CN114564880B - Method for constructing digital twin module in additive manufacturing process - Google Patents
Method for constructing digital twin module in additive manufacturing process Download PDFInfo
- Publication number
- CN114564880B CN114564880B CN202210094566.XA CN202210094566A CN114564880B CN 114564880 B CN114564880 B CN 114564880B CN 202210094566 A CN202210094566 A CN 202210094566A CN 114564880 B CN114564880 B CN 114564880B
- Authority
- CN
- China
- Prior art keywords
- prediction
- model
- machine learning
- molten pool
- additive manufacturing
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 73
- 239000000654 additive Substances 0.000 title claims abstract description 55
- 230000000996 additive effect Effects 0.000 title claims abstract description 55
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 55
- 238000010801 machine learning Methods 0.000 claims abstract description 66
- 230000007246 mechanism Effects 0.000 claims abstract description 49
- 230000008569 process Effects 0.000 claims abstract description 43
- 238000002474 experimental method Methods 0.000 claims abstract description 33
- 230000004927 fusion Effects 0.000 claims abstract description 9
- 238000007639 printing Methods 0.000 claims abstract description 9
- 230000008878 coupling Effects 0.000 claims abstract description 6
- 238000010168 coupling process Methods 0.000 claims abstract description 6
- 238000005859 coupling reaction Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 32
- 238000012360 testing method Methods 0.000 claims description 15
- 208000003098 Ganglion Cysts Diseases 0.000 claims description 13
- 208000005400 Synovial Cyst Diseases 0.000 claims description 13
- 238000012795 verification Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000014509 gene expression Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000574 ganglionic effect Effects 0.000 description 3
- 238000002844 melting Methods 0.000 description 3
- 230000008018 melting Effects 0.000 description 3
- 239000000843 powder Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 229910001338 liquidmetal Inorganic materials 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000007711 solidification Methods 0.000 description 1
- 230000008023 solidification Effects 0.000 description 1
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
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B22—CASTING; POWDER METALLURGY
- B22F—WORKING METALLIC POWDER; MANUFACTURE OF ARTICLES FROM METALLIC POWDER; MAKING METALLIC POWDER; APPARATUS OR DEVICES SPECIALLY ADAPTED FOR METALLIC POWDER
- B22F10/00—Additive manufacturing of workpieces or articles from metallic powder
- B22F10/20—Direct sintering or melting
- B22F10/28—Powder bed fusion, e.g. selective laser melting [SLM] or electron beam melting [EBM]
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B22—CASTING; POWDER METALLURGY
- B22F—WORKING METALLIC POWDER; MANUFACTURE OF ARTICLES FROM METALLIC POWDER; MAKING METALLIC POWDER; APPARATUS OR DEVICES SPECIALLY ADAPTED FOR METALLIC POWDER
- B22F10/00—Additive manufacturing of workpieces or articles from metallic powder
- B22F10/30—Process control
- B22F10/36—Process control of energy beam parameters
- B22F10/366—Scanning parameters, e.g. hatch distance or scanning strategy
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B22—CASTING; POWDER METALLURGY
- B22F—WORKING METALLIC POWDER; MANUFACTURE OF ARTICLES FROM METALLIC POWDER; MAKING METALLIC POWDER; APPARATUS OR DEVICES SPECIALLY ADAPTED FOR METALLIC POWDER
- B22F10/00—Additive manufacturing of workpieces or articles from metallic powder
- B22F10/80—Data acquisition or data processing
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B33—ADDITIVE MANUFACTURING TECHNOLOGY
- B33Y—ADDITIVE MANUFACTURING, i.e. MANUFACTURING OF THREE-DIMENSIONAL [3-D] OBJECTS BY ADDITIVE DEPOSITION, ADDITIVE AGGLOMERATION OR ADDITIVE LAYERING, e.g. BY 3-D PRINTING, STEREOLITHOGRAPHY OR SELECTIVE LASER SINTERING
- B33Y50/00—Data acquisition or data processing for additive manufacturing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Manufacturing & Machinery (AREA)
- Materials Engineering (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Plasma & Fusion (AREA)
- Automation & Control Theory (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
Abstract
The invention discloses a method for constructing a digital twin module in an additive manufacturing process, which comprises the following steps of 1: performing an additive manufacturing experiment, and collecting the size of a molten pool under corresponding process parameters as experimental data to form a physical entity in the additive manufacturing process; and 2, step: establishing a multi-physical-field coupling mechanism model in the additive manufacturing process, verifying the confidence coefficient of the mechanism model, virtually printing a digital platform based on the high-confidence-coefficient mechanism model, and obtaining a printing result and data to form a digital twin body in the additive manufacturing physical process; and 3, step 3: forward and backward prediction is carried out on process parameters and the size of the deep pool based on machine learning, and virtual-real fusion and data intercommunication of the additive manufacturing physical entity and the digital twin are realized; and 4, step 4: the obtained optimal machine learning model is used for new data prediction of an experiment and mechanism model, and additive manufacturing experiments, high-confidence mechanism model and machine learning fusion and efficient and accurate prediction of corresponding process parameters and molten pool sizes are achieved.
