A method for optimizing heat treatment process parameters of a bearing ring of GCr15 bearing steel

Abstract

This invention belongs to the field of heat treatment technology and provides a method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings. The method includes the following steps: S1, obtaining the bearing model, geometric parameters, performance constraints, and target heat treatment equipment capability boundaries of the GCr15 bearing steel bearing rings to be optimized; S2, establishing a set of collaborative process parameters composed of independent process parameters and co-coupled parameters; S3, establishing and verifying a four-field fully coupled dynamic simulation model; S4, establishing a training dataset; S5, establishing a two-level collaborative prediction model; S6, multi-objective collaborative optimization and three-level verification; S7, small-batch trial production and a closed-loop iterative mechanism with three-level triggering and two-dimensional updates; S8, updating the process database. This invention, by establishing a mapping relationship between bearing geometric parameters and collaborative process parameters, enables rapid migration and reuse of heat treatment processes for different types of bearing rings, significantly shortening the process development cycle and reducing trial-and-error costs.

Description

Technical Field

[0001] This invention relates to the field of heat treatment technology for metallic materials, specifically to a method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings. Background Technology

[0002] GCr15 high-carbon chromium bearing steel is the core steel for precision bearing rings due to its high hardness, excellent wear resistance, and good dimensional stability. For angular contact ball bearings operating under high-speed and high-reliability conditions, the microstructure, hardness distribution, and residual stress state of the rings after heat treatment directly determine the contact fatigue life and service reliability of the bearing. Salt bath isothermal quenching-deep cryogenic treatment-tempering synergistic heat treatment is the core process route for GCr15 bearing rings to achieve high hardness matching and long service life. It pre-establishes a bainite and martensite composite structure through salt bath isothermal treatment, controls residual austenite through deep cryogenic treatment, and stabilizes the structure and stress through tempering.

[0003] Currently, there are three major technical challenges in the application and optimization of this synergistic heat treatment process: First, the process parameters are strongly coupled with the bearing geometric parameters, making it difficult to transfer and reuse them across different models: the geometric parameters such as the inner diameter, outer diameter, effective wall thickness, and contact angle of the bearing rings directly affect the temperature field distribution, phase transformation process, and stress evolution during the heat treatment process. The performance of the same set of coordinated process parameters varies greatly on different models of bearings. Existing technologies mostly optimize the process for a single model of bearing, while new models of bearings still require engineers to rely on trial and error based on their experience. The process development cycle is long, costly, and inefficient, making it impossible to achieve rapid transfer and reuse of parameters. Second, existing optimization methods are not adapted to the coupling characteristics of synergistic processes and cannot achieve global optimization. Existing heat treatment parameter optimization schemes often treat salt bath quenching, cryogenic treatment, and tempering as independent processes, optimizing them step-by-step without considering the strong coupling and synergistic effects among the three. The final microstructure of salt bath quenching directly determines the cryogenic phase transformation process, and the dislocation density and retained austenite content after cryogenic treatment directly affect the precipitation of carbides and the stress relaxation effect during tempering. General simulation, experimental design, prediction, and optimization frameworks do not design specific parameter systems, model corrections, and constraints for synergistic processes. Therefore, the optimization results can only achieve local optima for single processes and cannot fully realize the performance potential of synergistic processes. Third, existing solutions lack a closed-loop mechanism for engineering adaptation and continuous iteration: existing parameter optimization solutions mostly remain at the laboratory simulation level, without considering the equipment capability boundaries and the robustness of process parameter fluctuations in industrial production; at the same time, the simple iterative logic of recalculating when the measured performance data does not meet the requirements has neither clear triggering rules nor a parameter accumulation and migration mechanism for bearing models, making it impossible to achieve continuous improvement in model accuracy and engineering implementation of the process.

