Fastener heat treatment process parameter optimization method based on big data analysis

Abstract

The invention discloses a fastener heat treatment process parameter optimization method based on big data analysis, and particularly relates to the technical field of metal heat treatment. An optimization model of a fastener hardness uniformity index, a fastener residual stress balance index and a fastener deformation index in the heat treatment process is constructed according to the preprocessing result, and fusion is carried out to obtain a multi-target comprehensive optimization model; adopting an improved multi-target intelligent optimization algorithm to obtain an optimal heat treatment process parameter combination; and finally, evaluating the multiple sub-optimization targets, and feeding back an abnormal evaluation result. According to the method, a multi-target comprehensive optimization model is constructed by fusing sub-optimization models of fastener hardness uniformity, residual stress distribution and deformation, and the problems of algorithm convergence and search capability in multi-target optimization are solved by adopting an improved multi-target intelligent optimization algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal heat treatment, in particular to a fastener heat treatment process parameter optimization method based on big data analysis. BACKGROUND

[0002] Fasteners (such as bolts, screws, nuts, etc.) are core and basic components in many fields such as mechanical manufacturing, construction, and automobiles, and their performance directly affects the reliability and safety of equipment. Heat treatment process is a key link to determine the performance of fasteners. By controlling the process parameters of quenching, tempering and other processes, the hardness, residual stress, deformation and other performance indicators of fasteners can be adjusted.

[0003] Currently, the formulation of fastener heat treatment process parameters mainly relies on: relying on the personal experience of process engineers, using the experience trial-and-error method to determine parameters through repeated tests; using physical simulation methods such as finite element analysis (FEA) software to simulate the heat treatment process; using response surface method (RSM) or Taguchi method, etc., to establish a simple linear or quadratic model based on a small amount of experimental data; traditional statistical model; fastener heat treatment process parameter optimization mostly uses single-objective optimization method, such as optimizing only for hardness or deformation, to achieve control of fastener heat treatment process parameters.

[0004] However, in the actual application of fasteners, it is necessary to meet multiple requirements such as hardness uniformity, reasonable distribution of residual stress, and small deformation. Traditional single-objective optimization methods cannot take into account multiple performance indicators, resulting in that after optimization, the fastener may not meet the requirements in other performances when one performance meets the requirements, for example, when pursuing hardness uniformity, the residual stress may increase, increasing the risk of fastener cracking; or when controlling deformation, the hardness uniformity may be affected. In addition, traditional optimization methods rely on experience or simple experimental design, lack effective use of big data, and are difficult to accurately capture the complex nonlinear relationship between process parameters and multiple performance indicators, so the optimization efficiency and accuracy need to be improved. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a fastener heat treatment process parameter optimization method based on big data analysis to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a fastener heat treatment process parameter optimization method based on big data analysis, comprising: S1: collecting process input parameters, equipment operation parameters, environmental parameters and fastener heat treatment performance detection parameters through a data acquisition device, determining the dimension of the collected data, and obtaining first heat treatment process multidimensional data; S2: Collecting multi-dimensional data of different batches of the first heat treatment process and preprocessing to obtain multi-dimensional data of the second heat treatment process; S3: Based on the first heat treatment process multi-dimensional data and the second heat treatment process multi-dimensional data after data normalization in the preprocessing, an optimization model of the fastener hardness uniformity index, the fastener residual stress balance index and the fastener deformation index in the heat treatment process is constructed; S4: Fusing the three sub-optimization models, a multi-objective comprehensive optimization model is constructed, and an improved multi-objective intelligent optimization algorithm is used to solve the multi-objective comprehensive optimization model to obtain an optimal heat treatment process parameter combination; S5: The optimal heat treatment process parameter combination is applied to the fastener heat treatment production process, the multi-sub-optimization objectives are evaluated, and the abnormal evaluation results are transmitted to the management end for human-computer interaction.

[0007] Technical effects and advantages of the present application: 1. The present application collects four-dimensional data of process, equipment, environment and performance in the heat treatment process through the data acquisition equipment, and constructs a high-quality standardized data set through Min-Max normalization and PCA reduction, solves the problem of single data dimension in the traditional data, and provides scientific data support for subsequent accurate capture of the complex relationship between process parameters and multiple performance indicators; 2. The present application realizes quantitative analysis of parameter influence by establishing the linear / quadratic function relationship between process parameters and performance indicators; at the same time, the multi-objective comprehensive optimization model is constructed by fusing the sub-optimization models of fastener hardness uniformity, residual stress distribution and deformation amount, and the improved multi-objective intelligent optimization algorithm is used to solve the problem of convergence and search ability in multi-objective optimization; 3. The present application uses quantitative improvement rate and reduction rate to evaluate the optimization effect through process parameter optimization evaluation and feedback, so that the evaluation result is objective and accurate, and it is convenient for enterprises to intuitively compare and judge the process optimization effect; a closed-loop optimization process of "optimized parameters-production application-effect evaluation-model updating" is formed, which can continuously improve the process parameter optimization model and continuously improve the performance of the fastener. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 It is the overall flowchart of the present application.

[0009] Figure 2 It is the method flowchart of the present application. DETAILED DESCRIPTION

[0010] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0011] Please refer to Figure 1 As shown in the drawings, the present application provides a fastener heat treatment process parameter optimization system based on big data analysis, comprising a heat treatment multi-dimensional big data acquisition system, a heat treatment process parameter preprocessing module, a multi-sub optimization model construction module, a multi-objective comprehensive optimization model construction module, and a heat treatment process parameter optimization evaluation and feedback module.

[0012] The heat treatment multi-dimensional big data acquisition system is connected with the heat treatment process parameter preprocessing module, the multi-sub optimization model construction module is connected with the heat treatment process parameter preprocessing module and the multi-objective comprehensive optimization model construction module, and the heat treatment process parameter optimization evaluation and feedback module is connected with the multi-objective comprehensive optimization model construction module.

[0013] The heat treatment multi-dimensional big data acquisition system: through a data acquisition device, process input parameters, equipment operation parameters, environmental parameters and fastener heat treatment performance detection parameters are collected, the data dimensions of the collected data are determined, first heat treatment process multi-dimensional data are obtained, and are transmitted to the heat treatment process parameter preprocessing module and the multi-sub optimization model construction module; The heat treatment process parameter preprocessing module: based on the big data acquisition system, multi-batch first heat treatment process multi-dimensional data are collected and preprocessed to obtain second heat treatment process multi-dimensional data; The multi-sub optimization model construction module: based on the first heat treatment process multi-dimensional data and the second heat treatment process multi-dimensional data after data normalization in preprocessing, optimization models of fastener hardness uniformity index, fastener residual stress balance index and fastener deformation index in the heat treatment process are constructed, and are transmitted to the multi-objective comprehensive optimization model construction module; The multi-objective comprehensive optimization model construction module: three sub-optimization models are fused to construct a multi-objective comprehensive optimization model, an improved multi-objective intelligent optimization algorithm is used to solve the multi-objective comprehensive optimization model, an optimal heat treatment process parameter combination is obtained, and is transmitted to the heat treatment process parameter optimization evaluation and feedback module; The heat treatment process parameter optimization evaluation and feedback module: the optimal heat treatment process parameter combination is applied to the fastener heat treatment production process, multi-sub optimization objectives are evaluated, and abnormal evaluation results are transmitted to the management end for human-computer interaction; Please refer to Figure 2As shown, the fastener heat treatment process parameter optimization method based on big data analysis comprises: S1: collecting process input parameters, equipment operation parameters, environmental parameters and fastener heat treatment performance detection parameters through a data acquisition device, determining the data dimension of the collected data, and obtaining first heat treatment process multidimensional data; S2: collecting different batches of first heat treatment process multidimensional data and preprocessing to obtain second heat treatment process multidimensional data; S3: based on the first heat treatment process multidimensional data and the second heat treatment process multidimensional data after data normalization in preprocessing, constructing an optimization model of fastener hardness uniformity index, fastener residual stress balance index and fastener deformation index in the heat treatment process; S4: fusing three sub-optimization models to construct a multi-objective comprehensive optimization model, using an improved multi-objective intelligent optimization algorithm to solve the multi-objective comprehensive optimization model, and obtaining an optimal heat treatment process parameter combination; S5: applying the optimal heat treatment process parameter combination to the fastener heat treatment production process, evaluating multiple sub-optimization objectives, and transmitting abnormal evaluation results to the management end for human-computer interaction.

[0014] S1: collecting process input parameters, equipment operation parameters and environmental parameters and fastener heat treatment performance detection parameters through a data acquisition device, determining the data dimension of the collected data, and obtaining first heat treatment process multidimensional data, comprising the following steps: S1.1: data acquisition: acquiring process input parameters through a data acquisition device, including quenching temperature (range usually 800-950℃), tempering time (range usually 1-3 hours), cooling rate (range usually 5-20℃ / s), heating rate (range usually 10-50℃ / min), quenching medium temperature (range usually 20-80℃); acquiring equipment operation parameters including heating power (range usually 50-100kW), conveyor belt speed (range usually 0.5-2m / min), cooling system pressure (range usually 0.3-0.8MPa); acquiring environmental parameters including workshop temperature (range usually 15-30℃), humidity (range usually 30%-70%); acquiring fastener heat treatment performance detection parameters including surface hardness H1 (unit HRC), core hardness H2 (unit HV), hardness standard deviation σh (unit HV), surface residual tensile stress σ1 (unit MPa), core residual compressive stress σ2 (unit MPa), stress gradient G (unit MPa), diameter shrinkage rate ΔD, length elongation rate ΔL, ovality ΔE; It is particularly pointed out in this embodiment that the data acquisition device includes temperature sensors, pressure sensors, speed sensors, etc.; the performance detection device includes Rockwell hardness testers, microhardness testers, X-ray diffractometers, drilling devices and high-precision measuring instruments, etc.; the X-ray diffractometer is used to measure the surface residual stress, and the drilling device is used to measure the core residual stress.

[0015] S1.2: Determine the data dimensions collected: 5 data dimensions of process input parameters, 3 data dimensions of equipment operation parameters, 2 data dimensions of environmental parameters, and 9 dimensions of fastener heat treatment performance detection parameters; finally, obtain the first heat treatment process multi-dimensional data, including various parameters and data dimensions; S2: Collect different batches of first heat treatment process multi-dimensional data and perform preprocessing to obtain second heat treatment process multi-dimensional data, the preprocessing including data cleaning, data normalization, and data reduction; data cleaning is to handle missing values by mean filling method, identify and remove outliers by 3σ principle, and delete duplicate data; use Min-Max normalization function to convert data to [0, 1] interval; data reduction uses principal component analysis algorithm, retains principal components with cumulative contribution rate greater than or equal to the corresponding threshold (for example, threshold is 0.92), and obtains the jth batch of heat treatment process principal component dataset PC, , pc i is the ith principal component, n represents the number of principal components, j = 1, 2,..., m, m is the number of different batches, and finally traverse all batches to obtain the second heat treatment process multi-dimensional data, including the principal component dataset and the corresponding principal component value dataset of all batches; It needs to be specifically explained in this embodiment that the principal component is a linear combination of the normalized parameters, for example, pc1=0.4×heating temperature+0.3×cooling speed+0.2×hardness+0.1×tensile strength; m batches represent that the heat treatment process parameters in each batch are not exactly the same.

[0016] S3: Based on the first heat treatment process multi-dimensional data and the second heat treatment process multi-dimensional data after data normalization in the preprocessing, construct the optimization model of the fastener hardness uniformity index, the fastener residual stress balance index, and the fastener deformation index in the heat treatment process, including the following steps: S3.1: Based on the first heat treatment process multi-dimensional data after data normalization, construct the fastener hardness uniformity index optimization target U, the fastener residual stress balance index optimization target B, and the fastener deformation index optimization target S, respectively, the hardness standard deviation σh is obtained by calculating the standard deviation of the core hardness of m1 fasteners in the same batch, and the smaller U is, the better the uniformity is; G is the stress gradient, G=|σ1-|σ2||, ΔD is obtained by the diameter change rate of the fastener before and after heat treatment (representing the transverse dimension of the fastener at the measurement position, and the diameter is the dimension perpendicular to the axis direction), and ΔL is obtained by the length change rate of the fastener before and after heat treatment (representing the dimension of the fastener along its axis direction), , D max and D min, respectively, the diameter of the long axis direction and the diameter of the short axis direction of the cross section of the fastener after heat treatment, D no is the design diameter; It needs to be specifically pointed out in this embodiment that in mechanics, it is known that tensile stress is positive and compressive stress is negative. In "stress balance", attention is paid to the "size" (not the direction) of the compressive stress - the greater the absolute value of the compressive stress, the stronger the ability to "balance the surface tensile stress", which meets the physical mechanism of "cooling rate difference → plastic deformation constraint", and this state has engineering necessity for preventing cracking and fatigue failure of the fastener.

[0017] S3.2: Build a fastener hardness uniformity index optimization model: S3.2.1: Correlation calculation: First, since the hardness standard deviation σh is a key indicator of hardness uniformity, the smaller σh is, the more uniform the hardness is, therefore based on the multi-dimensional data of the second heat treatment process, the Pearson correlation coefficient is used to calculate the correlation coefficient Y i (pc,σh) between the m batches of principal component values X j and the hardness standard deviation σh i of the m batches, each batch corresponds to a group of sample pairs containing principal component values-σh values, a total of m groups; Then traverse the n principal components, and filter out the candidate principal components pc i whose absolute value of the correlation coefficient Y i (pc,σh) is greater than or equal to the corresponding threshold value (for example, the threshold value is 0.8), to obtain the candidate principal component dataset pc , pc s,i is the candidate i-th principal component, and n1 represents the number of candidate principal components, n1 It needs to be specifically pointed out in this embodiment that the Pearson correlation coefficient is a prior art, which is obtained by calculating the ratio of the covariance and the standard deviation product of the sample pairs of the principal component values and σh values of the m batches.

[0018] S3.2.2: Correlation deduplication: First, calculate the correlation coefficient between each pair of candidate principal components. If the absolute value of the correlation coefficient between each pair is greater than or equal to the corresponding threshold value (for example, 0.7), it is considered that the two are highly redundant, and the principal component with a larger correlation coefficient value is retained to obtain the target principal component dataset pc m , , pc m,i is the target i-th principal component, and n2 represents the number of target principal components, n2 S3.2.3: Importance sorting: sort the target principal component dataset pc mAs an input feature, the hardness standard deviation σh is taken as the target variable, and a random forest regression model is constructed by the sample pairs of the principal component values and the σh values of the m batches, the number of decision trees in the random forest regression model = 100, the maximum depth = 5, the number of split features is set to sqrt(n2), if n2≤5, all principal component features are selected, and the feature importance score dataset w of each principal component to the hardness standard deviation σh is output m , , w i is the importance score of the ith target principal component; It needs to be specifically pointed out in this embodiment that the calculation of feature importance by random forest belongs to the prior art, and the principle of calculating feature importance by random forest is based on the influence degree of features on sample classification or regression results in the decision tree construction process. By averaging the feature importance of all decision trees in the forest, the importance score of each feature can be obtained.

[0019] S3.2.4: Constructing an optimization model: First, based on the fastener hardness uniformity index optimization target U, the target principal component dataset pc m of the m batches is fitted with the corresponding mapping relationship model H1(pc m ), H2(pc m ) and σh(pc m ), H1(pc m )=(a)×(w×pc m ) T +b1, b1 is a constant term, (a) is the coefficient matrix of each target principal component in the mapping relationship model H1(pc m ), (a)=(a1,a2,...,a n2 ), a n2 is the coefficient of the nth2 target principal component in this mapping relationship model, [w×pc m ] T is the matrix transpose of the product of the target principal component dataset pc m and the corresponding feature importance score dataset, H2(pc m )=(c)×(w×pc m ) T +b2, b2 is a constant term, (c) is the set of coefficients of each target principal component in the mapping relationship model H2(pc m ), (c)=(c1,c2,...,c n2 ), c n2 is the coefficient of the nth2 target principal component in this mapping relationship model, σh(pc m )=[(e),(f)]×[((w×pc m,i ) 2 )m , (w x pc m ) T + b3, b3 is a constant term, [(e), (f)] is a mapping relationship model σh(pc m ) each target principal component quadratic term coefficient matrix and the splicing of the first term coefficient matrix, (e) = (e1, e2,..., e n2 ), e n2 is the quadratic term coefficient of the n2th target principal component in this mapping relationship model, (f) = (f1, f2,..., f n2 ), f n2 is the first term coefficient of the n2th target principal component in this mapping relationship model c n2 is the coefficient of the n2th target principal component in this mapping relationship model; [((w x pc m,i ) 2 ) m , (w x pc m ) T is the matrix splicing of the quadratic term and the first term target principal component data set pc m and the corresponding characteristic importance score data set product; The embodiment needs to be specifically explained that H1(pc m ) = a1 x (w1 x pc m,1 ) +... + a n2 x (w n2 x pc m,n2 ) + b1, H2(pc m ) = c1 x (w1 x pc m,1 ) +... + c n2 x (w n2 x pc m,n2 ) + b2, σh(pc m ) = e1 x (w1 x pc m,1 ) 2 +... + e n2 x (w n2 x pc m,n2 ) 2 + f1 x (w1 x pc m,1 ) +... + f n2 x (w n2 x pc m,n2 ) + b3, for example, the principal component data set includes pc m,1 and pc m,6 , H1(pc m ) = 5.2 x (0.68 x pc m,1 ) + 4.8 x (0.32 x pc m,6 ) + 8.3, H2(pc m) = 4.8 × (0.68 × pc m,1 )+4.5×(0.32×pc m,6 )+7.9,σh(pc m ) = 0.002 × (0.68 × pc m,1 ) 2 +0.0018×(0.32×pc m,6 ) 2 +0.03×(0.68×pc m,1 )+0.028×(0.032×pc m,6 +1.2; S3.2.5: Constructing the objective function: First, combine the target principal component dataset pc m Construct the objective function U(pc) with the optimization objective U. m ), Then, statistically analyze the range of values ​​(pc) of the target principal components whose hardness standard deviation σh is less than or equal to the corresponding threshold among m batches. min ,pc max ), pc min and PC max These represent the minimum and maximum values ​​of each principal component; finally, set the collaborative constraint condition st, st=[(pc min ,pc max )1,(pc min ,pc max )2,...,(pc min ,pc max ) n2 ], (pc min ,pc max ) n2 The range of values ​​for the n2th objective principal component; for example, σh≤5HV, pc m,1 The effective interval is (0.35, 0.8), pc m,6 The effective interval is (0.25, 0.75). Set the collaborative constraint st, st = [(0.35, 0.8), (0.25, 0.75)]. S3.2.6: Model Validation: Collect multi-dimensional data of m1 batches of the new second heat treatment process, and extract the corresponding m1 actual hardness standard deviations σh. j At the same time, through the mapping relationship model σh(pc) m The standard deviation of hardness σh from the outputs of m1 models is obtained. j m Calculate the coefficient of determination R of the model 2 Then calculate the error rate Δ(σh) for a single batch. j ), through the j-th actual hardness standard deviation σh jthe hardness standard deviation σh of the jth model output j m The absolute value of the deviation rate of R 2 ≥ the corresponding threshold value (for example, 0.94) and the error rate Δ(σh j ) of all batches ≤ the corresponding threshold value (for example, 0.05), indicating that the target function U(pc m ) is qualified in terms of precision, and the optimized model U(pc m ) of the fastener hardness uniformity index is obtained; if R 2 < the corresponding threshold value, the principal component weight is adjusted again, and if there is an error rate Δ(σh j ) > the corresponding threshold value, the mapping relationship model fitting process is performed again. It is necessary to specifically explain that the coefficient of determination R 2 is an existing formula, and the value range is [0, 1]. The closer to 1, the higher the degree of explanation of the model to the hardness standard deviation σh.