Description
Technical Field
The invention belongs to the technical field of additive manufacturing, and particularly relates to a method for constructing a digital twin module in an additive manufacturing process.
Background
Laser powder bed melting is a typical metal additive manufacturing method, and relates to complex dynamic processes of rapid heating and melting of metal powder to form a molten pool, subsequent solidification and cooling. The characteristics of the molten pool such as shape and size are closely related to the additive manufacturing process parameters, and the additive manufacturing forming defects and the part performance are obviously influenced. Due to the multi-physical field coupling and multi-scale characteristics of additive manufacturing techniques, the formation and evolution of molten pools have a high degree of complexity, and prediction and control of the molten pools under different additive manufacturing conditions face many difficulties. Full-time-space data of a molten pool in the printing process are difficult to obtain only by adopting an experimental method, and heat transfer, liquid metal flow and mass transfer behaviors in the printing process need to be understood by combining a multi-physical-field coupling high-confidence-degree mechanism model.
The digital twin technology integrates multidisciplinary, multi-physical quantity and multi-scale methods, and integrates modules such as physical entities, digital twin bodies and interactive data. The analysis, prediction, training and the like can be carried out based on the digital twins, and the simulation results of the digital twins are transmitted to the physical object, so that the optimization and decision of the physical object are carried out. At present, the development of the additive manufacturing digital twinning technology is still in the starting stage, and particularly, the additive manufacturing process digital twinning technology which integrates an additive manufacturing experiment, a high-confidence mechanism model, a high-accuracy forward and reverse prediction machine learning is still blank.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a method for constructing a digital twin module in the additive manufacturing process, which provides an implementation means for establishing the digital twin module in the additive manufacturing process, can overcome the problems of difficult research and insufficient machine learning data source caused by the complex molten pool evolution process, and realizes the closed-loop research from the process parameters to the molten pool size and then from the molten pool size to the process parameters.
The technical solution for realizing the purpose of the invention is as follows:
a method for constructing a digital twin module in an additive manufacturing process comprises the following steps:
step 1: performing an additive manufacturing experiment, collecting a molten pool size under corresponding process parameters as experiment data, and forming a physical entity of the additive manufacturing process, wherein the process parameters comprise laser power and scanning speed, and the molten pool size comprises molten pool width and depth;
step 2: establishing a multi-physical-field coupling mechanism model in the additive manufacturing process, verifying the confidence coefficient of the mechanism model, further performing digital platform virtual printing based on the high-confidence-coefficient mechanism model, and obtaining corresponding printing results and data to form a digital twin body in the additive manufacturing physical process;
and step 3: the method comprises the steps of constructing an additive manufacturing machine learning model after analyzing and processing data obtained by an experiment and mechanism model, carrying out model training and verification by adopting the data obtained by the experiment and mechanism model to obtain an optimal machine learning model, carrying out forward and reverse prediction on process parameters and a deep pool size based on machine learning, and realizing virtual-real fusion and data intercommunication of an additive manufacturing physical entity and a digital twin;
and 4, step 4: the obtained optimal machine learning model is used for new data prediction of an experiment and mechanism model, and additive manufacturing experiments, high-confidence mechanism model and machine learning fusion and efficient and accurate prediction of corresponding process parameters and molten pool sizes are achieved.
Further, the step 1 specifically comprises: the laser power and the scanning speed are changed to obtain different process parameter combinations, square profile scanning is selected as an experimental scanning strategy for obtaining a plurality of single channels under the same parameter, and the width and the depth of a molten pool are obtained by measuring and averaging deposited channels on different sides of the printed square profile.
Further, the laser power value range in the step 1 is 220W-350W, and the scanning speed is 300mm/s-1500 mm/s.
Further, the step 2 specifically includes: the specific verification method for verifying the confidence coefficient of the mechanism model is to take a plurality of groups of laser power and scanning speed combinations used in the experiment as the input of the mechanism model, take the width and the depth of a molten pool in a steady state as corresponding output, compare the output with the result of the width and the depth of the molten pool obtained in the experiment, ensure that the deviation is within a required error range, obtain more process parameter combinations by taking a plurality of intervals in the preset laser power and scanning speed after ensuring the effectiveness of the mechanism model, and further input the parameter combinations into the mechanism model to obtain more molten pool size data.
Further, the laser power and the scanning speed predetermined in step 2 are: 100W-350W of laser power and 300mm/s-1500mm/s of scanning speed.