[0004] In summary, existing technologies have not yet developed a parameter prediction and recommendation scheme for the salt bath-deep cryogenic-tempering synergistic process of GCr15 bearing rings, which is adaptable to different bearing geometric parameters and has engineering closed-loop iteration capabilities. This cannot meet the core needs of the precision bearing industry for rapid and low-cost development of new heat treatment processes. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, the present invention aims to construct a parameter optimization framework deeply integrated with the salt bath-deep cryogenic-tempering synergistic process. This framework solves the problems that existing general solutions cannot adapt to the coupling characteristics of the synergistic process and cannot achieve global optimization. By establishing a mapping relationship between bearing geometric parameters and synergistic process parameters, the heat treatment processes for different types of bearing rings can be rapidly transferred and reused, significantly shortening the process development cycle and reducing trial and error costs.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings includes the following steps: S1. Obtain the bearing model, geometric parameters, performance constraints, and target heat treatment equipment capacity boundary of the bearing rings of the GCr15 bearing steel to be optimized. S2. Establish a set of collaborative process parameters for salt bath isothermal treatment, cryogenic treatment and tempering treatment. The set of collaborative process parameters includes independent process parameters for characterizing the set values ​​of each process and collaborative coupling parameters for characterizing the connection relationship between processes. S3. Based on the aforementioned geometric parameters, establish a continuous four-field fully coupled dynamic simulation model of temperature field, phase transition field, stress field, and carbide precipitation field for the entire process of salt bath-deep cryogenic-tempering, and verify the accuracy of the four-field fully coupled dynamic simulation model using physical experimental data; S4. Generate sample parameter combinations within the parameter space defined by the collaborative process parameter set, call the verified four-field fully coupled dynamic simulation model to obtain the simulation performance response data corresponding to each sample parameter combination, and form a training dataset. S5. Based on the training dataset, establish a two-level collaborative prediction model, which includes a first-level prediction model with bearing ring geometric parameters as input and the optimal collaborative process parameter range as output, and a second-level prediction model with the optimal collaborative process parameter range as input and the bearing ring predicted performance response data as output. S6. Taking the maximization of bearing ring hardness as the core optimization objective, a multi-objective collaborative optimization function is constructed, which includes residual austenite content, tangential residual compressive stress amplitude on raceway surface, and maximum residual tensile stress on raceway subsurface. Combining performance constraints and collaborative process parameter set, a particle swarm optimization algorithm with collaborative constraint penalty factor is used to perform global optimization on the verified two-level prediction model. During the optimization process, a three-level verification mechanism is implemented, which includes pre-verification of optimal collaborative process parameter range, verification of predicted performance response data, and verification of equipment capability and process robustness. The recommended combination of collaborative heat treatment process parameters is then output. S7. Conduct small-batch trial production according to the recommended combination of synergistic heat treatment process parameters, collect the measured performance data after the trial production, and feed the measured performance data back to the process database. When the measured performance data meets the performance constraints, output the final synergistic heat treatment process parameters. When the measured performance data does not meet the performance constraints, adjust the optimization algorithm parameters, constraints, or synergistic process parameter set and re-execute S6, while simultaneously feeding the measured performance data back to the process database, and execute a closed-loop iteration mechanism of three-level triggering and two-dimensional updating. S8. Write the bearing model, geometric parameters, final co-process heat treatment parameters, simulation performance data, measured performance data, and model iteration records into the process database for subsequent retrieval of similar models and parameter migration.

[0007] Furthermore, in step S1, the geometric parameters include at least the inner diameter, outer diameter, width, contact angle, and effective wall thickness; the performance constraints include at least the lower limit of hardness, the range of retained austenite content, the lower limit of the amplitude of tangential residual compressive stress on the raceway surface, and the upper limit of the maximum residual tensile stress on the secondary surface of the raceway; the capability boundary of the target heat treatment equipment includes at least the temperature range, the heating rate range, and the cooling rate range.

[0008] Furthermore, in step S2, the independent process parameters include at least the austenitizing temperature, austenitizing holding time, salt bath isothermal temperature, salt bath holding time, salt bath cooling rate, cryogenic temperature, cryogenic heating and cooling rate, cryogenic holding time, tempering temperature, tempering heating rate, and tempering holding time. The synergistic coupling parameters include at least the difference between the final salt bath cooling temperature and the initial cryogenic temperature, the interval between the end of cryogenic cooling and the start of tempering, the matching coefficient between the salt bath cooling rate and the cryogenic temperature drop rate, the ratio of the cryogenic holding time to the tempering holding time, and the matching coefficient between the cryogenic temperature and the tempering temperature.

[0009] Furthermore, in step S3: The temperature field adopts the transient heat conduction equation that takes into account the effective wall thickness coupling, and heat transfer coefficient models are set for rapid heat transfer in salt bath, cryogenic heat transfer and tempering isothermal heat transfer respectively. The phase transformation field adopts a segmented collaborative modified phase transformation dynamics model, wherein the diffusion-type phase transformation from austenite to bainite is described by the Johnson-Mehl-Avrami equation, and the non-diffusion-type phase transformation from austenite to martensite during salt bath isothermal treatment and cryogenic treatment is described by the Koistinen-Marburger equation with the introduction of a bainite volume fraction correction term after salt bath isothermal treatment. The stress field incorporates carbide precipitation-induced strain into the total strain decomposition and uses the Norton creep model to describe stress relaxation during the tempering stage. The carbide precipitation field adopts a carbide precipitation kinetic model that incorporates a matrix dislocation density correction term after cryogenic treatment.

[0010] Furthermore, in step S4, orthogonal experiments are first used to generate basic samples within the set of collaborative process parameters, and then Latin hypercube sampling is performed within the preset set of collaborative process parameters to generate extended samples.

[0011] Furthermore, the second-level prediction model includes independently trained sub-models for hardness prediction, residual austenite content prediction, raceway surface tangential residual compressive stress amplitude prediction, and raceway subsurface maximum residual tensile stress prediction. Each sub-model takes the optimal collaborative process parameter range as input and outputs corresponding single prediction performance response data.

[0012] Furthermore, each sub-model of the second-level prediction model is constructed using a BP neural network with an attention mechanism, which is used to assign weights to the normalized optimal collaborative process parameter range features; the two-level collaborative prediction model is trained using the trainlm algorithm, and the model is deemed to have passed validation when the determination coefficient R² of each sub-model is not less than 0.95.

