CN112464541B - Multi-scale uncertainty considered mixed composite material layering method - Google Patents

Multi-scale uncertainty considered mixed composite material layering method Download PDF

Info

Publication number
CN112464541B
CN112464541B CN202011504349.0A CN202011504349A CN112464541B CN 112464541 B CN112464541 B CN 112464541B CN 202011504349 A CN202011504349 A CN 202011504349A CN 112464541 B CN112464541 B CN 112464541B
Authority
CN
China
Prior art keywords
layer
alternative
parameters
materials
neural network
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
Application number
CN202011504349.0A
Other languages
Chinese (zh)
Other versions
CN112464541A (en
Inventor
彭翔
郭玉良
王明博
高启龙
李吉泉
姜少飞
景立挺
吴化平
张征
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202011504349.0A priority Critical patent/CN112464541B/en
Publication of CN112464541A publication Critical patent/CN112464541A/en
Application granted granted Critical
Publication of CN112464541B publication Critical patent/CN112464541B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/26—Composites

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Geometry (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computer Hardware Design (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses a mixed composite material layering method considering multi-scale uncertainty. Selecting an alternative material set, obtaining microscopic attributes of the alternative material set, and obtaining distribution types of macroscopic attributes of the alternative material set through a multi-scale analysis model; determining a macroscopic finite element model of the alternative material set by utilizing macroscopic attributes and distribution types thereof; constructing a neural network model of microscopic attributes and mixed composite material parameters of the candidate material set, calculating a mean square error MSE for the trained neural network model, judging the accuracy of the trained neural network model according to the mean square error MSE, and optimizing the selection of the candidate material set and the selection of the layer stacking sequence by using a genetic algorithm. The method of the invention uses a plurality of optimization algorithms, greatly reduces the burden of designers and computers, greatly improves the calculation speed, improves the optimization accuracy, and simultaneously accords with the engineering practical application.