Further, the step 3 specifically includes: dividing the analyzed and processed experimental data and the data obtained by the mechanism model into a training set, a verification set and a test set, and carrying out normalization processing on the data in the training set in order to avoid the influence on prediction caused by the inconsistency of the laser power and the scanning speed dimension and the variation range, wherein the specific operational expression is as shown in a formula (1):
wherein:
denotes normalized data, x i Representing the original data, E [ x ] i ]Represents the average of the data in the training set,
representing the standard deviation of the data in the training set;
the machine learning model is a multilayer perceptron model, the multilayer perceptron model is used for forward prediction from a process parameter to a molten pool size and reverse prediction from the molten pool size to the process parameter, in the forward prediction and the reverse prediction, the machine learning model is divided into 6 types, wherein the forward prediction is 2 types, the reverse prediction is 4 types, and the 2 types of forward prediction models are as follows: (1) inputting: laser power and scanning speed, output: the width of the molten pool; (2) inputting: laser power and scanning speed, output: the 4 types of models of the molten pool depth and the reverse prediction are as follows: scheme 1: (1) inputting: molten pool width and depth, output: laser power; (2) inputting: molten pool width and depth, output: a scanning speed; scheme 2: (1) inputting: molten pool width, depth and scanning speed, output: laser power; (2) inputting: molten pool width, depth and laser power, output: the speed of the scan is such that,
then constructing a multilayer perceptron model, wherein the parameters of the multilayer perceptron model comprise an activation function, an optimization algorithm, a training round number, a hidden layer number and a hidden layer ganglion point number, wherein the activation function selects a rectifying linear unit function (ReLU), the optimization algorithm selects a root mean square back propagation method (RMSprop), a forward prediction training round number selection 200 and a reverse prediction training round number selection 300,
the number of hidden layers and the number of ganglion points are obtained by an increasing method, namely, the optimal machine learning model is found by gradually increasing the number of hidden layers and the number of ganglion points, and the process of obtaining the optimal machine learning model is as follows: step A, firstly selecting the number of hidden layers and the number of ganglionic points of a multilayer perceptron model, training a training set by adopting a K-fold cross validation method, step B, then importing input data of a validation set into the multilayer perceptron model in the step A to obtain corresponding prediction output, obtaining prediction error and accuracy according to actual output of the validation set under the same input and prediction output comparison of the multilayer perceptron, finally gradually increasing the number of ganglionic points and the number of hidden layers to obtain different multilayer perceptron models, repeating the step A and the step B to obtain prediction error and accuracy of different multilayer perceptron models, comparing the prediction error and accuracy of different multilayer perceptron models, finding the multilayer perceptron model with the optimal prediction performance, also called as an optimal machine learning model, wherein the prediction result of the multilayer perceptron model is measured by Absolute Percentage Error (APE) and Accuracy (AR), and the expressions are respectively formula (2) and formula (3):
AR＝(1-APE) (3)
wherein,
as a predictor of a machine learning model, y i Is the actual value.
Further, after the optimal machine learning model is obtained, the generalization ability of the optimal machine learning model needs to be tested, the mode of testing the generalization ability is to input test set data into 4 optimal machine learning models obtained in the step S3 and check the result of the optimal machine learning model, and the new data is the test set, wherein the 4 optimal machine learning models correspond to the optimal machine learning model of the molten pool width and the molten pool depth in forward prediction and the optimal machine learning model of the laser power and the scanning speed of the scheme 2 in reverse prediction.
Compared with the prior art, the invention has the remarkable advantages that:
the invention provides a new method for the digital twin construction in the additive manufacturing process by combined application of a printing experiment, a mechanism model and machine learning, provides a new technology for forward and backward prediction of process parameters and molten pool characteristics and multi-target bidirectional prediction of additive manufacturing, and provides a new idea for data acquisition, processing and prediction in additive manufacturing by machine learning.
Drawings
FIG. 1 is a digital twinning frame diagram of the present invention.
FIG. 2 is a schematic diagram of a multi-layer perceptron optimization.
Fig. 3 is a schematic diagram of machine learning for forward prediction and reverse prediction, wherein (a) forward prediction, (b) reverse prediction scheme 1, and (c) reverse prediction scheme 2.
FIG. 4 is a graph of the optimization results of the forward prediction example, in which (a) the molten pool width prediction result and (b) the molten pool depth prediction result are obtained.
Fig. 5 is a graph of comparison results of two schemes for inverse prediction, in which (a) laser power prediction results and (b) scanning speed prediction results.
FIG. 6 is a graph showing the test results of the optimum width prediction model in forward prediction, in which (a) the target molten pool width, (b) the predicted molten pool width, and (c) the prediction accuracy of different molten pool widths.
FIG. 7 is a graph of the test results of the optimal depth prediction model in forward prediction, wherein (a) the target bath depth, (b) the predicted bath depth, and (c) the prediction accuracy of different bath depths.