[0013] Furthermore, in step S6, the multi-objective collaborative optimization function is: ; In the formula: I max (X) is the overall objective function, d H (X) is the desired hardness function, d τ (X) is the expected function of the residual austenite content, d γ (X) is the desired function of the amplitude of the tangential residual compressive stress on the raceway surface, d δ (X) is the expected function of the maximum residual tensile stress on the secondary surface of the raceway; ω1, ω2, ω3, and ω4 are the synergistic weighting coefficients, assigned based on the weighting of the influence of salt bath, cryogenic, and tempering processes on each property, and ω1+ω2+ω3+ω4=1; the design variable X is the independent process parameter and synergistic coupling parameter in step S2.

[0014] Furthermore, in step S6, the three-level verification includes: The optimal collaborative process parameter range pre-verification is used to determine whether the optimal collaborative process parameter range meets the value range requirements of the preset collaborative process parameter set in step S2. If it does not meet the requirements, the parameter combination is directly regenerated. Predicted performance response data verification is used to determine whether the predicted performance response data output in step S5 meets the performance constraints in step S1. If it does not meet the constraints, the algorithm iterative optimization will proceed. Equipment capability and process robustness verification is used to determine whether the optimal range of synergistic process parameters is within the capability range of the target heat treatment equipment. At the same time, it simulates the performance stability under ±5% process parameter fluctuations and outputs the process parameter combination with the best robustness as the recommended synergistic heat treatment process parameter combination.

[0015] Furthermore, the three-level triggering includes: When the deviation between the measured performance data and the predicted performance response data of the same type of bearing exceeds a preset threshold, the fine-tuning of the two-level collaborative prediction model is triggered. Adding geometric parameters, collaborative process parameters, and measured performance data of three or more new bearing models triggers the expansion of the training dataset and retraining of the two-level collaborative prediction model. When newly verified measured performance data is added to the collaborative process parameter set and the measured performance data indicates that the boundary of the collaborative process parameter set needs to be adjusted, the collaborative process parameter set boundary update is triggered. The two-dimensional update includes: In terms of dataset dimension, actual performance data is added to the training dataset, and training weights are assigned according to bearing model similarity and cooperative matching degree; In terms of mapping relationships, update the mapping relationship between geometric parameters and the optimal collaborative process parameter range in the process database.

[0016] The technical solution provided by this invention has the following advantages compared with the prior art: 1. A full-chain parameter optimization framework deeply integrated with the salt bath-deep cryogenic-tempering synergistic process was constructed, abandoning the basic framework of general heat treatment optimization in existing technologies. The whole scheme takes the salt bath-deep cryogenic-tempering synergistic process as the core. In addition to independent process parameters, synergistic coupling parameters such as temperature difference, process interval time, cooling rate matching coefficient, holding time ratio and temperature matching coefficient are further incorporated into process characteristics and constraints. 2. A continuous four-field fully coupled simulation model of the entire process of salt bath-deep cryogenic-tempering was established. The model incorporates a correction term for the salt bath bainite content on the deep cryogenic phase transformation, a correction term for the deep cryogenic dislocation density on the tempering carbide precipitation, and a coupling term for the bearing geometry and thermal conductivity. The two-way feedback between the fields can describe the continuous evolution of microstructure, temperature and residual stress in the entire heat treatment process. Compared with the existing step-by-step model, the simulation prediction deviation is reduced and the accuracy of parameter optimization is further improved. 3. By constructing a two-level prediction model of "geometric parameters - optimal collaborative process parameter range - predicted performance data", the optimal collaborative process parameter range can be quickly matched by inputting bearing geometric parameters. This solves the pain point of the difficulty in transferring process parameters of different bearing models in the existing technology, reduces the workload of process development that relies entirely on manual trial and error, and improves the efficiency of the heat treatment process development cycle for new bearing models. 4. The second-level prediction model in the two-level prediction model is split into four independently trained BP neural network sub-models with attention mechanism. Each sub-model receives collaborative process parameters, but only outputs a single performance index, so that the model can identify different combinations of sensitive parameters for different performance indices and avoid prediction interference caused by differences in units and optimization directions. 5. Through a closed-loop iteration mechanism of three-level triggering and two-dimensional updating, model fine-tuning, model retraining, or boundary updates are triggered according to the prediction deviation, the amount of new model data accumulation, and the amount of optimized sub-interval verification data accumulation, respectively. The update object covers both the training dataset and the mapping relationship between geometric parameters and optimal collaborative process parameters. The process database can be continuously accumulated with trial production and mass production verification, gradually improving the model prediction accuracy, the reliability of the optimal collaborative process parameter interval, and the cross-model transfer efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the framework of a two-level collaborative prediction model; Figure 3 This is a schematic diagram of a three-level triggering + two-dimensional closed-loop iteration logic. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] like Figures 1-3 As shown, a method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings includes the following steps: S1. Obtain the bearing model, geometric parameters, performance constraints, and target heat treatment equipment capability boundary of the GCr15 bearing steel bearing ring to be optimized.

