CN115481488A - Machine learning-based multi-target optimization method for collision resistance of square cone type energy absorption structure - Google Patents
Machine learning-based multi-target optimization method for collision resistance of square cone type energy absorption structure Download PDFInfo
- Publication number
- CN115481488A CN115481488A CN202211151523.7A CN202211151523A CN115481488A CN 115481488 A CN115481488 A CN 115481488A CN 202211151523 A CN202211151523 A CN 202211151523A CN 115481488 A CN115481488 A CN 115481488A
- Authority
- CN
- China
- Prior art keywords
- energy absorption
- optimization
- absorption structure
- cone type
- square cone
- 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.)
- Granted
Links
- 238000010521 absorption reaction Methods 0.000 title claims abstract description 158
- 238000005457 optimization Methods 0.000 title claims abstract description 101
- 238000000034 method Methods 0.000 title claims abstract description 85
- 238000010801 machine learning Methods 0.000 title claims abstract description 50
- 238000004088 simulation Methods 0.000 claims abstract description 30
- 230000004044 response Effects 0.000 claims abstract description 9
- 238000013401 experimental design Methods 0.000 claims abstract description 8
- 238000005070 sampling Methods 0.000 claims abstract description 6
- 238000012952 Resampling Methods 0.000 claims abstract description 4
- 238000006073 displacement reaction Methods 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 23
- 238000002474 experimental method Methods 0.000 claims description 16
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 claims description 14
- 229910052782 aluminium Inorganic materials 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 12
- 230000000306 recurrent effect Effects 0.000 claims description 10
- 230000006835 compression Effects 0.000 claims description 6
- 238000007906 compression Methods 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000005192 partition Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 210000000629 knee joint Anatomy 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 19
- 230000008569 process Effects 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000011160 research Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000009863 impact test Methods 0.000 description 3
- 238000012821 model calculation Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000008649 adaptation response Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012827 research and development 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]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
-
- 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
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Automation & Control Theory (AREA)
- Aviation & Aerospace Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a machine learning-based multi-target optimization method for collision resistance of a square cone type energy absorption structure, which comprises the following steps of: establishing a finite element simulation model of a square cone type energy absorption structure of the subway train; extracting the structural parameters and the energy absorption characteristic curve of the subway train energy absorption structure based on a method of combining the established finite element simulation model with experimental design; sampling according to a Latin hypercube method, and determining an optimal energy absorption characteristic curve prediction model of a subway train pyramid type energy absorption structure, and an optimized variable and an optimized target through experimental Design (DOE); establishing an optimization theoretical model according to the optimal energy absorption characteristic curve prediction model, the optimization variables and the optimization target; resampling the optimization theoretical model by adopting a Ha Mosi Lei Caiyang method, and calculating corresponding absorption energy and peak force by utilizing an optimal energy absorption characteristic curve prediction model to generate a new DOE; performing multi-objective optimization by adopting a Global Response Surface Method (GRSM) to obtain an optimization result; and obtaining a pareto solution set of an optimization target based on an optimization result, and performing optimal decision on the pareto solution set obtained by optimization by adopting a minimum distance method to obtain an optimal solution.
Description
Technical Field
The invention relates to the technical field of train collision resistance detection, in particular to a machine learning-based multi-objective optimization method for collision resistance of a square cone type energy absorption structure.
Background
In the collision process of the urban rail train, energy absorption is carried out through the car coupler device and the square cone type energy absorption structure. The anti-climbing energy-absorbing structure is the last defense line of passive safety protection of the train, the collision resistance of the anti-climbing energy-absorbing structure has important significance for safety guarantee of the train body, and once the anti-climbing energy-absorbing structure fails to cause a collision accident, serious casualties can be caused. Therefore, the method has important significance for the research of the collision resistance of the subway train pyramid type energy absorption structure.
The traditional method for researching the energy absorption structure mainly adopts a finite element method, an experimental method and a multi-body dynamics method to research the collision resistance of the train, the experimental method needs to consume more material resources and financial resources for research, the experiment has larger uncertainty, the finite element method and the multi-body dynamics method are adopted for research, the performance requirement on a computer is higher, and the calculation time is longer.
