CN116611322A - Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder - Google Patents
Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder Download PDFInfo
- Publication number
- CN116611322A CN116611322A CN202310525021.4A CN202310525021A CN116611322A CN 116611322 A CN116611322 A CN 116611322A CN 202310525021 A CN202310525021 A CN 202310525021A CN 116611322 A CN116611322 A CN 116611322A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- gbdt
- parameters
- powder
- 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.)
- Pending
Links
- 239000000843 powder Substances 0.000 title claims abstract description 100
- 238000000034 method Methods 0.000 title claims abstract description 97
- 238000013461 design Methods 0.000 title claims abstract description 32
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 29
- RTAQQCXQSZGOHL-UHFFFAOYSA-N Titanium Chemical compound [Ti] RTAQQCXQSZGOHL-UHFFFAOYSA-N 0.000 title claims abstract description 25
- 239000000654 additive Substances 0.000 title claims abstract description 24
- 230000000996 additive effect Effects 0.000 title claims abstract description 24
- 229910052719 titanium Inorganic materials 0.000 title claims abstract description 20
- 239000010936 titanium Substances 0.000 title claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 53
- 239000000463 material Substances 0.000 claims abstract description 48
- 230000002787 reinforcement Effects 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 36
- 239000003795 chemical substances by application Substances 0.000 claims description 28
- 238000012360 testing method Methods 0.000 claims description 26
- 230000009471 action Effects 0.000 claims description 19
- 238000005457 optimization Methods 0.000 claims description 18
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 16
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 16
- 238000012795 verification Methods 0.000 claims description 15
- 238000002360 preparation method Methods 0.000 claims description 13
- 238000005516 engineering process Methods 0.000 claims description 9
- 230000009286 beneficial effect Effects 0.000 claims description 7
- 230000006872 improvement Effects 0.000 claims description 7
- 238000002844 melting Methods 0.000 claims description 7
- 230000008018 melting Effects 0.000 claims description 7
- 230000000977 initiatory effect Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000000110 selective laser sintering Methods 0.000 claims description 3
- 238000012407 engineering method Methods 0.000 claims 8
- 238000009826 distribution Methods 0.000 claims 1
- 238000012827 research and development Methods 0.000 abstract description 3
- 239000002994 raw material Substances 0.000 description 7
- 238000011160 research Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 229910001069 Ti alloy Inorganic materials 0.000 description 2
- 229910000883 Ti6Al4V Inorganic materials 0.000 description 2
- 238000000889 atomisation Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000000498 ball milling Methods 0.000 description 1
- 238000010923 batch production Methods 0.000 description 1
- 238000005266 casting Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 238000010411 cooking Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000006356 dehydrogenation reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005242 forging Methods 0.000 description 1
- 238000005984 hydrogenation reaction Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010309 melting process Methods 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000004482 other powder Substances 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000003746 surface roughness Effects 0.000 description 1
- 239000013077 target material Substances 0.000 description 1
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/10—Additive manufacturing, e.g. 3D printing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P10/00—Technologies related to metal processing
- Y02P10/25—Process efficiency
Abstract
The invention provides a reverse design method of titanium-based workpiece additive manufacturing process parameters based on near-spherical powder, which aims at constructing powder characteristic parameters-process parameters-material performance gradient lifting regression tree (GBDT) models based on data sets for near-spherical powder with different powder characteristics, constructing an intelligent model capable of outputting the process parameters and the powder characteristic parameters by utilizing a Q-Learning reinforcement Learning algorithm, fusing and applying the GBDT model and the intelligent model, and further realizing rapid and accurate design of an optimal process path according to given performance requirements, and solving the problems of long material research and development period and high cost in the prior art.
Description
Technical Field
The invention relates to the technical field of metal material preparation, in particular to a reverse design method of additive manufacturing process parameters of a titanium-based workpiece based on near spherical powder.
Background
Compared with the prior art of forging, casting and the like for forming the workpiece, the laser powder bed additive manufacturing titanium and titanium alloy workpiece has more excellent mechanical property and design freedom, and has wide application prospect in the strategic emerging industry fields of aerospace, national defense and military industry, biomedical treatment, marine equipment and the like, and is highly focused by various national governments and scientific industries. At present, raw material powder used in the additive manufacturing technology is mainly spherical powder prepared by an air atomization or plasma fuse atomization method, however, the price of the raw material powder is high because the yield of fine powder is generally low, and the price of the raw material powder is generally higher than 100 ten thousand yuan/ton by taking Ti-6Al-4V powder used in the selective laser melting technology as an example, so that the raw material powder becomes a big 'bottleneck' which restricts the technical progress and the wide application in the field. Therefore, a great deal of research work related to low cost of titanium powder for additive manufacturing is developed at home and abroad in recent years, and early research shows that low-cost and irregular Hydrogenation Dehydrogenation (HDH) titanium powder is taken as a raw material, and a high-temperature ball milling technology is utilized to prepare low-cost nearly spherical titanium powder meeting the requirements of powder for additive manufacturing.