[0021] The geometric parameters include at least the inner diameter, outer diameter, width, contact angle, and effective wall thickness. The performance constraints include at least the lower limit of hardness, the range of retained austenite content, the lower limit of the amplitude of tangential residual compressive stress on the raceway surface, and the upper limit of the maximum residual tensile stress on the secondary surface of the raceway. The amplitude of tangential residual compressive stress on the raceway surface represents the absolute value of the tangential residual compressive stress on the raceway surface. The capability boundary of the target heat treatment equipment includes at least the temperature range, the heating rate range, and the cooling rate range.

[0022] Taking the 7008C type angular contact ball bearing race as an example, its geometric parameters are: inner diameter 40mm, outer diameter 68mm, width 15mm, and contact angle 15°; performance constraints are: hardness ≥60HRC, retained austenite content 1%~5%, tangential residual compressive stress amplitude on raceway surface ≥200MPa, maximum residual tensile stress on raceway subsurface ≤50MPa, quenching furnace temperature ≤1000℃, cryogenic furnace temperature ≥-150℃, salt bath cooling rate range 5~20℃ / s, cryogenic furnace cooling rate range 0.3~1.5℃ / min, and tempering furnace heating rate range 1~5℃ / min.

[0023] S2. Establish a set of collaborative process parameters for salt bath isothermal treatment, cryogenic treatment and tempering treatment. The set of collaborative process parameters includes independent process parameters for characterizing the set values ​​of each process and collaborative coupling parameters for characterizing the connection relationship between processes.

[0024] The independent process parameters include at least the austenitizing temperature, austenitizing holding time, salt bath isothermal temperature, salt bath holding time, salt bath cooling rate, cryogenic temperature, cryogenic heating and cooling rate, cryogenic holding time, tempering temperature, tempering heating rate, and tempering holding time. The synergistic coupling parameters include at least the difference between the final salt bath cooling temperature and the initial cryogenic temperature, the interval between the end of cryogenic treatment and the start of tempering, the matching coefficient between the salt bath cooling rate and the cryogenic temperature drop rate, the ratio of the cryogenic holding time to the tempering holding time, and the matching coefficient between the cryogenic temperature and the tempering temperature.

[0025] In this embodiment, the 7008C type angular contact ball bearing ring is also used, with the following independent process parameters: austenitizing temperature 800~850℃, austenitizing holding time 20~40min, salt bath isothermal temperature 160~200℃, salt bath holding time 15~45min, salt bath cooling rate 5~15℃ / s, cryogenic temperature -90~-70℃, cryogenic heating and cooling rate 0.3~1.0℃ / min, cryogenic holding time 90~150min, tempering temperature 160~200℃, tempering heating rate 2~4℃ / min, and tempering holding time 180~300min.

[0026] The range of values ​​for the synergistic coupling parameters is as follows: the difference between the final salt bath cooling temperature and the initial cryogenic temperature is 0–20℃; the interval between the end of cryogenic cooling and the start of tempering is 5–30 min; the matching coefficient K1 between the salt bath cooling rate and the cryogenic temperature drop rate is 300–3000; the ratio of the cryogenic holding time to the tempering holding time is 0.4–0.56; and the matching coefficient K2 between the cryogenic temperature and the tempering temperature is -0.5–-0.35.

[0027] S3. Based on the geometric parameters, establish a full-coupled dynamic simulation model of four fields—temperature field, phase transformation field, stress field, and carbide precipitation field—for the entire process of salt bath-deep cryogenic-tempering, and verify the accuracy of the full-coupled dynamic simulation model of four fields using physical experimental data.

[0028] Specifically, the temperature field employs a transient heat conduction equation considering the coupling of bearing geometric parameters, the expression of which is: ; In the formula: T is temperature, t is time, ρ is the density of GCr15 steel, Cp is specific heat capacity, k(h) is the thermal conductivity coupled with the effective wall thickness h of the bearing ring, and Q 相变 For the latent heat of phase transition, Q 塑性 For heat generated by plastic deformation, Q 析出 To address the heat of carbide precipitation, dedicated heat transfer coefficient models were established for rapid heat exchange in the salt bath, cryogenic heat exchange, and tempering isothermal heat exchange. A continuous temperature evolution control equation was constructed for the entire process of salt bath-cryogenic-tempering, replacing the step-by-step modeling approach of existing technologies.