Description

Multi-scale uncertainty considered mixed composite material layering method
Technical Field
The invention belongs to the field of methods for optimizing stacking sequence and material distribution of fiber reinforced composite materials, and particularly relates to a multi-scale uncertainty layering optimization method for a mixed composite material.
Background
Fiber reinforced composites have the advantages of high strength to weight ratio, high stiffness to weight ratio and low cost under various loads and grow exponentially in automotive, marine, civilian, aerospace, construction applications. In contrast to conventional composites composed of single fibers and a single matrix, hybrid composites are manufactured by combining two or more fibers into one matrix. The performance of the hybrid composite is a weighted sum of the individual components, and the advantage of one fiber can compensate for the lack of characteristics of other types of components in the hybrid composite. Meanwhile, aiming at multi-scale analysis of composite materials, certain research results exist at home and abroad, and stability and reliability of micro-scale analysis output results can be realized. In addition, in actual engineering application, the multi-scale analysis is more in line with the actual application of the composite material, and the analysis result is more matched with the actual use condition.
The specific performance parameters are optimized by carrying out proper multi-scale uncertainty engineering optimization design on the mixed composite material, so that the practical application range of the mixed composite material can be better expanded, and the application prospect of the mixed composite material is expanded.
Disclosure of Invention
The invention aims to solve the problem that the design process of a mixed composite material is complex and tedious, and provides a simple multi-scale uncertainty mixed composite material layering arrangement method based on stacking sequence and material distribution.
The technical scheme of the invention is as follows:
1) Selecting a plurality of different composite materials as an alternative material set for a hybrid composite material Wherein/>Representing the first alternative material,/>Representing a second alternative material,/>Representing a third alternative material,/>Represents the n-th candidate material, n represents the sum of the candidate materials, and the sum of the candidate materials/>Uncertainty quantization is carried out on microscopic attributes of the material, and an alternative material set/> is obtained after uncertainty quantizationWherein the set of candidate materials/>, is of a distribution type of microscopic propertiesIncluding microscopic elastic properties and microscopic strength properties;
2) Based on the set of alternative materials obtained in step 1) By alternative material set/>Is to collect/join the candidate materials into a multi-scale analysis modelIs propagated to the macro-level to obtain a set of alternative materialsThe distribution type of macro elastic attribute and macro strength attribute;
3) Obtaining a set of alternative materials After the distribution type of the macro elastic attribute and the macro strength attribute, the macro elastic attribute and the macro strength attribute and the distribution type thereof are utilized to determine the set of the alternative materialsFor a macroscopic finite element model of an alternative material set/>And alternative Material set/>Latin hypercube sampling is carried out on a layer stack sequence (theta 1,θ2,θ3、、、θx) of the macroscopic finite element model, and then a laminated plate method is used for collecting/carrying out alternative materialsIs analyzed to obtain a set of candidate materials/>Performance parameters/>Failure parameters FI and overall cost parameters C; wherein θ 1 represents the stacking angle of the first layer material in the hybrid composite; θ 2 represents the stacking angle of the second layer material in the hybrid composite; θ 3 represents the stacking angle of the third layer material in the hybrid composite; θ x represents the stacking angle of the xth layer material in the hybrid composite, x represents the sum of the layer stacking sequences;
4) Build alternative material set The built neural network model comprises an input layer, an intermediate layer and an output layer, wherein the input layer inputs a required candidate material set/>Desired set of alternative materials/>Micro elastic properties, micro strength properties and layer stack sequence (θ 1,θ2,θ3、、、θx),/>, of the respective candidate materialsRepresenting the first alternative material,/>Representing a second alternative material of interest,/>Representing a third alternative material required,/>Represents the required mth candidate material, m represents the sum of the required candidate materials; output layer output Performance parameter/>Failure parameters FI and overall cost parameters C;
5) Calculating a mean square error MSE for the trained neural network model;
6) Judging the accuracy of the trained neural network model according to the mean square error MSE;
If the obtained MSE is smaller than a preset MSE threshold, the built neural network model is accurate, and step 7) is performed;
if the obtained mean square error MSE is greater than or equal to a mean square error threshold, the constructed neural network model is inaccurate, and then parameter learning is carried out on the neural network model based on the performance prediction variance, and then the step 5) is returned;
7) Using Genetic Algorithm (GA) for candidate Material set And (3) selecting a layer stacking sequence (theta 1,θ2,θ3、、、θx) to obtain the material selection and the stacking angle of each layer in the mixed composite material layer.
In the step 1), uncertainty quantization is performed on microscopic attributes of the candidate material set, specifically:
