CN113486588A - Metal material performance calculation method based on machine learning model - Google Patents
Metal material performance calculation method based on machine learning model Download PDFInfo
- Publication number
- CN113486588A CN113486588A CN202110766139.7A CN202110766139A CN113486588A CN 113486588 A CN113486588 A CN 113486588A CN 202110766139 A CN202110766139 A CN 202110766139A CN 113486588 A CN113486588 A CN 113486588A
- Authority
- CN
- China
- Prior art keywords
- model
- machine learning
- metal material
- space
- determining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a metal material performance calculation method based on a machine learning model, which specifically comprises the following steps: determining an input space of an experimental sample according to material component distribution, process and experience of existing data; step two, determining the extracted features of each input sampleStep three, determining the extracted characteristics of the input sampleThen, an iterative sequence learning loop is performed. The invention relates to the technical field of machine learning algorithms, and particularly provides a metal material performance calculation method based on a machine learning modelCompared with the prior art, the method has the advantages that under the condition that the same performance index is achieved, the number of experimental points is greatly reduced, and the time and cost for material development are saved; compared with a manual trial and error method, the method does not need abundant development experience, and reduces the requirements on qualified research and development personnel.
Description
Technical Field
The invention relates to the technical field of machine learning algorithms, in particular to an optimization algorithm of metal material performance indexes in the field of material informatics based on a machine learning model, and specifically relates to a metal material performance calculation method based on the machine learning model.
Background
The metal material is a material having properties such as luster, ductility, easy electrical conduction, and heat transfer. The metal materials are various in types and play a vital role in national economic development.
Current manufacturers typically design products or components under certain material property standards. In order to ensure that the indexes meet expectations, manufacturers can perform simulation, experimental tests and trial production. Manufacturers usually choose the appropriate material label from a manual or design a new material, and ask suppliers to develop a material meeting the performance according to the requirement. However, the new materials are not known to the manufacturer or supplier, and some performance indexes are difficult to be realized by the supplier.
Disclosure of Invention
Aiming at the situation, in order to make up for the existing defects, the invention provides a metal material performance calculation method based on a machine learning model based on the defects of the existing experimental trial-and-error method, and the method realizes automatic machine learning model parameter adjustment through an algorithm model; and one or more material component ratios which are most likely to realize the material performance improvement under the current training set are found out by calculating the model optimization confidence.
The invention provides the following technical scheme: the invention relates to a metal material performance calculation method based on a machine learning model, which specifically comprises the following steps:
step one, determining an input space of an experimental sample according to material component distribution, process and experience of existing data, wherein the determination of the input space comprises the following steps:
(1) inputting a parameter space M of the components;
(2) inputting a parameter space C of the processing technology;
(3) inputting a parameter space P of a post-treatment process;
(4) after three subspaces are determined, the input space X is M multiplied by C multiplied by P (Cartesian product), and the output space Y is output;
for example, for titanium alloys, the reference number is Ti6Al4V, then the determination of the input space comprises the steps of:
(1) parameter space M of input components
For the existing reference number Ti6Al4V, the input space should contain the composition ratio of Ti, Al and V; according to experience, the component proportion of C and B should be considered, so the input space of the material components should be M ═ Ti%, Al%, V%, C%, B% }, wherein Ti% is the component proportion of Ti, other components are similar, and the production process and post-treatment process parameters of the material also need to be considered;
(2) parameter space C of input machining process
The machining process is a means of material addition, material reduction, deformation and the like performed by the manufacturing side on the basis of raw materials, so that the final form meets the expectations of the manufacturing side, the machining type comprises but is not limited to forging, casting, cutting and 3D printing, taking 3D printing as an example, the machining process comprises but is not limited to laser power C1Scanning rate C2Laser size C3Energy density C4Etc., which can be expressed as C ═ C1,C2,C3,C4,…}。
(3) Inputting a parameter space P of a post-processing process