[0029] For the phase transition field, a segmented, collaboratively modified phase transition dynamics model is established for the entire phase transition process of salt bath-deep cryogenic-tempering, wherein: The austenitizing process and the diffusion-type phase transformation from austenite to bainite are described by the Johnson-Mehl-Avrami (JMA) equation: ; In the formula, ξ is the volume fraction of the transformed phase, b(T) is the temperature-dependent rate parameter, n is the Avrami exponent, and t is the isothermal time; The non-diffusional phase transformation from austenite to martensite during isothermal salt bath treatment and cryogenic treatment is described by the Koistinen-Marburger equation, which is modified based on the salt bath microstructure. ; In the formula: ξ M ξ is the volume fraction of martensite. B σ represents the volume fraction of bainite after salt bath isothermal treatment (a co-correction term), C is the current austenite carbon content, C0 is the nominal carbon content of the material, and σ eq The equivalent stress is represented by φ1 to φ5, which are model correction coefficients. This equation introduces for the first time a correction term for the salt bath bainite content on the cryogenic martensitic phase transformation, accurately characterizing the microstructural synergy effect of salt bath-cryogenic transformation. The austenite inversion transformation is described using a simplified diffusion-type model, expressed as: ; Where ξ r T represents the volume fraction of austenite, and T is the current temperature. S and T e Let A and D be the onset and termination temperatures of the phase transition, respectively. A and D are kinetic exponents, with A = -4 and D = 2, used to characterize the phase transition rate and the shape of the transition curve. This model is based on a diffusion-controlled mechanism, assuming that at T... S and T e Within the temperature range, the volume fraction of austenite increases monotonically with temperature.

[0030] For the stress field, establish a continuous mechanical equilibrium equation throughout the entire process, and decompose the total strain as follows: ; In the formula, Let S be the total strain generated in the bearing ring micro-element at a certain moment. For elastic strain, For plastic strain, For thermal strain, For phase transition strain, Induced strain for carbide precipitation (synergistic new item); Stress relaxation during the tempering stage is described using the Norton creep model, and the expression is: ; In the formula, A and n are material constants, Q is the creep activation energy, R is the gas constant, and T is the tempering absolute temperature.

[0031] For carbide precipitation sites, a carbide precipitation kinetic model correlated with cryogenic dislocation density is established, expressed as: ; Where, ξ 碳化物 ρ is the carbide precipitation fraction, k0 is the frequency factor, Q is the activation energy, R is the gas constant, T is the tempering temperature, ρ is the matrix dislocation density after cryogenic treatment, ρ0 is the matrix dislocation density in the quenched state, and m is the dislocation influence coefficient (synergistic correction term). This model accurately characterizes the microstructural synergy of cryogenic-tempering by incorporating the effect of cryogenic dislocation density on tempered carbide precipitation.

[0032] Furthermore, the accuracy verification method for the four-field fully coupled dynamic simulation model is as follows: Select a standard GCr15 bearing ring, and under the same collaborative process conditions, compare the simulation calculation results with the physical test performance data. The comparison indicators include hardness, residual austenite content, and residual stress. If the deviation of each indicator meets the following conditions: hardness ≤ ±0.8HRC, residual austenite content ≤ ±1.5%, and residual stress on the raceway surface ≤ ±40MPa, then the model verification is successful.

[0033] Salt bath-deep cryogenics-tempering is not simply three independent processes, but a continuous and inherited process. The bainite, martensite, and temperature gradient formed in the salt bath stage will affect the continued transformation of the retained austenite in the deep cryogenic stage. The martensitic transformation, dislocation density, and residual stress generated in the deep cryogenic stage will affect the carbide precipitation and stress relaxation in the tempering stage. Therefore, establishing a temperature field, phase transformation field, or stress field separately cannot accurately reflect the actual heat treatment process. The four-field fully coupled model calculates the temperature field, phase transformation field, stress / strain field, and carbide precipitation field under the same continuous framework. It can simultaneously consider the driving effect of temperature change on phase transformation and carbide precipitation, the inverse effect of latent heat of phase transformation on the temperature field, the volumetric strain caused by martensitic transformation and carbide precipitation, the influence of microstructure strain on residual stress, the influence of stress state on phase transformation process and carbon diffusion / precipitation, and the influence of carbide precipitation in the tempering stage on microstructure stability and residual stress relaxation.

[0034] S4. Generate sample parameter combinations within the parameter space defined by the collaborative process parameter set, and call the verified four-field fully coupled dynamic simulation model to obtain the simulation performance response data corresponding to each sample parameter combination, forming a training dataset.

[0035] More specifically: Orthogonal experimental full-space coverage: using all 11 independent process parameters and 5 synergistic coupling parameters in step S2 as variables, and adopting L81(3 16 The basic experiment of orthogonal array design generated 81 sets of sample parameter combinations, covering the main effects and second-order interaction effects of the entire 16-dimensional collaborative process parameter space. Targeted and encrypted sampling of collaborative process parameter set: Within the 16 collaborative process parameter sets defined in step S2, 243 sets of extended sample parameter combinations are generated using the Latin hypercube sampling method, so that the sample density in this optimal sub-interval reaches more than 3 times the average sample density of the entire space. Finally, the 81 basic samples and 243 extended samples were merged to form a training dataset containing 324 samples.

[0036] S5. Based on the training dataset, establish a two-level collaborative prediction model, which includes a first-level prediction model with bearing ring geometric parameters as input and the optimal collaborative process parameter range as output, and a second-level prediction model with the optimal collaborative process parameter range as input and the bearing ring predicted performance response data as output.

[0037] The first-level prediction model is a geometric parameter-optimal collaborative process parameter interval prediction model. It takes the bearing geometric parameters (inner diameter, outer diameter, width, contact angle, and effective wall thickness) as input and the optimal collaborative process parameter interval (i.e., the sub-interval of the corresponding 16 collaborative process parameters) of the bearing model as output. It is constructed using a BP neural network. The function of this model is to quickly narrow the search range of process parameters based on the bearing geometric characteristics, so as to realize the rapid migration of process parameters for different bearing models.