First, the candidate materials are assembled The various attribute parameters in the micro elastic attribute and the micro intensity attribute are regarded as different sparse variables y, and the probability density function f y(y|ε,θk of the sparse variables y expressed by the following formula is obtained through processing statistics under different distribution types theta k and different distribution parameters epsilon of sampling distribution of the sparse variables y:
Wherein L (ε, θ k) is the likelihood estimate of the sparse variable y under the distribution type θ k and the distribution parameter ε, ε L (ε, θ k) dε is the integral of the likelihood estimate of the sparse variable y with respect to the distribution parameter ε, θ k represents the kth distribution type;
Obtaining a maximum likelihood estimated value L max(ε,θk of the sparse variable y according to statistics of all likelihood estimated values of the sparse variable y under different distribution types theta k and different distribution parameters epsilon, and then calculating information entropy AIC k of different distribution types theta k by adopting a minimum information criterion method:
AICk＝2numk-2ln(Lmax(ε,θk))
Where num k represents the number of distribution parameters ε, L max(ε,θk) represents the maximum likelihood estimation value of the sparse variable y under the distribution type θ k and the distribution parameters ε, AIC k represents the information entropy of the kth distribution type, and k represents the ordinal number of the distribution type;
Then, the probability values P θ_k of different distribution types are obtained using the following formula:
Pθ_k＝exp((AICmin-AICk)/2)
Wherein AIC min represents the minimum information entropy;
Finally, the distribution type theta k with the probability value P θ_k being larger than the distribution type probability threshold value P Δ is used as the distribution type of the sparse variable y, namely the distribution type of the microscopic attribute of the candidate material set corresponding to the sparse variable y.
In said step 7), a Genetic Algorithm (GA) is used for the collection of alternative materialsThe selection of the layer stack sequence (θ 1,θ2,θ3、、、θx) and the selection of the layer stack sequence (θ 1,θ2,θ3、、、θx), in particular: the optimization variables are the layer stack sequence (θ 1,θ2,θ3、、、θx) and the set of alternative materials/>
Output of obtained performance parameters using accurate neural network modelAnd construct performance parameters/>Is a robust optimization function f 1:
θ＝(θ1,θ2,θ3、、、θx)
wherein mean represents taking the mean, std represents taking the standard deviation, p is the weight coefficient, Representing a final set of schemes for alternative material selection, θ representing a final scheme for the layer stack sequence; θ 1 represents the stacking angle of the first layer material in the hybrid composite; θ 2 represents the stacking angle of the second layer material in the hybrid composite; θ 3 represents the stacking angle of the third layer material in the hybrid composite; θ x represents the stacking angle of the xth layer material in the hybrid composite, x represents the sum of the layer stacking sequences; representing the first material in the final set of schemes,/> Representing the second material in the final set of schemes,/>Representing the third material in the final set of schemes,/>Represents the j-th material in the final schema set, j representing the sum of the materials in the final schema set;
by penalty function method and establishing the following penalty objective function
Wherein λ represents a penalty parameter, α represents a small positive tolerance coefficient, g represents an inequality constraint of a failure parameter FI and an overall cost parameter C, and h represents a force balance constraint of the hybrid composite;
simultaneously setting constraint conditions of failure parameters FI and overall cost parameters C; to punish the objective function Maximizing the objective solution to obtain the final set of schemes for alternative material selection/>And two parameter results of the final scheme θ of the layer stack sequence;
in the step 6), the method for learning the neural network model parameters based on the performance prediction variance comprises the following specific steps:
6.1 A) the set of candidate materials required in the neural network model input from step 4) The micro elastic property, the micro strength property, the layer stacking sequence (theta 1,θ2,θ3、、、θx) and the required alternative material set of each alternative material are randomly selected to form an input data set as a sample point; the sample points are predicted and obtained through neural network model processing to obtain the performance parameter/>, of each sample pointFailure parameters FI and overall cost parameters C;
6.2 Repeating step 6.1) for multiple times to obtain multiple sample points and corresponding performance parameters Failure parameters FI and overall cost parameters C;
6.3 Using a formula to calculate a characteristic value gamma i of each sample point:
γi＝ωilog(u(i))
u(i)＝rand(0,1)
Wherein ω i is a performance parameter value obtained by calculating the ith sample point, γ i represents a characteristic value of the ith sample point, rand (0, 1) represents a random number between 0 and 1, and u (i) represents a random number corresponding to the ith sample point;
6.4 Selecting a sample point with the maximum characteristic value, and retraining the neural network model by using an input data set of the sample point so as to update parameters of the neural network model;
6.5 Returning to step 5).
The beneficial effects of the invention are as follows:
1) The invention optimizes the stacking sequence and material distribution of the mixed composite material from the microscopic scale, and compared with the mixed composite material optimized from the macroscopic scale, the invention has more accurate analysis and better accords with engineering practical application.
2) The single composite material is extremely complex in design and manufacture due to unique composition components, and further, the performance of a computer is very tested in the aspect of computer design optimization, but the mixed composite material is more complex than the single composite material, so that various optimization algorithms are used, the burden of a designer and the computer is greatly reduced, and meanwhile, the optimization accuracy is increased.
