CN112507625A - Method for predicting deformation of stratospheric airship skin material by utilizing neural network - Google Patents
Method for predicting deformation of stratospheric airship skin material by utilizing neural network Download PDFInfo
- Publication number
- CN112507625A CN112507625A CN202011515090.XA CN202011515090A CN112507625A CN 112507625 A CN112507625 A CN 112507625A CN 202011515090 A CN202011515090 A CN 202011515090A CN 112507625 A CN112507625 A CN 112507625A
- Authority
- CN
- China
- Prior art keywords
- neural network
- skin material
- deformation
- stress
- predicting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000000463 material Substances 0.000 title claims abstract description 103
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 65
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012360 testing method Methods 0.000 claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000006870 function Effects 0.000 claims description 38
- 210000002569 neuron Anatomy 0.000 claims description 25
- 230000006399 behavior Effects 0.000 claims description 18
- 238000003062 neural network model Methods 0.000 claims description 12
- 239000005437 stratosphere Substances 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 7
- 239000002131 composite material Substances 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004519 manufacturing process Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000009941 weaving Methods 0.000 claims description 2
- 238000013531 bayesian neural network Methods 0.000 abstract description 18
- 238000013461 design Methods 0.000 abstract description 8
- 238000004458 analytical method Methods 0.000 abstract description 4
- 239000010410 layer Substances 0.000 description 49
- 238000009864 tensile test Methods 0.000 description 15
- 238000004422 calculation algorithm Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 5
- 238000007781 pre-processing Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012897 Levenberg–Marquardt algorithm Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 239000003733 fiber-reinforced composite Substances 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000013329 compounding Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010030 laminating Methods 0.000 description 1
- 239000002346 layers by function Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
本发明涉及一种利用神经网络预测平流层飞艇蒙皮材料变形的方法，属于损伤分析技术领域。本发明实现方法为：针对平流层飞艇蒙皮材料在复杂工作环境下的真实变形接近非比例双轴拉伸，开展多种应力比条件下的双轴拉伸试验，收集神经网络所需的训练样本数据；构建包含输入层、隐含层和输出层的贝叶斯神经网络，从而建立蒙皮材料的变形行为简单表述模型；采用训练好的神经网络对蒙皮材料的变形行为进行实时预测。本发明提供的一种利用神经网络预测平流层飞艇蒙皮材料变形的方法预测精度较高，稳定性较好，推广能力强，能够满足实时、准确预测蒙皮材料变形行为的要求，为优化材料设计和指导平流层飞艇结构设计提供新的方法。
The invention relates to a method for predicting the deformation of a stratospheric airship skin material by using a neural network, and belongs to the technical field of damage analysis. The realization method of the present invention is as follows: aiming at the real deformation of the stratospheric airship skin material in a complex working environment close to the non-proportional biaxial stretching, carrying out biaxial stretching tests under various stress ratio conditions, and collecting the training required by the neural network Sample data; construct a Bayesian neural network including input layer, hidden layer and output layer, so as to establish a simple expression model of skin material deformation behavior; use trained neural network to predict skin material deformation behavior in real time. The method for predicting the deformation of a stratospheric airship skin material by using a neural network provided by the invention has high prediction accuracy, good stability and strong generalization ability, and can meet the requirements of real-time and accurate prediction of the deformation behavior of the skin material. Provides new methods for designing and directing the structural design of stratospheric airships.
Description
Technical Field
The invention relates to a method for predicting deformation of a skin material of an airship on a stratosphere by utilizing a neural network, and belongs to the technical field of damage analysis.
Background
The stratospheric airship mainly overcomes the gravity of the earth by using static buoyancy and flies at the height of 20-50km of the stratosphere. Because of the capability of long-term sky staying, fixed point and high-resolution earth observation, the stratospheric airship has higher economic and strategic values and becomes a research hotspot of scientific workers in recent years. The complex environmental characteristics of the stratosphere and the unique appearance design of the airship provide higher requirements for the mechanical properties of the stratosphere airship skin material, and the stratosphere airship which can be parked in the air for a long time and is practical in engineering is not developed at present.
The characteristics of low stratosphere atmospheric density and small convective heat transfer enable the temperature of the floating gas sealed in the airship to have a temperature difference of up to 70K caused by the solar radiation changing in a day-night period, and have great influence on the mechanical property of the airship flexible airbag material, and the overpressure, thermal fatigue and the like can cause the airship flexible airbag to generate initial cracks or defects to tear and expand rapidly, thereby influencing the safety reliability and control of the airship. The method for effectively predicting the deformation behavior of the airship skin material in the complex load state has important significance for optimizing material design, improving the damage resistance of the airship skin material and designing the structure layout of the airship.