[0038] The second-level prediction model is a process parameter-performance prediction model. It takes the optimal collaborative process parameter range output from the first level as input and the predicted performance response data of the bearing ring as output. It establishes four sub-prediction models for hardness, residual austenite content, tangential residual compressive stress amplitude on the raceway surface, and maximum residual tensile stress on the raceway subsurface. It is constructed using a BP neural network with attention mechanism.

[0039] For the aforementioned sub-prediction models, each sub-prediction model is constructed using a BP neural network with an attention mechanism. The input layer receives 16 collaborative process parameters (i.e., within the range of the optimal collaborative process parameters). After normalization, the parameters enter the attention weight allocation layer, which automatically identifies the contribution of different process parameters to the corresponding performance indicators. The parameters then enter the hidden layer of the BP neural network to complete the nonlinear mapping and finally output a single performance prediction value. The four sub-models are trained independently, which can avoid prediction interference caused by differences in the dimensions and optimization directions between different performance indicators.

[0040] The connection method between the two-level models: In the subsequent multi-objective optimization process of step S6, the optimal cooperative process parameter range output by the first level is used as the search boundary of the particle swarm optimization algorithm. Within this boundary, the process parameter combination to be optimized is generated and then input into the second-level prediction model for performance prediction. The model training adopts the trainlm algorithm, and the dataset is normalized in the interval [-1,1] before training. After training, the coefficient of determination R is used. 2 To verify the model's fitting effect, the R-values ​​of each sub-model were calculated. 2 If all values ​​are ≥0.95, the model is validated.

[0041] S6. Taking the maximization of bearing ring hardness as the core optimization objective, a multi-objective collaborative optimization function is constructed, which includes the residual austenite content, the amplitude of tangential residual compressive stress on the raceway surface, and the maximum residual tensile stress on the raceway subsurface. Combining the performance constraints in step S1 with the preset collaborative process parameter set in step S2, a particle swarm optimization algorithm with a collaborative constraint penalty factor is used to globally optimize the verified two-level prediction model. During the optimization process, a three-level verification mechanism is implemented, which includes pre-verification of the optimal collaborative process parameter range, verification of predicted performance response data, and verification of equipment capability and process robustness. The recommended combination of collaborative heat treatment process parameters that meets the requirements is output.

[0042] The multi-objective collaborative optimization function is as follows: ; Among them: I max (X) is the overall objective function; d H (X) is the expected hardness function. In the formula, H min H is the lower limit of hardness. tar For the target hardness value, a H The shape factor is the desired function of hardness. d τ (X) is the expected value function of the residual austenite content. In the formula, L This represents the lower limit of the residual austenite content. U This represents the upper limit of the residual austenite content. 0 represents the target value for retained austenite, which can be the midpoint of the interval or the optimal value determined experimentally. τ b τ The shape factor is the desired function of hardness. d γ (X) is the expected value function of the tangential residual compressive stress amplitude on the raceway surface. In the formula: min This represents the lower limit of the tangential residual compressive stress amplitude on the raceway surface. tar For the target compressive stress amplitude, a γ The shape factor is the expected function of compressive stress. d δ (X) is the expected function of the maximum residual tensile stress in the subsurface layer of the raceway. In the formula: max The upper limit of the maximum allowable residual tensile stress, tar The maximum residual tensile stress in the target subsurface layer, The shape factor is the expected function of tensile stress. ω1, ω2, ω3, and ω4 are synergistic weighting coefficients, assigned based on the weighting of the effects of salt bath, cryogenic, and tempering processes on each performance, and ω1+ω2+ω3+ω4=1; The design variable X is the independent process parameter and the collaborative coupling parameter in step S2.

[0043] In addition, the three-level verification includes: The optimal collaborative process parameter range pre-verification is used to determine whether the optimal collaborative process parameter range meets the value range requirements of the preset collaborative process parameter set in step S2. If it does not meet the requirements, the parameter combination is directly regenerated. Predicted performance response data verification is used to determine whether the predicted performance response data output in step S5 meets the performance constraints in step S1. If it does not meet the constraints, the algorithm iterative optimization will proceed. Equipment capability and process robustness verification is used to determine whether the optimal range of synergistic process parameters is within the capability range of the target heat treatment equipment. At the same time, it simulates the performance stability under ±5% process parameter fluctuations and outputs the process parameter combination with the best robustness as the recommended synergistic heat treatment process parameter combination.

[0044] S7. Conduct small-batch trial production according to the recommended combination of synergistic heat treatment process parameters, collect measured performance data after trial production, and feed the measured performance data back to the process database. When the measured performance data meets the performance constraints, output the final synergistic heat treatment process parameters. When the measured performance data does not meet the performance constraints, adjust the optimization algorithm parameters, constraints, or synergistic process parameter set, and re-execute S6, while simultaneously feeding the measured performance data back to the process database, and execute a closed-loop iteration mechanism of three-level triggering and two-dimensional updating.