3) Compared with the traditional Genetic Algorithm (GA), the novel punishment objective function is provided, and aims to reduce optimization variables which do not meet constraint conditions, so that the calculation load of a computer is further reduced, the calculation speed is greatly improved, and the optimization accuracy is improved.
In a word, the method of the invention uses a plurality of optimization algorithms, thereby greatly reducing the burden of designers and computers, greatly improving the calculation speed, improving the optimization accuracy and being more suitable for engineering practical application.
Drawings
FIG. 1 is a schematic diagram of a hybrid composite model according to an embodiment of the present invention;
FIG. 2 is an iterative process of stacking order and material patch optimization in accordance with an embodiment of the present invention;
FIG. 3 is a finite element analysis flow of a hybrid composite material according to the present invention;
FIG. 4 is a flow chart of the hybrid composite optimization in the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and examples, but the scope of the invention is not limited thereto, and the method of the invention specifically comprises the following steps, as shown in fig. 3 and 4:
1. Determining basic information of the mixed composite material to be analyzed: the boundary condition is nine layers of mixed composite materials with four sides simply supported, the length L=0.508 m, the width W=0.406 m, and the thickness of each layer is deltaT=0.125 mm. The maximum overall cost parameter c=9.5 USD, the minimum value of failure probability is The relationship between the reliability parameter P f and the failure parameter FI is: p f = P [1-FI ]. Three alternative materials were selected: material I, material II and material III will be used in subsequent calculations to represent material I by "1", material II by "2" and material III by "3"; the layer stack sequence is selected from the angles [ -45,45,0,90] and recombined. Performance parameters/>Is defined as the natural frequency F f. An example of modeling a hybrid composite material is shown in fig. 1.
2. Uncertainty quantization is performed on microscopic properties of alternative materials (material I, material II, material III): the method specifically comprises the following steps:
Firstly, regarding each attribute parameter in the microscopic elastic attribute and the microscopic strength attribute of the candidate materials (material I, material II and material III) as different sparse variables y, and processing statistics under different distribution types theta k and different distribution parameters epsilon of sampling distribution of the sparse variables y to obtain a probability density function f y(y|ε,θk of the sparse variables y expressed by the following formula:
Wherein L (ε, θ k) is the likelihood estimate of the sparse variable y under the distribution type θ k and the distribution parameter ε, ε L (ε, θ k) dε is the integral of the likelihood estimate of the sparse variable y with respect to the distribution parameter ε, θ k represents the kth distribution type;
Obtaining a maximum likelihood estimated value L max(ε,θk of the sparse variable y according to statistics of all likelihood estimated values of the sparse variable y under different distribution types theta k and different distribution parameters epsilon, and then calculating information entropy AIC k of different distribution types theta k by adopting a minimum information criterion method:
AICk＝2numk-2ln(Lmax(ε,θk))
Where num k represents the number of distribution parameters ε, L max(ε,θk) represents the maximum likelihood estimation value of the sparse variable y under the distribution type θ k and the distribution parameters ε, AIC k represents the information entropy of the kth distribution type, and k represents the ordinal number of the distribution type;
Then, the probability values P θ_k of different distribution types are obtained using the following formula:
PA_k＝exp((AICmin-AICk)/2)
Wherein AIC min represents the minimum information entropy;
Finally, the distribution type theta k with the probability value P θ_k being larger than the distribution type probability threshold value P Δ is used as the distribution type of the sparse variable y, namely the distribution type of the microscopic attribute of the candidate material set corresponding to the sparse variable y.
And obtaining the distribution type of the microscopic attributes of the alternative materials (material I, material II and material III) after the uncertainty is quantified, wherein the microscopic attributes of the alternative materials (material I, material II and material III) have microscopic elastic attributes and microscopic strength attributes.
3. Based on the distribution type of the microscopic properties of the alternative materials (materials I, II and III) obtained in the step 2, the uncertainty of the microscopic properties of the alternative materials (materials I, II and III) is transmitted to a macroscopic level through a multi-scale analysis model of the alternative materials (materials I, II and III), and the distribution type of the macroscopic elastic properties and the macroscopic strength properties of the alternative materials (materials I, II and III) is obtained;
4. After obtaining the distribution types of macroscopic elasticity properties and macroscopic strength properties of the alternative materials (materials I, II and III), determining macroscopic finite element models of the alternative materials (materials I, II and III) by utilizing the macroscopic elasticity properties, the macroscopic strength properties and the distribution types thereof, carrying out Latin hypercube sampling on a layer stack sequence (theta 1,θ2,θ3、、、Ax) of the macroscopic finite element models of the alternative materials (materials I, II and III) and the macroscopic finite element models of the alternative materials (materials I, II and III), and then analyzing the macroscopic finite element models of the alternative materials (materials I, II and III) by a laminate method to obtain the inherent frequency F f, the failure parameter FI and the overall cost parameter C of the alternative materials (materials I, II and III) after analysis;