Fig. 8 is a graph of the test results of an inverse predictive optimum model, in which (a) laser power and (b) scan speed.
FIG. 9 is a flow chart of a digital twinning module construction method of the additive manufacturing process of the present invention.
Detailed Description
Referring to fig. 9, a method for constructing a digital twin module in an additive manufacturing process is mainly described in fig. 1, and comprises the following steps:
step 1: performing an additive manufacturing experiment, collecting a molten pool size under corresponding process parameters as experiment data, and forming a physical entity of the additive manufacturing process, wherein the process parameters comprise laser power and scanning speed, the molten pool size comprises a molten pool width and a molten pool depth, and the method specifically comprises the following steps:
considering that the laser power and the scanning speed jointly determine the input energy density and have a great relationship with the shape and the size of a molten pool, the laser power and the scanning speed are mainly changed to obtain different process parameter combinations, and other process parameters such as the scanning strategy, the laser radius and the like are kept unchanged. Wherein the laser power value range is between 220W and 350W, and the scanning speed is between 300mm/s and 1500 mm/s. In order to obtain a plurality of single channels under the same parameter, square profile scanning is selected as a scanning strategy of an experiment. The molten pool width and depth may be obtained by measuring and averaging the deposited roads on different sides of the printed square profile.
Step 2: establishing a multi-physical-field coupling mechanism model in the additive manufacturing process, verifying the confidence coefficient of the mechanism model, further performing digital platform virtual printing based on the high-confidence-coefficient mechanism model, and obtaining corresponding printing results and data to form a digital twin body in the additive manufacturing physical process; (ii) a
To ensure that the mechanism model is accurate and effective, it needs to be verified. The specific verification method is that 16 groups of laser power and scanning speed combinations used in the experiment are used as the input of a digital twin body (namely a mechanism model), the width and the depth of the molten pool in a steady state are used as corresponding output, and the output is compared with the result of the width and the depth of the molten pool obtained in the experiment, so that the deviation is ensured to be within a required error range. After the effectiveness of the mechanism model is ensured, more process parameter combinations are obtained by taking small intervals in the laser power of 100W-350W and the scanning speed of 300mm/s-1500mm/s, and then the parameter combinations are input into the mechanism model to obtain more weld pool size data. In total, 99 sets of bath widths and depths were obtained.
And step 3: the method comprises the steps of analyzing and processing data obtained by an experiment and mechanism model, constructing an additive manufacturing machine learning model, carrying out model training and verification by adopting the data obtained by the experiment and mechanism model to obtain an optimal machine learning model, carrying out forward and reverse prediction on process parameters and a deep pool size based on machine learning, and realizing virtual-real fusion and data intercommunication of an additive manufacturing physical entity and a digital twin;
before the machine learning model is constructed, the experimental data obtained in the first two steps and the data obtained by the mechanism model need to be analyzed and processed so as to achieve the fusion of virtual and real data. The processed data remained 103 sets, of which experimental data 16 set and mechanism model data 87 set. The 103 groups of data are divided into a training set, a verification set and a test set, and the proportion is 70:20:13.. All data are used for training, optimizing and verifying machine learning models for forward prediction and reverse prediction, and the rules and behaviors of the digital twins are explored. In order to avoid the influence on the prediction caused by the inconsistency of the dimension and the variation range of the laser power and the scanning speed, the data in the training set needs to be normalized, and the specific operational expression is as shown in formula (1):
wherein:
denotes normalized data, x i Representing the original data, E [ x ] i ]Represents the average of the data in the training set,
representing the standard deviation of the data in the training set.
Referring to fig. 3, the machine learning model used in the present invention is a multilayer perceptron model, which is used in the forward prediction of process parameters to bath size and the reverse prediction of process parameters from bath size. In forward prediction and reverse prediction, machine learning models are classified into 6 classes, wherein the forward prediction is 2 classes, and the reverse prediction is 4 classes. The 2-class model for forward prediction is: (1) inputting: laser power and scanning speed, output: the width of the molten pool; (2) inputting: laser power and scanning speed, output: the depth of the molten pool. The 4-class model for reverse prediction is: scheme 1: (1) inputting: molten pool width and depth, output: laser power; (2) inputting: molten pool width and depth, output: a scanning speed; scheme 2: (1) inputting: molten pool width, depth and scanning speed, output: laser power; (2) inputting: molten pool width, depth and laser power, output: the scanning speed. Each model includes a multi-layer sensor model with 9 different parameters, such as table 1 and table 2.
