Bayesian optimization-based low-temperature environment gear fatigue strength enhancement heat treatment method

Abstract

The invention discloses a low-temperature environment gear fatigue strength enhancing heat treatment method based on Bayesian optimization. The low-temperature environment gear fatigue strength enhancing heat treatment method comprises the following steps that S1, an alloy steel material suitable for a low-temperature environment is selected and optimized; s2, microstructure analysis is conducted on the material, and key features are quantified; s3, optimizing process parameters such as heating and cooling by adopting an improved Bayesian optimization method; s4, adjusting process parameters in real time through adaptive fuzzy reasoning; s5, predicting the fatigue life by using an extended Kalman filtering algorithm; s6, optimizing process parameters by adopting a self-adaptive multi-scale Bayesian optimization algorithm; and S7, applying the optimized process to batch production. According to the method, the anti-fatigue performance of the gear in the low-temperature environment is remarkably improved, the service life is prolonged, and the stability and consistency of the process are ensured through intelligent real-time adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of heat treatment technology for metal materials, and in particular to a heat treatment method for enhancing fatigue strength of gears in a low-temperature environment based on Bayesian optimization. Background Art

[0002] In the modern machinery industry, gears, as core components in mechanical transmission systems, are widely used in various equipment working in low-temperature environments, such as aerospace, marine engineering, and polar exploration equipment. The fatigue strength and service life of gears directly affect the performance and reliability of these equipment. Therefore, improving the fatigue strength of gears in low-temperature environments and extending their service life are key technical links to ensure the long-term stable operation of mechanical equipment.

[0003] Existing gear heat treatment technologies mainly improve the microstructure of gear materials by adjusting process parameters such as heating temperature, holding time, cooling rate, etc., thereby improving the hardness and fatigue resistance of gears. However, these traditional heat treatment technologies still have some obvious technical defects in the application of low temperature environments:

[0004] The first is that the parameter optimization of existing heat treatment processes mostly relies on experience and repeated tests, lacking systematic and intelligent optimization methods. In low-temperature environments, the microstructure and mechanical properties of gear materials change in a complex manner, and traditional heat treatment processes are difficult to accurately control various parameters to meet the needs of extreme environments. Experience and test methods are not only inefficient, but also easily cause unforeseen fatigue damage due to changes in the microstructure of the material, resulting in a significant shortening of the service life of the gear under low-temperature conditions.

[0005] The second is that existing heat treatment technologies are difficult to cope with the multi-scale effects of material performance in low-temperature environments. The organizational characteristics of materials at the nano, micro and macro scales may show significant differences under low-temperature conditions. Traditional single-scale heat treatment processes cannot fully optimize the organizational characteristics of different scales, and thus cannot achieve the best fatigue strength improvement effect. Especially under low-temperature conditions, the heterogeneity of microstructural characteristics increases the complexity of fatigue failure. Traditional processes are difficult to fully take into account the optimization needs of each scale, which often leads to a contradiction between the strengthening of local microstructural characteristics and the optimization of overall performance.

[0006] The third is that the parameter adjustment in the existing technology lacks intelligent and adaptive mechanisms. In the actual production process, changes in material properties and process conditions may lead to deviations in process parameters. Traditional process control methods rely on manual monitoring and adjustment, which makes it difficult to respond to dynamic changes in the process in a timely manner. This lagging adjustment mechanism is prone to fluctuations in material properties and cannot guarantee the stability and consistency of the heat treatment process. Especially in extreme low temperature environments, slight deviations in process parameters may cause a significant decline in gear performance.

[0007] Therefore, how to provide a heat treatment method for enhancing the fatigue strength of gears in low-temperature environments based on Bayesian optimization is an urgent problem that technicians in this field need to solve. Summary of the invention

[0008] One object of the present invention is to propose a heat treatment method for enhancing fatigue strength of gears in low temperature environment based on Bayesian optimization. The method systematically optimizes various parameters of gear heat treatment process by introducing Bayesian optimization algorithm and combining multi-scale microstructure feature analysis. Specifically, the present invention uses adaptive multi-scale Bayesian optimization algorithm to quantify and optimize the microstructure characteristics of materials at three scales: nanometer, micrometer and macro, and tracks stress-strain data in real time through extended Kalman filter algorithm, so as to dynamically adjust heat treatment process parameters and generate fatigue life prediction model. Compared with traditional empirical process optimization methods, the present invention has the following advantages: first, the intelligent parameter optimization process improves the efficiency and accuracy of the process and reduces the number and time of experiments; second, the multi-scale optimization method significantly improves the fatigue resistance of gear materials in low temperature environment and prolongs the service life; finally, the adaptive real-time adjustment mechanism ensures the stability and consistency of the process, and overcomes the performance fluctuation problem caused by environmental changes in traditional methods. The present invention shows significant advantages in improving the performance of gears in low temperature environment and provides an efficient and reliable solution for related fields.

