Welding process reliability optimization acceleration method and system based on machine learning

Abstract

The invention discloses a welding process reliability optimization acceleration method and system based on machine learning, and the method comprises the steps: obtaining a process parameter list, carrying out a welding process test of a test piece according to the process parameter list, and obtaining historical test data; constructing a welding seam multi-scale numerical simulation model, and performing numerical simulation analysis by adopting the welding seam multi-scale numerical simulation model to obtain a simulation calculation result; carrying out dimension reduction processing on the simulation calculation result, inputting the data after dimension reduction processing into a plurality of machine learning algorithms for training and verification, and preferentially obtaining an optimal machine learning model; performing welding process reliability prediction based on the optimal machine learning model to obtain optimal process parameters; welding line multi-scale numerical simulation verification is conducted on the optimal process parameters, and welding process verification is conducted on the optimal process parameters meeting verification requirements till the welding process is optimized. According to the method, the reliability of the formed welding seams with different welding process parameters is quickly predicted and optimized, and the process optimization period is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of welding process optimization and acceleration, and in particular to a welding process reliability optimization and acceleration method and system based on machine learning. Background Art

[0002] Industrial technology is booming in the new era. A large number of new structures, new materials and new processes need to be converted into products that can be used reliably for a long time. Welding, as a widely applicable, stable and reliable connection process, is widely used in component connection in various industrial fields. Welding process optimization is one of the most important links in the development of welding processes. The purpose of welding process parameter optimization is to obtain a process with the most optimized reliability. Traditional optimization methods rely on welding process tests, which are mainly aimed at the dimensions and mechanical properties of the structure formed by the welding process. They are costly and have a long cycle, making it difficult to evaluate their reliability. Multi-scale numerical simulation of welds, as an important auxiliary means of process optimization, can greatly reduce the number of process tests and realize reliability evaluation. It consists of four parts: phase field simulation, crystal plasticity finite element analysis, macroscopic finite element analysis and reliability calculation. However, due to the large amount of calculation and long cycle of multi-scale numerical simulation of welds, the number of combinations of process parameters and service conditions is large, and the computing resources and cycle are limited, which cannot meet the needs of process optimization. Therefore, in order to solve the problem of low computational efficiency of multi-scale numerical simulation of welds, machine learning is used to train the model using a limited number of multi-scale numerical simulation calculation results of welds, so as to realize the rapid prediction of the reliability of welds formed by different welding process parameters and accelerate the process optimization cycle.

[0003] The domestic patent titled "A method for determining the connection configuration and process parameters of a component package" discloses a method for obtaining solder joint reliability and inversely determining process parameters using a neural network model and simulation analysis. The method uses process parameters as input and connection configuration as output, trains a neural network model, obtains the relationship between connection configuration and solder joint reliability through simulation analysis, establishes a process parameter-connection configuration-reliability inversion equation, and adjusts the process parameters based on a particle swarm algorithm until the calculated results meet the reliability requirements. However, the method belongs to the field of electronic component packaging, which is completely different from the field of fusion welding. In addition, the role of machine learning in the method is to predict the connection configuration corresponding to the process parameters, and the connection configuration is not obtained by simulating the process parameters. The domestic patent title is "Weld Defect Identification Method, Device, Electronic Equipment and Storage Medium", which discloses a weld defect identification method using machine learning. The model is trained based on weld image samples and corresponding weld defect labels. The trained weld defect identification model is used to extract and fuse multi-scale features of the weld image to be identified. The weld defects are identified based on the fused features. The weld defect segmentation results in the weld image to be identified can be accurately identified, but it does not involve welding process optimization.

[0004] In view of this, this application is hereby filed. Summary of the invention

[0005] The technical problem to be solved by the present invention is the problem of low efficiency in multi-scale numerical simulation calculation of welds. The purpose of the present invention is to provide a welding process reliability optimization acceleration method and system based on machine learning. By means of machine learning, a limited number of multi-scale numerical simulation calculation results of welds are used to train the model, so as to realize rapid prediction and optimization of the reliability of welds formed by different welding process parameters, accelerate the process optimization cycle, and provide a fast and effective means for welding process optimization.