Compared with spherical powder, the spherical powder has the advantages that the spherical property, the surface roughness, the fluidity, the apparent density and other powder characteristics of the spherical powder are changed, so that the thermal conductivity, the laser absorptivity and the like of the spherical powder are different from those of the spherical powder, and further the density, the surface quality, the mechanical property and the like of a formed part are obviously changed under the same process parameters, and the existing mature process is basically aimed at the spherical powder raw material and lacks the process parameters of the spherical powder suitability, so that the research of the process suitability of the spherical powder is critical to the research and development of low-cost additive manufacturing of titanium-based parts. At present, the traditional dish frying method is still mainly relied on for carrying out process parameter design optimization, namely, a worker designs the process parameters empirically according to the target performance requirement to finish sample preparation, then repeatedly adjusts the process parameters until the target performance is realized by analyzing the microstructure and the performance of a finished piece, and finally, the process parameters suitable for batch production are required to be searched to realize the production target with high efficiency and low cost.
Therefore, the conventional method has the problems of long period and high cost, and because the additive manufacturing is a complex nonlinear process under the joint continuous action of multiple process parameters, the conventional method can only carry out parameter design in a relatively narrow range, and is difficult to flexibly and efficiently explore multiple parameter spaces, so that the rapid response preparation of the titanium workpiece facing the target performance requirement is further hindered.
Disclosure of Invention
In order to overcome the defects in the prior art, the main purpose of the invention is to provide a reverse design method of titanium-based workpiece additive manufacturing process parameters based on near-spherical powder, wherein the design method is used for constructing a powder characteristic parameter-process parameter-material performance GBDT model based on a data set aiming at near-spherical powder with different powder characteristics, constructing an agent model capable of outputting the process parameters and the powder characteristic parameters by utilizing a Q-Learning reinforcement Learning algorithm, and fusing and applying the GBDT model and the agent model, thereby realizing rapid and accurate design of the optimal process path and the powder characteristic parameters according to given performance requirements.
In order to achieve the above object, the present invention provides a reverse design method of additive manufacturing process parameters of a titanium-based product based on near spherical powder.
The reverse design method of the additive manufacturing process parameters of the titanium-based workpiece based on the near-spherical powder comprises the following steps:
s1, establishing a data set;
s2, establishing a GBDT optimal model of powder characteristic parameter-process parameter-material performance by using the data set;
s3, constructing an intelligent body model by utilizing a Q-learning reinforcement learning algorithm, inputting preset target performance data into the intelligent body model, and carrying out fusion application on the intelligent body model and the GBDT optimal model to obtain optimal technological parameters and powder characteristic parameters.
Further, the preset target performance data is one or a combination of more of compactness, hardness, tensile strength and elongation;
the powder characteristic parameters include, but are not limited to, sphericity, bulk density.
Further, in step S3, preset target performance data is input into the agent model, and the agent model and the GBDT best model are fused and applied, so as to obtain best process parameters and powder characteristic parameters, which specifically include:
s3-1, inputting preset target performance data into the intelligent body model, and enabling the intelligent body to obtain initial prediction process parameters and powder characteristic parameters corresponding to the input preset target performance data through interaction with the environment;
s3-2, inputting the initial prediction technological parameters and the powder characteristic parameters into the GBDT optimal model, outputting initial material performance prediction data corresponding to the initial prediction technological parameters and the powder characteristic parameters, and calculating the distance between the material performance prediction data and preset target performance data;
s3-3, the intelligent agent performs initial action selection at the current t moment according to the correlation of the process parameter, the powder characteristic parameter and the target performance; wherein the action is an adjustment action to a process parameter and a powder characteristic;
s3-4, the intelligent agent interacts with the environment according to the initial action at the moment t to obtain a new state and rewards R;
s3-5, the intelligent agent obtains a new current action guiding strategy at the moment t through a new state and rewarding R, if R >0 shows that the current action of the intelligent agent is beneficial to the result, the intelligent agent can execute the new action, if R is less than or equal to 0, the current action of the intelligent agent is not beneficial to the result, the intelligent agent can be returned to the original state, and S3-4-S3-5 are repeated;
s3-6, continuously repeating the steps S3-3 to S3-5 until the iterative training times P reach a preset time threshold value, and taking the corresponding technological parameters and powder characteristics from the current moment to the final moment when the cumulative reward is maximum as the optimal technological parameters and powder characteristic parameters; wherein, p=100-1000, T times are set in each iteration, each time T is updated once by the agent, until t=t, the iteration is ended.