[0009] A heat treatment method for enhancing fatigue strength of gears in a low temperature environment based on Bayesian optimization according to an embodiment of the present invention comprises the following steps:

[0010] S1. Select alloy steel materials suitable for low temperature environments and optimize the proportion of alloy elements to optimize the fatigue resistance and toughness of gear materials under low temperature conditions, while taking into account the machinability and cost factors of the materials;

[0011] S2. Analyze the microstructure of the selected material using electron backscatter diffraction technology to obtain grain structure, phase distribution and dislocation density data, and quantify the obtained grain structure, phase distribution and dislocation density data into numerical parameters;

[0012] S3. Using an improved Bayesian optimization method, based on the quantified grain structure, phase distribution and dislocation density numerical parameters, multiple rounds of intelligent iterative optimization of the heat treatment process parameters of heating temperature, heating rate, holding time, cooling medium and cooling speed are performed. During the optimization process, a hybrid model based on Latin hypercube sampling and Gaussian process regression is used to optimize the parameter search space;

[0013] S4. After each round of optimization, the temperature distribution and stress distribution of the gear are monitored in real time, the monitoring data are analyzed using the adaptive fuzzy reasoning method, and the heat treatment process parameters are dynamically adjusted according to the analysis results;

[0014] S5. When performing fatigue tests on gear samples after optimized heat treatment in a low temperature environment, an algorithm based on extended Kalman filtering is combined to track stress-strain data in real time, predict the critical points of crack initiation and propagation, and generate a fatigue life prediction model;

[0015] S6. Based on stress-strain data and fatigue life prediction model, an adaptive multi-scale Bayesian optimization algorithm is used to dynamically adjust the weights of heat treatment process parameters at the nano, micro and macro levels;

[0016] S7. Apply the adjusted heat treatment process parameters to mass production, monitor the parameter deviation in the production process in real time, and adjust the process parameters in time through an adaptive feedback mechanism.

[0017] Optionally, the S3 specifically includes:

[0018] S31, input the quantified grain structure, phase distribution and dislocation density numerical parameters into the improved Bayesian optimization model, denoted as input parameter set X = {x1, x2, x3, ..., x n}, where x i represents the value corresponding to the i-th microstructure feature;

[0019] S32, determine a heat treatment process parameter set, denoted as P = {p1, p2, p3, p4, p5}, where p1 is the heating temperature, p2 is the heating rate, p3 is the holding time, p4 is the cooling medium, and p5 is the cooling rate;

[0020] S33, using Latin hypercube sampling method to perform initial sampling of the sample space of the input parameter set X and the heat treatment process parameter set P, generating a sample set S = {(X1, P1), (X2, P2), ..., (X m ,P m )}, where each sample point (X j ,P j ) contains a set of input parameters X j And the corresponding heat treatment process parameters P j ;

[0021] S34, fitting the sample set S based on a Gaussian process regression model to construct a multidimensional prediction function f(X, P), wherein f(X, P) is used to predict the effects of different combinations of heat treatment process parameters on gear fatigue strength;

[0022] S35. Under the guidance of the improved Bayesian optimization method, the heat treatment process parameter set P is adjusted by dynamically selecting the optimal sample points for iteration through the expected improvement criterion combined with the hybrid model of Latin hypercube sampling and Gaussian process regression;

[0023] S36. Through the improved Bayesian optimization method, the weight parameters in the hybrid model combining Latin hypercube sampling and Gaussian process regression are adaptively adjusted, and the optimal heat treatment process parameter combination is generated by gradually converging in multiple rounds of iterations.

[0024] Optionally, the S35 specifically includes:

[0025] S351. Based on the current prediction function f(X,P), the improved efficiency function value IEF(X,P) of the input parameter set X and the heat treatment process parameter set P at all candidate sample points is calculated, wherein X represents the quantified microstructure characteristic parameters, including grain structure, phase distribution and dislocation density, and P represents the heat treatment process parameter set, including heating temperature, heating rate, holding time, cooling medium and cooling rate;

[0026] S352, in each iteration, a new set of candidate sample points is generated based on the Latin hypercube sampling method, denoted as S' = {(X'1, P'1), (X'2, P'2), ..., (X' k ,P' k )}, and combined with the current prediction model, the improved efficiency function value IEF(X' j ,P' j );

[0027] S353, the improved Bayesian optimization method is used to analyze the improved efficiency function value, and the comprehensive efficiency improvement index CEI (X' j ,P' j ):

[0028]

[0029] Among them, γ and δ are adaptive adjustment factors, λ is the adjustment coefficient, σ i (X' j ) represents the prediction uncertainty of the i-th microstructural feature, σ max (X' j ) is the value with the largest prediction uncertainty among all features;

[0030] S354, select the comprehensive performance improvement index CEI (X' j ,P' j ) is the largest sample point, denoted as (X * ,P *), take it as the sampling point for the next round of iteration, and update the sample set;

[0031] S355, the selected optimal sample point (X * ,P * ) is added to the existing sample set. After updating the sample space, the new sample set is refitted through the Gaussian process regression model to correct the prediction function f(X,P);

[0032] S356, evaluating the new prediction function, if the increase of the comprehensive performance improvement index gradually decreases, judging whether to continue the iteration according to the Bayesian optimization criterion, if continuing, returning to S352 to generate a new set of candidate sample points, until the comprehensive performance improvement index value converges or reaches the preset optimization standard;

[0033] S357, when the comprehensive efficiency improvement index value converges, determine the heat treatment process parameter combination P * .

[0034] Optionally, the S36 specifically includes:

[0035] S361. After each round of iteration, according to the current comprehensive performance improvement index CEI (X * ,P * ), dynamically adjust the weight parameters in the Bayesian optimization model, including the adjustment factor γ, the adjustment coefficient λ, and the uncertainty weight factor δ;

[0036] S362, adding the new sample points generated in each round of iteration to the sample set S to form an updated sample set S new =S∪(X * ,P * ), and recalculate the overall prediction uncertainty of the sample set, denoted as σ total (S new ),in is the jth microstructure feature in the sample set, ω j is an adaptive weight factor based on feature importance;

[0037] S363, based on the updated overall forecast uncertainty σ total (S new ), modify the Gaussian process regression model and adjust the confidence interval of the prediction function f(X,P);

[0038] S364, applying the modified prediction function f(X,P) to the next round of Bayesian optimization process, using the improved Bayesian optimization method, based on the new sample set S new Recalculate the expected improvement value and comprehensive performance improvement index, and dynamically generate the next round of candidate sample point sets;

[0039] S365. In each round of optimization, the corresponding comprehensive performance improvement index is calculated based on different scales of micro-structure characteristics, and a multi-scale interaction strategy is adopted to perform nonlinear weighted calculation on the CEI values ​​of different scales to form the final multi-scale performance interaction index MSI(X,P);

[0040] S366. According to the results of the multi-scale performance interaction index MSI (X, P), the adjustment parameters in the Bayesian optimization model are dynamically adjusted so that the optimization process gradually converges at the nano, micro and macro scales, and finally generates a global heat treatment process parameter combination P. * .