TABLE 1 ganglionic points of Single hidden layer
TABLE 2 number of ganglion points in each layer of multilayer hidden layer
And then constructing a multi-layer perceptron, wherein parameters needing to be considered comprise an activation function, an optimization algorithm, the number of training rounds, the number of hidden layers and the number of ganglion points of the hidden layers, the activation function selects a rectification linear unit function (ReLU), the optimization algorithm selects a root mean square back propagation method (RMSprop), and the algorithm well solves the problem of premature termination in deep learning. The forward prediction training round number is selected to be 200, and the reverse prediction training round number is selected to be 300.
The number of hidden layers and the number of ganglion points are obtained by using an increasing method, namely, an optimal machine learning model is found by gradually increasing the number of hidden layers and the number of ganglion points, and the parameters of the optimal machine learning model are changed as shown in tables 1 and 2. As shown in fig. 2, after data processing and analysis, the detailed process of optimizing the multi-layered perceptron model to obtain the optimal machine learning model is as follows: firstly, selecting the number of hidden layers and the number of ganglion points of a multilayer perceptron model, and training a training set by adopting a K-fold cross validation method. And then importing the input data of the verification set into a multi-layer perceptron model to obtain corresponding prediction output. And according to the actual output of the verification set and the prediction output of the multilayer perceptron, the prediction error and the accuracy are obtained through comparison. And finally, repeating the previous two steps for 9 different multilayer perceptron models in each category to obtain the prediction errors and the accuracy of the different models. And comparing the prediction errors and the accuracy rates of different models to find an optimal multilayer perceptron model, which is also called an optimal machine learning model or an optimal prediction model. The prediction result of the multilayer perceptron model is measured by Absolute Percentage Error (APE) and Accuracy (AR), and the expressions are respectively formula (2) and formula (3):
AR＝(1-APE) (3)
wherein, the predicted value is the machine learning model, and yi is the actual value.
According to the prediction results of different multilayer perceptron models (such as figure 4), a model 6 for selecting the optimal prediction model of the width of the molten pool in forward prediction and a model 8 for selecting the optimal prediction model of the depth of the molten pool are selected. In the reverse prediction, as can be seen from fig. 5, the prediction results of the case 2 are better than those of the case 1, regardless of the predicted laser power or scanning speed. Therefore, the inverse prediction model in the present invention refers to the model corresponding to scheme 2. From the highest error and the average error in fig. 5, it can be known that the optimal model for predicting the laser power is the model 8, and the optimal model for predicting the scanning speed is also the model 8.
And 4, step 4: the obtained optimal machine learning model is used for new data prediction of an experiment and mechanism model, and additive manufacturing experiments, high-confidence mechanism model and machine learning fusion and efficient and accurate prediction of corresponding process parameters and molten pool sizes are achieved. And S3, after 4 optimal prediction models are obtained, the generalization capability of the models needs to be tested. The generalization ability of a machine learning model refers to the ability of the model to accurately predict output under new inputs. The corresponding new input data refers to an independent test set, which is partitioned in S3 and never used since then. And the generalization ability is tested by inputting the test set data into 4 optimal prediction models obtained in the step S3 and checking the results of the optimal prediction models.
The prediction result obtained on the test set is shown in fig. 6, the gray level distribution of the target molten pool width and the predicted molten pool width are similar, which shows that the target molten pool width and the predicted molten pool width are similar in value, and the accurate prediction of the molten pool width can be realized on new data by obtaining the machine learning model in the S3. In fig. 7 (c), the accuracy of the depth is above 85%, the average accuracy reaches 93%, and the machine learning model realizes accurate prediction of the depth of the molten pool on the test set. From the predictions of the bath width and the bath depth, the data obtained in S1 and S2 and the data processing methods used therein can provide the machine learning model with data which meets the requirements both in quantity and quality. Meanwhile, the model parameters and the optimization method used in the invention can realize accurate prediction of specific process parameters and molten pool size in additive manufacturing. While the results of backward prediction of laser power and scan speed are shown in fig. 8, it can be seen that although the backward prediction results are not as good as the forward prediction results, the error is within the tolerable range.