Further, the distance between the material property prediction data and the preset target property data is specifically expressed as:
wherein ,is the distance between the predicted value and the target value; />Is a target value; y is a predicted value at the moment T; yt is a predicted value corresponding to the process parameter at time t predicted by the GBDT best model.
Further, the reward R is a difference between the reciprocal of the distance at the next time (t+1) and the reciprocal of the distance at the current time (t), and is specifically expressed as:
where γ is the discount factor for the future time step prize value, γ ε [0,1].
Further, in step S1, establishing the data set includes the steps of:
s1-1, acquiring powder characteristic parameters, process parameters and material performance data, and establishing a basic data set;
s1-2, classifying the data in the basic data set, and selecting the data of the corresponding type as an initial data set according to the material type and the preparation technology;
s1-3, carrying out normalization processing on the initial data set to obtain the data set.
Further, the material performance data is one or a combination of more of compactness, hardness, tensile strength and elongation;
preferably, the material types include, but are not limited to, pure titanium, ti-6Al-4V, TA;
preferably, the preparation techniques include, but are not limited to, selective laser melting techniques, selective laser sintering techniques.
Further, the data set is divided into a training set, a verification set and a test set; establishing a powder characteristic parameter-process parameter-material property GBDT optimal model using the data set includes the steps of:
s2-1, performing model training by using the training set and the verification set to generate a powder characteristic parameter-process parameter-material performance GBDT initial model;
s2-2, performing super-parameter optimization design on the GBDT initial model by adopting Random Search in combination with a K-Fold Cross Validation algorithm to obtain a GBDT improved model; wherein k=5, 10, the super parameters include the number of trees (n_counter), learning rate (learning_rate), maximum tree depth (max_depth), and sub-sampling scale (subsamples);
s2-3, performing prediction effect evaluation and optimization design on the GBDT improvement model by using the test set to obtain the GBDT optimal model.
Further, in step S2-3, the decision coefficient R is used 2 And judging whether the GBDT improvement model reaches model precision or not by using the average absolute percentage error MAPE so as to complete optimization design:
when determining the coefficient R 2 The average absolute percentage error MAPE is not less than a preset threshold value and not more than the preset threshold value, the model precision is achieved, and the GBDT optimal model is obtained;
when determining the coefficient R 2 And (3) the value of the error MAPE is less than a preset threshold value or the average absolute percentage error MAPE is greater than the preset threshold value, resetting the RS algorithm to optimize the super-parameter interval and the iteration number (n_iter) and continuing to optimize the model.
Further, the decision coefficient R 2 The preset thresholds for the mean absolute percentage error MAPE are 0.95 and 5% respectively;
preferably, the number (n_timer), learning rate (learning_rate), maximum tree depth (max_depth) and sub-sampling ratio (subsampled) optimization intervals of the tree in the super parameter are respectively 1-1000, 0.01-0.5, 1-10 and 0-1, and the iteration number (n_ter) is 200-1500.
Further, the data volume in the training set is 40-90% of the data volume in the data lump, the data volume in the verification set is 5-30% of the data volume in the data lump, the data volume in the test set is 5-30% of the data volume in the data lump, and the sum of the percentages of the training set, the verification set and the test set is always 100%.
The invention has the following advantages:
1) Compared with the traditional 'cooking' method, the machine learning method adopted by the invention fully utilizes the existing mass production and research data of the additive manufacturing products, and establishes an implicit relation model among powder characteristic parameters, additive manufacturing process parameters and workpiece performances without deep knowledge of internal mechanisms of the additive manufacturing technology, thereby greatly reducing the research and development period and cost of the near-spherical powder additive manufacturing suitability process.
2) The GBDT model and reinforcement learning process parameter reverse design method can be used for intelligently optimizing additive manufacturing process parameters and powder characteristic parameters for a given material performance to find an optimal path for achieving target performance, so that quick response preparation of the titanium-based workpiece for additive manufacturing facing performance requirements is achieved; compared with the traditional parameter optimizing model, the method has stronger migration capability and better parameter recommending effect, and is suitable for most preparation technologies.