5. Constructing a neural network model of microscopic properties of alternative materials (materials I, II and III) and parameters of mixed composite materials, wherein the constructed neural network model comprises an input layer, an intermediate layer and an output layer, the input layer inputs the alternative materials (materials I, II and III), and the microscopic elastic properties, the microscopic strength properties and the layer stacking sequence (theta 1,θ2,θ3、、、θx) of each alternative material in the alternative materials (materials I, II and III); the output layer outputs a natural frequency F f, a failure parameter FI and an overall cost parameter C;
6. calculating a mean square error MSE for the trained neural network model;
7. judging the accuracy of the trained neural network model according to the mean square error MSE;
if the obtained MSE is smaller than a preset MSE threshold, the built neural network model is accurate, and step 8) is performed;
if the obtained mean square error MSE is greater than or equal to a mean square error threshold, the constructed neural network model is inaccurate, and then parameter learning is carried out on the neural network model based on the performance prediction variance, and then the step 6) is returned; the method comprises the following specific steps:
7.1 Forming an input data set as a sample point after randomly selecting each of the micro elastic property, the micro strength property and the layer stacking sequence (theta 1,θ2,θ3、、、θx) of each of the candidate materials (material I, material II and material III) in the neural network model input in the step 5); sample points are predicted to obtain performance parameters of each sample point through neural network model processing Failure parameters FI and overall cost parameters C;
7.2 Repeating the step 7.1) for a plurality of times to obtain a plurality of sample points and corresponding natural frequencies F f, failure parameters FI and overall cost parameters C;
7.3 Using a formula to calculate a characteristic value gamma i of each sample point:
γi＝ωilog(u(i))
u(i)＝rand(0,1)
Wherein ω i is a performance parameter value obtained by calculating the ith sample point, γ i represents a characteristic value of the ith sample point, rand (0, 1) represents a random number between 0 and 1, and u (i) represents a random number corresponding to the ith sample point;
7.4 Selecting a sample point with the maximum characteristic value, and retraining the neural network model by using an input data set of the sample point so as to update parameters of the neural network model;
7.5 Returning to step 6).
8. And optimizing the selection of the alternative materials (materials I, II and III) and the selection of the layer stacking sequence (theta 1,θ2,A3、、、Ax) by using a Genetic Algorithm (GA) to obtain the selection of each layer material and each layer stacking angle in the mixed composite material layering. The method specifically comprises the following steps: the optimization variables are a layer stack sequence (theta 1,θ2,θ3、、、θx) and alternative materials (materials I, II and III);
output of obtained performance parameters using accurate neural network model And constructs a robust optimization function F 1 of the natural frequency F f:
θ＝(θ1,θ2,θ3、、、θx)
wherein mean represents taking the mean, std represents taking the standard deviation, p is the weight coefficient, Representing a final set of schemes selected for the alternative materials (material i, material ii, material iii), θ representing the final scheme of the layer stack sequence; θ 1 represents the stacking angle of the first layer material in the hybrid composite; θ 2 represents the stacking angle of the second layer material in the hybrid composite; θ 3 represents the stacking angle of the third layer material in the hybrid composite; θ x represents the stacking angle of the xth layer material in the hybrid composite, x represents the sum of the layer stacking sequences; to reduce the choice of unsuitable constraints, the layer stack sequence (θ 1,θ2,θ3、、、θx) and the candidate materials (Material I, material II, material III) are subjected to a penalty function method, while the following penalty objective function/>, is established
Wherein λ represents a penalty parameter, α represents a small positive tolerance coefficient, g represents an inequality constraint of a failure parameter FI and an overall cost parameter C, and h represents a force balance constraint of the hybrid composite; in the specific implementation, the punishment parameter lambda is set to be-100, and the small positive tolerance coefficient alpha is set to be 10 -4, so that the time required by optimizing a computer is shortened;
simultaneously setting constraint conditions of failure parameters FI and overall cost parameters C; to punish the objective function Maximizing the objective solution to obtain the final set of schemes for alternative material selection/>And two parameter results of the final scheme a of the layer stack sequence.
By means of the adaptive genetic algorithm in step 8, the layer stack sequence (θ 1,θ2,θ3、、、θx) is iterated continuously with the alternative composite materials (material I, material II, material III), resulting in an optimal layer stack sequence [ -45/-45/45/-45/90/-45/45/-45/-45] and alternative composite material distribution [ 1] in this example; 1, a step of; 2;1, a step of; 1], the corresponding natural frequency value is 36.58Hz, as shown in figure 2.
What has been described in this specification is merely an enumeration of possible forms of implementation for the inventive concept, and the scope of protection of the present invention should not be construed as limited to the specific forms set forth in the examples, nor is it intended that the scope of protection of the present invention be limited to only equivalent technical means as would occur to those skilled in the art based on the inventive concept.