[0041] Optionally, the S5 specifically includes:

[0042] S51. During the fatigue test of the gear sample, real-time stress-strain data is collected;

[0043] S52, using the extended Kalman filter algorithm to filter the real-time collected stress-strain data to predict the dynamic change trend of stress and strain. The filtered data is recorded as in and are the estimated values ​​of stress and strain after filtering at the i-th sampling point;

[0044] S53. Based on the filtered stress-strain data, a fatigue life prediction model is constructed, denoted as Where L f represents the fatigue life of the gear, T represents the temperature, and the model predicts the critical point of crack initiation by correlating stress, strain and temperature parameters;

[0045] S54. To optimize the accuracy of fatigue life prediction, the stress-strain data is decomposed according to different time scales to obtain a multi-scale stress signal set, denoted as D”(t)={D”1(t),D”2(t),...,D” m (t)}, where D” k (t) is the stress-strain signal on the kth time scale;

[0046] S55. Input the multi-scale stress signal set D”(t) into the fatigue life prediction model and introduce the adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t);

[0047] S56, by modulating the fatigue life prediction value L on different time scales f,k(t) Perform nonlinear superposition to generate multi-scale comprehensive fatigue life prediction values

[0048] S57, based on multi-scale comprehensive fatigue life prediction value Determine the critical points of crack initiation and propagation, generate crack prediction maps, and optimize and adjust heat treatment process parameters based on crack initiation trends.

[0049] Optionally, the S55 specifically includes:

[0050] S551, input the multi-scale stress signal set D" (t) into the fatigue life prediction model, where D" k (t) represents the stress-strain signal on the kth time scale;

[0051] S552, introduce adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t):

[0052]

[0053] in, is the phase shift factor on the kth time scale, which is used to capture the interaction between stress and strain at different time scales and dynamically adjust the shape of the stress-strain curve. and are the estimated values ​​of stress and strain after filtering on the kth time scale, respectively, and f k is the fatigue life prediction sub-model at the kth time scale, T is the temperature parameter in the low temperature environment;

[0054] S553, dynamically adjust the phase shift factor according to temperature T and time t and the phase correction factor δ k (T,t):

[0055]

[0056] Among them, δ k (T, t) is the phase correction factor related to temperature T, which is used to compensate for the nonlinear stress-strain behavior in low temperature environment.

[0057] Optionally, the S6 specifically includes:

[0058] S61. Based on stress-strain data and fatigue life prediction model, a multi-scale Bayesian optimization model is constructed. The multi-scale Bayesian optimization model combines the microstructure characteristics of nanometer, micrometer and macro scales for optimization, denoted as X = {X n ,Xμ ,X M}, where X n , X μ , X M They represent the microstructure characteristic parameter sets at nanometer, micrometer and macroscale, respectively;

[0059] S62, introduce adaptive multi-scale Bayesian optimization algorithm, through adaptive weight allocation mechanism, the heat treatment process parameter set P = {P1, P2,, P m} Optimize the heat treatment process parameters including heating temperature, heating rate, holding time, cooling medium and cooling speed;

[0060] S63. In the adaptive multi-scale Bayesian optimization algorithm, the weight allocation mechanism dynamically adjusts the weight of each scale by considering the stress-strain response characteristics at each scale. l (t);

[0061] S64, through multiple rounds of iterations, in each iteration, according to the feedback of stress-strain data, further optimize the process parameter set P, and dynamically adjust the weight ω of each scale l (t), until the objective function f(X,P) converges, and the final optimized process parameter combination P is obtained * .

[0062] Optionally, the S63 specifically includes:

[0063] S631. At the beginning of each iteration, obtain the current stress-strain data D(t) = {σ(t), ε(t)}, and calculate the stress-strain response characteristics associated with different time scales, denoted as R l (t), where l represents nanometer, micrometer, and macroscale;

[0064] S632, stress-strain response characteristics R l (t) Perform second-order time derivative calculations to obtain stress-strain acceleration responses at each scale And introduce the nonlinear modulation factor κ l (t) as an exponential function of time:

[0065]

[0066] in, is the time-dependent nonlinear modulation factor, λ l is the modulation index used to control the nonlinear change of the second-order derivative;

[0067] S633, based on the calculation of the second-order derivative in S632 Calculate the weight adjustment factor ω in the current iteration l(t), and combined with the phase modulation mechanism:

[0068]

[0069] Among them, ω l0 is the initial weight, α l is the weight adjustment coefficient, is the time-dependent phase shift factor, τ l represents the characteristic time period of the lth scale, represents the phase modulation factor at the lth scale;

[0070] S634, in each round of iteration, update the weight ω of each scale in real time l (t), and apply it to the objective function of the Bayesian optimization algorithm:

[0071]

[0072] Among them, f l (X l ,P) represents the sub-objective function of the lth scale, reflecting the microstructure characteristic parameter X at the lth scale l and the influence of heat treatment process parameters P on fatigue life;

[0073] S635: During the iterative optimization process, the weight adjustment coefficient α is dynamically adjusted according to the changes in the objective function and the feedback of real-time stress-strain data. l and the characteristic time period τ l .