Aiming at the problems in the prior art and the respective limitations of a printing experiment, a mechanism model and machine learning, the method disclosed by the invention is a construction method for closely fusing the three, predicting key parameters and molten pool characteristics in the melting of the laser powder bed and forming a digital twin module in the additive manufacturing process. The experiment is used for providing basic data and verifying a mechanism model and a machine learning model, the mechanism model is used for machine learning expansion data and explaining an evolution mechanism of a molten pool, and the machine learning is used for carrying out forward prediction on the size of the molten pool and reverse prediction on laser power and scanning speed from different directions, so that the development of intelligent additive manufacturing is promoted.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (4)
1. A method for constructing a digital twin module in an additive manufacturing process is characterized by comprising the following steps of:
step 1: performing an additive manufacturing experiment, collecting a molten pool size under corresponding process parameters as experiment data, and forming a physical entity of the additive manufacturing process, wherein the process parameters comprise laser power and scanning speed, and the molten pool size comprises molten pool width and depth;
the step 1 specifically comprises the following steps: changing laser power and scanning speed to obtain different process parameter combinations, selecting square outline scanning as an experimental scanning strategy for obtaining a plurality of single channels under the same parameter, and measuring and averaging the widths and depths of molten pools on deposited channels on different sides of the printed square outline;
step 2: establishing a multi-physical-field coupling mechanism model in the additive manufacturing process, verifying the confidence coefficient of the mechanism model, further performing digital platform virtual printing based on the high-confidence-coefficient mechanism model, and obtaining corresponding printing results and data to form a digital twin body in the additive manufacturing physical process;
the step 2 specifically comprises the following steps: the specific verification method for verifying the confidence coefficient of the mechanism model is that a plurality of groups of laser power and scanning speed combinations used in an experiment are used as input of the mechanism model, the width and the depth of a molten pool in a steady state are used as corresponding output, the output is compared with the result of the width and the depth of the molten pool obtained in the experiment, the deviation is ensured to be within a required error range, after the validity of the mechanism model is ensured, a plurality of intervals are taken in the preset laser power and scanning speed to obtain more process parameter combinations, and then the parameter combinations are input into the mechanism model to obtain more molten pool size data;
and 3, step 3: analyzing and processing data obtained by the experiment and mechanism model, constructing an additive manufacturing machine learning model, performing model training and verification by adopting the data obtained by the experiment and mechanism model to obtain an optimal machine learning model, and performing forward and reverse prediction on process parameters and a molten pool size based on machine learning to realize virtual-real fusion and data intercommunication of an additive manufacturing physical entity and a digital twin;
the step 3 specifically comprises the following steps: dividing the analyzed and processed experimental data and the data obtained by the mechanism model into a training set, a verification set and a test set, and carrying out normalization processing on the data in the training set in order to avoid the influence on prediction caused by the inconsistency of the laser power and the scanning speed dimension and the variation range, wherein the specific operational expression is as shown in a formula (1):
wherein:
denotes normalized data, x i Representing the original data, E [ x ] i ]Represents the average of the data in the training set,
represents the standard deviation of the data in the training set,
the machine learning model is a multilayer perceptron model, the multilayer perceptron model is used for forward prediction from a process parameter to a molten pool size and reverse prediction from the molten pool size to the process parameter, in the forward prediction and the reverse prediction, the machine learning model is divided into 6 types, wherein the forward prediction is 2 types, the reverse prediction is 4 types, and the 2 types of forward prediction models are as follows: (1) inputting: laser power and scanning speed, output: the width of the molten pool; (2) inputting: laser power and scanning speed, output: the 4 types of models for the depth of the molten pool and the reverse prediction are as follows: scheme 1: (1) inputting: molten pool width and depth, output: laser power; (2) inputting: molten pool width and depth, output: a scanning speed; scheme 2: (1) inputting: molten pool width, depth and scanning speed, output: laser power; (2) inputting: molten pool width, depth and laser power, output: the speed of the scan is such that,
then a multilayer perceptron model is constructed, the parameters of the multilayer perceptron model comprise an activation function, an optimization algorithm, a training round number, a hidden layer number and a hidden layer ganglion point number, wherein the activation function selects a rectification linear unit function ReLU, the optimization algorithm selects a root-mean-square back propagation method RMSprop, a forward prediction training round number selection 200 and a reverse prediction training round number selection 300,
the number of hidden layers and the number of ganglion points are obtained by an increasing method, namely, the optimal machine learning model is found by gradually increasing the number of hidden layers and the number of ganglion points, and the process of obtaining the optimal machine learning model is as follows: step A, firstly selecting the number of hidden layers and the number of ganglion points of a multilayer perceptron model, training a training set by adopting a K-fold cross validation method, step B, then importing input data of a validation set into the multilayer perceptron model in the step A to obtain corresponding prediction output, obtaining prediction error and accuracy according to the actual output of the validation set under the same input and the prediction output of the multilayer perceptron, finally gradually increasing the number of ganglion points and the number of hidden layers to obtain different multilayer perceptron models, repeating the step A and the step B to obtain the prediction error and accuracy of different multilayer perceptron models, comparing the prediction error and accuracy of different multilayer perceptron models, finding the multilayer perceptron model with the optimal prediction performance, also called as an optimal machine learning model, wherein the prediction result of the multilayer perceptron model is measured by an absolute percentage error APE and accuracy AR, and the expressions are respectively formula (2) and formula (3):
AR＝(1-APE) (3)
wherein,
as a predictor of a machine learning model, y i Is an actual value;
and 4, step 4: the obtained optimal machine learning model is used for new data prediction of an experiment and mechanism model, and additive manufacturing experiments, high-confidence mechanism model and machine learning fusion and efficient and accurate prediction of corresponding process parameters and molten pool sizes are achieved.
