the degree of martensite transformation and the content of retained austenite based on the key node data in order to determine the carbide precipitation behavior. The carbide precipitation behavior determined by the neural network model is obtained, and the finite element simulation is updated in combination with real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution. If the dynamic hardness uniform distribution prediction indicates an increased risk of internal defects, the quality fluctuation trend is analyzed using a support vector machine model to determine the process intervention threshold. Using the process intervention threshold determined by the support vector machine model, a cooling rate adjustment command is generated to obtain an optimized microstructure evolution path. The final stress distribution data is extracted from the optimized tissue evolution path, and the hardness uniformity and defect formation probability are determined based on the final stress distribution data to obtain a stable product quality control scheme.

2. The method according to claim 1, characterized in that, The process of acquiring multi-point temperature data and cooling medium flow parameters within the furnace through a sensor array to obtain an initial temperature distribution map and a cooling rate curve includes: The sensor array acquires temperature data at multiple points inside the furnace and the flow parameters of the cooling medium. An initial temperature distribution map is generated based on the temperature data; The temperature sequence of each point is extracted from the initial temperature distribution map, and the cooling rate sequence is calculated based on the temperature difference between adjacent time points; A cooling rate curve is plotted based on the cooling rate sequence.

3. The method according to claim 1, characterized in that, Based on the initial temperature distribution diagram and cooling rate curve, the finite element method is used to calculate the internal heat conduction process of the saw blade, determine the initial state of microstructure evolution and preliminary stress distribution values, including: Based on the initial temperature distribution map, a thermal field model of the saw blade is constructed, and the heat conduction process is meshed and numerically calculated using finite element simulation technology to obtain preliminary results of the thermal field distribution. Based on the preliminary results of the thermal field distribution and the cooling rate curve, the variation law of temperature gradient with time is analyzed to determine the key time nodes in the tissue evolution process. For the key time points, the finite element method was used to dynamically simulate the heat conduction process inside the saw blade to obtain the initial state parameters of the tissue evolution. Based on the initial state parameters of the tissue evolution, the stress distribution in different regions inside the saw blade is calculated to obtain preliminary stress distribution values.

4. The method according to claim 1, characterized in that, If the initial state of microstructure evolution exceeds a preset threshold, the simulation parameters are adjusted to match the actual alloy element distribution to obtain a corrected stress distribution map, including: Acquire initial state data of organizational evolution and determine whether the initial state data exceeds a preset threshold; If the deviation exceeds the limit, the simulation parameters are adjusted based on the deviation between the initial state data and the preset threshold to obtain the adjusted parameter configuration. Based on the adjusted parameter configuration and the actual alloy element distribution data, the corresponding simulation results are generated. Based on the comparison and analysis between the simulation results and the actual alloy properties, the stress calculation model is modified, and a modified stress distribution map is generated.

5. The method according to claim 1, characterized in that, The process of extracting key node data from the corrected stress distribution map and using a neural network model to predict the degree of martensite transformation and the content of retained austenite based on the key node data to determine carbide precipitation behavior includes: Obtain the stress values ​​at key node locations in the corrected stress distribution map; The stress values ​​are input into a pre-trained neural network model to obtain the corresponding values ​​of martensite transformation degree and retained austenite content. Based on whether the value of the degree of martensite transformation is higher than a first preset threshold, the region where significant martensite transformation occurs is determined; The carbide precipitation behavior is determined based on whether the residual austenite content is lower than a second preset threshold and in conjunction with the region where significant martensitic transformation occurs.

6. The method according to claim 1, characterized in that, The process of obtaining the carbide precipitation behavior determined by the neural network model and updating the finite element simulation with real-time monitoring data to obtain a prediction of dynamic hardness uniform distribution includes: Obtain the amount and location distribution of carbide precipitation as determined by the neural network model; Based on the amount and location distribution of carbide precipitation, the phase transition kinetic parameters in the finite element model are corrected to obtain an updated finite element simulation model. Real-time temperature and stress field data are collected by real-time monitoring equipment to obtain real-time temperature and stress sequences. The real-time temperature sequence and stress sequence are input into the updated finite element simulation model, and incremental calculations are performed to obtain the current hardness field distribution and dynamic hardness uniform distribution prediction.

7. The method according to claim 1, characterized in that, The process intervention threshold determined by the support vector machine model is used to generate cooling rate adjustment commands and obtain an optimized tissue evolution path, including: Obtain the original cooling process parameter sequence; The classification hyperplane is obtained by training the tissue state samples to classify them using a support vector machine model. The distance from the current tissue state to the intervention boundary is calculated based on the classification hyperplane; If the distance is less than a preset safety margin, a cooling speed adjustment command is generated; The original cooling process parameter sequence is modified according to the cooling rate adjustment command, and the microstructure evolution simulation calculation is performed on the adjusted cooling process parameter sequence to obtain the optimized microstructure evolution path.

8. The method according to claim 1, characterized in that, The process of extracting final stress distribution data from the optimized microstructure evolution path and determining hardness uniformity and defect formation probability based on the final stress distribution data to obtain a stable product quality control scheme includes: Obtain the final stress distribution data in the optimized tissue evolution path; The final stress distribution data is spatially discretized to obtain a set of discrete stress points, and the hardness value corresponding to each discrete stress point is calculated. The uniformity of hardness distribution is determined based on the statistical characteristics of each hardness value; The discrete stress point set is classified using a support vector machine to obtain high-risk and low-risk defect regions. The probability of defect formation is determined based on the characteristics of stress concentration points within the high-risk defect area. Based on the uniformity of hardness distribution and the probability of defect formation, a stable product quality control scheme is generated.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images