3) The distance function provided by the invention is to calculate the accumulated distance from the initial moment to a certain moment, fully considers the errors of the current moment and the previous moment, and is helpful for searching the optimal process path; the proposed reward function fully considers errors before and after moments to obtain denser learning reward information, and is beneficial to accurate optimization of the reinforcement learning process.
4) Compared with spherical titanium powder, the near-spherical powder preparation equipment and process adopted by the invention are simple, the cost is low, the yield is high and is close to 100%, the fluidity is better to meet the powder requirement for material increase, and the low cost requirement of material increase manufacturing of titanium-based parts can be met.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
FIG. 1 is a flow chart of a reverse design method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a reverse design method in accordance with an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The invention develops the process suitability research of the near-spherical titanium powder laser powder bed melting technology to establish an efficient reverse design method of the additive manufacturing process parameters of the low-cost titanium-based workpiece facing the target performance, thereby realizing the efficient low-cost preparation of the additive manufactured titanium-based workpiece with the given target performance.
The reverse design method of the additive manufacturing process parameters facing the target performance takes any given material performance as a target, outputs the predicted process parameters and the powder characteristic parameters, and the predicted process parameters and the powder characteristic can be used as the optimal process parameters and the optimal powder characteristic parameters. Specifically, the intelligent agent model and the GBDT optimal model are subjected to fusion application, the system automatically inputs the predicted technological parameters into the GBDT optimal model before outputting the predicted technological parameters, the predicted material performance values are output, the distance between the predicted material performance values and the target material performance values is calculated, the technological parameters and the adjustment strategies of the powder characteristic parameters are obtained, the iteration is repeatedly performed until the maximum iteration number P is reached, the technological parameters and the powder characteristic parameters corresponding to the maximum accumulated rewards are output as the optimal technological parameters and the powder characteristic parameters, and intelligent accurate regulation and control of the performance of the workpiece are realized.
The reverse design method of the titanium-based workpiece additive manufacturing process parameters based on the near-spherical powder comprises the following steps:
s1, establishing a data set; wherein the data set is divided into a training set, a verification set and a test set.
The training set is used for GBDT model training; the validation set is used to control the training process to prevent the GBDT model from overfitting; the test set is used to test model accuracy. The training set and the verification set are used in the model training process, and the test set is used for model optimization improvement after model training is finished.
In the embodiment of the invention, the data volume in the training set accounts for 40-90% of the data volume in the data lump, the data volume in the verification set accounts for 5-30% of the data volume in the data lump, the data volume in the test set accounts for 5-30% of the data volume in the data lump, and the sum of the percentages of the training set, the verification set and the test set is always 100%.
The method for establishing the data set comprises the following steps of:
s1-1, acquiring powder characteristic parameters, process parameters and material performance data from a material database, high-quality documents and experimental data published by public publications, and establishing a basic data set;
s1-2, classifying data in the basic data set, and selecting data of a corresponding type as an initial data set according to the material type and the preparation technology;
s1-3, carrying out normalization processing on the initial data set to obtain the data set.
In embodiments of the invention, the material property data is a combination of one or more of density, hardness, tensile strength, elongation.
In embodiments of the present invention, the powder characteristic parameters include, but are not limited to, sphericity, bulk density.
In embodiments of the present invention, the material types include, but are not limited to, pure titanium, ti-6Al-4V, TA.
In embodiments of the present invention, the fabrication techniques include, but are not limited to, selective laser melting techniques, selective laser sintering techniques.
S2, establishing a powder characteristic parameter-process parameter-material performance GBDT optimal model by utilizing the data set, wherein the method specifically comprises the following steps of:
s2-1, performing model training by using a training set and a verification set to generate a GBDT initial model of powder characteristic parameter-process parameter-material performance.
In the embodiment of the invention, the verification set control training process adopts an early stop strategy, namely, after the error of the verification set is continuously reduced for a plurality of times, the training is stopped, so that the overfitting can be effectively prevented and the training time can be saved.
In an embodiment of the present invention, the training set contains a plurality of sets of training data, and each set of training data includes powder characteristic parameters, process parameters, and material property data corresponding thereto. Taking process parameter data and powder characteristic data in a training set as input, taking material performance data as output data, setting values of super parameters n_ estimator, learning _rate, max_depth and subsamples, and establishing a powder characteristic parameter-process parameter-material performance GBDT initial model.