[0045] Specifically, prioritize adjusting and optimizing algorithm parameters, including the inertia weight, learning factor, population size, and maximum number of iterations for the particle swarm optimization algorithm, to expand the algorithm's global search capability; If the constraints still cannot be met after adjusting the algorithm parameters, the non-core performance constraints should be relaxed appropriately, but the two core constraints of the lower limit of hardness and the lower limit of the amplitude of tangential residual compressive stress on the raceway surface should not be relaxed. If the above adjustments are ineffective, return to step S3 to recalibrate the correction coefficients of the four-field fully coupled simulation model, and then re-execute the subsequent process.

[0046] The three-level triggering includes: Level 1 trigger: Triggered when the deviation between the measured performance data of a single batch of bearings of the same model and the predicted performance response data of the second-level prediction model exceeds a preset threshold. The preset thresholds are: hardness deviation > ±0.8HRC, residual austenite content deviation > ±1.5%, tangential residual compressive stress amplitude deviation on raceway surface > ±40MPa, and maximum residual tensile stress deviation on raceway subsurface > ±20MPa. Level 2 trigger: Model retraining is triggered when the complete geometric parameters, collaborative process parameter set, and measured performance data of 3 or more new bearing models are added to the process database. Level 3 trigger: When 10 or more sets of verified measured performance data are added to the collaborative process parameter set, and the data indicates that the boundaries of the existing collaborative process parameter set can be further optimized, the collaborative process parameter set will be updated.

[0047] The two-dimensional update rules are as follows: Dataset dimension update: New measured performance data is added to the training dataset, and training weights are assigned according to the reliability of the data. The weight of measured performance data of the same model is set to 1.0, the weight of measured performance data of similar models is set to 0.7, and the weight of simulation data is set to 0.3. The first level trigger only performs the dataset dimension update. Mapping relationship dimension update: Update the mapping relationship between bearing geometric parameters and optimal collaborative process parameter intervals in the process database, and retrain the first-level geometric parameter-optimal collaborative process parameter interval prediction model. The second and third levels trigger simultaneous updates of the dataset dimension and mapping relationship dimension.

[0048] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing heat treatment process parameters of GCr15 bearing steel bearing rings, characterized in that, Includes the following steps: S1. Obtain the bearing model, geometric parameters, performance constraints, and target heat treatment equipment capacity boundary of the bearing rings of the GCr15 bearing steel to be optimized. S2. Establish a set of collaborative process parameters for salt bath isothermal treatment, cryogenic treatment and tempering treatment. The set of collaborative process parameters includes independent process parameters for characterizing the set values ​​of each process and collaborative coupling parameters for characterizing the connection relationship between processes. S3. Based on the aforementioned geometric parameters, establish a continuous four-field fully coupled dynamic simulation model of temperature field, phase transition field, stress field, and carbide precipitation field for the entire process of salt bath-deep cryogenic-tempering, and verify the accuracy of the four-field fully coupled dynamic simulation model using physical experimental data; S4. Generate sample parameter combinations within the parameter space defined by the collaborative process parameter set, call the verified four-field fully coupled dynamic simulation model to obtain the simulation performance response data corresponding to each sample parameter combination, and form a training dataset. S5. Based on the training dataset, establish a two-level collaborative prediction model, which includes a first-level prediction model with bearing ring geometric parameters as input and the optimal collaborative process parameter range as output, and a second-level prediction model with the optimal collaborative process parameter range as input and the bearing ring predicted performance response data as output. S6. Taking the maximization of bearing ring hardness as the core optimization objective, a multi-objective collaborative optimization function is constructed, which includes residual austenite content, tangential residual compressive stress amplitude on raceway surface, and maximum residual tensile stress on raceway subsurface. Combining performance constraints and collaborative process parameter set, a particle swarm optimization algorithm with collaborative constraint penalty factor is used to perform global optimization on the verified two-level prediction model. During the optimization process, a three-level verification mechanism is implemented, which includes pre-verification of optimal collaborative process parameter range, verification of predicted performance response data, and verification of equipment capability and process robustness. The recommended combination of collaborative heat treatment process parameters is then output. S7. Conduct small-batch trial production according to the recommended combination of synergistic heat treatment process parameters, collect the measured performance data after the trial production, and feed the measured performance data back to the process database. When the measured performance data meets the performance constraints, output the final synergistic heat treatment process parameters. When the measured performance data does not meet the performance constraints, adjust the optimization algorithm parameters, constraints, or synergistic process parameter set and re-execute S6, while simultaneously feeding the measured performance data back to the process database, and execute a closed-loop iteration mechanism of three-level triggering and two-dimensional updating. S8. Write the bearing model, geometric parameters, final co-process heat treatment parameters, simulation performance data, measured performance data, and model iteration records into the process database for subsequent retrieval of similar models and parameter migration.

2. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S1, the geometric parameters include at least the inner diameter, outer diameter, width, contact angle, and effective wall thickness; the performance constraints include at least the lower limit of hardness, the range of residual austenite content, the lower limit of the amplitude of tangential residual compressive stress on the raceway surface, and the upper limit of the maximum residual tensile stress on the secondary surface of the raceway; the capability boundary of the target heat treatment equipment includes at least the temperature range, the heating rate range, and the cooling rate range.

3. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S2, the independent process parameters include at least the austenitizing temperature, austenitizing holding time, salt bath isothermal temperature, salt bath holding time, salt bath cooling rate, cryogenic temperature, cryogenic heating and cooling rate, cryogenic holding time, tempering temperature, tempering heating rate, and tempering holding time. The synergistic coupling parameters include at least the difference between the final salt bath cooling temperature and the initial cryogenic temperature, the interval between the end of cryogenic cooling and the start of tempering, the matching coefficient between the salt bath cooling rate and the cryogenic temperature drop rate, the ratio of the cryogenic holding time to the tempering holding time, and the matching coefficient between the cryogenic temperature and the tempering temperature.

4. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S3: The temperature field adopts the transient heat conduction equation that takes into account the effective wall thickness coupling, and heat transfer coefficient models are set for rapid heat transfer in salt bath, cryogenic heat transfer and tempering isothermal heat transfer respectively. The phase transformation field adopts a segmented collaborative modified phase transformation dynamics model, wherein the diffusion-type phase transformation from austenite to bainite is described by the Johnson-Mehl-Avrami equation, and the non-diffusion-type phase transformation from austenite to martensite during salt bath isothermal treatment and cryogenic treatment is described by the Koistinen-Marburger equation with the introduction of a bainite volume fraction correction term after salt bath isothermal treatment. The stress field incorporates carbide precipitation-induced strain into the total strain decomposition and uses the Norton creep model to describe stress relaxation during the tempering stage. The carbide precipitation field adopts a carbide precipitation kinetic model that incorporates a matrix dislocation density correction term after cryogenic treatment.

5. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S4, orthogonal experiments are first used to generate basic samples within the set of collaborative process parameters, and then Latin hypercube sampling is performed within the preset set of collaborative process parameters to generate extended samples.

6. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, The second-level prediction model includes independently trained sub-models for hardness prediction, residual austenite content prediction, raceway surface tangential residual compressive stress amplitude prediction, and raceway subsurface maximum residual tensile stress prediction. Each sub-model takes the optimal collaborative process parameter range as input and outputs corresponding single predicted performance response data.

7. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 6, characterized in that, Each sub-model of the second-level prediction model is constructed using a BP neural network with an attention mechanism, which is used to assign weights to the normalized optimal collaborative process parameter range features. The two-level collaborative prediction model is trained using the trainlm algorithm, and the model is deemed to have passed validation when the determination coefficient R² of each sub-model is not less than 0.

95.

8. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S6, the multi-objective collaborative optimization function is: ； In the formula: I max (X) is the overall objective function, d H (X) is the desired hardness function, d τ (X) is the expected function of the residual austenite content, d γ (X) is the desired function of the amplitude of the tangential residual compressive stress on the raceway surface, d δ (X) is the expected function of the maximum residual tensile stress on the secondary surface of the raceway; ω1, ω2, ω3, and ω4 are the synergistic weighting coefficients, assigned based on the weighting of the influence of salt bath, cryogenic, and tempering processes on each property, and ω1+ω2+ω3+ω4=1; the design variable X is the independent process parameter and synergistic coupling parameter in step S2.

9. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S6, the three-level verification includes: The optimal collaborative process parameter range pre-verification is used to determine whether the optimal collaborative process parameter range meets the value range requirements of the preset collaborative process parameter set in step S2. If it does not meet the requirements, the parameter combination is directly regenerated. Predicted performance response data verification is used to determine whether the predicted performance response data output in step S5 meets the performance constraints in step S1. If it does not meet the constraints, the algorithm iterative optimization will proceed. Equipment capability and process robustness verification is used to determine whether the optimal range of synergistic process parameters is within the capability range of the target heat treatment equipment. At the same time, it simulates the performance stability under ±5% process parameter fluctuations and outputs the process parameter combination with the best robustness as the recommended synergistic heat treatment process parameter combination.

10. The method for optimizing the heat treatment process parameters of GCr15 bearing steel bearing rings according to claim 1, characterized in that, In step S7, The three-level triggering includes: When the deviation between the measured performance data and the predicted performance response data of the same type of bearing exceeds a preset threshold, the fine-tuning of the two-level collaborative prediction model is triggered. Adding geometric parameters, collaborative process parameters, and measured performance data of three or more new bearing models triggers the expansion of the training dataset and retraining of the two-level collaborative prediction model. When newly verified measured performance data is added to the collaborative process parameter set and the measured performance data indicates that the boundary of the collaborative process parameter set needs to be adjusted, the collaborative process parameter set boundary update is triggered. The two-dimensional update includes: In terms of dataset dimension, actual performance data is added to the training dataset, and training weights are assigned according to bearing model similarity and cooperative matching degree; In terms of mapping relationships, update the mapping relationship between geometric parameters and the optimal collaborative process parameter range in the process database.

Images