Claims (3)

1. A hybrid composite lay-up method taking into account multi-scale uncertainty, characterized by: the method specifically comprises the following steps:
1) Selecting a plurality of different composite materials as an alternative material set for a hybrid composite material Wherein/>Representing the first alternative material,/>Representing a second alternative material,/>Representing a third alternative material,/>Represents the n-th candidate material, n represents the sum of the candidate materials, and the sum of the candidate materials/>Uncertainty quantization is carried out on microscopic attributes of the material, and an alternative material set/> is obtained after uncertainty quantizationWherein the set of candidate materials/>, is of a distribution type of microscopic propertiesIncluding microscopic elastic properties and microscopic strength properties;
2) Based on the set of alternative materials obtained in step 1) By alternative material set/>Is to collect/join the candidate materials into a multi-scale analysis modelIs propagated to the macro-level to obtain the candidate material set/>The distribution type of macro elastic attribute and macro strength attribute;
3) Obtaining a set of alternative materials After the distribution type of the macro elastic attribute and the macro strength attribute, the macro elastic attribute and the macro strength attribute and the distribution type thereof are utilized to determine the set of the alternative materialsFor a macroscopic finite element model of an alternative material set/>And alternative Material set/>Latin hypercube sampling is carried out on a layer stack sequence (theta 1,θ2,θ3、、、θx) of the macroscopic finite element model, and then a laminated plate method is used for collecting/carrying out alternative materialsIs analyzed to obtain a set of candidate materials/>Performance parameters/>Failure parameters FI and overall cost parameters C; wherein θ 1 represents the stacking angle of the first layer material in the hybrid composite; θ 2 represents the stacking angle of the second layer material in the hybrid composite; θ 3 represents the stacking angle of the third layer material in the hybrid composite; θ x represents the stacking angle of the xth layer material in the hybrid composite, x represents the sum of the layer stacking sequences;
4) Build alternative material set The built neural network model comprises an input layer, an intermediate layer and an output layer, wherein the input layer inputs a required candidate material set/>The required set of alternative materialsMicro elastic properties, micro strength properties and layer stack sequence (θ 1,θ2,θ3、、、θx),/>, of the respective candidate materialsRepresenting the first alternative material,/>Representing a second alternative material of interest,/>Representing a third alternative material required,/>Represents the required mth candidate material, m represents the sum of the required candidate materials; output layer output Performance parameter/>Failure parameters FI and overall cost parameters C;
5) Calculating a mean square error MSE for the trained neural network model;
6) Judging the accuracy of the trained neural network model according to the mean square error MSE;
If the obtained MSE is smaller than a preset MSE threshold, the built neural network model is accurate, and step 7) is performed;
if the obtained mean square error MSE is greater than or equal to a mean square error threshold, the constructed neural network model is inaccurate, and then parameter learning is carried out on the neural network model based on the performance prediction variance, and then the step 5) is returned;
7) Use of genetic algorithm GA for candidate Material set Optimizing the selection of each layer of material and each layer of stacking angle in the mixed composite material layer, and finally, performing mixed composite material layer according to the selection of each layer of material and each layer of stacking angle in the mixed composite material layer;
in said step 7), a genetic algorithm GA is used for the collection of alternative materials The selection of the layer stack sequence (θ 1,θ2,θ3、、、θx) and the selection of the layer stack sequence (θ 1,θ2,θ3、、、θx), in particular: the optimization variables are the layer stack sequence (θ 1,θ2,θ3、、、θx) and the set of alternative materials/>
Output of obtained performance parameters using accurate neural network modelAnd construct performance parameters/>Is a robust optimization function f 1:
θ＝(θ1,θ2,θ3、、、θx)
wherein mean represents taking the mean, std represents taking the standard deviation, p is the weight coefficient, Representing a final set of schemes for alternative material selection, θ representing a final scheme for the layer stack sequence; θ 1 represents the stacking angle of the first layer material in the hybrid composite; θ 2 represents the stacking angle of the second layer material in the hybrid composite; θ 3 represents the stacking angle of the third layer material in the hybrid composite; θ x represents the stacking angle of the xth layer material in the hybrid composite, x represents the sum of the layer stacking sequences; representing the first material in the final set of schemes,/> Representing the second material in the final set of schemes,/>Representing the third material in the final set of schemes,/>Represents the j-th material in the final schema set, j representing the sum of the materials in the final schema set;
by penalty function method and establishing the following penalty objective function
Wherein λ represents a penalty parameter, α represents a small positive tolerance coefficient, g represents an inequality constraint of a failure parameter FI and an overall cost parameter C, and h represents a force balance constraint of the hybrid composite;
To punish the objective function Maximizing the objective solution to obtain the final set of schemes for alternative material selection/>And two parameter results of the final scheme θ of the layer stack sequence.
2. A hybrid composite lay-up method taking into account multi-scale uncertainty as defined in claim 1, wherein:
In the step 1), uncertainty quantization is performed on microscopic attributes of the candidate material set, specifically:
First, the candidate materials are assembled The various attribute parameters in the micro elastic attribute and the micro intensity attribute are regarded as different sparse variables y, and the probability density function f y(y|ε,θk of the sparse variables y expressed by the following formula is obtained through processing statistics under different distribution types theta k and different distribution parameters epsilon of sampling distribution of the sparse variables y:
Wherein L (ε, θ k) is the likelihood estimate of the sparse variable y under the distribution type θ k and the distribution parameter ε, ε L (ε, θ k) dε is the integral of the likelihood estimate of the sparse variable y with respect to the distribution parameter ε, θ k represents the kth distribution type;
Obtaining a maximum likelihood estimated value L max(ε,θk of the sparse variable y according to statistics of all likelihood estimated values of the sparse variable y under different distribution types theta k and different distribution parameters epsilon, and then calculating information entropy AIC k of different distribution types theta k by adopting a minimum information criterion method:
AICk＝2numk-2ln(Lmax(ε,θk))
Wherein num k represents the number of distribution parameters ε, L max(ε,θk) represents the maximum likelihood estimation value of the sparse variable y under the distribution type θ and the distribution parameters ε, AIC k represents the information entropy of the kth distribution type, and k represents the ordinal number of the distribution type;
Then, the probability values P θ_k of different distribution types are obtained using the following formula:
Pθ_k＝exp((AICmin-AICk)/2)
Wherein AIC min represents the minimum information entropy;
Finally, the distribution type theta k with the probability value P θ_k being larger than the distribution type probability threshold value P Δ is used as the distribution type of the sparse variable y, namely the distribution type of the microscopic attribute of the candidate material set corresponding to the sparse variable y.
3. A hybrid composite lay-up method taking into account multi-scale uncertainty as defined in claim 1, wherein:
in the step 6), the method for learning the neural network model parameters based on the performance prediction variance comprises the following specific steps:
6.1 A) the set of candidate materials required in the neural network model input from step 4) The micro elastic property, the micro strength property, the layer stacking sequence (theta 1,θ2,θ3、、、θx) and the required alternative material set of each alternative material are randomly selected to form an input data set as a sample point; the sample points are predicted and obtained through neural network model processing to obtain the performance parameter/>, of each sample pointFailure parameters FI and overall cost parameters C;
6.2 Repeating step 6.1) for multiple times to obtain multiple sample points and corresponding performance parameters Failure parameters FI and overall cost parameters C;
6.3 Using a formula to calculate a characteristic value gamma i of each sample point:
γi＝ωilog(u(i))
u(i)＝rand(0,1)
Wherein ω i is a performance parameter value obtained by calculating the ith sample point, γ i represents a characteristic value of the ith sample point, rand (0, 1) represents a random number between 0 and 1, and u (i) represents a random number corresponding to the ith sample point;
6.4 Selecting a sample point with the maximum characteristic value, and retraining the neural network model by using an input data set of the sample point so as to update parameters of the neural network model;
6.5 Returning to step 5).
CN202011504349.0A 2020-12-18 2020-12-18 Multi-scale uncertainty considered mixed composite material layering method Active CN112464541B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011504349.0A CN112464541B (en) 2020-12-18 2020-12-18 Multi-scale uncertainty considered mixed composite material layering method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011504349.0A CN112464541B (en) 2020-12-18 2020-12-18 Multi-scale uncertainty considered mixed composite material layering method