2. The additive manufacturing process digital twinning module build method of claim 1,
in the step 1, the value range of the laser power is between 220W and 350W, and the scanning speed is between 300mm/s and 1500 mm/s.
3. The additive manufacturing process digital twinning module construction method of claim 1,
the preset laser power and scanning speed in the step 2 are as follows: 100W-350W laser power, 300mm/s-1500mm/s scanning speed.
4. The method for constructing the digital twin module in the additive manufacturing process according to claim 1, wherein after obtaining the optimal machine learning model, the generalization capability of the optimal machine learning model needs to be tested, the generalization capability is tested by inputting test set data into 4 optimal machine learning models obtained in step S3 and checking the result of the optimal machine learning model, and the new data is the test set, wherein the 4 optimal machine learning models correspond to the optimal machine learning model of the molten pool width and the molten pool depth in forward prediction and the optimal machine learning model of the laser power and the scanning speed in scheme 2 in reverse prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210094566.XA CN114564880B (en) | 2022-01-26 | 2022-01-26 | Method for constructing digital twin module in additive manufacturing process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210094566.XA CN114564880B (en) | 2022-01-26 | 2022-01-26 | Method for constructing digital twin module in additive manufacturing process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114564880A CN114564880A (en) | 2022-05-31 |
| CN114564880B true CN114564880B (en) | 2022-11-25 |
Family
ID=81713950
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210094566.XA Active CN114564880B (en) | 2022-01-26 | 2022-01-26 | Method for constructing digital twin module in additive manufacturing process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114564880B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114762915B (en) * | 2022-06-16 | 2022-08-16 | 吉林大学 | Intelligent manufacturing system based on digital twinning |
| CN115401696A (en) * | 2022-09-29 | 2022-11-29 | 中国科学院沈阳自动化研究所 | Data/model hybrid-driven robot remote driving method |
| CN115489125A (en) * | 2022-09-30 | 2022-12-20 | 吉林大学 | Additive manufacturing digital twin system construction method based on cloud service |
| CN115828654A (en) * | 2022-11-15 | 2023-03-21 | 南京林业大学 | Laser additive manufacturing online monitoring system and method for constructing component digital twin body |
| CN115822325A (en) * | 2022-11-29 | 2023-03-21 | 哈尔滨工程大学 | Intelligent maintenance mobile square cabin integrating material increase and decrease composite manufacturing technology |
| CN117150587B (en) * | 2023-08-08 | 2024-07-02 | 宁夏大学 | Arc fuse additive manufacturing digital twin system construction method and system based on mixed reality |
| CN118313274B (en) * | 2024-06-03 | 2024-08-06 | 武汉科技大学 | Method for constructing additive manufacturing process digital twin system with full-period data management |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019055538A1 (en) * | 2017-09-12 | 2019-03-21 | Arconic Inc. | Systems and methods for additive manufacture |
| CN110795891A (en) * | 2019-10-18 | 2020-02-14 | 武汉科技大学 | Method for learning operating parameters of trailing suction hopper dredger |
| CN111283192A (en) * | 2020-01-22 | 2020-06-16 | 南京理工大学 | Laser powder bed melting additive manufacturing molten pool monitoring and pore control method |
| CN112084569A (en) * | 2019-06-14 | 2020-12-15 | 通用电气公司 | Additive manufacturing coupled digital twin ecosystem based on multivariable distribution model of performance |
| CN112689546A (en) * | 2018-09-05 | 2021-04-20 | 通用电气公司 | Method and system for generating device-specific machine learning models |
| CN112859739A (en) * | 2021-01-15 | 2021-05-28 | 天津商业大学 | Digital twin-driven multi-axis numerical control machine tool contour error suppression method |
| CN113373295A (en) * | 2021-04-22 | 2021-09-10 | 上海孟伯智能物联网科技有限公司 | Annealing furnace digital twin intelligent alarm system and method based on multilayer perceptron |
| CN113762240A (en) * | 2021-08-30 | 2021-12-07 | 江苏大学 | Method and system for predicting geometrical characteristics of cladding layer based on deep learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10747202B2 (en) * | 2017-06-30 | 2020-08-18 | General Electric Company | Systems and method for advanced additive manufacturing |
| CN110682400B (en) * | 2019-11-04 | 2020-12-11 | 清华大学 | Digital twin structure test system and method based on 3D printing platform |
-
2022
- 2022-01-26 CN CN202210094566.XA patent/CN114564880B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019055538A1 (en) * | 2017-09-12 | 2019-03-21 | Arconic Inc. | Systems and methods for additive manufacture |