The stratospheric airship skin material is usually made of an orthogonally woven fiber reinforced composite material, and under the action of inflation rise and huge pressure difference of a service environment, the material cannot bear out-of-plane load due to the characteristic of flexibility, and the normal pressure difference load is converted into in-plane biaxial non-proportional tensile load usually through out-of-plane deformation, wherein the biaxial tensile direction is consistent with the orthogonal yarn direction. Therefore, the biaxial tension mechanical behavior of the stratospheric airship skin material under different stress ratios is researched, and the deformation behavior of the airship skin material is predicted by adopting a proper method, so that the material damage is avoided, the material design is optimized, and the method becomes a new challenge for analyzing the mechanical properties of the airship skin material.
The existing mathematical formula and basic theory face the problems of complex model and heavy calculation amount, the effect of predicting unknown deformation data of the skin material cannot be realized, machine learning is used as an important branch of artificial intelligence, and the machine can learn the rules from a large amount of historical data through an algorithm, so that intelligent identification is carried out on a new sample or prediction is carried out in the future. As a representative artificial neural network in the field of machine learning, the deformation characteristics of the skin material under the condition of biaxial non-proportional stretching can be associated and learned through a unified criterion, a simple expression model is constructed, and the deformation behavior under a brand-new stress ratio can be rapidly and accurately predicted. The invention aims to provide a method for predicting the deformation of a skin material of an airship on a stratosphere by using a Bayesian neural network, predicting the deformation characteristic of the skin material under different stress ratios, and providing a new method for guiding the structural design of the airship on the stratosphere.
Disclosure of Invention
The invention aims to provide a method for predicting deformation of a skin material of an airship on a stratosphere by utilizing a neural network, which solves the problems of heavy calculation and complex model of the traditional mathematical model, can construct a simple expression model to understand and rapidly predict the deformation characteristic of the skin material by independently learning biaxial tension test data of the skin material, is simple, reduces the test cost and improves the working efficiency.
The purpose of the invention is realized by the following technical scheme:
a method for predicting deformation of a skin material of an airship on a stratosphere by utilizing a neural network comprises the following steps:
step one, manufacturing a cross-shaped skin material sample, and chamfering the cross corners of the sample to avoid the phenomenon of stress concentration. And placing the extensometer in the middle area along the longitudinal and latitudinal directions, fixing the extensometer on a sample by using screws, and measuring the strain of the material in the longitudinal and latitudinal directions. Measuring the stress of the material by adopting a tension sensor, assembling the tension sensor on a stretching arm of a stretching testing machine, and carrying out a test by adopting a special biaxial stretching testing machine;
and step two, carrying out biaxial tensile tests on the samples under the conditions of various stress ratios, wherein the stress ratios comprise (1:1, 1.3:1, 1.5:1, 1.7:1 and 2:1), and the stress ratios represent the stress in the warp yarn direction divided by the stress in the weft yarn direction. Collecting and screening test result data, and carrying out preprocessing of disordering sequence and normalization on the data to obtain training sample data required by the Bayesian neural network;
and step three, determining an optimal neural network structure, including input and output parameters of the neural network, the number of hidden layers and the number of neurons of the hidden layers. Different from metal and other composite materials, the main deformation form of the airship skin material is in-plane biaxial tension, and the stress strain needs to comprise two directions of longitude and latitude. Therefore, the input parameters of the neural network are 2, namely the warp direction stress and the weft direction stress obtained by the biaxial tensile test. The output parameters are 2, namely warp direction strain and weft direction strain obtained by a biaxial tensile test. The number of hidden layers is 1 layer, and the number of neurons in the hidden layers is 8;
selecting proper activation functions and training functions to learn and train the neural network;
testing the model by adopting a test sample set, and evaluating the accuracy of the constructed skin material neural network constitutive model;
and sixthly, learning and predicting the constitutive relation of the stratospheric airship skin material under different stress ratios by using the Bayes neural network model determined in the step.
In the first step, the skin material is a laminated woven composite material Uretek3216LV, and the bearing woven layer is manufactured by a plain orthogonal weaving method.
In the first step, in order to realize the application of the multi-stress ratio load, the skin material sample is obtained by preparing a cross-shaped lancing sample, and the longitudinal and latitudinal directions of the main shaft of the test piece are consistent with those of the skin.
In the second step, the stress ratio is the ratio of the stress borne by the warp yarn and the weft yarn of the material, and the considered stress ratio load types comprise 1:1, 1.3:1, 1.5:1, 1.7:1 and 2: 1.
In step three, the optimal neural network structure comprises: the number of neurons in the input layer is 2, the number of neurons in the hidden layer is 8, and the number of neurons in the output layer is 2. The input layer neurons are respectively warp-wise stress and weft-wise stress, the output layer neurons are respectively warp-wise strain and weft-wise strain, and the number of the hidden layer neurons is determined by calculation of an empirical formula.
In the fourth step, the activation function is a hyperbolic tangent function, the input parameters are nonlinearly mapped into [ -1,1], the training function is a Bayesian Regression function (BR), compared with the traditional BP training function, the generalization capability is stronger, the learning rate is faster, the prediction precision is higher, and the occurrence of the local optimal solution can be effectively avoided.
And step five, comparing the deformation data trained by the neural network with the measured value, wherein the relative error is within 10%.
And step six, predicting the biaxial tensile deformation behavior of the skin material under the condition of more stress ratios in real time by adopting a trained Bayes neural network model to obtain the optimal prediction method of the neural network for skin material deformation of the stratospheric airship. The prediction speed and accuracy of the trained network model on the new deformation data meet the requirements.
Advantageous effects
By adopting the method, the constitutive relation of the skin material of the stratospheric airship under the biaxial action of different stress ratios can be conveniently and accurately constructed, the constitutive relation of the skin material under the condition of more stress ratios can be accurately predicted, the calculation difficulty caused by a plurality of parameters in the mathematical derivation process is avoided, and a large amount of time and cost are saved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly described below.
FIG. 1 is a macroscopic and microscopic structure of stratospheric airship skin material;
FIG. 2 is a skin material cross specimen design meeting biaxial stretching requirements;
FIG. 3 is a constitutive relation Bayesian network structure of a stratospheric airship skin material;
FIG. 4 is a regression result of a Bayesian neural network; wherein, the graph a is a training set regression result, the graph b is a test set regression result, and the graph c is a network final regression result;
FIG. 5 is a comparison of warp direction stress-strain data trained by a Bayesian neural network with measured data;
FIG. 6 is a comparison of weft direction stress-strain data trained by a Bayesian neural network with measured data;
FIG. 7 is a comparison of warp direction stress-strain data and measured data under Bayesian neural network prediction for full new stress ratio;
FIG. 8 is a comparison of weft direction stress-strain data and measured data under Bayesian neural network prediction for an all-new stress ratio.
Fig. 9 is a flow chart of bayesian neural network prediction.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, belong to the scope of the present invention.
As introduced in the background art, the traditional mathematical formula and basic theory can not effectively and accurately predict the deformation behavior of the material, and once the stratospheric airship skin material generates initial cracks and expands, the damage and damage speed is very high, which puts higher requirements on the safety, reliability and control of the airship. In order to solve the problems, the application provides a method for predicting deformation of a skin material of an stratospheric airship by utilizing a neural network, man-machine interaction is realized based on the principle of the neural network, redundant calculated quantities are processed by a computer, and corresponding strain parameters can be predicted according to warp-direction and weft-direction stress parameters measured in a biaxial stretching deformation process of the skin material after a large amount of training analysis. The biaxial tensile deformation data can be accurately predicted in real time in the material deformation process only by acquiring the biaxial tensile deformation data after the data analysis is completed by the early-stage neural network model. Before the skin material is subjected to fatigue deformation until an initial crack is generated, accurate prediction on the deformation behavior of the skin material by using the constructed neural network model provides reliable reference for the optimization design of the airship skin material and the improvement of the safety performance of the airship.
The method for predicting the deformation of the skin material of the stratospheric airship by using the neural network disclosed by the embodiment specifically comprises the following steps:
step one, manufacturing a skin material sample suitable for a biaxial tensile test, and carrying out a biaxial tensile deformation test.
Referring to fig. 1, the airship skin material is a fiber reinforced composite material, the bearing layer is a woven layer, and the fibers are usually made by an orthogonal plain weave method. Referring to fig. 2, a cross-shaped skin material sample is manufactured, and the cross corners of the sample are rounded to avoid the stress concentration phenomenon. A special biaxial tensile testing machine is adopted to carry out biaxial tensile test testing, a stretching arm of the testing machine clamps a cross-shaped sample, a extensometer is placed in the middle area along the longitudinal direction and the latitudinal direction and is fixed on the sample by screws, and the strain of the material in the longitudinal direction and the latitudinal direction is measured. The tension sensors are assembled on four stretching arms of the stretching testing machine, and the tension sensors are adopted to measure the stress of the material in the warp and weft directions;
and step two, carrying out biaxial tensile test on the test sample under the condition of various stress ratios, collecting and screening test result data, and carrying out preprocessing of disordering sequence and normalization on the data to obtain training sample data required by the neural network.
The sequencing mode of the disordered original data is to enable the neural network to learn the corresponding stress-strain data under different stress ratios more comprehensively and more uniformly, so that the robustness of the network is improved; the input parameters are normalized before being transferred to the hidden layer for training, so that the input parameters with any dimension and any size are nonlinearly mapped into a certain determined interval, and the learning and prediction of the neural network on any nonlinear function relation are realized.
And step three, determining an optimal neural network structure, including input and output parameters of the neural network, the number of hidden layers and the number of neurons of the hidden layers.
Referring to fig. 3, the biaxial tension test obtains the stress-strain data of the skin fiber composite material in the warp direction and the weft direction, so as to predict the deformation of the skin material in the subsequent process. The input parameters of the neural network are independent, and the stress ratio is the ratio of the warp stress to the weft stress, so that the stress ratio is not used as an independent input parameter. Determining the structure of the neural network as follows: the warp and weft stresses are used as 2 input parameters of the neural network input layer; taking warp and weft strain as 2 output parameters of the neural network output layer; the number of input and output parameters is small, so the number of hidden layers is set to 1 layer, and the number of neurons in the hidden layers is determined to be 8 by an empirical formula.
And step four, selecting a proper activation function and a proper training function, and learning and training the neural network.
Selecting a hyperbolic tangent function as an activation function of a neural network, wherein the function of the hyperbolic tangent function is to map input parameters of any dimension into [ -1,1] in a nonlinear manner, so that the neural network learns any nonlinear function relationship; a Bayesian Regression function (Bayesian Regression) is selected as a training function of the neural network, so that the training precision of the network is improved, and the defect that the traditional BP network is easy to fall into a local optimal solution is avoided.
Testing the model by adopting a test sample set, and evaluating the accuracy of the constructed skin material neural network model;
referring to fig. 4(a), samples are divided into training sets and test sets according to 80% and 20%, the regression accuracy of the bayesian neural network on the training sets reaches 0.99844, and the constructed bayesian neural network model is proved to be capable of accurately learning the training sets; referring to fig. 4(b) and (c), the regression precision of the constructed bayesian neural network on the test set reaches 0.99807, the overall training effect of the network model reaches 0.99838, and the constructed bayesian neural network model is proved to be capable of accurately learning the acquired biaxial tensile deformation data.
And sixthly, learning and predicting the deformation behavior of the stratospheric airship skin material under different stress ratios by using the Bayes neural network model determined in the step.
Referring to fig. 5 and 6, the comparison of the longitudinal and latitudinal stress-strain data learned by the constructed bayesian neural network model with the test data shows that the degree of coincidence between the learning result of the network model and the test result is good, and the network model can rapidly and accurately learn the biaxial tension deformation characteristics of the skin material. Referring to fig. 7 and 8, a biaxial tensile test with stress ratio conditions of (1:1, 1.5:1, 1.8:1 and 2:1) was conducted, and warp and weft stress data were input into a trained network model in real time, and completely new warp and weft strain data were predicted. The stress ratio condition is a brand-new stress ratio condition which is not contacted with the neural network, and the stress-strain condition of the stress-strain condition is predicted to test the generalization capability of the constructed Bayesian network model. The result shows that the network model can accurately predict the biaxial stretching deformation behavior of the skin material under different stress ratios in real time.
In the first step, the main deformation occurrence area of the skin material sample is in the central square area. In order to avoid the stress concentration action of a local area, the stretching arm of the sample is subjected to slitting treatment, and the cross-shaped area of the sample is subjected to rounding treatment.
And in the second step, carrying out biaxial tensile tests under different stress ratio conditions. Since the force-bearing layer of the skin material is mainly a woven layer, the stress ratio represents the stress in the warp direction divided by the stress in the weft direction. Before the test data are substituted into the neural network for training, the test data need to be preprocessed, wherein the disordered sequence is beneficial to improving the robustness of a network model, and the normalization is adopted to nonlinearly map input data of any dimension into a certain determined region, so that the learning capacity of the neural network to any function is realized.
In the third step, the optimal neural network structure includes: the number of neurons in the input layer is 2, the number of neurons in the hidden layer is 8, and the number of neurons in the output layer is 2. The determination of the input and output parameters of the network structure is mainly determined by the particularity of biaxial stretching. Compared with metal and traditional composite materials, the airship skin material is mainly subjected to non-proportional biaxial stretching. The stress ratio in the orthogonal direction is related to the stress value, so that the input parameter is determined to be the warp-wise stress and the weft-wise stress, and the output parameter is the warp-wise strain and the weft-wise strain; the number of hidden layers is determined by the quantity of network input and output parameters, so that a single hidden layer structure is adopted; the determination of the number of cryptic neurons does not currently have a uniform formula or theory, and is generally determined by the following empirical formula:
wherein n is1Representing the number of hidden layer neurons, n representing the number of input layer neurons, m representing the number of output layer neurons, a being [1,10 ]]Is constant.
In step four, the proper activation function and training function: the activation functions of the hidden layer and the output layer are respectively a hyperbola tandent and a logic sigmoid, and the training function is a Bayesian regression algorithm. Compared with traditional feedforward algorithms such as a Levenberg-Marquardt algorithm, a Scaled connected Gradient algorithm and the like, the Bayesian algorithm is a function containing a residual sum of squares and a weight sum of squares, so that the generalization capability is stronger, the learning rate is faster, the prediction accuracy is higher, and the occurrence of a local optimal solution can be effectively avoided.
In the fifth step, the iteration speed of the biaxial tensile test data under different stress ratios of the Bayesian neural network learning skin material is equivalent to that of a Levenberg-Marquardt algorithm, a Scaled concrete Gradient algorithm and other algorithms, and the accuracy of the skin material deformation data learned by the Bayesian neural network model meets the requirement.
And in the sixth step, a trained Bayesian neural network is adopted to predict the biaxial tensile deformation behavior of the skin material under the condition of more stress ratios in real time, so as to obtain the optimal prediction method of the neural network for skin material deformation of the stratospheric airship. The prediction speed and accuracy of the trained network model on the new deformation data meet the requirements.
One specific application process of the embodiment of the invention is as follows:
(1) experimental design and acquisition of stress-strain data
The laminated braided composite material Uretek3216LV is a commonly used stratospheric airship skin material. In order to meet the requirement of stratospheric environment on skin materials, Uretek3216LV is formed by laminating and compounding five functional layers, which are respectively: the coating comprises an abrasion-resistant layer, an ultraviolet layer, a fabric layer, a gas retaining layer and a sealing layer. A laminated and woven composite material Uretek3216LV is taken as a research object, a cross-shaped material sample required by a biaxial tensile test is designed, the size is 160mm x 160mm, the radius of a fillet is 25mm, and the effective cantilever length is 160 mm. The number of samples meeting the requirements of the 5 stress ratio test was produced. A biaxial tensile testing machine is adopted to carry out static tensile tests on the sample under different stress ratios, and warp direction and weft direction stress strain data of the sample under different stress ratios are measured and used as training and testing samples of the neural network.
(2) Preprocessing of raw test data
315 pairs of test data under 5 stress ratios are collected and screened through a biaxial tensile test, and training sample data needs to be preprocessed before neural network training. The preprocessing mainly comprises the processing and normalization of the data sequence. Neural networks, like biological brain nerves, have the property of being "forgotten", i.e., there is definite memory for the feature just learned, while memory for earlier learned features is gradually diluted. The original test data are arranged in an ascending order by taking the stress ratio as a reference according to the experiment development order, and a disorder arrangement method is adopted for the original data, so that the learning capacity and robustness of the neural network on the deformation characteristics of the skin material under different stress ratios are improved; in addition, normalization is a very important step of the neural network structure. The activation function in the input layer of the neural network is a nonlinear function, input data with different dimensions and any size can be nonlinearly mapped into a numerical value interval, and the output of the network is no longer linear mapping of input, so that the neural network has the capability of learning complex nonlinear relations. Taking a hyperbolic tangent function as an input layer activation function, and calculating two parameters of an input layer: the warp and weft stress data sets are non-linearly mapped into [ -1,1], i.e. normalized. The two normalized input parameters are then subsequently transmitted to the hidden layer and the output layer for network training.
(3) Determination of neural network structure
The actual deformation of the stratospheric airship skin material is close to non-proportional biaxial stretching, the existing neural network structure is not suitable for the skin material biaxial stretching deformation behavior considering the stress ratio condition, and the traditional mathematical formula and model have the limitations of heavy calculation amount, complex parameters, incapability of realizing real-time accurate prediction of material deformation and the like. A proper neural network structure is designed, so that the biaxial stretching deformation behavior of the skin material can be accurately learned, and the real-time prediction of the unknown deformation behavior provides reference for the optimization design of the airship skin material and the improvement of the safety performance of the airship. Because the skin material is subjected to non-proportional biaxial stretching, the structure of the neural network is determined as follows: warp and weft stresses are taken as 2 input parameters of the input layer; warp and weft strain as 2 output parameters of the output layer; the number of the hidden layers is set to be 1 layer, and the neuron number of the hidden layers is determined to be 8 by an empirical formula.
(4) Selecting proper training function, training neural network and predicting deformation behavior of skin material
According to the comparison and analysis of the feedforward neural network training function, a Bayesian regression function is adopted as the training function, and the preprocessed test data are substituted to train the network model. After 417 times of 2 seconds of iterative calculation, the training of the network model is finished, the mean square error is as low as 0.0010264, the network training value and the measured value are analyzed by a statistical regression method, the correlation coefficient between the network training value and the measured value can be found to reach 0.99838, and the result proves that the trained network model meets the precision requirement. In order to verify the capability of the trained network model for predicting the deformation behavior of the skin material, biaxial tensile tests are carried out on skin material samples under the conditions of different stress ratios, wherein the stress ratios include 1:1, 1.5:1, 1.8:1 and 2:1, and the stress ratio 1.8:1 is the stress ratio condition which is not identified by the neural network, so that the capability of the network model for predicting is favorably analyzed. And inputting the longitudinal and latitudinal stress data into the trained Bayesian neural network model in real time, and predicting brand-new longitudinal and latitudinal strain data. Referring to fig. 7 and 8, the network prediction result and the test result are compared, and the relative error of the prediction result is within 10%, which shows that the constructed network model has high prediction precision, good stability and strong popularization capability, and can meet the requirement of accurately predicting the deformation behavior of the skin material in real time.
The above detailed description is intended to illustrate the objects, aspects and advantages of the present invention, and it should be understood that the above detailed description is only exemplary of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A method for predicting deformation of a skin material of an airship on a stratosphere by utilizing a neural network is characterized by comprising the following steps of: the method comprises the following steps:
the method comprises the following steps: manufacturing a cross-shaped skin material sample, and rounding off the cross-shaped part of the sample to avoid stress concentration during stretching;
step two: carrying out biaxial tension tests on skin material samples of the stratospheric airship under different stress ratio conditions, collecting and screening test result data, and carrying out normalization pretreatment to obtain training sample data required by a neural network model;
step three: determining an optimal neural network model structure, which comprises input parameters, hidden layer numbers, hidden layer neuron numbers and output parameters of a neural network;
step four: selecting an activation function and a training function, and learning and training a neural network model;
step five: testing the model by adopting test sample data, and evaluating the accuracy of the established neural network model;
step six: and predicting the deformation behavior of the stratospheric airship skin material under the condition of more stress ratios by using the neural network model generated in the steps.
2. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: in the first step, the skin material is a laminated woven composite material Uretek3216LV, and the bearing woven layer is manufactured by a plain orthogonal weaving method.
3. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: in the first step, in order to realize the application of the multi-stress ratio load, the skin material sample is obtained by preparing a cross-shaped lancing sample, and the longitudinal and latitudinal directions of the main shaft of the test piece are consistent with those of the skin.
4. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: in the second step, the stress ratio is the ratio of the stress borne by the warp yarn and the weft yarn of the material, and the considered stress ratio load types comprise 1:1, 1.3:1, 1.5:1, 1.7:1 and 2: 1.
5. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: in step three, the optimal neural network structure comprises: the number of neurons in the input layer is 2, the number of neurons in the hidden layer is 8, and the number of neurons in the output layer is 2. The input layer neurons are respectively warp-wise stress and weft-wise stress, the output layer neurons are respectively warp-wise strain and weft-wise strain, and the number of the hidden layer neurons is determined by calculation of an empirical formula.
6. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: in the fourth step, the activation function is a hyperbolic tangent function, the input parameters are nonlinearly mapped into [ -1,1], the training function is a Bayesian Regression function (BR), compared with the traditional BP training function, the generalization capability is stronger, the learning rate is faster, the prediction precision is higher, and the occurrence of the local optimal solution can be effectively avoided.
7. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: and step five, comparing the deformation data trained by the neural network with the measured value, wherein the relative error is within 10%.
8. The method for predicting deformation of the skin material of the stratospheric airship by using the neural network as claimed in claim 1, wherein: and step six, predicting the biaxial tensile deformation behavior of the skin material under the condition of more stress ratios in real time by adopting a trained Bayes neural network model to obtain the optimal prediction method of the neural network for skin material deformation of the stratospheric airship. The prediction speed and accuracy of the trained network model on the new deformation data meet the requirements.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011515090.XA CN112507625A (en) | 2020-12-21 | 2020-12-21 | Method for predicting deformation of stratospheric airship skin material by utilizing neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011515090.XA CN112507625A (en) | 2020-12-21 | 2020-12-21 | Method for predicting deformation of stratospheric airship skin material by utilizing neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112507625A true CN112507625A (en) | 2021-03-16 |
Family
ID=74922666
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011515090.XA Pending CN112507625A (en) | 2020-12-21 | 2020-12-21 | Method for predicting deformation of stratospheric airship skin material by utilizing neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112507625A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113673031A (en) * | 2021-08-11 | 2021-11-19 | 中国科学院力学研究所 | A flexible airship service angle-of-attack identification method based on the fusion of strain response and deep learning |
| CN114619659A (en) * | 2022-05-16 | 2022-06-14 | 河南银金达新材料股份有限公司 | Low-density lightweight polyester film stretching process |
| CN114783550A (en) * | 2022-06-20 | 2022-07-22 | 数聚(山东)医疗科技有限公司 | Artificial intelligence-based new material online evaluation method and system |
| CN116108762A (en) * | 2023-04-13 | 2023-05-12 | 南京航空航天大学 | A Method for Predicting Assembly Deformation of Large Composite Components Using Force Sensors |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| CN107340184A (en) * | 2017-05-26 | 2017-11-10 | 上海交通大学 | Biaxial stretch-formed intensity test piece of diesis type compound fabric membrane material and preparation method thereof |
| CN109163972A (en) * | 2018-08-27 | 2019-01-08 | 上海交通大学 | A kind of determination method of the biaxial stretch-formed intensity of fabric membrane material |
| US20190072467A1 (en) * | 2016-03-28 | 2019-03-07 | Mitsubishi Heavy Industries, Ltd. | Biaxial load test specimen, biaxial load test apparatus, and biaxial load test method |
| CN111159863A (en) * | 2019-12-17 | 2020-05-15 | 上海交通大学 | Neural network algorithm-based material deformation and failure prediction method and system |
| CN111950099A (en) * | 2020-08-03 | 2020-11-17 | 中国石油大学(华东) | Method, system, medium and computer equipment for testing mechanical property of equipment material |
-
2020
- 2020-12-21 CN CN202011515090.XA patent/CN112507625A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| US20190072467A1 (en) * | 2016-03-28 | 2019-03-07 | Mitsubishi Heavy Industries, Ltd. | Biaxial load test specimen, biaxial load test apparatus, and biaxial load test method |
| CN107340184A (en) * | 2017-05-26 | 2017-11-10 | 上海交通大学 | Biaxial stretch-formed intensity test piece of diesis type compound fabric membrane material and preparation method thereof |
| CN109163972A (en) * | 2018-08-27 | 2019-01-08 | 上海交通大学 | A kind of determination method of the biaxial stretch-formed intensity of fabric membrane material |
| CN111159863A (en) * | 2019-12-17 | 2020-05-15 | 上海交通大学 | Neural network algorithm-based material deformation and failure prediction method and system |
| CN111950099A (en) * | 2020-08-03 | 2020-11-17 | 中国石油大学(华东) | Method, system, medium and computer equipment for testing mechanical property of equipment material |
Non-Patent Citations (3)
| Title |
|---|
| JUNHUI MENG 等: "Fatigue Damage Mechanical Model of the Envelope Material for Stratospheric Airships", 《APPL COMPOS MATER》 * |
| JUNHUI MENG 等: "Mechanical Properties and Strength Criteria of Fabric Membrane for the Stratospheric Airship Envelope", 《APPL COMPOS MATER》 * |
| 陈建稳 等: "飞艇用层压织物膜材料在双向应力作用下的弹性参数分析", 《上海交通大学学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113673031A (en) * | 2021-08-11 | 2021-11-19 | 中国科学院力学研究所 | A flexible airship service angle-of-attack identification method based on the fusion of strain response and deep learning |
| CN113673031B (en) * | 2021-08-11 | 2024-04-12 | 中国科学院力学研究所 | Flexible airship service attack angle identification method integrating strain response and deep learning |
| CN114619659A (en) * | 2022-05-16 | 2022-06-14 | 河南银金达新材料股份有限公司 | Low-density lightweight polyester film stretching process |
| CN114619659B (en) * | 2022-05-16 | 2022-09-02 | 河南银金达新材料股份有限公司 | Low-density lightweight polyester film stretching process |
| CN114783550A (en) * | 2022-06-20 | 2022-07-22 | 数聚(山东)医疗科技有限公司 | Artificial intelligence-based new material online evaluation method and system |
| CN114783550B (en) * | 2022-06-20 | 2022-08-23 | 数聚(山东)医疗科技有限公司 | Artificial intelligence-based new material online evaluation method and system |
| CN116108762A (en) * | 2023-04-13 | 2023-05-12 | 南京航空航天大学 | A Method for Predicting Assembly Deformation of Large Composite Components Using Force Sensors |
| CN116108762B (en) * | 2023-04-13 | 2023-06-27 | 南京航空航天大学 | Large composite material component assembly deformation prediction method by using force sensor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112507625A (en) | Method for predicting deformation of stratospheric airship skin material by utilizing neural network | |
| Silva et al. | Statistical models applied to service life prediction of rendered façades | |
| Liu et al. | Data-driven approaches for characterization of delamination damage in composite materials | |
| CN110263429B (en) | Method for adjusting shape and surface precision of cable net reflecting surface antenna based on machine learning | |
| CN103884593B (en) | Construction damage positionings based on a small amount of static measurement data and degree recognition methods | |
| CN112001110A (en) | Structural damage identification monitoring method based on vibration signal space real-time recursive graph convolutional neural network | |
| CN105092374A (en) | Test method suitable for elastic modulus of coated fabric membrane material in model test | |
| CN112199891A (en) | A method and system for formulating early warning index of gravity dam safety deformation | |
| CN113504768B (en) | High-precision product digital twin computability method for assembly quality prediction | |
| CN113688465A (en) | A digital twin method of aircraft structural strength based on the combination of load and state | |
| CN111027016A (en) | Rock mass structural plane advantage and occurrence clustering analysis method based on netting algorithm | |
| Chen et al. | Approximate Bayesian assisted inverse method for identification of parameters of variable stiffness composite laminates | |
| Ding et al. | Jaya-based long short-term memory neural network for structural damage identification with consideration of measurement uncertainties | |
| CN109241649A (en) | A kind of fiber filament method for testing performance and system based on decision-tree model | |
| CN112784331A (en) | Soil stress-strain relation determination method based on improved LSTM deep learning method | |
| CN113973403B (en) | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology | |
| CN112487712B (en) | Intelligent identification method for buried suspended pipeline working condition parameters based on genetic algorithm and multidimensional dynamic time regularity | |
| Yang et al. | A novel damage identification algorithm by combining the boundary element method and a series connection neural network | |
| Wang et al. | Failure envelope prediction of 2D SiCf/SiC composites based on XGBoost model | |
| Meng et al. | Prediction of tear propagation path of stratospheric airship envelope material based on deep learning | |
| Yu et al. | An improved multi-objective genetic algorithm and data fusion in structural damage identification | |
| CN112560312B (en) | Topology Optimization Method for Structural Defect Identification | |
| Mohanty et al. | An Artificial Intelligence/Machine Learning Based Regression-Clustering-Sensor-Fusion Framework for Predicting Unmeasurable Time-Series Strain From Other Sensor Measurements | |
| Elenchezhian et al. | Damage precursor identification in composite laminates using data driven approach | |
| CN105699043A (en) | Method for improving measuring stability and precision of wind tunnel sensor |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210316 |