[0006] The present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a welding process reliability optimization acceleration method based on machine learning, the method comprising:

[0008] Obtain a list of process parameters, carry out welding process tests on test pieces according to the list of process parameters, and obtain historical test data;

[0009] Construct a multi-scale numerical simulation model of the weld, and use it to perform numerical simulation analysis based on historical test data to obtain simulation calculation results;

[0010] Perform dimensionality reduction on the simulation calculation results, and input the reduced-dimensionality data into a variety of machine learning algorithms for training and verification, and select the best machine learning model;

[0011] Welding process reliability prediction is performed based on the optimal machine learning model to obtain the optimal process parameters;

[0012] The optimal process parameters are verified by multi-scale numerical simulation of the weld, and the welding process is verified for the optimal process parameters that meet the verification requirements until the welding process is optimized.

[0013] Furthermore, a process parameter list is obtained, and a welding process test of the test piece is carried out according to the process parameter list to obtain historical test data, including:

[0014] Determine and formulate the peak current of initial parameters based on actual project conditions;

[0015] Based on the initial parameters, the peak current is increased or decreased to form a process parameter list;

[0016] Carry out welding process test of test pieces according to the process parameter list to obtain process test pieces after welding;

[0017] The welded process test pieces are subjected to EBSD characterization, base material tensile test and fatigue test to obtain test results, which are used as historical test data.

[0018] Furthermore, the upper and lower limits of the process parameter list envelop possible optimal parameters.

[0019] Furthermore, the weld multi-scale numerical simulation model includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model;

[0020] Among them, the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

[0021] Furthermore, a multi-scale numerical simulation model of the weld is constructed, and according to historical test data, the multi-scale numerical simulation model of the weld is used to perform numerical simulation analysis to obtain simulation calculation results, including:

[0022] Construct a finite element analysis sub-model of the welding process in the finite element analysis software; calculate the residual stress field based on the finite element analysis sub-model of the welding process according to the historical test data and the first data, and extract the thermal cycle curves of the weld, heat affected zone and base material;

[0023] Constructing a phase field simulation sub-model; performing phase field simulation based on the phase field simulation sub-model according to the residual stress field, the thermal cycle curve and the second data, and obtaining the microstructure evolution results of the weld, the heat affected zone and the base material;

[0024] Construct a crystal plasticity analysis sub-model; according to the organizational evolution results and the third data, simulate based on the crystal plasticity analysis sub-model to obtain the SN curves of the weld, heat-affected zone and base material, as well as the stress-strain curves of the weld, heat-affected zone and base material;

[0025] Construct a macroscopic finite element analysis submodel of the working condition; perform calculations based on the macroscopic finite element analysis submodel of the working condition according to the SN curve, the stress-strain curve, the residual stress field and the fourth data, and obtain the stress distribution and load spectrum under the working condition;

[0026] A fatigue life macro-finite element analysis sub-model is constructed; according to the load spectrum and the SN curve, calculations are performed based on the fatigue life macro-finite element analysis sub-model to obtain the fatigue life cloud diagram of the process test piece.

[0027] Furthermore, the simulation calculation results are processed by dimensionality reduction, and the data after dimensionality reduction are input into a variety of machine learning algorithms for training and verification, and the optimal machine learning model is obtained by selection, including:

[0028] Perform dimension reduction processing on the simulation calculation results to obtain the data after dimension reduction processing;

[0029] The data after dimensionality reduction is divided into a training set and a validation set, and the training set is input into a variety of machine learning algorithms for training to obtain fatigue training results;

[0030] The mean square error of the fatigue training results is verified based on the validation set, and the machine learning algorithm corresponding to the fatigue training result with the smallest mean square error is obtained as the optimal machine learning model.

[0031] Furthermore, the optimal process parameters are verified by multi-scale numerical simulation of the weld, and the welding process is verified for the optimal process parameters that meet the verification requirements until the welding process is optimized, including:

[0032] Perform multi-scale numerical simulation of the weld for the optimal process parameters to obtain the simulated value; and verify whether the deviation between the predicted value and the simulated value of the optimal machine learning model meets the requirements;

[0033] If the deviation meets the requirements, the welding process verification is continued for the optimal process parameters that meet the requirements to verify whether each parameter meets the requirements. If each parameter meets the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iteration until the welding process is optimized;

[0034] If the deviation does not meet the requirements, multiple machine learning algorithms are retrained and verified to select the best one.

[0035] In a second aspect, the present invention further provides a welding process reliability optimization acceleration system based on machine learning, the system comprising:

[0036] The welding process test unit is used to obtain a list of process parameters, carry out welding process tests on test pieces according to the list of process parameters, and obtain historical test data;

[0037] The multi-scale numerical simulation model unit is used to construct a multi-scale numerical simulation model of the weld, and to perform numerical simulation analysis using the multi-scale numerical simulation model of the weld based on historical test data to obtain simulation calculation results;

[0038] The model optimization unit is used to perform dimensionality reduction processing on the simulation calculation results, and input the data after dimensionality reduction processing into a variety of machine learning algorithms for training and verification, and select the best machine learning model;

[0039] Welding process prediction unit, used to predict welding process reliability based on the optimal machine learning model and obtain the optimal process parameters;

[0040] The simulation and welding process verification unit conducts multi-scale numerical simulation verification of the weld for the optimal process parameters, and conducts welding process verification for the optimal process parameters that meet the verification requirements until the welding process is optimized.

[0041] Furthermore, the weld multi-scale numerical simulation model includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model;

[0042] Among them, the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

[0043] Furthermore, the simulation and welding process verification unit includes:

[0044] The model simulation verification subunit is used to perform multi-scale numerical simulation of the weld for the optimal process parameters to obtain the simulation value; and verify whether the deviation between the predicted value and the simulation value of the optimal machine learning model meets the requirements; if the deviation meets the requirements, the welding process verification is continued for the optimal process parameters that meet the requirements; if the deviation does not meet the requirements, multiple machine learning algorithms are retrained and verified to select the best one;

[0045] The welding process verification subunit is used to verify the welding process and verify whether each parameter meets the requirements. If each parameter meets the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iterations until the welding process is optimized.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] 1. The welding process reliability optimization acceleration method and system based on machine learning of the present invention adopts machine learning means and uses a limited number of multi-scale numerical simulation calculation results of welds to train the model, so as to realize the rapid prediction and optimization of the reliability of welds formed by different welding process parameters, accelerate the process optimization cycle, and provide a fast and effective means for welding process optimization.

[0048] 2. The present invention provides a welding process reliability optimization acceleration method and system based on machine learning. This method can assist the weld multi-scale numerical simulation method to achieve process optimization. If all weld multi-scale numerical simulations are used to calculate reliability, it takes a long time, is inefficient, and occupies a large amount of resources. It takes about 1 week to calculate the reliability of a set of structural parts corresponding to a set of process parameters on a high-performance workstation, including about 1 day for finite element analysis of the welding process, about 3 days for phase field simulation, about 2 days for crystal plasticity finite element analysis, and about 1 day for macroscopic finite element analysis of working conditions and fatigue reliability. This method can reduce the multi-scale numerical simulations that originally required dozens or hundreds of times to more than ten times or even less, and the remaining times are replaced by machine learning model predictions, greatly improving the efficiency of process optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0050] Figure 1 This is a flow chart of the welding process reliability optimization acceleration method based on machine learning of the present invention;

[0051] Figure 2 A detailed flow chart of the welding process reliability optimization acceleration method based on machine learning of the present invention;

[0052] Figure 3 This is the EBSD characterization result of the weld-heat-affected zone of the present invention;

[0053] Figure 4 It is a graph of input and output of multi-scale numerical simulation data of the present invention;

[0054] Figure 5 This is an example of the residual stress field of the present invention;

[0055] Figure 6 This is an example of a thermal cycle curve of the present invention;

[0056] Figure 7 This is an example of the results of tissue evolution of the present invention;

[0057] Figure 8 This is an example of the stress-strain curve of the present invention;

[0058] Fig. 9 This is an example of the change of plastic work per cycle with fatigue cycles corresponding to different stress amplitudes of the present invention;

[0059] Fig.10 This is an example of the stress field of the working condition of the present invention;

[0060] Fig.11 This is an example of a fatigue life prediction cloud chart of the present invention;

[0061] Fig.12 is the mean square error of the training group and the validation group during the iterative training process of the artificial neural network model of the present invention;

[0062] Fig.13 is the mean square error of the training group and the validation group in the iterative process of Bayesian linear regression training of the present invention;

[0063] Fig.14 This is a structural block diagram of the welding process reliability optimization acceleration system based on machine learning of the present invention. DETAILED DESCRIPTION

[0064] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0065] The welding process reliability optimization acceleration method and system based on machine learning in this application are used for the accelerated optimization of welding processes. On the basis of obtaining effective data through welding process tests and multi-scale numerical simulation methods of welds, rapid prediction and optimization of weld reliability are achieved, providing a fast and effective means for welding process optimization.

[0066] like Figure 2 As shown, the specific steps are: Step A, welding process design and test; Step B, multi-scale numerical simulation calculation of welds; Step C, combined model training of multiple dimensionality reduction algorithms and multiple machine learning algorithms; Step D, variance comparison and selection of multiple combined models; Step E, rapid prediction and selection of weld reliability based on machine learning; Step F: multi-scale numerical simulation verification of welds; Step G: welding process test verification; repeat steps BG until the welding process is optimized.

[0067] Example 1

[0068] like Figure 1 As shown, the welding process reliability optimization acceleration method based on machine learning of the present invention takes a single tube butt weld as an example, the materials on both sides are TA16 and TA17 respectively, the welding method is automatic wire-filling argon arc welding, and the welding sequence is one primer and one filler. The method of the present invention comprises:

[0069] S1: Obtain a process parameter list, carry out a welding process test of the test piece according to the process parameter list, and obtain historical test data;

[0070] Step S1 specifically includes:

[0071] S11: The peak current of the initial parameters is determined to be 52A based on the actual project situation;

[0072] S12: Based on the initial parameters, the peak current is increased or decreased to form a process parameter list; as shown in Table 1, Table 1 is a nine-level process parameter list, and the upper and lower limits of the process parameter list envelop possible optimal parameters.

[0073] Table 1 Process parameters list

[0074]

[0075] Note: XXX is the process test piece number, starting from 001.

[0076] S13: Conduct a welding process test of the test piece according to the process parameter list to obtain a process test piece after welding;

[0077] S14: Perform EBSD (electron backscatter diffraction technology) characterization, base material tensile test and fatigue test on the process test pieces after welding to obtain test results; and use the test results as historical test data.

[0078] The test results will be used to verify the multi-scale numerical simulation model of the weld in step S2, such as Figure 3 As shown, Figure 3 This is the EBSD characterization result of weld-heat-affected zone.

[0079] S2: Construct a multi-scale numerical simulation model of the weld, and use the multi-scale numerical simulation model of the weld to perform numerical simulation analysis based on historical test data to obtain simulation calculation results;

[0080] The multi-scale numerical simulation model of the weld includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model, wherein the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

[0081] like Figure 4 As shown in the figure, each sub-model describes in detail the input and output relationship of the data between each part. The blue data box indicates that the data is obtained by numerical simulation calculation, and the data in the white data box comes from the historical test data in step 1, material parameter database or literature and other channels.

[0082] It should be noted that the welding process finite element analysis sub-model, the working condition macro finite element analysis sub-model and the fatigue life macro finite element analysis sub-model are macro analysis, the phase field simulation sub-model is micro analysis, and the crystal plasticity analysis sub-model is meso analysis. The present invention integrates multi-scale numerical simulation analysis of welds.

[0083] Step S2 specifically includes:

[0084] S21: Construct a finite element analysis sub-model of the welding process in the finite element analysis software; calculate the residual stress field based on the finite element analysis sub-model of the welding process according to the historical test data and the first data, and extract the thermal cycle curves of the weld, heat affected zone and base material; that is, the input of the finite element analysis sub-model of the welding process: process parameters, model structure, material thermal conductivity, specific heat capacity, thermal expansion coefficient; the output of the finite element analysis sub-model of the welding process is the residual stress field and thermal cycle curve.

[0085] Specifically, a weld structure model is established in the finite element analysis software, a welding heat source model is established according to the process parameters and the motion trajectory is defined, and the residual stress field is calculated. Figure 5 As shown in the figure, the temperature history of several points in the weld, heat affected zone and base material is extracted to obtain the thermal cycle curve. Figure 6 shown.

[0086] S22: Construct a phase field simulation sub-model; perform phase field simulation based on the phase field simulation sub-model according to the residual stress field, thermal cycle curve and the second data, and obtain the microstructure evolution results of the weld, heat affected zone and base material; that is, the input of the phase field simulation sub-model: the residual stress field, thermal cycle curve, phase change driving force curve, phase fraction curve, and solid phase fraction curve output by step S21; the output of the phase field simulation sub-model: microstructure evolution results.

[0087] Specifically, a phase field model of solidification structure and solid phase transformation is established, and a phase field simulation is performed based on the thermal cycle curve obtained by finite element analysis of the welding process to obtain the results of the weave evolution. Figure 7 The phase field simulation model was calibrated based on the EBSD characterization data of the process test piece.

[0088] S23: Construct a crystal plasticity analysis sub-model; according to the microstructure evolution results and the third data, perform simulation based on the crystal plasticity analysis sub-model to obtain the SN curves of the weld, heat-affected zone and base material, as well as the stress-strain curves of the weld, heat-affected zone and base material; that is, the input of the crystal plasticity analysis sub-model: microstructure evolution results and elastic constant matrix, crystal plane index, crystal direction index; the output of the crystal plasticity analysis sub-model: SN curve and stress-strain curve.

[0089] Specifically, a crystal plasticity analysis sub-model was established, and the orientation data of 100 grains were extracted based on the microstructure evolution results obtained by the phase field simulation sub-model to establish a model. The inverse pole figures of the microstructure model after modeling and the microstructure evolution results before extraction were plotted to verify the consistency between the two. Tensile and tension-compression fatigue simulations were performed on the established crystal plasticity analysis sub-model, and the stress-strain curves of the local microstructure were calculated as follows: Figure 8 As shown, the SN curve is composed of Fig. 9 The plastic work per cycle corresponding to the different stress amplitudes shown is fitted with the life cycle value calculated by the fatigue cycle. The crystal plasticity analysis sub-model is verified according to the results of the tensile test and fatigue test of the base material of the process test piece. Figure 8 The horizontal axis represents the true strain value, and the vertical axis represents the true stress value; Fig. 9 The horizontal axis represents the number of cyclic loading, and the vertical axis represents the plastic work per week.

[0090] S24: Construct a macroscopic finite element analysis sub-model of the working condition; perform calculations based on the macroscopic finite element analysis sub-model of the working condition according to the SN curve, stress-strain curve, residual stress field and the fourth data, and obtain the stress distribution and load spectrum under the working condition; that is, the input of the macroscopic finite element analysis sub-model of the working condition: SN curve and stress-strain curve, residual stress field, working condition, JC constitutive parameters; the output of the macroscopic finite element analysis sub-model of the working condition: stress distribution and load spectrum under the working condition.

[0091] Specifically, taking the tensile working condition as an example, the welding residual stress field obtained by the finite element analysis sub-model of the welding process is imported into the initial state, and the weld joint is divided into three parts: weld, heat affected zone, and base material. The stress-strain curves obtained by the crystal plasticity analysis sub-model are assigned to each part. The stress field of the working condition is calculated as follows: Fig.10 In addition, the load spectrum is obtained by adding the time variation factor to the stress field.

[0092] S25: construct a fatigue life macro finite element analysis sub-model; according to the load spectrum and SN curve, calculate based on the fatigue life macro finite element analysis sub-model to obtain the fatigue life cloud map of the process test piece; that is, the input of the fatigue life macro finite element analysis sub-model: load spectrum and SN curve; the output of the fatigue life macro finite element analysis sub-model: fatigue life cloud map of the process test piece.

[0093] Specifically, an appropriate fatigue life macroscopic finite element analysis sub-model is selected, such as the Brown-Miller model for predicting metal fatigue life at room temperature, and the SN curves of the three regions of the weld, heat-affected zone, and base material obtained by the crystal plasticity analysis sub-model are imported. Different regions of the joint have different fatigue properties. The fatigue life prediction cloud diagram corresponding to the stress field and residual stress field under different working conditions is calculated. For example Fig.11 shown.

[0094] S3: Perform dimensionality reduction processing on the simulation calculation results, and input the data after dimensionality reduction processing into multiple machine learning algorithms for training and verification, and select the best machine learning model;

[0095] Step S3 specifically includes:

[0096] S31: Performing dimensionality reduction processing on the simulation calculation results to obtain data after dimensionality reduction processing;

[0097] Specifically, data dimensionality reduction methods such as principal component analysis, linear discriminant analysis, kernel principal component analysis, isometric mapping, local linear embedding, random neighbor embedding, t-SNE, etc. can be used to reduce the data that can be processed in the simulation calculation results;

[0098] S32: dividing the data after the dimension reduction process into a training set and a validation set, and inputting the training set into a plurality of machine learning algorithms for training to obtain fatigue training results;

[0099] Specifically, various machine learning algorithms include multivariate linear regression, locally weighted linear regression, polynomial regression, Lasso regression & Ridge regression, elastic net regression, Bayesian ridge regression, Huber regression, artificial neural network, KNN, SVM, CART tree, random forest, GBDT, XGBOOST, time series-ARIMA, hidden Markov, etc.

[0100] S33: Verify the mean square error of the fatigue training results based on the validation set, and obtain the machine learning algorithm corresponding to the fatigue training result with the smallest mean square error as the optimal machine learning model.

[0101] The following are the training and validation mean square errors of two models with better prediction effects. The artificial neural network model contains two closely connected hidden layers, and the output layer can return a single continuous value. The mean square error of the validation group as a loss function tends to oscillate and diverge as the number of iterations increases. Although the mean square error is small, it is not as stable as the Bayesian linear regression model. Fig.12 and Fig.13 As shown, Fig.12 is the mean square error of the training group and the validation group during the iterative training of the artificial neural network model, Fig.13 is the mean square error of the training group and the validation group during the Bayesian linear regression training iteration. Fig.12 and Fig.13 The horizontal axis represents the number of iterations, and the vertical axis represents the mean square error.

[0102] S4: Predict the welding process reliability based on the optimal machine learning model and obtain the optimal process parameters;

[0103] Specifically, a large number of process parameters within the upper and lower limits of the proposed process parameter list are imported into the optimal machine learning model, and the process parameters with the best reliability are selected through machine learning prediction.

[0104] S5: Perform multi-scale numerical simulation verification of the weld for the optimal process parameters, and perform welding process verification on the optimal process parameters that meet the verification requirements until the welding process is optimized.

[0105] Step S5 specifically includes:

[0106] S51: Perform multi-scale numerical simulation of the weld for the optimal process parameters to obtain a simulation value; and verify whether the deviation between the predicted value and the simulation value of the optimal machine learning model meets the preset requirements;

[0107] S52: If the deviation meets the preset requirements, the welding process verification is continued for the optimal process parameters that meet the requirements to verify whether the parameters such as size and performance meet the requirements. If the parameters meet the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iteration until the welding process is optimized;

[0108] S53: If the deviation does not meet the requirements, retrain and verify multiple machine learning algorithms to select the best one.

[0109] The present invention is to design and test welding process, calculate multi-scale numerical simulation of welds, train a combination model of multiple dimensionality reduction algorithms and multiple machine learning algorithms, compare and select the variances of multiple combined models, quickly predict and select the reliability of welds based on machine learning, verify the multi-scale numerical simulation of welds, and verify the welding process by testing; repeat the above steps until the welding process is optimized. This method aims at the problem of low efficiency of multi-scale numerical simulation calculation of welds, adopts machine learning means, and uses a limited number of multi-scale numerical simulation calculation results of welds to train the model, so as to realize the rapid prediction of the reliability of welds formed by different welding process parameters and accelerate the process optimization cycle.

[0110] Example 2

[0111] like Fig.14 As shown, the difference between this embodiment and embodiment 1 is that this embodiment provides a welding process reliability optimization acceleration system based on machine learning, which corresponds to the welding process reliability optimization acceleration method based on machine learning in embodiment 1 in a one-to-one manner; the system includes:

[0112] The welding process test unit is used to obtain a list of process parameters, carry out welding process tests on test pieces according to the list of process parameters, and obtain historical test data;

[0113] The multi-scale numerical simulation model unit is used to construct a multi-scale numerical simulation model of the weld, and to perform numerical simulation analysis using the multi-scale numerical simulation model of the weld based on historical test data to obtain simulation calculation results;

[0114] The model optimization unit is used to perform dimensionality reduction processing on the simulation calculation results, and input the data after dimensionality reduction processing into a variety of machine learning algorithms for training and verification, and select the best machine learning model;

[0115] Welding process prediction unit, used to predict welding process reliability based on the optimal machine learning model and obtain the optimal process parameters;

[0116] The simulation and welding process verification unit conducts multi-scale numerical simulation verification of the weld for the optimal process parameters, and conducts welding process verification for the optimal process parameters that meet the verification requirements until the welding process is optimized.

[0117] As a further implementation, the weld multi-scale numerical simulation model includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model;

[0118] Among them, the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

[0119] As a further implementation, the simulation and welding process verification unit includes:

[0120] The model simulation verification subunit is used to perform multi-scale numerical simulation of the weld for the optimal process parameters to obtain the simulation value; and verify whether the deviation between the predicted value and the simulation value of the optimal machine learning model meets the requirements; if the deviation meets the requirements, the welding process verification is continued for the optimal process parameters that meet the requirements; if the deviation does not meet the requirements, multiple machine learning algorithms are retrained and verified to select the best one;

[0121] The welding process verification subunit is used to verify the welding process and verify whether each parameter meets the requirements. If each parameter meets the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iterations until the welding process is optimized.

[0122] Among them, the execution process of each unit can be executed according to the process steps of the welding process reliability optimization acceleration method based on machine learning in Example 1, and will not be repeated one by one in this embodiment.

[0123] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0124] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0127] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A welding process reliability optimization acceleration method based on machine learning, characterized in that: The method includes: Obtain a process parameter list, conduct a welding process test of a test piece according to the process parameter list, and obtain historical test data; Constructing a multi-scale numerical simulation model of a weld, and performing numerical simulation analysis using the multi-scale numerical simulation model of the weld according to the historical test data to obtain simulation calculation results; Performing dimensionality reduction processing on the simulation calculation results, and inputting the dimensionality reduced data into a variety of machine learning algorithms for training and verification, and selecting the best machine learning model; Based on the optimal machine learning model, the reliability of the welding process is predicted to obtain the optimal process parameters; The optimal process parameters are verified by multi-scale numerical simulation of the weld, and the welding process verification is performed on the optimal process parameters that meet the verification requirements until the welding process is optimized.

2. The welding process reliability optimization acceleration method based on machine learning according to claim 1 is characterized in that: Obtain a process parameter list, conduct welding process tests on the test piece according to the process parameter list, and obtain historical test data, including: Determine and formulate the peak current of initial parameters based on actual project conditions; Based on the initial parameters, the peak current is increased or decreased to form a process parameter list; Carry out welding process test of the test piece according to the process parameter list to obtain the process test piece after welding; The welded process test piece is subjected to EBSD characterization, base material tensile test and fatigue test to obtain test results; and the test results are used as historical test data.

3. The welding process reliability optimization acceleration method based on machine learning according to claim 2 is characterized in that: The upper and lower limits of the process parameter list envelop possible optimal parameters.

4. The welding process reliability optimization acceleration method based on machine learning according to claim 1 is characterized in that: The multi-scale numerical simulation model of the weld includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model; Among them, the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

5. The welding process reliability optimization acceleration method based on machine learning according to claim 4 is characterized in that: Constructing a multi-scale numerical simulation model of the weld, and performing numerical simulation analysis using the multi-scale numerical simulation model of the weld according to the historical test data to obtain simulation calculation results, including: Constructing a finite element analysis sub-model for the welding process; calculating the residual stress field based on the finite element analysis sub-model for the welding process according to the historical test data and the first data, and extracting thermal cycle curves of the weld, the heat affected zone and the base material; Constructing a phase field simulation sub-model; performing phase field simulation based on the phase field simulation sub-model according to the residual stress field, the thermal cycle curve and the second data, and obtaining the microstructure evolution results of the weld, the heat affected zone and the base material; Constructing a crystal plasticity analysis sub-model; performing simulation based on the crystal plasticity analysis sub-model according to the organizational evolution result and the third data, and obtaining SN curves of the weld, the heat affected zone and the base material, and stress-strain curves of the weld, the heat affected zone and the base material; Constructing a macroscopic finite element analysis submodel of working conditions; performing calculations based on the macroscopic finite element analysis submodel of working conditions according to the SN curve, stress-strain curve, residual stress field and fourth data to obtain stress distribution and load spectrum under working conditions; A fatigue life macroscopic finite element analysis submodel is constructed; and according to the load spectrum and the SN curve, a fatigue life macroscopic finite element analysis submodel is used to perform calculations to obtain a fatigue life cloud diagram of the process test piece.

6. The welding process reliability optimization acceleration method based on machine learning according to claim 1 is characterized in that: The simulation calculation results are subjected to dimensionality reduction processing, and the data after dimensionality reduction processing are input into a variety of machine learning algorithms for training and verification, and the optimal machine learning model is obtained by selection, including: Performing dimensionality reduction processing on the simulation calculation results to obtain data after dimensionality reduction processing; Dividing the data after the dimension reduction process into a training set and a validation set, and inputting the training set into a plurality of machine learning algorithms for training to obtain fatigue training results; The mean square error of the fatigue training results is verified based on the verification set, and the machine learning algorithm corresponding to the fatigue training results with the smallest mean square error is obtained as the optimal machine learning model.

7. The welding process reliability optimization acceleration method based on machine learning according to claim 1 is characterized in that: The optimal process parameters are verified by multi-scale numerical simulation of the weld, and the welding process is verified for the optimal process parameters that meet the verification requirements until the welding process is optimized, including: Performing multi-scale numerical simulation of the weld on the optimal process parameters to obtain a simulation value; and verifying whether the deviation between the predicted value of the optimal machine learning model and the simulation value meets the requirements; If the deviation meets the requirements, the welding process verification is continued for the optimal process parameters that meet the requirements to verify whether each parameter meets the requirements. If each parameter meets the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iteration until the welding process is optimized; If the deviation does not meet the requirements, multiple machine learning algorithms are retrained and verified to select the best one.

8. The welding process reliability optimization acceleration system based on machine learning is characterized by: The system includes: A welding process test unit is used to obtain a process parameter list, carry out a welding process test of a test piece according to the process parameter list, and obtain historical test data; A multi-scale numerical simulation model unit is used to construct a multi-scale numerical simulation model of a weld, and to perform numerical simulation analysis using the multi-scale numerical simulation model of the weld according to the historical test data to obtain simulation calculation results; A model optimization unit is used to perform dimensionality reduction processing on the simulation calculation results, and input the data after dimensionality reduction processing into a variety of machine learning algorithms for training and verification, so as to obtain the optimal machine learning model; A welding process prediction unit, used to predict the reliability of the welding process based on the optimal machine learning model to obtain optimal process parameters; The simulation and welding process verification unit performs multi-scale numerical simulation verification of the weld on the optimal process parameters, and performs welding process verification on the optimal process parameters that meet the verification requirements until the welding process is optimized.

9. The welding process reliability optimization acceleration system based on machine learning according to claim 8 is characterized in that: The multi-scale numerical simulation model of the weld includes a welding process finite element analysis sub-model, a phase field simulation sub-model, a crystal plasticity analysis sub-model, a working condition macro finite element analysis sub-model and a fatigue life macro finite element analysis sub-model; Among them, the output of the welding process finite element analysis sub-model is used as the input of the phase field simulation sub-model, the output of the phase field simulation sub-model is used as the input of the crystal plasticity analysis sub-model, the output of the crystal plasticity analysis sub-model is used as the input of the working condition macro finite element analysis sub-model, and the output of the working condition macro finite element analysis sub-model is used as the input of the fatigue life macro finite element analysis sub-model.

10. The welding process reliability optimization acceleration system based on machine learning according to claim 8 is characterized in that: The simulation and welding process verification unit includes: The model simulation verification subunit is used to perform multi-scale numerical simulation of the weld for the optimal process parameters to obtain simulation values; and verify whether the deviation between the predicted value of the optimal machine learning model and the simulation value meets the requirements; if the deviation meets the requirements, the welding process verification is continued for the optimal process parameters that meet the requirements; if the deviation does not meet the requirements, multiple machine learning algorithms are retrained and verified to select the best one; The welding process verification subunit is used to verify the welding process and verify whether each parameter meets the requirements. If each parameter meets the requirements, the welding process has been optimized; otherwise, the multi-scale numerical simulation model of the weld is used to perform welding process optimization iteration until the welding process is optimized.

Images