[0074] The beneficial effects of the present invention are:

[0075] (1) The present invention introduces an adaptive multi-scale Bayesian optimization algorithm to systematically quantify and optimize the microstructure characteristics of gear materials at three scales: nanometer, micrometer and macro. This multi-scale optimization method effectively overcomes the limitations of traditional single-scale heat treatment processes, allowing gear materials to exhibit higher fatigue resistance and toughness in low-temperature environments, thereby significantly extending the service life of gears.

[0076] (2) The present invention dynamically tracks stress-strain data and generates a fatigue life prediction model by real-time monitoring and analysis of temperature distribution and stress distribution during heat treatment, combined with an extended Kalman filter algorithm. Compared with the traditional parameter adjustment method that relies on experience and experiments, the intelligent real-time adjustment mechanism of the present invention can respond to changes in material properties in a timely manner during the process, ensure the stability and consistency of the process, and reduce process deviations caused by environmental factors.

[0077] (3) The present invention monitors and adjusts process parameters in real time through an adaptive feedback mechanism in batch production, effectively avoiding the performance fluctuation problem caused by the lag of manual monitoring in traditional processes. Through intelligent process adjustment, the present invention not only improves production efficiency, but also ensures the consistency of product quality, so that the overall performance of gears in low temperature environments is fully optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0079] Figure 1 A flowchart of a heat treatment method for enhancing fatigue strength of gears in a low temperature environment based on Bayesian optimization proposed by the present invention;

[0080] Figure 2 This is a flow chart of heat treatment process parameter optimization based on Bayesian optimization in a heat treatment method for fatigue strength enhancement of gears in a low-temperature environment based on Bayesian optimization proposed by the present invention. DETAILED DESCRIPTION

[0081] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0082] refer to Figure 1-2 , a heat treatment method for enhancing fatigue strength of gears in a low temperature environment based on Bayesian optimization, comprising the following steps:

[0083] S1. Select alloy steel materials suitable for low temperature environments and optimize the proportion of alloy elements to optimize the fatigue resistance and toughness of gear materials under low temperature conditions, while taking into account the machinability and cost factors of the materials;

[0084] S2. Analyze the microstructure of the selected material using electron backscatter diffraction technology to obtain grain structure, phase distribution and dislocation density data, and quantify the obtained grain structure, phase distribution and dislocation density data into numerical parameters;

[0085] S3. Using an improved Bayesian optimization method, based on the quantified grain structure, phase distribution and dislocation density numerical parameters, multiple rounds of intelligent iterative optimization of the heat treatment process parameters of heating temperature, heating rate, holding time, cooling medium and cooling speed are performed. During the optimization process, a hybrid model based on Latin hypercube sampling and Gaussian process regression is used to optimize the parameter search space;

[0086] S4. After each round of optimization, the temperature distribution and stress distribution of the gear are monitored in real time, the monitoring data are analyzed using the adaptive fuzzy reasoning method, and the heat treatment process parameters are dynamically adjusted according to the analysis results;

[0087] S5. When performing fatigue tests on gear samples after optimized heat treatment in a low temperature environment, an algorithm based on extended Kalman filtering is combined to track stress-strain data in real time, predict the critical points of crack initiation and propagation, and generate a fatigue life prediction model;

[0088] S6. Based on stress-strain data and fatigue life prediction model, an adaptive multi-scale Bayesian optimization algorithm is used to dynamically adjust the weights of heat treatment process parameters at the nano, micro and macro levels;

[0089] S7. Apply the adjusted heat treatment process parameters to mass production, monitor the parameter deviation in the production process in real time, and adjust the process parameters in time through an adaptive feedback mechanism.

[0090] In this implementation, S3 specifically includes:

[0091] S31, input the quantified grain structure, phase distribution and dislocation density numerical parameters into the improved Bayesian optimization model, denoted as input parameter set X = {x1, x2, x3, ..., x n}, where x i represents the value corresponding to the i-th microstructure feature;

[0092] S32, determine a heat treatment process parameter set, denoted as P = {p1, p2, p3, p4, p5}, where p1 is the heating temperature, p2 is the heating rate, p3 is the holding time, p4 is the cooling medium, and p5 is the cooling rate;

[0093] S33, using Latin hypercube sampling method to perform initial sampling of the sample space of the input parameter set X and the heat treatment process parameter set P, generating a sample set S = {(X1, P1), (X2, P2), ..., (X m ,P m )}, where each sample point (X j ,P j ) contains a set of input parameters X j And the corresponding heat treatment process parameters P j ;

[0094] S34, fitting the sample set S based on a Gaussian process regression model to construct a multidimensional prediction function f(X, P), wherein f(X, P) is used to predict the influence of different heat treatment process parameter combinations on the fatigue strength of the gear;

[0095] S35. Under the guidance of the improved Bayesian optimization method, the heat treatment process parameter set P is adjusted by dynamically selecting the optimal sample points for iteration through the expected improvement criterion combined with the hybrid model of Latin hypercube sampling and Gaussian process regression;

[0096] S36. Through the improved Bayesian optimization method, the weight parameters in the hybrid model combining Latin hypercube sampling and Gaussian process regression are adaptively adjusted, and the optimal heat treatment process parameter combination is generated by gradually converging in multiple rounds of iterations.