[0020] S3.3: The principle of constructing the optimized model of the fastener hardness uniformity index is the same. Since the stress gradient G is an important indicator for evaluating the uniformity of residual stress distribution, the smaller the G, the more uniform the residual stress distribution, therefore, based on the principal components and the corresponding principal component values in the multi-dimensional data of the second heat treatment process and the stress gradient G, the optimized model B(pc1 m ) of the fastener residual stress balance index is obtained, B(pc1 m ), σ1(pc1 m ), |σ2|(pc1 m ) and G(pc1 m ) are respectively the fastener residual stress balance index, the surface residual tensile stress, the core residual compressive stress and the compressive stress gradient under the target principal component data set pc1 m . S3.4: The principle of constructing the optimized model of the fastener hardness uniformity index is the same. Since the diameter shrinkage rate ΔD is an indicator for comprehensively measuring the deformation degree, the smaller the ΔD, the smaller the deformation, therefore, based on the principal components and the corresponding principal component values in the multi-dimensional data of the second heat treatment process and the diameter shrinkage rate ΔD, the optimized model S(pc2 m ) of the fastener deformation index is obtained, S(pc2 m ), ΔD(pc2 m ), ΔL(pc2 m ) and ΔE(pc2 m ) are respectively the fastener deformation index, the diameter shrinkage rate, the length elongation rate and the ovality under the target principal component data set pc2 m . S4: fuse the three sub-optimization models, construct a multi-objective comprehensive optimization model, solve the multi-objective comprehensive optimization model by using an improved multi-objective intelligent optimization algorithm, and obtain an optimal heat treatment process parameter combination, including the following steps: S4.1: fuse the three sub-optimization models, construct a multi-objective comprehensive optimization model maxF, maxF = ω1 × [1 / U(pc m )] + ω2 × [B(pc1 m )] + ω3 × [1 / S(pc2 m )], ω1, ω2 and ω3 are respective weights, which are set according to the priority of each target in actual production, for example, ω1 = 0.4 for hardness uniformity priority, ω2 = 0.3 for residual stress, and ω3 = 0.3 for deformation; S4.2: improved multi-objective intelligent optimization algorithm: introduce the elite preservation strategy into the NSGA-II and PSO hybrid algorithm to avoid local optimum, and the algorithm settings include: S4.2.1: population size: the population size is set to 100 to ensure sufficient individual diversity in the search process and cover a wide solution space; S4.2.2: iteration number: the iteration number is set to 200, and in each iteration, the particle velocity update function is used to calculate the new velocity, and then the particle position update function is used to move the particle to the new position, the particle velocity update function is used to calculate the velocity of the next iteration, and the particle position update function is used to calculate the position of the next iteration according to the updated velocity, so as to realize the optimization of the fastener heat treatment process parameters; for example, Y = (quenching temperature T q , tempering time t t , cooling rate v c , heating rate v h , quenching medium temperature T m ); S4.2.3: elite preservation ratio: the elite preservation ratio is 20%, and the individual selection function S el is used to retain excellent individuals in each iteration, so as to accelerate the convergence speed of the algorithm and improve the quality of the solution, S el = {Y | Y ∈ Pareto optimal solution set, F(Y) ≥ F avg}, F avg is the average fitness of the population, and Y is the process parameter combination; S4.2.4: adaptive inertia weight: the adaptive inertia weight ranges from 0.4 to 0.9, and the global search ability and local development ability of the algorithm are dynamically balanced according to the iteration process and search state of the population; S4.2.5: obtain the optimal process parameter combination: when the change of the comprehensive optimization model maxF in continuous K iterations is less than or equal to a preset threshold ε, it is preliminarily determined that the algorithm converges, maxk=1,2,..,K |F (t) -F (t-k) |≤ε,F (t) is the target comprehensive optimization function value of the tth iteration; when the Lth process parameter x L (e.g. quenching temperature T q , tempering time t t , etc.), x L ∈Y, all process parameters in K consecutive iterations satisfy the variation Δ(x L )≤ the corresponding preset threshold δ L , the search is determined to be approaching stable, Δ(x L ) k=1,2,..,K |x L (t) -x L (t-k) |≤δ L , x L (t) is the value of the Lth process parameter in the tth iteration; finally, when the preliminary determination algorithm converges and the search is approaching stable, or the maximum number of iterations is reached, the final determination algorithm converges, and the optimal heat treatment process parameter combination Y is output, for example Y=(quenching temperature T q y , tempering time t t y , cooling rate v c y , heating rate v h y , quenching medium temperature T m y ); It needs to be specifically pointed out in this embodiment that the hybrid algorithm based on NSGA-II and PSO (particle swarm optimization) is prior art, which is a collaborative method combining global search and multi-objective optimization, and is widely used in process parameter optimization, engineering design and other fields; PSO optimization proxy model: adjust the weights and biases of BP neural network through particle swarm algorithm, and construct a high-precision nonlinear mapping relationship model between process parameters and target variables; NSGA-II automatically balances conflicting goals (such as hardness, residual stress, deformation index), and generates a Pareto solution set covering a wider production scenario.

[0021] S5: apply the optimal heat treatment process parameter combination to the fastener heat treatment production process, evaluate the multi-objective optimization target, and transmit the abnormal evaluation result to the management end for human-computer interaction, including the following steps: S5.1: apply the optimal heat treatment process parameter combination to the fastener heat treatment production process, respectively calculate the fastener hardness uniformity index improvement rate Delta U, the fastener residual stress balance index improvement rate Delta B and the fastener deformation index reduction rate Delta S; the improvement rate is the difference value after optimization and before optimization, and then the ratio of the difference value after optimization and before optimization is obtained, and the reduction rate is the difference value before optimization and after optimization, and then the ratio of the difference value before optimization and after optimization is obtained; S5.2: if Delta U and Delta B meet the corresponding expectations and Delta S also meets the corresponding expectations, it indicates that the heat treatment process parameter optimization is good, otherwise there is at least one that does not meet the expectation, which indicates that the evaluation result is abnormal, and feedback is fed back to the management end for human-computer interaction, such as periodically updating model parameters, optimizing model construction process, etc.

[0022] Secondly: the drawings in the disclosed embodiments of the application only involve the structures involved in the disclosed embodiments of the application, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the application can be combined with each other; Finally: the above only describes the preferred embodiments of the application and is not used to limit the application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. A method for optimizing fastener heat treatment process parameters based on big data analysis, characterized in that: include: S1: Through data acquisition equipment, process input parameters, equipment operating parameters, environmental parameters, and fastener heat treatment performance testing parameters are collected to determine the dimensions of the collected data and obtain multi-dimensional data of the first heat treatment process; S2: Collect multi-dimensional data of the first heat treatment process from different batches and preprocess them to obtain multi-dimensional data of the second heat treatment process; S3: Based on the multi-dimensional data of the first heat treatment process and the multi-dimensional data of the second heat treatment process after data normalization in the preprocessing, an optimization model is constructed for the fastener hardness uniformity index, fastener residual stress balance index and fastener deformation index during the heat treatment process. S4: Integrate the three sub-optimization models to construct a multi-objective comprehensive optimization model, and use an improved multi-objective intelligent optimization algorithm to solve the multi-objective comprehensive optimization model to obtain the optimal combination of heat treatment process parameters; S5: Apply the optimal heat treatment process parameter combination to the fastener heat treatment production process, evaluate the multi-sub-optimization objectives, and transmit the abnormal evaluation results to the management terminal for human-machine interaction.

2. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The S3 implementation includes: S3.1: Based on the multi-dimensional data of the first heat treatment process after data normalization, optimize the fastener hardness uniformity index U, the fastener residual stress balance index B, and the fastener deformation index S respectively.

3. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The implementation of S3 also includes: constructing an optimization model for the uniformity of fastener hardness index: S3.2.1: Correlation calculation: based on multi-dimensional data of the second heat treatment process, using the Pearson correlation coefficient, calculate the i-th principal component value X of m batches. i The standard deviation of hardness σh of m batches j Correlation coefficient Y i (pc, σh), each batch corresponds to one sample pair containing the principal component value - σh value, for a total of m groups; then, iterate through the n principal components and select the correlation coefficient Y. i Candidate principal components whose absolute value of (pc,σh) is greater than or equal to the corresponding threshold are pc. i The candidate principal component dataset pc is obtained. pc s,i Let n be the candidate i-th principal component, and n1 represent the number of candidate principal components, where n1 < n; S3.2.2: Correlation Deduplication: First, calculate the correlation coefficient between each pair of candidate principal components. If the absolute value of the correlation coefficient between any two pairs is greater than or equal to the corresponding threshold, then both are considered highly redundant. The principal component with the larger correlation coefficient is retained, resulting in a target principal component dataset pc without redundancy. m , pc m,i Let n be the target i-th principal component, and n2 represent the number of target principal components, where n2 ≤ n1; S3.2.3: Importance Ranking: Ranking the target principal component dataset pc m Using the hardness standard deviation σh as the input feature and the hardness standard deviation σh as the target variable, a random forest regression model is constructed using m batches of principal component values ​​minus σh values. In the random forest regression model, the number of decision trees is 100, the maximum depth is 5, and the number of split features is set to sqrt(n²). If n² ≤ 5, all principal component features are selected. The output dataset w shows the feature importance score of each principal component to the hardness standard deviation σh. m , w i The importance score of the i-th target principal component.

4. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The implementation of S3 also includes: constructing an optimization model: S3.2.4: constructing an optimization model for the fastener hardness uniformity index: firstly, optimize the target U based on the fastener hardness uniformity index, using m batches of target principal component datasets pc m The corresponding mapping relationship model H1(pc) is fitted to H1, H2 and σh. m H2(pc) m ) and σh(pc m ); S3.2.5: Constructing the objective function: First, combine the target principal component dataset pc m Construct the objective function U(pc) with the optimization objective U. m Then, statistically analyze the range of values ​​(pc) of the target principal components whose hardness standard deviation σh is less than or equal to the corresponding threshold among m batches. min ,pc max ), pc min and PC max These represent the minimum and maximum values ​​of each principal component; finally, set the collaborative constraint condition st, st=[(pc min ,pc max )1,(pc min ,pc max )2,...,(pc min ,pc max ) n2 ], (pc min ,pc max ) n2 The range of values ​​for the n2th target principal component.

5. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The implementation of S3 also includes: constructing an optimization model: S3.2.6: Model verification: collecting multi-dimensional data of m1 batches of new second heat treatment processes, and extracting the corresponding m1 actual hardness standard deviations σh. j At the same time, through the mapping relationship model σh(pc) m The standard deviation of hardness σh from the outputs of m1 models is obtained. j m Calculate the coefficient of determination R of the model 2 Then calculate the error rate Δ(σh) for a single batch. j ), through the j-th actual hardness standard deviation σh j The standard deviation of hardness σh output by the j-th model j m The absolute value of the deviation rate is obtained; if R 2 ≥ the corresponding threshold and the error rate Δ(σh) of all batches j If ) ≤ the corresponding threshold, it means that the objective function U(pc) m The construction accuracy is qualified, and the optimized model U(pc) of the fastener hardness uniformity index is obtained. m If R 2 If the corresponding threshold is exceeded, the principal component weights are readjusted. If an error rate Δ(σh) exists... j If the value exceeds the corresponding threshold, the mapping relationship model fitting process is repeated.

6. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The implementation of S3 further includes: S3.3: Similar to the principle of constructing the fastener hardness uniformity index optimization model, based on the principal components and corresponding principal component values ​​in the multi-dimensional data of the second heat treatment process and the stress gradient G, an optimization model B(pc1) of the fastener residual stress balance index is obtained. m ); S3.4: Similar to the principle of constructing the fastener hardness uniformity index optimization model, based on the principal components and corresponding principal component values ​​in the multi-dimensional data of the second heat treatment process, as well as the diameter shrinkage rate ΔD, the optimization model S(pc2) of the fastener deformation index is obtained. m ).

7. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The multi-objective integrated optimization model constructed in S4 includes: S4.1: Integrate the three sub-optimization models to construct a multi-objective comprehensive optimization model maxF, maxF=ω1×[1 / U(pc m )]+ω2×[B(pc1 m )]+ω3×[1 / S(pc2 m )], where ω1, ω2 and ω3 are the corresponding weights.

8. The method for optimizing fastener heat treatment process parameters based on big data analysis according to claim 1, characterized in that: The improved multi-objective intelligent optimization algorithm in S4 includes: S4.2.1: Population size: Set the population size to 100; S4.2.2: Number of iterations: The number of iterations is set to 200. In each iteration, the new velocity is first calculated by the particle velocity update function, and then the particle is moved to the new position by the position update function. The particle velocity update function is used to calculate the velocity of the particle in the next iteration, and the particle position update function is used to calculate the position of the particle in the next iteration based on the updated velocity. S4.2.3: Elite Retention Ratio: The elite retention ratio is 20%, determined by the individual selection function S. el Retaining the best individuals from each iteration accelerates the algorithm's convergence speed and improves the quality of the solution. el ={Y|Y∈Pareto optimal solution set, F(Y)≥F avg }, F avg Y represents the average fitness of the population, and Y represents the combination of process parameters. S4.2.4: Adaptive inertia weight: The adaptive inertia weight ranges from 0.4 to 0.9; S4.2.5: Obtaining the optimal combination of process parameters: When the change in the comprehensive optimization model maxF in K consecutive iterations is ≤ the preset threshold ε, the algorithm is initially determined to be converged; when the Lth process parameter x L x L ∈Y, in K consecutive iterations all process parameters satisfy the change Δ(x) L )≤ the corresponding preset threshold δ L The algorithm is determined to be close to stabilizing when the initial determination is that the algorithm has converged and the search has approached stabilization, or when the maximum number of iterations is reached, the algorithm is finally determined to be converged and the optimal combination of heat treatment process parameters Y is output.

Images