Publications (2)

Publication Number Publication Date
CN112464541A CN112464541A (en) 2021-03-09
CN112464541B true CN112464541B (en) 2024-05-24

Family

ID=74804791

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011504349.0A Active CN112464541B (en) 2020-12-18 2020-12-18 Multi-scale uncertainty considered mixed composite material layering method

Country Status (1)

Country Link
CN (1) CN112464541B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106991212A (en) * 2017-03-07 2017-07-28 西安交通大学 A kind of blade root Strength Safety characteristic prediction method for optimizing GRNN network algorithms based on GA_PSO
CN110083900A (en) * 2019-04-17 2019-08-02 大连理工大学 A kind of fast synergistic optimization method towards fiber hybrid composite plate and shell structure

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106033555A (en) * 2015-03-13 2016-10-19 中国科学院声学研究所 Big data processing method based on depth learning model satisfying K-dimensional sparsity constraint

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106991212A (en) * 2017-03-07 2017-07-28 西安交通大学 A kind of blade root Strength Safety characteristic prediction method for optimizing GRNN network algorithms based on GA_PSO
CN110083900A (en) * 2019-04-17 2019-08-02 大连理工大学 A kind of fast synergistic optimization method towards fiber hybrid composite plate and shell structure

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
试验设计法与BP-GA算法在复合材料结构优化中的应用;李洪波;徐元铭;黄英兰;;飞机设计;20070615(03);全文 *

Also Published As

Publication number Publication date
CN112464541A (en) 2021-03-09

Similar Documents

Publication Publication Date Title
CN109918708B (en) Material performance prediction model construction method based on heterogeneous ensemble learning
CN107886161A (en) A kind of global sensitivity analysis method for improving Complex Information System efficiency
CN110544011B (en) Intelligent system combat effectiveness evaluation and optimization method
CN110070145A (en) LSTM wheel hub single-item energy consumption prediction based on increment cluster
CN106056127A (en) GPR (gaussian process regression) online soft measurement method with model updating
CN111461286B (en) Spark parameter automatic optimization system and method based on evolutionary neural network
CN103226728B (en) High density polyethylene polymerization cascade course of reaction Intelligent Measurement and yield optimization method
CN111832839B (en) Energy consumption prediction method based on sufficient incremental learning
CN113103535A (en) GA-ELM-GA-based injection molding part mold parameter optimization method
CN113887694A (en) Click rate estimation model based on characteristic representation under attention mechanism
CN114706840A (en) Load perception-oriented method for optimizing parameters of ArangoDB of multi-mode database
CN111598313A (en) Wind power prediction method
Chen et al. Distributed text feature selection based on bat algorithm optimization
Wang et al. Data-driven inverse method with uncertainties for path parameters of variable stiffness composite laminates
CN114564787A (en) Bayesian optimization method, device and storage medium for target-related airfoil design
CN112464541B (en) Multi-scale uncertainty considered mixed composite material layering method
CN111310974A (en) Short-term water demand prediction method based on GA-ELM
CN116258234A (en) BP neural network model-based energy enterprise carbon emission measuring and predicting method
CN113011014B (en) Composite material layering optimization method and system
Liu et al. Warpage optimization of the GFRP injection molding process parameters
Tian et al. An active learning Kriging‐based multipoint sampling strategy for structural reliability analysis
Liao et al. Multiobjective design of equivalent accelerated life testing plans
Yang et al. Power demand forecast based on optimized neural networks by improved genetic algorithm
Shonkwiler et al. Deep reinforcement learning for stacking sequence optimization of composite laminates
CN117828744A (en) BP neural network-based wing structure optimization design method

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