[0097] In this implementation manner, the S35 specifically includes:

[0098] S351. Based on the current prediction function f(X,P), the improved efficiency function value IEF(X,P) of the input parameter set X and the heat treatment process parameter set P at all candidate sample points is calculated, wherein X represents the quantified microstructure characteristic parameters, including grain structure, phase distribution and dislocation density, and P represents the heat treatment process parameter set, including heating temperature, heating rate, holding time, cooling medium and cooling rate;

[0099] S352, in each iteration, a new set of candidate sample points is generated based on the Latin hypercube sampling method, denoted as S' = {(X'1, P'1), (X'2, P'2), ..., (X' k ,P' k )}, and combined with the current prediction model, the improved efficiency function value IEF(X' j ,P' j );

[0100] S353, the improved Bayesian optimization method is used to analyze the improved efficiency function value, and the comprehensive efficiency improvement index CEI (X' j ,P' j ):

[0101]

[0102] Among them, γ and δ are adaptive adjustment factors, λ is the adjustment coefficient, σ i (X' j ) represents the prediction uncertainty of the i-th microstructural feature, σ max (X' j ) is the value with the largest prediction uncertainty among all features;

[0103] S354, select the comprehensive performance improvement index CEI (X' j ,P' j ) is the largest sample point, denoted as (X * ,P *), take it as the sampling point for the next round of iteration, and update the sample set;

[0104] S355, the selected optimal sample point (X * ,P * ) is added to the existing sample set. After updating the sample space, the new sample set is refitted through the Gaussian process regression model to correct the prediction function f(X,P);

[0105] S356, evaluating the new prediction function, if the increase of the comprehensive performance improvement index gradually decreases, judging whether to continue the iteration according to the Bayesian optimization criterion, if continuing, returning to S352 to generate a new set of candidate sample points, until the comprehensive performance improvement index value converges or reaches the preset optimization standard;

[0106] S357, when the comprehensive efficiency improvement index value converges, determine the heat treatment process parameter combination P * .

[0107] In this implementation manner, the S36 specifically includes:

[0108] S361. After each round of iteration, according to the current comprehensive performance improvement index CEI (X * ,P * ), dynamically adjust the weight parameters in the Bayesian optimization model, including the adjustment factor γ, the adjustment coefficient λ, and the uncertainty weight factor δ;

[0109] S362, adding the new sample points generated in each round of iteration to the sample set S to form an updated sample set S new =S∪(X * ,P * ), and recalculate the overall prediction uncertainty of the sample set, denoted as σ total (S new ),in is the jth microstructure feature in the sample set, ω j is an adaptive weight factor based on feature importance;

[0110] S363, based on the updated overall forecast uncertainty σ total (S new ), modify the Gaussian process regression model and adjust the confidence interval of the prediction function f(X,P);

[0111] S364, applying the modified prediction function f(X,P) to the next round of Bayesian optimization process, using the improved Bayesian optimization method, based on the new sample set S new Recalculate the expected improvement value and comprehensive performance improvement index, and dynamically generate the next round of candidate sample point sets;

[0112] S365. In each round of optimization, the corresponding comprehensive performance improvement index is calculated based on different scales of micro-structure characteristics, and a multi-scale interaction strategy is adopted to perform nonlinear weighted calculation on the CEI values ​​of different scales to form the final multi-scale performance interaction index MSI(X,P);

[0113] S366. According to the results of the multi-scale performance interaction index MSI (X, P), the adjustment parameters in the Bayesian optimization model are dynamically adjusted so that the optimization process gradually converges at the nano, micro and macro scales, and finally generates a global heat treatment process parameter combination P. * .

[0114] In this implementation manner, S5 specifically includes:

[0115] S51. During the fatigue test of the gear sample, real-time stress-strain data is collected;

[0116] S52, using the extended Kalman filter algorithm to filter the real-time collected stress-strain data to predict the dynamic change trend of stress and strain. The filtered data is recorded as in and are the estimated values ​​of stress and strain after filtering at the i-th sampling point;

[0117] S53. Based on the filtered stress-strain data, a fatigue life prediction model is constructed, denoted as Where L f represents the fatigue life of the gear, T represents the temperature, and the model predicts the critical point of crack initiation by correlating stress, strain and temperature parameters;

[0118] S54. To optimize the accuracy of fatigue life prediction, the stress-strain data is decomposed according to different time scales to obtain a multi-scale stress signal set, denoted as D”(t)={D”1(t),D”2(t),...,D” m (t)}, where D” k (t) is the stress-strain signal on the kth time scale;

[0119] S55. Input the multi-scale stress signal set D”(t) into the fatigue life prediction model and introduce the adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t);

[0120] S56, by modulating the fatigue life prediction value L on different time scales f,k(t) Perform nonlinear superposition to generate multi-scale comprehensive fatigue life prediction values

[0121] S57, based on multi-scale comprehensive fatigue life prediction value Determine the critical points of crack initiation and propagation, generate crack prediction maps, and optimize and adjust heat treatment process parameters based on crack initiation trends.

[0122] In this implementation manner, the S55 specifically includes:

[0123] S551, input the multi-scale stress signal set D" (t) into the fatigue life prediction model, where D" k (t) represents the stress-strain signal on the kth time scale;

[0124] S552, introduce adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t):

[0125]

[0126] in, is the phase shift factor on the kth time scale, which is used to capture the interaction between stress and strain at different time scales and dynamically adjust the shape of the stress-strain curve. and are the estimated values ​​of stress and strain after filtering on the kth time scale, respectively, and f k is the fatigue life prediction sub-model at the kth time scale, T is the temperature parameter in the low temperature environment;

[0127] S553, dynamically adjust the phase shift factor according to temperature T and time t and the phase correction factor δ k (T,t):

[0128]

[0129] Among them, δ k (T, t) is the phase correction factor related to temperature T, which is used to compensate for the nonlinear stress-strain behavior in low temperature environment.

[0130] In this implementation, it is characterized in that the S6 specifically includes:

[0131] S61. Based on stress-strain data and fatigue life prediction model, a multi-scale Bayesian optimization model is constructed. The multi-scale Bayesian optimization model combines the microstructure characteristics of nanometer, micrometer and macro scales for optimization, denoted as X = {X n,X μ ,X M}, where X n , X μ , X M They represent the microstructure characteristic parameter sets at nanometer, micrometer and macroscale, respectively;

[0132] S62, introduce adaptive multi-scale Bayesian optimization algorithm, through adaptive weight allocation mechanism, the heat treatment process parameter set P = {P1, P2,, P m} Optimize the heat treatment process parameters including heating temperature, heating rate, holding time, cooling medium and cooling speed;

[0133] S63. In the adaptive multi-scale Bayesian optimization algorithm, the weight allocation mechanism dynamically adjusts the weight of each scale by considering the stress-strain response characteristics at each scale. l (t);

[0134] S64, through multiple rounds of iterations, in each iteration, according to the feedback of stress-strain data, further optimize the process parameter set P, and dynamically adjust the weight ω of each scale l (t), until the objective function f(X,P) converges, and the final optimized process parameter combination P is obtained * .

[0135] In this implementation manner, the S63 specifically includes:

[0136] S631. At the beginning of each iteration, obtain the current stress-strain data D(t) = {σ(t), ε(t)}, and calculate the stress-strain response characteristics associated with different time scales, denoted as R l (t), where l represents nanometer, micrometer, and macroscale;

[0137] S632, stress-strain response characteristics R l (t) Perform second-order time derivative calculations to obtain stress-strain acceleration responses at each scale And introduce the nonlinear modulation factor κ l (t) as an exponential function of time:

[0138]

[0139] in, is the time-dependent nonlinear modulation factor, λ l is the modulation index used to control the nonlinear change of the second-order derivative;

[0140] S633, based on the calculation of the second-order derivative in S632 Calculate the weight adjustment factor ω in the current iteration l(t), and combined with the phase modulation mechanism:

[0141]

[0142] Among them, ω l0 is the initial weight, α l is the weight adjustment coefficient, is the time-dependent phase shift factor, τ l represents the characteristic time period of the lth scale, represents the phase modulation factor at the lth scale;

[0143] S634, in each round of iteration, update the weight ω of each scale in real time l (t), and apply it to the objective function of the Bayesian optimization algorithm:

[0144]

[0145] Among them, f l (X l ,P) represents the sub-objective function of the lth scale, reflecting the microstructure characteristic parameter X at the lth scale l and the influence of heat treatment process parameters P on fatigue life;

[0146] S635: During the iterative optimization process, the weight adjustment coefficient α is dynamically adjusted according to the changes in the objective function and the feedback of real-time stress-strain data. l and the characteristic time period τ l .

[0147] Embodiment 1:

[0148] The Bayesian optimization-based heat treatment method for enhancing fatigue strength of gears in low-temperature environments of the present invention was implemented in Company B, a low-temperature manufacturing base in Northeast China. Company B mainly produces and manufactures high-strength gears for use in extreme low-temperature environments, which are widely used in aerospace, polar exploration equipment, marine engineering and other fields.

[0149] In the traditional gear heat treatment process, Company B faced problems such as low efficiency in process parameter optimization and large fluctuations in product quality. In particular, the fatigue resistance of gears was difficult to guarantee in low-temperature environments. In order to improve product performance and consistency, Company B decided to introduce the intelligent heat treatment method of the present invention, and achieved comprehensive optimization and dynamic adjustment of process parameters through Bayesian optimization and multi-scale microstructure feature analysis.

[0150] During the implementation process, Company B conducted multiple rounds of process parameter optimization and fatigue life prediction tests for gears of different materials and sizes. By applying the system of the present invention, Company B significantly improved the fatigue resistance of gears, especially in low temperature environments, and the service life of gears was greatly improved.

[0151] Table 1 Comparison of heat treatment methods for gear fatigue strength enhancement in low temperature environments before and after implementation

[0152]

[0153] As shown in Table 1, after implementing the method of the present invention, the average fatigue life of the gear in a low temperature environment is increased from the original 5000 hours to 8500 hours, an increase of 70%. The process parameter optimization time is shortened from 45 days to 15 days, and the efficiency is increased by 3 times. Through real-time monitoring and dynamic adjustment of process parameters, the process parameter deviation rate is reduced from the original 5.5% to 0.7%, ensuring the accuracy and consistency of the production process. In addition, the product qualification rate is also increased from 89% to 97.5%.

[0154] By implementing the present invention, the fatigue resistance of the gears of Company B in low-temperature environments has been significantly improved, the product quality and production efficiency have been greatly improved, and the company's market competitiveness has also been enhanced.

[0155] In summary, the low-temperature environment gear fatigue strength enhancement heat treatment method based on Bayesian optimization of the present invention not only achieves innovative breakthroughs in technology, but also demonstrates significant economic and social benefits in practical applications.

[0156] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A heat treatment method for enhancing fatigue strength of gears in low temperature environment based on Bayesian optimization, characterized in that: The steps include: S1. Select alloy steel materials suitable for low temperature environments and optimize the proportion of alloy elements to optimize the fatigue resistance and toughness of gear materials under low temperature conditions, while taking into account the machinability and cost factors of the materials; S2. Analyze the microstructure of the selected material using electron backscatter diffraction technology to obtain grain structure, phase distribution and dislocation density data, and quantify the obtained grain structure, phase distribution and dislocation density data into numerical parameters; S3. Using an improved Bayesian optimization method, based on the quantified grain structure, phase distribution and dislocation density numerical parameters, multiple rounds of intelligent iterative optimization of the heat treatment process parameters of heating temperature, heating rate, holding time, cooling medium and cooling speed are performed. During the optimization process, a hybrid model based on Latin hypercube sampling and Gaussian process regression is used to optimize the parameter search space; S4. After each round of optimization, the temperature distribution and stress distribution of the gear are monitored in real time, the monitoring data are analyzed using the adaptive fuzzy reasoning method, and the heat treatment process parameters are dynamically adjusted according to the analysis results; S5. When performing fatigue tests on gear samples after optimized heat treatment in a low temperature environment, an algorithm based on extended Kalman filtering is combined to track stress-strain data in real time, predict the critical points of crack initiation and propagation, and generate a fatigue life prediction model; S6. Based on stress-strain data and fatigue life prediction model, an adaptive multi-scale Bayesian optimization algorithm is used to dynamically adjust the weights of heat treatment process parameters at the nano, micro and macro levels; S7. Apply the adjusted heat treatment process parameters to mass production, monitor the parameter deviation in the production process in real time, and adjust the process parameters in time through an adaptive feedback mechanism.

2. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 1, characterized in that: The S3 specifically includes: S31, input the quantified grain structure, phase distribution and dislocation density numerical parameters into the improved Bayesian optimization model, denoted as input parameter set X = {x1, x2, x3, ..., x n }, where x i represents the value corresponding to the i-th microstructure feature; S32, determine a heat treatment process parameter set, denoted as P = {p1, p2, p3, p4, p5}, where p1 is the heating temperature, p2 is the heating rate, p3 is the holding time, p4 is the cooling medium, and p5 is the cooling rate; S33, using Latin hypercube sampling method to perform initial sampling of the sample space of the input parameter set X and the heat treatment process parameter set P, generating a sample set S = {(X1, P1), (X2, P2), ..., (X m ,P m )}, where each sample point (X j ,P j ) contains a set of input parameters X j And the corresponding heat treatment process parameters P j ; S34, fitting the sample set S based on a Gaussian process regression model to construct a multidimensional prediction function f(X, P), wherein f(X, P) is used to predict the effects of different combinations of heat treatment process parameters on gear fatigue strength; S35. Under the guidance of the improved Bayesian optimization method, the heat treatment process parameter set P is adjusted by dynamically selecting the optimal sample points for iteration through the expected improvement criterion combined with the hybrid model of Latin hypercube sampling and Gaussian process regression; S36. Through the improved Bayesian optimization method, the weight parameters in the hybrid model combining Latin hypercube sampling and Gaussian process regression are adaptively adjusted, and the optimal heat treatment process parameter combination is generated by gradually converging in multiple rounds of iterations.

3. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 2, characterized in that: The S35 specifically includes: S351. Based on the current prediction function f(X,P), the improved efficiency function value IEF(X,P) of the input parameter set X and the heat treatment process parameter set P at all candidate sample points is calculated, wherein X represents the quantified microstructure characteristic parameters, including grain structure, phase distribution and dislocation density, and P represents the heat treatment process parameter set, including heating temperature, heating rate, holding time, cooling medium and cooling rate; S352, in each iteration, a new set of candidate sample points is generated based on the Latin hypercube sampling method, denoted as S' = {(X'1, P'1), (X'2, P'2), ..., (X' k ,P' k )}, and combined with the current prediction model, the improved efficiency function value IEF(X' j ,P' j ); S353, the improved Bayesian optimization method is used to analyze the improved efficiency function value, and the comprehensive efficiency improvement index CEI (X' j ,P' j ): Among them, γ and δ are adaptive adjustment factors, λ is the adjustment coefficient, σ i (X' j ) represents the prediction uncertainty of the i-th microstructural feature, σ max (X' j ) is the value with the largest prediction uncertainty among all features; S354, select the comprehensive performance improvement index CEI (X' j ,P' j ) is the largest sample point, denoted as (X * ,P * ), take it as the sampling point for the next round of iteration, and update the sample set; S355, the selected optimal sample point (X * ,P * ) is added to the existing sample set. After updating the sample space, the new sample set is refitted through the Gaussian process regression model to correct the prediction function f(X,P); S356, evaluating the new prediction function, if the increase of the comprehensive performance improvement index gradually decreases, judging whether to continue the iteration according to the Bayesian optimization criterion, if continuing, returning to S352 to generate a new set of candidate sample points, until the comprehensive performance improvement index value converges or reaches the preset optimization standard; S357, when the comprehensive efficiency improvement index value converges, determine the heat treatment process parameter combination P * .

4. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 2, characterized in that: The S36 specifically includes: S361. After each round of iteration, according to the current comprehensive performance improvement index CEI (X * ,P * ), dynamically adjust the weight parameters in the Bayesian optimization model, including the adjustment factor γ, the adjustment coefficient λ, and the uncertainty weight factor δ; S362, adding the new sample points generated in each round of iteration to the sample set S to form an updated sample set S new =S∪(X * ,P * ), and recalculate the overall prediction uncertainty of the sample set, denoted as σ total (S new ),in is the jth microstructure feature in the sample set, ω j is an adaptive weight factor based on feature importance; S363, based on the updated overall forecast uncertainty σ total (S new ), modify the Gaussian process regression model and adjust the confidence interval of the prediction function f(X,P); S364, applying the modified prediction function f(X,P) to the next round of Bayesian optimization process, using the improved Bayesian optimization method, based on the new sample set S new Recalculate the expected improvement value and comprehensive performance improvement index, and dynamically generate the next round of candidate sample point sets; S365. In each round of optimization, the corresponding comprehensive performance improvement index is calculated based on different scales of micro-organism characteristics, and a multi-scale interaction strategy is adopted to perform nonlinear weighted calculation on the CEI values ​​of different scales to form the final multi-scale performance interaction index MSI(X,P); S366. According to the results of the multi-scale performance interaction index MSI (X, P), the adjustment parameters in the Bayesian optimization model are dynamically adjusted so that the optimization process gradually converges at the nano, micro and macro scales, and finally generates a global heat treatment process parameter combination P. * .

5. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 1, characterized in that: The S5 specifically includes: S51. During the fatigue test of the gear sample, real-time stress-strain data is collected; S52, using the extended Kalman filter algorithm to filter the real-time collected stress-strain data to predict the dynamic change trend of stress and strain. The filtered data is recorded as in and are the estimated values ​​of stress and strain after filtering at the i-th sampling point; S53. Based on the filtered stress-strain data, a fatigue life prediction model is constructed, denoted as Where L f represents the fatigue life of the gear, T represents the temperature, and the model predicts the critical point of crack initiation by correlating stress, strain and temperature parameters; S54. To optimize the accuracy of fatigue life prediction, the stress-strain data is decomposed according to different time scales to obtain a multi-scale stress signal set, denoted as D”(t)={D”1(t),D”2(t),...,D” m (t)}, where D” k (t is the stress-strain signal on the kth time scale; S55. Input the multi-scale stress signal set D”(t) into the fatigue life prediction model and introduce the adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t); S56, by modulating the fatigue life prediction value L on different time scales f,k (t) Perform nonlinear superposition to generate multi-scale comprehensive fatigue life prediction values S57, based on multi-scale comprehensive fatigue life prediction value Determine the critical points of crack initiation and propagation, generate crack prediction maps, and optimize and adjust heat treatment process parameters based on crack initiation trends.

6. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 5, characterized in that: The S55 specifically includes: S551, input the multi-scale stress signal set D" (t) into the fatigue life prediction model, where D" k (t) represents the stress-strain signal on the kth time scale; S552, introduce adaptive phase shift factor Adjust the stress-strain relationship on each time scale to generate a phase-modulated fatigue life prediction value L f,k (t): in, is the phase shift factor on the kth time scale, which is used to capture the interaction between stress and strain at different time scales and dynamically adjust the shape of the stress-strain curve. and are the estimated values ​​of stress and strain after filtering on the kth time scale, respectively, and f k is the fatigue life prediction sub-model at the kth time scale, T is the temperature parameter in the low temperature environment; S553, dynamically adjust the phase shift factor according to temperature T and time t and the phase correction factor δ k (T,t): Among them, δ k (T, t) is the phase correction factor related to temperature T, which is used to compensate for the nonlinear stress-strain behavior in low temperature environment.

7. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 1, characterized in that: The S6 specifically includes: S61. Based on stress-strain data and fatigue life prediction model, a multi-scale Bayesian optimization model is constructed. The multi-scale Bayesian optimization model combines the microstructure characteristics of nanometer, micrometer and macro scales for optimization, denoted as X = {X n ,X μ ,X M }, where X n , X μ , X M They represent the microstructure characteristic parameter sets at nanometer, micrometer and macroscale, respectively; S62, introduce adaptive multi-scale Bayesian optimization algorithm, through adaptive weight allocation mechanism, the heat treatment process parameter set P = {P1, P2,, P m } Optimize the heat treatment process parameters including heating temperature, heating rate, holding time, cooling medium and cooling speed; S63. In the adaptive multi-scale Bayesian optimization algorithm, the weight allocation mechanism dynamically adjusts the weight of each scale by considering the stress-strain response characteristics at each scale. l (t); S64, through multiple rounds of iterations, in each iteration, according to the feedback of stress-strain data, further optimize the process parameter set P, and dynamically adjust the weight ω of each scale l (t), until the objective function f(X,P) converges, and the final optimized process parameter combination P is obtained * .

8. The method for enhancing the fatigue strength of gears in a low temperature environment based on Bayesian optimization according to claim 7, characterized in that: The S63 specifically includes: S631. At the beginning of each iteration, obtain the current stress-strain data D(t) = {σ(t), ε(t)}, and calculate the stress-strain response characteristics associated with different time scales, denoted as R l (t), where l represents nanometer, micrometer, and macroscale; S632, stress-strain response characteristics R l (t) Perform second-order time derivative calculations to obtain stress-strain acceleration responses at each scale And introduce the nonlinear modulation factor κ l (t) as an exponential function of time: in, is the time-dependent nonlinear modulation factor, λ l is the modulation index used to control the nonlinear change of the second-order derivative; S633, based on the calculation of the second-order derivative in S632 Calculate the weight adjustment factor ω in the current iteration l (t), and combined with the phase modulation mechanism: Among them, ω l0 is the initial weight, α l is the weight adjustment coefficient, is the time-dependent phase shift factor, τ l represents the characteristic time period of the lth scale, represents the phase modulation factor at the lth scale; S634, in each round of iteration, update the weight ω of each scale in real time l (t), and apply it to the objective function of the Bayesian optimization algorithm: Among them, f l (X l ,P) represents the sub-objective function of the lth scale, reflecting the microstructure characteristic parameter X at the lth scale l and the influence of heat treatment process parameters P on fatigue life; S635: During the iterative optimization process, the weight adjustment coefficient α is dynamically adjusted according to the changes in the objective function and the feedback of real-time stress-strain data. l and the characteristic time period τ l .

Images