Post-treatment processThe method is a means for further improving the performance of the metal part by appropriate change of temperature or environment in order to eliminate internal stress or defects after the metal part is manufactured. This step is optional (when the post-treatment process is the same for all data points, the effect of the post-treatment process need not be considered). Post-treatment processes include, but are not limited to, annealing temperature P1Carbonization time P2And normalizing temperature P3Etc., which can be expressed as P ═ P1,P2,P3,…};
Step two, determining the extracted features of each input sampleExtracted featuresThe method comprises the following steps:
(1) the parameters of the input space themselves: such as the proportion of each component and the quenching temperature of the post-treatment;
(2) microscopic and mesoscopic parameters: such as elemental mass, bandwidth energy, dislocations;
(3) data relating to material organization: such as a golden photo, etc.;
step three, determining the extracted characteristics of the input sampleThen, an iterative sequence learning loop is performed:
(1) searching model hyper-parameters on a training set/a verification set through an algorithm model to obtain a model hyper-parameter combination with the best performance index on the verification set, wherein the performance index comprises MSE and MAE, and the used model comprises but is not limited to one or combination of a Bayesian optimization algorithm model (hyper t/optuna and the like), a random forest algorithm model, a gradient lifting tree algorithm model and a depth integration algorithm model;
(2) according to experience and experimental settings, obtaining an alternative test set under the minimum step length, and operating the optimal model obtained in the previous step on the test set to obtain the prediction performance index and corresponding uncertainty corresponding to each test data;
(3) according to the predicted performance index and the corresponding uncertainty of each test data, calculating model optimization confidence, selecting one or more test data with the highest confidence for real experiment and measuring the performance index, wherein the available model optimization confidence includes but is not limited to maximum improvement likelihood, maximum expected improvement, maximum uncertainty and the like, and taking the maximum improvement likelihood as an example, firstly calculating the single-sample uncertainty:
wherein CovjRepresenting the variance calculated over the jth tree, ni,jRepresents the number of samples i, t, used in training the jth class treej(x) Representing the predicted outcome of the jth tree,represents the average of the predicted results of all trees for which sample i did not participate in the training,is the mean of the predicted results of all trees, e and v represent the variance of the Euler constant and all trees, respectively, and T is the number of trees. The uncertainty is then:
let FXIs normally distributed N (x, sigma)2(x) ) a cumulative distribution function with a maximum improvement likelihood equal to 1-FX(xbest) Wherein x isbestThe optimal result of the current experiment is obtained;
(4) if the performance indexes of the data obtained in the previous step meet the requirements, the cycle is ended, otherwise, the iteration is continued in the first step.
The invention with the structure has the following beneficial effects: according to the metal material performance calculation method based on the machine learning model, the model is used for improving the confidence coefficient to find the optimal alternative experimental points, compared with a random sampling method, under the condition that the same performance index is achieved, the number of the experimental points is greatly reduced, and the time and the cost of material development are saved; compared with a manual trial and error method, the method does not need abundant development experience, and reduces the requirements on qualified research and development personnel. Meanwhile, the whole process realizes digitization and automation, and the overall efficiency of product development is also improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a first sample model branch diagram obtained by optimizing a random forest algorithm model and a Bayesian optimization algorithm model in a metal material performance calculation method based on a machine learning model according to the invention;
FIG. 2 is a sample model branch diagram II obtained by optimizing a random forest algorithm model and a Bayesian optimization algorithm model in the metal material performance calculation method based on the machine learning model;
FIG. 3 is a third sample model branch diagram obtained by optimizing a random forest algorithm model and a Bayesian optimization algorithm model in the metal material performance calculation method based on the machine learning model of the invention;
fig. 4 is a schematic diagram of the predicted performance and uncertainty distribution of a test sample according to an embodiment of the metal material performance calculation method based on a machine learning model.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "front," "back," "left," "right," "upper" and "lower" used in the following description refer to directions in the drawings, and the terms "inner" and "outer" refer to directions toward and away from, respectively, the geometric center of a particular component.
Examples
Assuming that the objective of the optimization is Ti6Al4V, the existing training samples are:
(1) 0.907% Ti, 0.05% Al, 0.04% V, 0.001% C, 0.002% B, 100 laser power, 5 scan rate, 1 laser size, 2 energy density, 500MPa performance index.