| CN112689546A (en) * | 2018-09-05 | 2021-04-20 | 通用电气公司 | Method and system for generating device-specific machine learning models |
| CN112084569A (en) * | 2019-06-14 | 2020-12-15 | 通用电气公司 | Additive manufacturing coupled digital twin ecosystem based on multivariable distribution model of performance |
| CN110795891A (en) * | 2019-10-18 | 2020-02-14 | 武汉科技大学 | Method for learning operating parameters of trailing suction hopper dredger |
| CN111283192A (en) * | 2020-01-22 | 2020-06-16 | 南京理工大学 | Laser powder bed melting additive manufacturing molten pool monitoring and pore control method |
| CN112859739A (en) * | 2021-01-15 | 2021-05-28 | 天津商业大学 | Digital twin-driven multi-axis numerical control machine tool contour error suppression method |
| CN113373295A (en) * | 2021-04-22 | 2021-09-10 | 上海孟伯智能物联网科技有限公司 | Annealing furnace digital twin intelligent alarm system and method based on multilayer perceptron |
| CN113762240A (en) * | 2021-08-30 | 2021-12-07 | 江苏大学 | Method and system for predicting geometrical characteristics of cladding layer based on deep learning |
Non-Patent Citations (6)
| Title |
|---|
| "Building blocks for a digital twin of additive manufacturing";G.L. Knapp et al.;《Acta Materialia》;20170622;第390-399页 * |
| "Metallurgy, mechanistic models and machine learning in metal printing";T. DebRoy et al.;《Nature Reviews Materials》;20201002;第48-68页 * |
| "Mechanistic models for additive manufacturing of metallic components";H.L. Wei et al.;《Progress in Materials Science》;20200619;第1-113页 * |
| "基于高速图像特征提取的粉末床激光选区熔化翘曲缺陷识别技术";彭浩等;《计算机应用》;20211231;第265-269页 * |
| "综述：增材制造中的机器学习";EngineeringForLife;《知乎，链接：https://zhuanlan.zhihu.com/p/401657784》;20210824;第1-9页 * |
| "虚实融合与新一代信息技术";李京燕等;《第十七届中国CAE工程分析技术年会论文集》;20211112;第27-32页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114564880A (en) | 2022-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114564880B (en) | Method for constructing digital twin module in additive manufacturing process | |
| CN113255229B (en) | Fuel assembly multidisciplinary structural design optimization method based on joint simulation | |
| CN110976830B (en) | Control method for casting defects of aluminum alloy gear shifting hub | |
| CN106529051A (en) | Method for determining heat source model parameters of single wire submerged arc welding numerical simulation | |
| CN115186557B (en) | Additive manufacturing component microstructure homogenization regulation and control method based on multi-physical field coupling solution heat accumulation effect | |
| CN112651562B (en) | Laser cladding process optimization method based on signal-to-noise ratio and improved gray correlation degree | |
| CN114492173B (en) | Abnormal pipeline forming benchmark optimization method based on additive manufacturing | |
| CN114398824A (en) | Motor multi-target robustness optimization method based on local agent model | |
| Feng et al. | Application of machine learning to optimize process parameters in fused deposition modeling of PEEK material | |
| CN114970371A (en) | Method for predicting deformation of SLM (Selective laser melting) formed titanium alloy thin-wall part by applying GA-BP (genetic algorithm-Back propagation) neural network | |
| CN111444619B (en) | Online analysis method and equipment for injection mold cooling system | |
| Bahnini et al. | Accuracy investigation of fused deposition modelling (FDM) processed ABS and ULTRAT parts | |
| CN112685958A (en) | SiC MOSFET blocking voltage determination method based on neural network | |
| CN115563736A (en) | Turbine blade arc additive real-time temperature field prediction method | |
| CN116796546A (en) | High-temperature alloy additive manufacturing process optimization prediction method based on small sample learning | |
| CN115130239B (en) | Mechanical property prediction method for metal additive manufacturing based on multi-scale modeling | |
| CN114091304B (en) | Intelligent decision-making method for processing hull plate by oxyhydrogen gas heat source | |
| CN114997038B (en) | Selective laser melting area temperature prediction and variable parameter scanning method | |
| CN113435083B (en) | Additive manufacturing residual stress and deformation prediction method and system based on machine learning | |
| CN114997004A (en) | Storage box internal support clamp assembling quality prediction method based on finite element simulation and RNN neural network | |
| CN115600492A (en) | Laser cutting process design method and system | |
| Yang et al. | Research on Optimization of Injection Molding Process Parameters of Automobile Plastic Front‐End Frame | |
| CN115455787A (en) | Workpiece casting processing method, device, electronic device and storage medium | |
| CN118297098A (en) | Self-adaptive dotted neural network model construction method suitable for SLM process optimization | |
| CN118180411B (en) | Intelligent additive manufacturing flow control method and system based on data analysis |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |