CN113343566B - Nickel-based alloy fracture toughness prediction method and system based on deep learning - Google Patents
Nickel-based alloy fracture toughness prediction method and system based on deep learning Download PDFInfo
- Publication number
- CN113343566B CN113343566B CN202110601155.0A CN202110601155A CN113343566B CN 113343566 B CN113343566 B CN 113343566B CN 202110601155 A CN202110601155 A CN 202110601155A CN 113343566 B CN113343566 B CN 113343566B
- Authority
- CN
- China
- Prior art keywords
- feature vector
- fracture toughness
- vector
- nickel
- experimental data
- 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
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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/26—Composites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
-
- 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
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Hardware Design (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Geometry (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention relates to a nickel-based alloy fracture toughness prediction method and system based on deep learning, which are characterized in that a deep confidence network and a support vector regression network based on an attention mechanism are combined together to predict fracture toughness values, and the strong feature extraction capacity of a deep learning model and the regression prediction capacity of support vector regression are organically combined. Meanwhile, feature vectors and original features obtained by the deep confidence network based on the attention mechanism are spliced and input into a support vector regression model, so that the fracture toughness prediction method disclosed by the invention simultaneously considers high-order combined features and low-order linear features related to fracture toughness, and has higher accuracy.
Description
Technical Field
The invention relates to the technical field of toughness prediction, in particular to a nickel-based alloy fracture toughness prediction method and system based on deep learning.
Background
With the continuous development of the aerospace industry, the requirements on the performance and reliability of an aeroengine are continuously improved, and the requirements on the comprehensive performance, the temperature bearing capacity and the like of materials are also higher and higher. The nickel-based superalloy has higher strength and excellent oxidation resistance and corrosion resistance under the high-temperature condition, so that the nickel-based superalloy becomes a preferred material for key hot end components such as modern aeroengines, spacecrafts, rocket engines and the like.
Fracture toughness is a toughness parameter of a material against brittle failure, and is a critical value for judging whether a crack of the material reaches an unstable level, and plays a critical role in damage tolerance design and structural integrity evaluation. For parts operating at high temperatures, the temperature distribution tends to be graded greatly and fracture toughness measured at room temperature is far from satisfactory for practical engineering design. Therefore, in the design of nickel-base superalloys, there is an urgent need for a method that can accurately predict the fracture toughness of nickel-base alloys.
Disclosure of Invention
The invention aims to provide a nickel-based alloy fracture toughness prediction method and system based on deep learning, which can accurately predict the fracture toughness of nickel-based superalloy with different temperatures and different components according to the existing experimental data.
In order to achieve the above object, the present invention provides the following solutions:
a method for predicting fracture toughness of a nickel-based alloy based on deep learning, the method comprising:
preprocessing the experimental data of fracture toughness of the nickel-base alloy to obtain an experimental data set;
performing feature learning and optimization on the experimental data set by using a deep confidence network based on an attention mechanism to obtain an optimized feature vector;
splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector;
and predicting the fracture toughness of the nickel-based alloy through a trained support vector regression model according to the prediction feature vector.
The invention also provides a nickel-based alloy fracture toughness prediction system based on deep learning, which comprises:
the input module is used for preprocessing the nickel-based alloy fracture toughness experimental data to obtain an experimental data set;
the deep learning module based on the attention is used for learning and optimizing the characteristics of the experimental data set by using a deep confidence network based on an attention mechanism to obtain an optimized characteristic vector;
the feature splicing module is used for splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector;
and the output module is used for predicting the fracture toughness of the nickel-based alloy through a trained support vector regression model according to the prediction feature vector.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a nickel-based alloy fracture toughness prediction method and system based on deep learning, which combines a deep confidence network based on an attention mechanism and a support vector regression model to predict fracture toughness values, organically combines strong feature extraction capacity of the deep learning model and regression prediction capacity of support vector regression, and compared with the traditional machine learning method which needs to rely on manual experience to select features, the method can extract features more closely related to the nickel-based alloy fracture toughness, and improves model prediction accuracy. Meanwhile, the feature vector obtained by the deep confidence network based on the attention mechanism is spliced with the original feature and is input into the support vector regression model, so that the prediction method of the invention considers the high-order combination feature and the low-order linear feature related to the fracture toughness of the nickel-based alloy at the same time, and has higher accuracy.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a nickel-based alloy fracture toughness prediction method based on deep learning provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of a deep belief network structure based on an attention mechanism according to an embodiment of the present invention;
FIG. 3 is a block diagram of a nickel-base alloy fracture toughness prediction system based on deep learning according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The fracture toughness of superalloys is strongly dependent on the alloying elements and experimental conditions, but is difficult to characterize with specific mathematical expressions due to its complex and highly nonlinear relationship. Thus, the diversity and interplay of alloying elements often makes it difficult to determine suitable fracture toughness. In solving the problem, the material genetic engineering technology has great potential.
The material genetic engineering is a subversion leading edge technology rising in the international material field in recent years, the basic idea is to integrate material high-flux calculation, high-flux experiment and material big data technology, and through collaborative innovation, the speed of the research and development process from discovery, manufacture and application of the material is accelerated, and the cost is reduced. The data and artificial intelligence are the core of material genetic engineering, and through the application of big data and artificial intelligence technology, the correlation between the tissue structure of a new material and performance and process optimization problem can be realized, and the performance of the material is improved.
The invention aims to provide a nickel-based alloy fracture toughness prediction method and system based on deep learning, which are based on the idea and method of material genetic engineering, can reduce the experimental quantity and the experimental cost, and have great theoretical significance and application value for promoting the research and development of nickel-based superalloy technology.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Example 1
The direct method for obtaining the relationship between the fracture toughness and the temperature of the high-temperature alloy is to conduct fracture toughness measurement tests on standard samples at different temperatures, and then fitting the standard samples by using an empirical model. However, such empirical models are not widely applicable and can only be used with the materials under investigation. In addition, more data at high temperatures is required to improve the accuracy of the model, which results in significant time and cost consumption.
To overcome the limitations of the fitting method based on empirical data, a learner proposed a main curve method based on physical meaning, but the method is only applicable to ductile-brittle transition temperature range, which is usually lower than 0 ℃ for ferritic steels, and obviously the method is not applicable to nickel-base superalloys operating at high temperature.
Still other temperature-dependent models have been proposed to predict fracture toughness values of superalloys at high temperatures based on stress-strain curves and the linear expansion coefficient α of the superalloy, but the models need to be based on other mechanical property data and the accuracy of the predicted data is not high.
In summary, experience-based physical models not only require significant time and cost, but also do not provide high accuracy of the predicted data.
In recent years, machine learning models have been widely used for predicting mechanical properties of materials. The mechanical properties (ultimate strength, yield strength, strain hardening index, elongation) of nickel-base superalloy IN718 were predicted using a feed-forward neural network. The three-layer feedforward neural network is utilized to establish an average static fracture energy prediction model of the nickel-based superalloy IN738LC based on experimental data, so that the time required by the experiment is reduced.
The relationship between the fracture toughness and the crack geometry, the load type and the working temperature of the 7075-T651 aluminum alloy is analyzed by utilizing a multilayer feedforward perception artificial neural network, or the quantitative relationship between the microstructure and the fracture toughness of the niobium-silicon alloy is established by utilizing a three-layer BP neural network, and the technological parameters of the alloy are optimized by utilizing an artificial neural network model. A3-layer BP neural network is adopted to establish a prediction model of mechanical properties (tensile strength, yield strength, elongation, reduction of area and fracture toughness) of the Ti-10V-2Fe-3Al alloy under thermal processing technological parameters such as deformation temperature, deformation degree, solid solution temperature, aging temperature and the like and a thermal treatment system. Still other people utilize BP neural networks with double hidden layers to build fracture toughness prediction models under different alloy compositions and low alloy yield stress, and provide possibility for designing new alloys with higher fracture toughness by modifying alloy chemical compositions.
The Generalized Regression Neural Network (GRNN) based on the Drosophila optimization algorithm (FOA) can be adopted to predict the fracture toughness of the pearlitic steel so as to improve the prediction accuracy of a small number of samples, and on the basis of the established model, the influence of alloy elements on the fracture toughness is further researched, and the composition parameters of the steel are optimized. The learner uses the artificial neural network to establish the relationship between the composition and processing parameters and the mechanical properties (yield strength, tensile strength, elongation and ductility) of the steel sheet to find the importance of different variables. A part of the scholars used 5 machine learning methods to predict four mechanical properties (fatigue strength, tensile strength, breaking strength and hardness) of steel selected from the database of the national institute of materials science.
The fracture energy of polymer nanocomposites was also predicted using ANN and ANFIS and compared to the Huang and Kinloch model and three linear regression models. Experimental results show that the ANN and ANFIS have better prediction effect, and the ANFIS method is best. A 3-layer BP neural network was used to predict fracture toughness of composite ceramic materials. Three layers of feedforward neural networks have been used to study the effect of single phase composition, morphology, size and volume fraction on the fracture toughness of Nb-silicide in situ composites and the model was used to predict the optimal microstructure that can improve fracture toughness.
However, the above methods all adopt a shallow learning model based on machine learning, which has a better prediction effect than an equation or model based on experience, but the shallow learning model needs to rely on prior knowledge in a specific field to extract features, and cannot fully reflect essential features of data.
Therefore, the deep learning thought is combined with the support vector regression method, the traffic flow prediction method based on the deep learning regression machine is provided, the data input of influence factors and the learning and conversion of layer-by-layer information are realized by constructing a multi-layer limited Boltzmann machine structure, and equivalent key information is extracted. And then, inputting the converted equivalent key information into a support vector regression model to realize short-time traffic flow prediction. The short-time traffic flow prediction method combining the deep confidence network model with the support vector regression is also provided, the feature learning is carried out on traffic flow data in the road network by using the deep confidence network model, so that the essential features of the data are mined, and then the traffic flow prediction is carried out by adopting the support vector regression method. There is also a model combining a gated loop unit network (GRU) with support vector regression for predicting short-term traffic flow, which uses the GRU to extract features of the data, and inputs it into the support vector regression model at the top layer of the model to obtain the final predicted output of the model.
Compared with a single support vector regression model, the prediction model combining the deep learning and the support vector regression has the advantages that the deep learning model can be used for extracting the characteristics of the data so as to reduce the dependence of the characteristic extraction on manpower, thereby describing the essential characteristics among the data and improving the prediction precision of the model. In addition, compared with a single deep neural network model, the combination model can realize the generalization performance of data to a greater degree by using the nonlinear kernel function of the support vector regression model, so that the generalization capability of the model is improved.
Specifically, as shown in fig. 1, the embodiment provides a nickel-based alloy fracture toughness prediction method based on deep learning, which includes:
step 101: preprocessing the experimental data of fracture toughness of the nickel-base alloy to obtain an experimental data set;
step 102: performing feature learning and optimization on the experimental data set by using a deep confidence network based on an attention mechanism to obtain an optimized feature vector;
step 103: splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector;
step 104: and predicting the fracture toughness of the nickel-based alloy through a trained support vector regression model according to the prediction feature vector.
In step 101, firstly, downloading 877 pieces of experimental data about fracture toughness of the nickel-base superalloy from a material genetic engineering database; then, the collected data are integrated together according to the sequence of the component contents (C, W, mo, nb, zr, co, al, cr, ti, ni), the quenching temperature and the fracture toughness values to form a two-dimensional matrix (877, 12); then, in order to eliminate the dimension influence between indexes, the data is subjected to the maximum normalization processing, and all the data are mapped between 0 and 1, as shown in the following formula:
wherein x is scale For the normalized value, x is the original value, x max 、x min Respectively the maximum and minimum of the original data set.
Finally, the component contents (C, W, mo, nb, zr, co, al, cr, ti, ni) and quenching temperatures of each piece of data are taken as input vectors of the model, and the corresponding fracture toughness values are taken as output vectors of the model. And according to 9:1 randomly dividing an experimental training set and an experimental testing set to obtain an experimental data set.
Step 102 is then performed: inputting the experimental data set into a deep confidence network to obtain a first feature vector;
inputting the first feature vector into an attention mechanism module to obtain a second feature vector;
and taking the second characteristic vector as an optimized characteristic vector.
The structure of the deep belief network (a-DBN) based on the attention mechanism provided by the present embodiment is shown in fig. 2, wherein the Deep Belief Network (DBN) is a deep neural network composed of two Restricted Boltzmann Machines (RBMs) and one BP neural network unit.
The Boltzmann machine (RBM) is a randomly generated neural network structure consisting of a display layer and an underlying layer, wherein the display layer consists of display elements, used as input training data. The hidden layer is composed of hidden elements and serves as a feature detector. When the fracture toughness prediction of the nickel-based superalloy is carried out, firstly, a characteristic matrix is formed by the content and the temperature of each component of the nickel-based superalloy and is input into a display layer of a first RBM network, each RBM network is independently trained in an unsupervised mode, so that when feature vectors are mapped to different feature spaces, feature information is kept as much as possible, and the output of the previous RBM network is used as the input of the next RBM network. After passing through the two layers of RBM networks, a first characteristic vector X of the fracture toughness of the nickel-based superalloy is obtained.
Because each layer of RBM network can only ensure that the weight in the layer reaches the optimum for the characteristic vector mapping of the layer, but not for the characteristic vector mapping of the whole DBN, the BP neural network is added in the last layer of the DBN, and the network model is trained in a supervision mode. Furthermore, to improve the accuracy of model prediction, we introduce an attention mechanism module before the BP neural network. The attention mechanism refers to the processing mode of human vision, focuses attention on a key area, and is essentially to select information playing a key role in a task from a plurality of pieces of information, so that the complexity of the task is reduced, and meanwhile, the accuracy of a model is improved. In this module we input the first feature vector X learned by RBM network to the attention mechanism layer, resulting in the second feature vector a. The calculation formula of the second feature vector a is shown in (1):
wherein alpha is i For the attention distribution, the degree of correlation of the ith information in the input information vector X with the query vector q is expressed given the query vector q. The calculation formulas are shown as (2) and (3):
wherein s (x i Q) is a attention scoring function. W, U and V are weight matrices.
Then we input the second eigenvector a into the BP neural network layer and propagate the error information top-down into the RBM of each layer using the back propagation network, thereby trimming the entire A-DBN network. Thereby, the optimized feature vector F is obtained by using the fine-tuned deep belief network based on the attention mechanism.
And then splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector, and inputting the predicted feature vector into a support vector regression model to predict the fracture toughness of the nickel-based superalloy.
The support vector regression algorithm is based on the VC dimension theory and the structural risk minimization principle, and can fully call limited sample information to find the best balance between the complexity and learning ability of the model so as to obtain the best generalization ability. Conventional regression models typically calculate the loss directly based on the difference between the model output f (x) and the true output y, which is zero if and only if the two are identical. In contrast, support vector regression assumes that a maximum of ε of deviation between the two can be tolerated, i.e., the loss is calculated when the absolute value of the difference between f (x) and y is greater than ε. This corresponds to the construction of a spacing band of width 2 epsilon centered on f (x). If the training samples fall within this interval band, the prediction is considered correct. Thus, the support vector regression problem can be converted to equation (4):
c in the formula (4) is penalty factor, l ε Is an epsilon-insensitive loss function, epsilon is a deviation threshold, W is a weight matrix, f (x) i ) For model output, y i For true output, i is the count of training samplesAnd the variable, m, is the total number of training samples.
Support vector regression maps data from a low dimensional space to a high dimensional space by means of a suitable kernel function and performs regression in that space, fitting a continuous function such that the loss function is minimized. There are four common kernel functions, namely a linear kernel function, a polynomial kernel function, a radial basis kernel function and a sigmoid kernel function. Because the radial basis function has the advantages of small parameter variable and selection calculation amount and high calculation efficiency, the radial basis function is selected as the kernel function for supporting vector regression, and the mathematical expression is shown as the formula (5):
in equation (5), σ is the radial basis function width. C and sigma are important parameters related to the generalization performance of the SVR model, and the optimal combination parameters of the (C and sigma) are optimally selected by using a grid search algorithm in a K-fold cross validation mode so as to ensure that the built model has optimal performance. The basic principle of K-fold cross validation is that a data set is divided into K parts with the same size by K times in turn, the K-1 parts are used as training sets, the rest 1 parts are used as validation sets, and the average value of the accuracy of K times of validation results is used as an estimated value of modeling accuracy.
The optimization selection of the optimal combination of the penalty factor and the radial basis function width of the support vector regression model by using a grid search algorithm comprises the following steps:
listing all possible combinations of penalty factors and radial basis function widths of the support vector regression model, the all possible combinations constituting a grid;
sequentially carrying out support vector regression modeling on possible combinations in the grid;
and selecting the possible combination with highest modeling precision by using a k-fold cross validation method as the optimal combination of the penalty factor and the radial basis function width of the support vector regression model.
The time complexity T (n) =o (n 2) of the algorithm of the nickel-base superalloy fracture toughness prediction method based on deep learning provided in this embodiment mainly originates from a double-layer for loop of 1-10 lines, and the code is as follows:
input: c, the value of sigma is a new training set formed by splicing a high-order combined feature vector on the training set obtained by the A-DBN model with an original training set;
and (3) outputting: a support vector regression model constructed using the optimal parameter combinations;
1:for C in[C 1 ,C 2 …C m ]:
2:forσin[σ 1 ,σ 2 …σ n ]:
constructing a support vector regression model;
4, performing 10-fold cross validation;
5:if(score>best_score)：
6:best_score＝score；
optimal parameter combination = { 'C': C, 'σ': σ };
8:end if
9:end for
10:end for
11, constructing a support vector regression model by using the optimal parameter combination;
in summary, in the regression prediction process of the fracture toughness of the nickel-based superalloy, firstly, inputting a data set into a trained A-DBN model to obtain a feature vector F obtained by the A-DBN model; then, the feature vector F and the original features are spliced to form a new training set and a new testing set, and the new training set and the new testing set are input into a support vector regression model for training; and finally, predicting the fracture toughness value of the nickel-base superalloy by using a trained support vector regression model.
The nickel-base superalloy fracture toughness prediction method based on deep learning provided by the embodiment can rapidly and accurately predict the fracture toughness of nickel-base superalloys with different temperatures and different components according to the existing experimental data, and effectively solves the problems of complexity, time consumption and low precision existing in obtaining the nickel-base superalloy fracture toughness value by using an experimental and experience-based method.
Aiming at the problem that the extracted features are insufficient based on manual experience, the embodiment utilizes the deep confidence network to extract the features, compared with the traditional machine learning method which needs to rely on manual experience to select the features and can not fully reflect the problem of essential features of data, the deep confidence network has excellent feature learning capability, and the learned features describe the data more essentially. In addition, in order to optimize the quality of feature extraction, an attention mechanism module is introduced into the deep confidence network, different weights are distributed according to the influence degree of each feature on the fracture toughness of the nickel-based superalloy, so that the model can extract the features more relevant to the fracture toughness value of the nickel-based superalloy in a feature learning stage, and the prediction performance of the model is improved.
In order to further improve the regression prediction capability of the model, feature vectors obtained by a deep belief network module (A-DBN) based on an attention mechanism are spliced with original features and input into a support vector regression model, so that the model simultaneously considers high-order combined features and low-order linear features.
Therefore, the nickel-base superalloy fracture toughness prediction method provided by the embodiment not only can solve the problems of complexity, time consumption and low precision existing in the process of obtaining the nickel-base superalloy fracture toughness value by using an experiment and an experience-based method, but also can effectively solve the problem that the traditional machine learning method needs to rely on manual experience to extract features, optimizes the feature extraction compared with a single deep learning model, considers the high-order combination feature and the low-order linear feature at the same time, and improves the prediction precision of the model.
Example 2
A deep learning-based nickel-based alloy fracture toughness prediction system, as shown in fig. 3, the system comprising:
the input module M1 is used for preprocessing the fracture toughness experimental data of the nickel-based alloy to obtain an experimental data set;
the attention-based deep learning module M2 is used for learning and optimizing the characteristics of the experimental data set by using an attention-mechanism-based deep confidence network to obtain an optimized characteristic vector;
the feature splicing module M3 is used for splicing the optimized feature vector and the original feature vector of the experimental data set to obtain a predicted feature vector;
and the output module M4 is used for predicting the fracture toughness of the nickel-based alloy through a trained support vector regression model according to the prediction feature vector.
In this specification, each embodiment is mainly described in the specification as a difference from other embodiments, and the same similar parts between the embodiments are referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (6)
1. A method for predicting fracture toughness of a nickel-based alloy based on deep learning, the method comprising:
preprocessing the experimental data of fracture toughness of the nickel-base alloy to obtain an experimental data set;
performing feature learning and optimization on the experimental data set by using a deep confidence network based on an attention mechanism to obtain an optimized feature vector;
splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector;
predicting the fracture toughness of the nickel-based alloy through a trained support vector regression model according to the prediction feature vector;
the learning and optimizing of the characteristics of the experimental data set are performed by using a deep confidence network based on an attention mechanism, and the obtaining of the optimized characteristic vector comprises the following steps:
inputting the experimental data set into a deep confidence network to obtain a first feature vector;
inputting the first feature vector into an attention mechanism module to obtain a second feature vector;
taking the second feature vector as an optimized feature vector;
the second feature vector isWherein alpha is i For the attention distribution, the degree of correlation between the ith information in the input information vector X and the query vector q is expressed when the query vector q is given; />S(X i ,q)＝V T tanh(W Xi +Uq), wherein S (X i Q) is a attention scoring function, W, U and V are weight matrices;
after the second feature vector is obtained, the method further comprises:
inputting the second feature vector into a BP neural network to obtain error information;
the error information is transmitted to the limited Boltzmann machine network of each layer by utilizing a back transmission network, so that fine adjustment of the deep confidence network based on the attention mechanism is completed, and the fine-adjusted deep confidence network based on the attention mechanism is obtained;
obtaining an optimized feature vector by using the fine-tuned deep confidence network based on the attention mechanism;
the trained support vector regression model comprises:
wherein C is penalty factor, l ε Is an epsilon-insensitive loss function, epsilon is a deviation threshold, W is a weight matrix, f (x) i ) Is thatModel output, y i For true output, i is the count variable of the training samples, and m is the total number of training samples.
2. The method of claim 1, wherein the preprocessing the experimental data of the fracture toughness of the nickel-base alloy to obtain the experimental data set comprises:
integrating the nickel-based alloy fracture toughness experimental data into a two-dimensional matrix;
and taking the component content and the quenching temperature in the two-dimensional matrix as input, and taking the corresponding fracture toughness value as output to construct an experimental training set and an experimental testing set, wherein the experimental training set and the experimental testing set form an experimental data set.
3. The method of claim 1, wherein the kernel function of the trained support vector regression model is a radial basis kernel function.
4. The method for predicting fracture toughness of nickel-base alloy based on deep learning according to claim 1, wherein the training method of the support vector regression model comprises:
and optimizing and selecting the optimal combination of the penalty factors and the radial basis function width of the support vector regression model by using a grid search algorithm in a K-fold cross validation mode to obtain a trained support vector regression model.
5. The method of claim 4, wherein optimizing the optimal combination of the penalty factor and the radial basis function width of the support vector regression model using a grid search algorithm comprises:
listing all combinations of penalty factors and radial basis function widths of a support vector regression model, wherein all combinations form a grid;
sequentially carrying out support vector regression modeling on the combinations in the grids;
and selecting the combination with highest modeling precision by using a k-fold cross validation method as the optimal combination of the penalty factor and the radial basis function width of the support vector regression model.
6. A deep learning-based nickel-based alloy fracture toughness prediction system, the system comprising:
the input module is used for preprocessing the nickel-based alloy fracture toughness experimental data to obtain an experimental data set;
the deep learning module based on the attention is used for learning and optimizing the characteristics of the experimental data set by using a deep confidence network based on an attention mechanism to obtain an optimized characteristic vector;
the learning and optimizing of the characteristics of the experimental data set are performed by using a deep confidence network based on an attention mechanism, and the obtaining of the optimized characteristic vector comprises the following steps:
inputting the experimental data set into a deep confidence network to obtain a first feature vector;
inputting the first feature vector into an attention mechanism module to obtain a second feature vector;
taking the second feature vector as an optimized feature vector;
the second feature vector isWherein alpha is i For the attention distribution, the degree of correlation between the ith information in the input information vector X and the query vector q is expressed when the query vector q is given; />S(X i ,q)＝V T tanh(W Xi +Uq), wherein S (X i Q) is a attention scoring function, W, U and V are weight matrices;
after the second feature vector is obtained, the method further comprises:
inputting the second feature vector into a BP neural network to obtain error information;
the error information is transmitted to the limited Boltzmann machine network of each layer by utilizing a back transmission network, so that fine adjustment of the deep confidence network based on the attention mechanism is completed, and the fine-adjusted deep confidence network based on the attention mechanism is obtained;
obtaining an optimized feature vector by using the fine-tuned deep confidence network based on the attention mechanism;
the feature splicing module is used for splicing the optimized feature vector with the original feature vector of the experimental data set to obtain a predicted feature vector;
the output module is used for predicting the fracture toughness of the nickel-base alloy through a trained support vector regression model according to the prediction feature vector;
the trained support vector regression model comprises:
wherein C is penalty factor, l ε Is an epsilon-insensitive loss function, epsilon is a deviation threshold, W is a weight matrix, f (x) i ) For model output, y i For true output, i is the count variable of the training samples, and m is the total number of training samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110601155.0A CN113343566B (en) | 2021-05-31 | 2021-05-31 | Nickel-based alloy fracture toughness prediction method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110601155.0A CN113343566B (en) | 2021-05-31 | 2021-05-31 | Nickel-based alloy fracture toughness prediction method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113343566A CN113343566A (en) | 2021-09-03 |
| CN113343566B true CN113343566B (en) | 2023-09-01 |
Family
ID=77472834
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110601155.0A Active CN113343566B (en) | 2021-05-31 | 2021-05-31 | Nickel-based alloy fracture toughness prediction method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113343566B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114383958B (en) * | 2022-01-11 | 2024-05-24 | 沈阳工业大学 | Method and system for determining key-based near-field dynamics critical elongation of deep rock mass |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709640A (en) * | 2016-12-15 | 2017-05-24 | 华南理工大学 | Method for predicting ship energy consumption based on deep learning and support vector regression |
| CN109543352A (en) * | 2018-12-29 | 2019-03-29 | 西南石油大学 | Shale fracture toughness prediction technique, device and electronic equipment |
| CN110378035A (en) * | 2019-07-19 | 2019-10-25 | 南京工业大学 | Hydrocracking soft measurement modeling method based on deep learning |
| CN110852273A (en) * | 2019-11-12 | 2020-02-28 | 重庆大学 | Behavior identification method based on reinforcement learning attention mechanism |
| CN110929958A (en) * | 2019-12-10 | 2020-03-27 | 西安邮电大学 | Short-term traffic flow prediction method based on deep learning parameter optimization |
| CN111210633A (en) * | 2020-02-09 | 2020-05-29 | 北京工业大学 | Short-term traffic flow prediction method based on deep learning |
| CN111341386A (en) * | 2020-02-17 | 2020-06-26 | 大连理工大学 | Attention-introducing multi-scale CNN-BilSTM non-coding RNA interaction relation prediction method |
| WO2020144417A1 (en) * | 2019-01-09 | 2020-07-16 | Constellium Issoire | Method for checking a damage tolerance property of a part made of an aluminium alloy |
| CN112597921A (en) * | 2020-12-28 | 2021-04-02 | 杭州电子科技大学 | Human behavior recognition method based on attention mechanism GRU deep learning |
| CN112668804A (en) * | 2021-01-11 | 2021-04-16 | 中国海洋大学 | Method for predicting broken track of ground wave radar ship |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11915105B2 (en) * | 2019-02-05 | 2024-02-27 | Imagars Llc | Machine learning to accelerate alloy design |
| US11386541B2 (en) * | 2019-08-22 | 2022-07-12 | Saudi Arabian Oil Company | System and method for cyber-physical inspection and monitoring of nonmetallic structures |
-
2021
- 2021-05-31 CN CN202110601155.0A patent/CN113343566B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709640A (en) * | 2016-12-15 | 2017-05-24 | 华南理工大学 | Method for predicting ship energy consumption based on deep learning and support vector regression |
| CN109543352A (en) * | 2018-12-29 | 2019-03-29 | 西南石油大学 | Shale fracture toughness prediction technique, device and electronic equipment |
| WO2020144417A1 (en) * | 2019-01-09 | 2020-07-16 | Constellium Issoire | Method for checking a damage tolerance property of a part made of an aluminium alloy |
| CN110378035A (en) * | 2019-07-19 | 2019-10-25 | 南京工业大学 | Hydrocracking soft measurement modeling method based on deep learning |
| CN110852273A (en) * | 2019-11-12 | 2020-02-28 | 重庆大学 | Behavior identification method based on reinforcement learning attention mechanism |
| CN110929958A (en) * | 2019-12-10 | 2020-03-27 | 西安邮电大学 | Short-term traffic flow prediction method based on deep learning parameter optimization |
| CN111210633A (en) * | 2020-02-09 | 2020-05-29 | 北京工业大学 | Short-term traffic flow prediction method based on deep learning |
| CN111341386A (en) * | 2020-02-17 | 2020-06-26 | 大连理工大学 | Attention-introducing multi-scale CNN-BilSTM non-coding RNA interaction relation prediction method |
| CN112597921A (en) * | 2020-12-28 | 2021-04-02 | 杭州电子科技大学 | Human behavior recognition method based on attention mechanism GRU deep learning |
| CN112668804A (en) * | 2021-01-11 | 2021-04-16 | 中国海洋大学 | Method for predicting broken track of ground wave radar ship |
Non-Patent Citations (1)
| Title |
|---|
| 基于注意力机制BiLSTM的设备智能故障诊断方法;王太勇 等;天津大学学报(自然科学与工程技术版);第53卷(第06期);601-608 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113343566A (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhan et al. | Machine learning based fatigue life prediction with effects of additive manufacturing process parameters for printed SS 316L | |
| Mahalle et al. | Neural network modeling for anisotropic mechanical properties and work hardening behavior of Inconel 718 alloy at elevated temperatures | |
| Anijdan et al. | Flow stress optimization for 304 stainless steel under cold and warm compression by artificial neural network and genetic algorithm | |
| Wang et al. | High-throughput map design of creep life in low-alloy steels by integrating machine learning with a genetic algorithm | |
| CN110472349A (en) | A kind of hot-rolled steel performance prediction method based on EEMD and depth convolutional network | |
| CN113073267B (en) | High-strength high-temperature-resistant RAFM steel and machine learning-based design method thereof | |
| Wei et al. | On the use of transfer modeling to design new steels with excellent rotating bending fatigue resistance even in the case of very small calibration datasets | |
| Quan et al. | Artificial neural network modeling to evaluate the dynamic flow stress of 7050 aluminum alloy | |
| CN113343566B (en) | Nickel-based alloy fracture toughness prediction method and system based on deep learning | |
| Mohanty et al. | Application of artificial neural network for fatigue life prediction under interspersed mode-I spike overload | |
| CN114897227A (en) | Multi-steel-grade mechanical property forecasting method based on improved random forest algorithm | |
| Jinlong et al. | VHCF evaluation with BP neural network for centrifugal impeller material affected by internal inclusion and GBF region | |
| McBride et al. | Modelling tensile properties of gamma-based titanium aluminides using artificial neural network | |
| Jenab et al. | The use of ANN to predict the hot deformation behavior of AA7075 at low strain rates | |
| Wen et al. | Machine learning-assisted constitutive modeling of a novel powder metallurgy superalloy | |
| Tan et al. | Creep lifetime prediction of 9% Cr martensitic heat-resistant steel based on ensemble learning method | |
| CN106202937B (en) | M50 steel forgings make carbide size prediction technique in tissue | |
| Zhang et al. | Fatigue database of complex metallic alloys | |
| Mohanty et al. | An adaptive neuro fuzzy inference system model for studying free in plane and out of plane vibration behavior of curved beams | |
| Sun et al. | Prediction of tensile property of hydrogenated Ti600 titanium alloy using artificial neural network | |
| Xu et al. | Prediction of static globularization of Ti-17 alloy with starting lamellar microstructure during heat treatment | |
| Fang et al. | Machine learning assisted materials design of high-speed railway wheel with better fatigue performance | |
| Chen et al. | Power-exponent function model for low-cycle fatigue life prediction and its applications–Part I: Models and validations | |
| Wei et al. | High-cycle fatigue SN curve prediction of steels based on a transfer learning-guided convolutional neural network | |
| Chen et al. | Modeling of flow stress of high titanium content 6061 aluminum alloy under hot compression |
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 |