CN-122021258-A - Metal constitutive relation prediction model based on machine learning and modeling method and application

CN122021258ACN 122021258 ACN122021258 ACN 122021258ACN-122021258-A

Abstract

The invention belongs to the technical field of material mechanical property modeling and machine learning intersection, and particularly relates to a metal constitutive relation prediction model based on machine learning, a modeling method and application thereof, wherein the modeling method of the metal constitutive relation prediction model based on machine learning comprises the following steps: acquiring multi-working condition experimental data, constructing training sample sets corresponding to different deformation stages, and respectively performing machine learning model training by utilizing each training sample set to obtain a stage model corresponding to each deformation stage; and optimizing the super parameters of the models in each stage by using a Bayesian optimization algorithm respectively, and obtaining the metal constitutive relation prediction model through weighted fusion. Prediction precision of each deformation stage of a metal constitutive relation prediction model constructed by the modeling method provided by the invention The full curve MAE is less than 2.5 MPa, the calculation efficiency is improved by 50% compared with a fixed characteristic model, and the method is suitable for multi-stage stress prediction of the thermoforming of the aerospace structural part.

Inventors

ZHANG MEIJUAN
QIE XIWANG
YANG DONGLI
WANG LIJUAN
ZHANG TONG

Assignees

北京航空材料研究院股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. The modeling method of the metal constitutive relation prediction model based on machine learning is characterized by comprising the following steps of: Acquiring multi-working condition experimental data, and dividing the multi-working condition experimental data into a plurality of original data sets according to different deformation stages, wherein one original data set corresponds to one deformation stage; Constructing a plurality of training sample sets corresponding to different deformation stages, wherein one training sample set corresponds to one deformation stage, each training sample set independently comprises an input parameter and an output parameter, the input parameter comprises at least two of physical derivative characteristics, temperature, strain and strain rate, and the output parameter comprises stress; training the machine learning model by using each training sample set to obtain a phase model corresponding to each deformation phase, and And respectively optimizing the super parameters of each stage model by using a Bayesian optimization algorithm, and obtaining a metal constitutive relation prediction model through weighted fusion.
2. The modeling method of claim 1, wherein constructing a plurality of training sample sets corresponding to different deformation phases comprises: and processing the acquired multi-working condition experimental data by using a multi-physical coupling constitutive equation to generate the physical derivative characteristics corresponding to each deformation stage.
3. The modeling method of claim 1, wherein the physically derived features include normalized values of elastic modulus Normalized value of strain Temperature correction factor Rate of logarithmic strain Inverse normalized temperature First-order strain hardening rate Cumulative amount of hardening rate Strain rate-temperature coupling term At least three of (a) and (b).
4. The modeling method of claim 2, wherein the equation of the multi-physical coupling constitutive equation is: ; In the elastic modulus normalization value , The elastic modulus is the elastic modulus at the current temperature, and the unit is MPa; An elastic modulus of 25 ℃ in MPa; , Is the limit strain of the first deformation stage, and the temperature correction factor ; As a temperature influence coefficient of the temperature, =25°C; Initial strain for the third deformation stage, yield strength First-order strain hardening rate Hardening rate accumulation amount Strain rate-temperature coupling term Wherein, C is the coupling coefficient, unit MPa.K; optionally, the first deformation stage and the second deformation stage are each independently selected from one of an elastic stage, a plastic stage, a recrystallization stage; Optionally, the first deformation stage is selected from an elastic stage and the second deformation stage is selected from a plastic stage.
5. The modeling method of claim 1, wherein using a bayesian optimization algorithm to stage an objective function for optimizing a hyper-parameter of the machine learning model comprises: ; ; In the formula, ; ; ; Representing the i-th stress value predicted by the predictive model, Representing the experimentally measured ith stress value, Representing the ith stress value generated by the multiple physically coupled constitutive equation, Determining coefficients representing a multi-physical coupling constitutive equation; Optionally, the super parameters comprise 8-20 layers of maximum tree depth, 2-8 minimum sample numbers of leaf nodes, 3-7 dimensions of feature subset and 5-15 minimum sample numbers of node splitting.
6. The modeling method of claim 1, wherein in the transition region of two adjacent deformation phases, the phase models are integrated by using a linear weight fusion method, and a fusion equation is expressed as: Wherein, the method comprises the steps of, , 。
7. The modeling method as defined in claim 1, further comprising performing model verification on the metal constitutive relation prediction model; optionally, each deformation stage adopts 5-fold cross validation to evaluate the performance of the metal constitutive relation prediction model, and each deformation stage evaluation index is: , MPa; optionally, the evaluation index of the first deformation stage is: , The evaluation indexes of the second deformation stage are as follows: , MPa; Optionally, the evaluation index of the metal constitutive relation prediction model is: 、 MPa。
8. A metal constitutive relation prediction model based on machine learning, characterized in that the prediction model is constructed based on the modeling method according to any one of claims 1 to 7, and the prediction model comprises: The data acquisition module is used for acquiring multi-working-condition experimental data; The system comprises a phase division module, a phase analysis module and a phase analysis module, wherein the phase division module is used for dividing multi-working condition experimental data into a plurality of original data sets according to different deformation phases, and one original data set corresponds to one deformation phase; The system comprises a whole data set construction module, a test sample set and a data analysis module, wherein the whole data set construction module is used for constructing a plurality of whole data sets corresponding to different deformation stages, one whole data set corresponds to one deformation stage, each whole data set is independently divided into a training sample set and a test sample set, the training sample set comprises an input parameter and an output parameter, the input parameter comprises at least two of physical derivative characteristics, temperature, strain and strain rate, and the output parameter comprises stress; the model construction and optimization module is used for respectively performing machine learning model training by utilizing each training sample set to obtain a phase model corresponding to each deformation phase; and the fusion module is used for obtaining a metal constitutive relation prediction model through transition weighted fusion.
9. The predictive model of claim 8, The prediction system also comprises a verification module, a prediction module and a prediction module, wherein the verification module is used for carrying out model verification on the metal constitutive relation prediction model; Optionally, the prediction system further comprises a prediction module, a prediction module and a prediction module, wherein the prediction module is used for inputting the working condition temperature, the working condition strain and the working condition strain rate data of the working condition to be detected into a metal constitutive relation prediction model to obtain a stress predicted value of the working condition to be detected, which is output by the prediction model, and obtaining a metal stress-strain constitutive relation under the working condition to be detected; Optionally, the phase division module comprises a derivative calculation sub-module for calculating a first derivative of the stress-strain curve using a central difference method to divide to obtain a plurality of raw data sets.
10. Use of the machine-learning-based metal constitutive relation prediction model according to any one of claims 8 to 9 or constructed by the modeling method according to any one of claims 1 to 7, characterized in that it comprises: and inputting the working condition temperature, working condition strain and working condition strain rate data of the target metal in the working condition to be detected into a metal constitutive relation prediction model to obtain a stress predicted value of the working condition to be detected, which is output by the prediction model, and obtaining a metal stress-strain constitutive relation under the working condition to be detected.

Description

Metal constitutive relation prediction model based on machine learning and modeling method and application Technical Field The invention belongs to the technical field of material mechanical property modeling and machine learning intersection, and particularly relates to a metal constitutive relation prediction model based on machine learning, a modeling method and application. Background The TA15 titanium alloy is a core material of key components such as an aerospace engine blade, a fuselage frame and the like by virtue of excellent specific strength, high-temperature stability and corrosion resistance. Titanium alloy thermoforming processes often undergo multi-stage deformation, an elastic stage (strain is small, stress is proportional to strain, interatomic elastic force dominates), a plastic flow stage (strain is increased, work hardening dominates, dislocation motion is the core physical mechanism), and a dynamic recrystallization stage (high temperature domain, stress softens with strain increase, dislocation density decreases as a key feature). The physical mechanisms at different stages differ significantly, resulting in a strong piecewise nonlinear behavior of the stress-strain constitutive relationship. The traditional constitutive model (such as Johnson-Cook and Zerill-Armstrong models) adopts a single empirical formula to represent full-stage deformation, and physical characteristics of each stage are difficult to consider, for example, the Johnson-Cook model has higher prediction precision (error is less than 5%) in an elastic stage, but the error is often more than 20% because the stress softening effect cannot be accurately captured in a dynamic recrystallization stage, and the parameter calibration needs to be carried out in a large number of high-temperature multi-strain rate experiments, and the period is as long as several months and the cost is high. The existing machine learning modeling method has the obvious defects that 1) a fixed derivative feature set is adopted, stage physical differences are not considered, for example, dynamic recrystallization related features are introduced in an elastic stage, so that feature redundancy is caused, the calculation efficiency is reduced by more than 30%, 2) the derivative features and stage physical mechanism are weak in relevance, for example, the work hardening features in a plastic stage have no practical significance in the elastic stage, the model fitting risk is increased, and 3) under the working condition of small samples and multi-stage intersection, the model generalization capability is obviously reduced due to mismatching of the features and the stages, and the error of the non-experimental working condition is often more than 8%. Therefore, the existing traditional constitutive model and machine learning model still need to be improved. Disclosure of Invention The invention aims to provide a metal constitutive relation prediction model based on machine learning, a modeling method and application, and the metal constitutive relation prediction model constructed by the modeling method can output high-precision stress prediction results of each deformation stage and prediction precision of each deformation stageThe full curve MAE is less than 2.5 MPa, the calculation efficiency is improved by 50% compared with a fixed characteristic model, the method is suitable for multi-stage stress prediction of hot forming of an aerospace structural member, and at least the problems of poor multi-stage suitability of a traditional model, redundancy of existing machine learning characteristics and weak physical interpretability are solved. The invention provides a modeling method of a metal constitutive relation prediction model based on machine learning, which comprises the following steps of obtaining multi-working-condition experimental data, dividing the multi-working-condition experimental data into a plurality of original data sets according to different deformation stages, wherein one original data set corresponds to one deformation stage, each original data set independently comprises temperature, strain rate and stress, constructing a plurality of training sample sets corresponding to different deformation stages, one training sample set corresponds to one deformation stage, each training sample set independently comprises input parameters and output parameters, the input parameters comprise physical derivative characteristics and at least two of temperature, strain and strain rate, the output parameters comprise stress, machine learning model training is conducted by utilizing each training sample set to obtain a stage model corresponding to each deformation stage, and super parameters of each stage model are optimized by utilizing a Bayesian optimization algorithm and are transitionally weighted and fused to obtain the metal constitutive relation prediction model. In some embodiments, constructing a plurality of training sample sets cor