S2-2, performing super-parameter optimization design on the GBDT initial model by adopting Random Search in combination with a K-Fold Cross Validation algorithm to obtain a GBDT improved model; where k=5, 10, the super parameters include the number of trees (n_counter), learning rate (learning_rate), maximum tree depth (max_depth), and sub-sampling scale (subsamples).
It should be noted that the K value may be 5 or 10, and of course, the K value may be designed according to actual needs, and is not specifically limited.
S2-3, performing prediction effect evaluation and optimization design on the GBDT improvement model by using a test set to obtain a GBDT optimal model; wherein the test set contains a plurality of sets of test data, and each set of test data includes a powder characteristic parameter, a process parameter (for better distinction, denoted as a second process parameter), and material property data (denoted as second material property data) corresponding thereto.
In the embodiment of the invention, the optimized design of the GBDT improvement model by using the test set comprises the following steps:
s2-3-1, inputting technological parameters (second technological parameters) in the test set into the GBDT improved model to obtain second predicted data of the material performance corresponding to the technological parameters;
s2-3-2, calculating a determination coefficient R by using the second predicted data of the material performance and the second material performance data 2 Average absolute percent error MAPE;
s2-3-3, using a determining coefficient R 2 And judging whether the GBDT improved model reaches model precision or not by using the average absolute percentage error MAPE so as to complete the optimization design:
when determining the coefficient R 2 The average absolute percentage error MAPE is not more than a preset threshold value and not more than the preset threshold value, the model precision is achieved, and the GBDT optimal model is obtained;
when determining the coefficient R 2 And (3) resetting the values of the RS algorithm optimization super parameter (n_ estimator, learning _rate, max_ depth, subsample) interval and the iteration number (n_iter) and continuing the step S2 to perform model optimization.
In embodiments of the invention, R 2 And the preset threshold for MAPE may be 0.95 and 5%, respectively.
In the present invention, the determination coefficient R is used 2 Determining whether the GBDT improved model reaches the model precision or not by averaging absolute percentage error MAPE, R 2 Values are in the range of 0-1, and closer to 1 indicates higher model accuracy, and smaller MAPE values indicate higher model accuracy.
S3, constructing an intelligent body model by utilizing a Q-learning reinforcement learning algorithm, inputting preset target performance data into the intelligent body model, and carrying out fusion application on the intelligent body model and a GBDT optimal model to obtain optimal technological parameters and powder characteristic parameters, wherein the method specifically comprises the following steps of:
s3-1, inputting preset target performance data into an intelligent body model, and obtaining initial prediction technological parameters and powder characteristic parameters corresponding to the input preset target performance data by the intelligent body through interaction with the environment;
s3-2, inputting the initial prediction technological parameters and the powder characteristic parameters into a GBDT optimal model, and outputting material performance prediction data corresponding to the initial prediction technological parameters and the powder characteristic parameters;
s3-3, the intelligent agent performs initial action selection at the current t moment according to the environment observation condition, namely according to the correlation between the process parameter, the powder characteristic parameter and the target performance; wherein the action is an adjustment action on the process parameter and the powder characteristic data;
s3-4, the intelligent agent interacts with the environment according to the initial action at the moment t to obtain a new state and rewards R;
s3-5, the intelligent agent obtains a new current action guiding strategy at the moment t through a new state and rewarding R, if R >0 shows that the current action of the intelligent agent is beneficial to the result, the intelligent agent can execute the new action, if R is less than or equal to 0, the current action of the intelligent agent is not beneficial to the result, the intelligent agent can be returned to the original state, and S3-4-S3-5 are repeated;
s3-6, continuously repeating the steps S3-3 to S3-5 until the iterative training times P reach a preset time threshold value, and taking the corresponding technological parameters and powder characteristics from the current moment to the final moment when the cumulative reward is maximum as the optimal technological parameters and powder characteristic parameters; wherein, p=100-1000, T times are set in each iteration, each time T is updated once, until t=t, the iteration is ended.
In the embodiment of the present invention, the distance between the material property prediction data and the preset target property data is specifically expressed as:
wherein ,is the distance between the predicted value and the target value; />Is a target value; y is a predicted value at the moment T; yt is a predicted value corresponding to the process parameter at time t predicted by the GBDT best model.
In the embodiment of the invention, the reward R is the difference between the reciprocal of the distance at the next moment (t+1) and the reciprocal of the distance at the current moment (t), and is specifically expressed as:
gamma is a discount factor for the prize value of a future time step, gamma e 0, 1.
The reverse design method of the process parameters for manufacturing titanium-based articles based on near spherical powders according to the present invention will be described in detail below by way of specific examples.
Example 1:
taking Ti-6Al-4V workpiece with density of 99.9% formed by selective laser melting as target, predicting/designing printing technological parameters and selecting powder raw materials with certain characteristics as an example for explanation, the specific steps are as follows:
1) Establishing an initial data set: and grabbing powder characteristic data, selective laser melting technical process parameters and Ti-6Al-4 workpiece performance data from published literature, experimental and production data to establish an initial data set.
2) Normalization: respectively carrying out normalization treatment (0-1) on the powder characteristic data, the process parameters and the workpiece performance data in the initial data set to obtain a powder characteristic and process parameter-Ti-6 Al-4 workpiece performance data set; the data set is divided into a training set, a verification set and a test set, wherein the data volume in the training set accounts for 90% of the data volume in the data set, and the data volume in the verification set and the test set accounts for 5% of the data volume in the data set.
3) Establishing a GBDT initial model of powder characteristics and selective laser melting process parameters-Ti-6 Al-4 workpiece performance: taking the process parameters and the powder characteristic parameters in the training set and the verification set as input data and the titanium alloy workpiece performance data as output data, and setting super parameters n_counter (200), learning_rate (0.02), max_depth (5) and subsamples (1) for model training;
then introducing Random Search and K-Fold Cross Validation algorithm to automatically optimize the super parameters, improving the model precision and generalization capability, and obtaining a GBDT improved model; wherein, K=10, the super parameters n_ estimator, learning _rate, max_depth and subsampled optimization interval in the random Search algorithm are respectively set to be 1-1000, 0.01-0.5, 1-10 and 0-1, and the iteration times n_iter are 400.
4) The accuracy of the GBDT improved model is evaluated by adopting a test set, specifically, a second technological parameter and second powder characteristic data of each group of test data are taken as input data to obtain density second prediction data, and then the density second prediction data and density data (corresponding to second material performance data) in each group of test data are utilized to calculate and obtain a determination coefficient R 2 Because the condition of not less than 0.94 is satisfied, the optimization intervals of the super parameters n_ estimator, learning _rate, max_depth and subsamples are respectively set to be 200-1000, 0.01-0.5, 3-10 and 0-1, the iteration times n_iter are 600, and model training and testing are carried out to obtain a decision coefficient R 2 =0.98 and mean absolute percentage error mape=2.1% (R is satisfied 2 And the conditions are more than or equal to 0.95 percent and MAPE is less than or equal to 5 percent), thus obtaining the GBDT optimal model.
5) And taking the density of 99.9% as target performance data, outputting an optimal path for realizing the target performance by the fusion application of an intelligent body model based on a Q-learning reinforcement learning algorithm and the GBDT optimal model, so as to realize intelligent accurate regulation and control of the performance of the workpiece, wherein the maximum iteration frequency P=800.
Examples 2 to 3 used the same preparation technique and materials as in example 1, except that the target properties were inputted, and the optimum process parameters and powder properties obtained in examples 1 to 3 were as shown in table 1.
Table 1 summary of the best process parameters corresponding to the target performance in examples 1-3
It should be noted that the term "comprising" in the description of the invention and in the claims, as well as any variants thereof, is intended to cover a non-exclusive inclusion, for example, comprising a series of elements not necessarily limited to those elements explicitly listed, but may include other elements not explicitly listed or inherent to elements.
The description of "first," "second," etc. in this disclosure is for descriptive purposes only and is not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature.
In addition, the technical solutions of the embodiments may be combined with each other, but it is necessary to base that the technical solutions can be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent and not within the scope of protection claimed in the present invention.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A reverse design method of additive manufacturing process parameters of a titanium-based workpiece based on near spherical powder, which is characterized by comprising the following steps:
s1, establishing a data set;
s2, establishing a GBDT optimal model of powder characteristic parameter-process parameter-material performance by using the data set;
s3, constructing an intelligent body model by utilizing a Q-learning reinforcement learning algorithm, inputting preset target performance data into the intelligent body model, and carrying out fusion application on the intelligent body model and the GBDT optimal model to obtain optimal technological parameters and powder characteristic parameters.
2. The reverse engineering method of claim 1, wherein the pre-set target performance data is a combination of one or more of density, hardness, tensile strength, elongation;
preferably, the powder characteristic parameters include, but are not limited to, sphericity, bulk density.
3. The reverse engineering method according to claim 1, wherein in step S3, the preset target performance data is input into the intelligent body model, and the intelligent body model and the GBDT optimal model are fused and applied, so as to obtain the optimal process parameters and the powder characteristic parameters specifically include:
s3-1, inputting preset target performance data into the intelligent body model, and enabling the intelligent body to obtain initial prediction process parameters and powder characteristic parameters corresponding to the input preset target performance data through interaction with the environment;
s3-2, inputting the initial prediction technological parameters and the powder characteristic parameters into the GBDT optimal model, outputting initial material performance prediction data corresponding to the initial prediction technological parameters and the powder characteristic parameters, and calculating the distance between the material performance prediction data and preset target performance data;
s3-3, the intelligent agent performs initial action selection at the current t moment according to the correlation of the process parameter, the powder characteristic parameter and the target performance; wherein the action is an adjustment action to a process parameter and a powder characteristic;
s3-4, the intelligent agent interacts with the environment according to the initial action at the moment t to obtain a new state and rewards R;
s3-5, the intelligent agent obtains a new current action guiding strategy at the moment t through a new state and rewarding R, if R >0 shows that the current action of the intelligent agent is beneficial to the result, the intelligent agent can execute the new action, if R is less than or equal to 0, the current action of the intelligent agent is not beneficial to the result, the intelligent agent can be returned to the original state, and S3-4-S3-5 are repeated;
s3-6, continuously repeating the steps S3-3 to S3-5 until the iterative training times P reach a preset time threshold value, and taking the corresponding technological parameters and powder characteristics from the current moment to the final moment when the cumulative reward is maximum as the optimal technological parameters and powder characteristic parameters; wherein, p=100-1000, T times are set in each iteration, each time T is updated once by the agent, until t=t, the iteration is ended.
4. The reverse engineering method of claim 3 wherein the distance between the predicted material property data and the predetermined target property data is expressed as:
wherein ,is the distance between the predicted value and the target value; />Is a target value; y is a predicted value at the moment T; yt is a predicted value corresponding to a process parameter at time t predicted by the GBDT optimal model;
preferably, the reward R is a difference between the reciprocal of the distance at the next time (t+1) and the reciprocal of the distance at the current time (t), and is specifically expressed as:
where γ is the discount factor for the future time step prize value, γ ε [0,1].
5. The reverse engineering method of claim 1, wherein in step S1, creating the data set comprises the steps of:
s1-1, acquiring powder characteristic parameters, process parameters and material performance data, and establishing a basic data set;
s1-2, classifying the data in the basic data set, and selecting the data of the corresponding type as an initial data set according to the material type and the preparation technology;
s1-3, carrying out normalization processing on the initial data set to obtain the data set.
6. The reverse engineering method of claim 5, wherein the material property data is a combination of one or more of density, hardness, tensile strength, elongation;
preferably, the material types include, but are not limited to, pure titanium, ti-6Al-4V, TA;
preferably, the preparation techniques include, but are not limited to, selective laser melting techniques, selective laser sintering techniques.
7. The reverse design method of claim 1, wherein the data set is divided into a training set, a validation set, and a test set; establishing a powder characteristic parameter-process parameter-material property GBDT optimal model using the data set includes the steps of:
s2-1, performing model training by using the training set and the verification set to generate a powder characteristic parameter-process parameter-material performance GBDT initial model;
s2-2, performing super-parameter optimization on the GBDT initial model by adopting a Random Search and K-Fold Cross Validation algorithm to obtain a GBDT improved model; wherein k=5, 10, the super parameters include the number of trees, learning rate, maximum tree depth, and sub-sampling ratio;
s2-3, performing prediction effect evaluation and optimization design on the GBDT improvement model by using the test set to obtain the GBDT optimal model.
8. The reverse engineering method according to claim 7, wherein in the step S2-3, the determination coefficient R is used 2 And judging whether the GBDT improvement model reaches a preset precision or not by using the average absolute percentage error MAPE so as to complete the optimization design:
when determining the coefficient R 2 The average absolute percentage error MAPE is not less than a preset threshold value and not more than the preset threshold value, the model precision is achieved, and the GBDT optimal model is obtained;
when determining the coefficient R 2 < preset threshold orAnd (3) the average absolute percentage error MAPE is larger than a preset threshold, resetting the RS algorithm to optimize the super-parameter interval and the iteration times, and continuing to optimize the model.
9. The reverse engineering method of claim 8, wherein the determining factor R 2 The preset threshold distribution for the mean absolute percentage error MAPE is 0.95 and 5%;
preferably, the number of the trees in the super parameter, the learning rate, the maximum tree depth and the sub-sampling proportion optimization interval are respectively 1-1000, 0.01-0.5, 1-10 and 0-1, and the iteration times are 200-1500.
10. The reverse engineering method of claim 7, wherein the amount of data in the training set is 40-90% of the total amount of data in the data set, the amount of data in the validation set is 5-30% of the total amount of data in the data set, the amount of data in the test set is 5-30% of the total amount of data in the data set, and the sum of the percentages of the training set, the validation set, and the test set is always 100%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310525021.4A CN116611322A (en) | 2023-05-10 | 2023-05-10 | Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310525021.4A CN116611322A (en) | 2023-05-10 | 2023-05-10 | Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116611322A true CN116611322A (en) | 2023-08-18 |
Family
ID=87681065
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310525021.4A Pending CN116611322A (en) | 2023-05-10 | 2023-05-10 | Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116611322A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117352109A (en) * | 2023-12-04 | 2024-01-05 | 宝鸡富士特钛业（集团）有限公司 | Virtual modeling method, device, equipment and medium applied to titanium alloy forging |
-
2023
- 2023-05-10 CN CN202310525021.4A patent/CN116611322A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117352109A (en) * | 2023-12-04 | 2024-01-05 | 宝鸡富士特钛业（集团）有限公司 | Virtual modeling method, device, equipment and medium applied to titanium alloy forging |
| CN117352109B (en) * | 2023-12-04 | 2024-03-08 | 宝鸡富士特钛业（集团）有限公司 | Virtual modeling method, device, equipment and medium applied to titanium alloy forging |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Singh et al. | Optimization of bead geometry of submerged arc weld using fuzzy based desirability function approach | |
| CN116611322A (en) | Reverse design method of additive manufacturing process parameters of titanium-based part based on near-spherical powder | |
| Knust et al. | Preform optimization for hot forging processes using genetic algorithms | |
| CN105740467B (en) | A kind of method for digging of C-Mn steel industry big data | |
| Ming et al. | Soft computing models and intelligent optimization system in electro-discharge machining of SiC/Al composites | |
| EP3916651A1 (en) | Design assistance method for metal material, prediction model generation method, metal material manufacturing method, and design assistance device | |
| Benn et al. | Additively manufactured INCONEL® alloy 718 | |
| CN110669963A (en) | Aluminum alloy casting batching system | |
| CN113128124B (en) | Multi-grade C-Mn steel mechanical property prediction method based on improved neural network | |
| CN116629107A (en) | Reverse design method of titanium-based workpiece additive manufacturing process parameters based on bimodal particle size distribution powder | |
| Shi et al. | Comparison of flow behaviors of near beta Ti-55511 alloy during hot compression based on SCA and BPANN models | |
| CN116542153A (en) | Reverse design method of metal material processing technological parameters based on reinforcement learning | |
| Fang et al. | Optimization of forging process parameters and prediction model of residual stress of Ti-6Al-4V alloy | |
| Joseph et al. | Optimization of pulsed current GTAW process parameters for sintered hot forged AISI 4135 P/M steel welds by simulated annealing and genetic algorithm | |
| Zhang et al. | Directed energy deposition combining high-throughput technology and machine learning to investigate the composition-microstructure-mechanical property relationships in titanium alloys | |
| Bhadrakali et al. | A hybrid approach consisting of multi-objective and multivariate analyses for WAAM specimens | |
| Quan et al. | Multi-variable and bi-objective optimization of electric upsetting process for grain refinement and its uniform distribution | |
| Dong et al. | Just-in-time learning-based soft sensor for mechanical properties of strip steel via multi-block weighted semisupervised models | |
| CN111861041B (en) | Method for predicting dynamic recrystallization type rheological stress of Nb microalloyed steel | |
| Yamada et al. | Prediction on microstructure and mechanical properties of hot forged Ni-based super alloy by optimization using genetic algorithms | |
| CN115456264A (en) | Method for predicting end point carbon content and end point temperature of small and medium-sized converters | |
| CN114117684A (en) | Selective laser melting forming 316L stainless steel abrasion prediction method based on machine learning | |
| Gupta et al. | Multi-objective optimization of friction stir welding of aluminium alloy using grey relation analysis with entropy measurement method | |
| Zhu et al. | Multiobjective optimization of precision forging process parameters based on response surface method | |
| CN110814342A (en) | Preparation method of computer-formalized 3D printing metal material |
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 |