(2) 0.916% Ti, 0.04% Al, 0.04% V, 0.002% C, 0.002% B, 50 laser power, 3 scan rate, 1 laser size, 3 energy density, 600MPa performance index.
(3) 0.918 wt% Ti, 0.05 wt% Al, 0.03 wt% V, 0.001 wt% C, 0.001 wt% B, 100 laser power, 2 scan rate, 1 laser size, 3 energy density, 400MPa performance index.
Because the post-treatment process of each sample is the same, the influence of the post-treatment process parameters is not considered during training.
The method comprises the following steps: the input space X for the problem is determined. Setting the input space to M C P, the input space for the problem is { Ti%, Al%, V%, C%, B%, P%1,P2,P3}。
Step two: determining input features for each sampleOnly the parameters of the input space are selected as the input features, and the input features of each training sample are as follows:
(1){0.907,0.05,0.04,0.001,0.002,100,5,1,2}
(2){0.916,0.04,0.04,0.002,0.002,50,3,1,3}
(3){0.918,0.05,0.03,0001,0.001,100,2,1,3}
step three: entering a sequence training loop:
and obtaining an optimal model by using a random forest algorithm model and a Bayesian optimization algorithm model, wherein the optimal model is respectively shown as an attached figure 1, an attached figure 2 and an attached figure 3.
Several alternative test samples were generated as follows:
{0.908,0.04,0.05,0.001,0.00,100,5,1,2}
{0.92,0.04,0.04,0.000,0.000,50,3,1,3}
{0.916,0.05,0.03,0.002,0.001,100,2,1,3}
{0.91,0.05,0.035,0.003,0.002,100,5,1,3}
and calculating the prediction performance indexes of the test samples to be 400MPa,425MPa,475MPa and 500 MPa.
calculating the uncertainty of prediction of each sample, and meanwhile, knowing that the current optimal performance index is 480MPa, calculating the maximum improvement likelihood of each sample to be 0,0.1,0.5 and 0.7, referring to the attached figure 4, wherein the area of a shaded part is the maximum improvement likelihood, selecting the group of parameters for real experiment because the maximum improvement likelihood of the 4 th test sample is maximum, wherein the performance index of the experiment result is 470MPa, and the performance index is not improved, and returning to the step 1 to continue iteration.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (4)
1. A metal material performance calculation method based on a machine learning model is characterized by comprising the following steps:
step one, determining an input space of an experimental sample according to material component distribution, process and experience of existing data, wherein the determination of the input space comprises the following steps:
(1) inputting a parameter space M of the components;
(2) inputting a parameter space C of the processing technology;
(3) inputting a parameter space P of a post-treatment process;
(4) after the three subspaces are determined, the input space X is MXC multiplied by P, and the output space Y is output;
step two, determining the extracted features of each input sampleExtracted featuresThe method comprises the following steps:
(1) inputting parameters of the space itself;
(2) microscopic and mesoscopic parameters;
(3) data relating to the organization of the material;
step three, determining the extracted characteristics of the input sampleThen, an iterative sequence learning loop is performed:
(1) searching model hyper-parameters on a training set/a verification set through an algorithm model to obtain a model hyper-parameter combination with the best performance index on the verification set;
(2) according to experience and experimental settings, obtaining an alternative test set under the minimum step length, and operating the optimal model obtained in the previous step on the test set to obtain the prediction performance index and corresponding uncertainty corresponding to each test data;
(3) calculating model optimization confidence according to the predicted performance index and corresponding uncertainty of each test data, selecting one or more test data with highest confidence to perform a real experiment, and measuring the performance index;
(4) if the performance indexes of the data obtained in the previous step meet the requirements, the cycle is ended, otherwise, the iteration is continued in the first step.
2. The machine learning model-based metal material performance calculation method according to claim 1, wherein the algorithm model in step three (1) includes but is not limited to one or more combinations of a bayesian optimization algorithm model, a random forest algorithm model, a gradient boosting tree algorithm model and a depth integration algorithm model.
3. The method for calculating the performance of the metal material based on the machine learning model according to claim 1, wherein the performance index in the step three (1) comprises MSE and MAE.
4. The machine learning model-based metal material property computation method of claim 1, wherein step three (3) the model optimization confidence includes but is not limited to maximum likelihood of improvement, maximum expected improvement, and maximum uncertainty.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110766139.7A CN113486588A (en) | 2021-07-07 | 2021-07-07 | Metal material performance calculation method based on machine learning model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110766139.7A CN113486588A (en) | 2021-07-07 | 2021-07-07 | Metal material performance calculation method based on machine learning model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113486588A true CN113486588A (en) | 2021-10-08 |
Family
ID=77941597
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110766139.7A Withdrawn CN113486588A (en) | 2021-07-07 | 2021-07-07 | Metal material performance calculation method based on machine learning model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113486588A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114674997A (en) * | 2022-03-29 | 2022-06-28 | 创材深造(苏州)科技有限公司 | Algorithm for testing composition of alloy powder |
CN116484228A (en) * | 2023-05-04 | 2023-07-25 | 小米汽车科技有限公司 | Model training method, process determining method, device, electronic equipment and medium |
-
2021
- 2021-07-07 CN CN202110766139.7A patent/CN113486588A/en not_active Withdrawn
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114674997A (en) * | 2022-03-29 | 2022-06-28 | 创材深造(苏州)科技有限公司 | Algorithm for testing composition of alloy powder |
CN114674997B (en) * | 2022-03-29 | 2024-04-30 | 创材深造(苏州)科技有限公司 | Algorithm for testing composition combination of alloy powder |
CN116484228A (en) * | 2023-05-04 | 2023-07-25 | 小米汽车科技有限公司 | Model training method, process determining method, device, electronic equipment and medium |
CN116484228B (en) * | 2023-05-04 | 2024-02-06 | 小米汽车科技有限公司 | Model training method, process determining method, device, electronic equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113486588A (en) | Metal material performance calculation method based on machine learning model | |
CN110175386B (en) | Method for predicting temperature of electrical equipment of transformer substation | |
CN109063234B (en) | High-speed press force application part reliability design method considering multiple types of uncertainty | |
CN108985335B (en) | Integrated learning prediction method for irradiation swelling of nuclear reactor cladding material | |
JP6950119B2 (en) | Material design system, material design method, and material design program | |
CN107038292B (en) | Multi-wind-field output correlation modeling method based on self-adaptive multivariate nonparametric kernel density estimation | |
EP2778990A2 (en) | Method and system for designing a material | |
CN111931983B (en) | Precipitation prediction method and system | |
Wu et al. | Solving the set-union knapsack problem by a novel hybrid Jaya algorithm | |
CN111310348A (en) | Material constitutive model prediction method based on PSO-LSSVM | |
CN114298134A (en) | Wind power prediction method and device and electronic equipment | |
CN107894710B (en) | Principal component analysis modeling method for temperature of cracking reaction furnace | |
CN110611334A (en) | Copula-garch model-based multi-wind-farm output correlation method | |
CN115237878A (en) | Process database construction method and medium based on additive manufacturing | |
CN113073267A (en) | High-strength high-temperature-resistant RAFM steel and machine learning-based design method thereof | |
CN112651153B (en) | Method for determining material parameters of crystal plasticity finite element model | |
CN111931990B (en) | Power load prediction method based on optimized least square support vector machine | |
CN113590748B (en) | Emotion classification continuous learning method based on iterative network combination and storage medium | |
Gu et al. | A cross domain feature extraction method for bearing fault diagnosis based on balanced distribution adaptation | |
CN115410671A (en) | Method and system for predicting low-cycle fatigue life of nickel-based superalloy | |
Abdelfettah et al. | Synergy between fractional order control and industry 4.0: a bibliometric analysis | |
Virdi et al. | Feature selection with LASSO and VSURF to model mechanical properties for investment casting | |
Bharati et al. | A performance prediction of worm-worm wheel and its multi objective design parameter optimization based on genetic algorithm | |
CN112508250B (en) | Incremental analysis method, system, medium and terminal for command information system generation scheme | |
Lagos et al. | Decision Trees+ Evolutionary Algorithm for Predict then Optimize |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20211008 |