The square cone type energy absorption structure is used as a main energy absorption element of the subway train, and the energy absorption characteristic (force-displacement) curve of the square cone type energy absorption structure has great influence on the running safety of the subway train, so that the force-displacement curve of the energy absorption structure is obtained in a reasonable mode, and the square cone type energy absorption structure has important significance on the optimization of the collision resistance of the subway train. At present, most of the methods of combining a finite element model with a test are adopted to obtain a force-displacement curve of a square cone type energy absorption structure, but the method needs long time and expensive expenses. Therefore, it is desirable to optimize the crashworthiness of the pyramid-shaped energy absorbing structure in a more efficient way.
Disclosure of Invention
In order to reduce the calculation time while ensuring the calculation accuracy, the invention provides a machine learning-based multi-objective optimization method for the crashworthiness of a square cone type energy absorption structure. By adopting the method, the accuracy of energy absorption characteristic prediction is high, the calculation time can be greatly reduced, and the collision resistance multi-target optimization of the square cone type energy absorption structure of the metro vehicle is subjected to large data energization.
In order to achieve the aim, the invention provides a machine learning-based multi-objective optimization method for collision resistance of a square cone type energy absorption structure, which comprises the following steps of:
establishing a finite element simulation model of a square cone type energy absorption structure of the subway train;
extracting structure parameters and energy absorption characteristic curves of the subway train energy absorption structure based on a method of combining the established finite element simulation model with experimental design; according to the method, a Latin hypercube method is adopted for sampling, virtual experiment design is carried out, an optimal energy absorption characteristic curve prediction model of a subway train pyramid type energy absorption structure is determined, and variables and an optimization target are optimized;
establishing an optimization theoretical model according to the optimal energy absorption characteristic curve prediction model, the optimization variables and the optimization target;
resampling the optimization theoretical model by adopting a Ha Mosi Lei Caiyang method, and calculating corresponding absorption energy and peak force by utilizing an optimal energy absorption characteristic curve prediction model to generate a new DOE; performing multi-objective optimization by adopting a global response surface method to obtain an optimization result;
and obtaining a pareto solution set of an optimization target based on an optimization result, and performing optimal decision on the pareto solution set obtained by optimization by adopting a minimum distance method to obtain an optimal solution.
Preferably, a full-size dynamic impact experiment of the pyramid energy-absorbing structure is carried out based on the same boundary conditions of finite element simulation, and the accuracy of the established finite element model is verified by comparing a force-displacement curve, a displacement-energy curve and a deformation sequence mode between the experiment and the simulation in a mode of combining the experiment and the simulation.
Preferably, extracting the structural parameters and the energy absorption characteristic curve of the subway train energy absorption structure based on a method of combining the established finite element simulation model with experimental design; according to the method, a Latin hypercube method is adopted for sampling, virtual experiment design is carried out, an optimal energy absorption characteristic curve prediction model of a subway train pyramid type energy absorption structure is determined, and variables and an optimization target are optimized, and the method specifically comprises the following steps:
taking five input variables of the DOE as input features of network training in machine learning, and taking the output of the DOE as actual output of the network training in the machine learning;
randomly dividing the input and the actual output of the network training into a training set and a test set;
normalizing the input data and the actual output data;
mapping an input geometric figure of the energy absorption structure into a Machine Learning (ML) system structure of a required output response, using a normalized training set and a normalized test set as samples, respectively using four machine learning network models of MLP, RNN, LSTM and GRU to predict an energy absorption characteristic curve of the energy absorption structure, adopting comparison analysis to predict the accuracy of different models, and using the network model with the highest accuracy of the predicted output curve as an optimal energy absorption characteristic curve prediction model of the square cone type energy absorption structure of the subway train.
Preferably, the optimal energy absorption characteristic curve prediction model of the subway train pyramid type energy absorption structure is a gated cyclic neural network model, and the thickness T of the outer wall is used A Outer wall thickness T B Thickness T of the partition board gb Strength delta of aluminum honeycomb A A And strength delta of aluminum honeycomb B B As optimization variables, the absorption energy and the peak force of the square cone type energy absorption structure are optimization targets.
Preferably, the established optimization theoretical model is as follows:
PCF＝max(F(s))
s is the compression displacement of the square cone type energy absorption structure; f(s) is the axial force of the square cone type energy absorption structure; t is A And T B The thickness of the outer wall of the square cone type energy absorption structure; t is a unit of gb The thickness of the baffle plate is of a square cone type energy absorption structure; delta A And delta B The strength of the aluminum honeycomb A and the strength of the aluminum honeycomb B are respectively; EA is the energy absorption of the square cone type energy absorption structure; the PCF is the peak force of the square cone type energy absorption structure.
Preferably, a pareto solution set of an optimization target is obtained based on an optimization result of the GRSM, and an optimal decision is performed on the pareto solution set obtained by optimization by using a minimum distance method to obtain an optimal solution, specifically:
absorbing energy EA, specific energy SEA and average force F mean And the energy absorption efficiency IFE is used as an energy absorption evaluation index, and the feasibility of the optimization method is verified by comparing an optimal solution with an experimental result by adopting a radar map:
wherein D is the distance between the knee joint point and the pareto solution point; m is the number of optimization objectives, f i k An optimal point k of the ith optimization target;
wherein m is the mass of the square cone type energy absorption structure; s is the compression displacement of the square cone type energy absorption structure; f mean The PCF is the peak force of the square cone type energy absorption structure; IFE is the energy absorption efficiency of the square cone type energy absorption structure.
The invention has the following beneficial effects:
1. the invention provides a machine learning-based multi-objective optimization method for collision resistance of a square cone type energy absorption structure, which extracts structural parameters and an energy absorption characteristic curve of a subway train energy absorption structure based on a method combining experimental design and finite element simulation. Meanwhile, an energy absorption characteristic curve (force-displacement curve) of the energy absorption structure is predicted by using four neural network models, namely a multilayer perceptron (MLP), a Recurrent Neural Network (RNN), a long-time memory method (LSTM) and a gated recurrent neural network (GRU), and the accuracy of the predicted output curve of the GRU model is high. Therefore, the invention adopts the gated recurrent neural network model (GRU) as the optimal energy-absorbing characteristic curve prediction model of the square cone type energy-absorbing structure of the subway train, so that the accuracy of energy-absorbing characteristic prediction is high, and the reliability of the prediction characteristic curve is ensured. On the premise of ensuring the calculation efficiency, the prediction precision can be improved, and the reliability of a multi-objective optimization result is ensured.
2. The invention provides a machine learning-based multi-objective optimization method for collision resistance of a square cone type energy absorption structure, which adopts a method of combining a global adaptive response surface (GRSM) and machine learning to carry out multi-objective optimization on the collision resistance of the energy absorption structure of a subway train. When the data needing to be calculated are large, the calculation time can be greatly reduced, and the crashworthiness of the square cone type energy absorption structure of the metro vehicle is enabled by large data through multi-objective optimization.
When the calculated data volume is large, the traditional method of combining the finite element and the optimization algorithm is adopted to carry out multi-objective optimization, and a long time is needed to generate a data set; the time for generating the data set can be shortened by adopting the machine learning method, the GRSM and the machine learning method are combined, the parameter optimization range of the energy absorption structure can be expanded, the mode identification is carried out on the energy absorption structure, the calculation efficiency is greatly improved, the upgrading and updating capacity of a product is improved, and the research and development cost of the product is reduced.
The optimization algorithm can not improve too much calculation speed, mainly adopts machine learning to reduce the time for generating a data set, and the optimization process is based on the generated data set for optimization, so that the method combining the machine learning and the optimization algorithm can improve the calculation efficiency.
3. The invention provides a machine learning-based multi-objective optimization method for collision resistance of a square cone type energy absorption structure, and provides a new idea for multi-objective optimization of the energy absorption structure. At present, the square cone type energy absorption structure is mainly characterized in that a finite element modeling and optimization algorithm are combined to optimize the energy absorption structure, the method is single, and a new thought can be provided for the optimization of the energy absorption structure by adopting the method provided by the invention.
In addition to the objects, features and advantages described above, other objects, features and advantages of the present invention are also provided. The present invention will be described in further detail below with reference to the drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a technical flow diagram of the present invention;
FIG. 2 is a diagram of the ML (Machine Learning) framework of the predictive model of the present invention;
FIG. 3 is a comparison graph between predicted output value curves and true output curves for different deep learning models;
FIG. 4 is a radar comparison graph of experimental results and optimized results of the present invention;
FIG. 5 is a time-contrast plot of DOEs with GRU and FE models calculating different sample numbers;
FIG. 6 is a diagram of an experimental scenario of the present invention;
FIG. 7 is a comparison of the experiment and simulation of the present invention: (ii) (a) a force-displacement curve; (b) an energy-displacement curve;
FIG. 8 is a comparison of experimental and simulated deformation sequences according to the present invention.
Detailed Description
Embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways, which are defined and covered by the claims.
The process of the invention is shown in figure 1: firstly, verifying the accuracy of a finite element model by adopting a mode of combining experiments and finite element simulation, and then changing the structural parameters of a square cone type energy absorption structure by utilizing an experimental Design (DOE) based on the finite element model to generate a training set and a test set for machine learning; introducing four machine models of a multilayer perceptron (MLP), a Recurrent Neural Network (RNN), a long-time memory method (LSTM) and a gated recurrent neural network (GRU), predicting the energy absorption characteristic of a pyramid energy absorption structure, comparing the prediction accuracy of the four models through different accuracy indexes, and comparing the machine learning model pair shown in FIG. 2; finally, taking a gated recurrent neural network (GRU) as a model which is most suitable for predicting the energy absorption characteristic of the pyramid energy absorption structure; and finally, taking a gated recurrent neural network (GRU) as a proxy model, and performing multi-objective optimization on the collision resistance of the pyramid energy-absorbing structure by adopting a self-adaptive Global Response Surface (GRSM) method to obtain an optimal solution.
Specifically, the invention provides a machine learning-based multi-target optimization method for collision resistance of a square cone type energy absorption structure, which comprises the following steps of:
step one, establishing a finite element simulation model of a square cone type energy absorption structure of the subway train. And based on the same boundary conditions of finite element simulation, carrying out full-size dynamic impact experiment of the square cone type energy absorption structure, and verifying the accuracy of the established finite element model by comparing a force-displacement curve, a displacement-energy curve and a deformation sequence mode between the experiment and the simulation in a mode of combining the experiment and the simulation.
Step two, the thickness T of the outer wall of the square cone type energy absorption structure A Outer and innerWall thickness T B Thickness T of the partition board gb Strength delta of aluminum Honeycomb A A And strength delta of aluminum honeycomb B B As a design variable, a force-displacement output curve of the pyramid type energy absorption structure is used as output, a Ding Chao cubic pulling method is adopted for sampling, virtual experiment Design (DOE) is carried out, and the DOE times are set to be 1000 groups.
Taking five input variables of the DOE as input features of network training in machine learning, and taking output of the DOE as actual output of the network training in the machine learning; then, randomly dividing the input and the actual output of the network training into a training set and a test set by utilizing a train _ test _ split () function, wherein 75% of data is used as the training set, and 25% of data is used as the test set; finally, the MinMaxscale () function of the Pythrch is adopted to respectively carry out normalization processing on the input data and the actual output data, and the calculation equation of the normalization is as follows:
wherein, i in the formula is the ith input characteristic, i =1, …,5; d represents the number of sample samples, d =1, …,1000; t represents the dimension of the output result, t =1, …,121;
is the regularization value of the ith input feature of the d sample; x d,i The value of the ith input feature for the d sample;
is the regularization value of the corresponding output of the ith input feature of the d sample; y is d,i Is the true output corresponding to the ith input feature of the d sample; x and Y represent input and output samples; min (X) and min (Y) are respectively the minimum values of the sample input and the real output; max (X) andmax (Y) is the maximum of the sample input and the true output, respectively.
Mapping an input geometric figure of the energy absorption structure into an ML system structure of a required output response, using a normalized training set and a normalized test set as samples, and dividing the training samples and the test samples into batches, wherein the number of the samples in each batch is 10; then, using the Mean Square Error (MSE) as the calculation mode of the loss function MSE of the optimization process, as shown in formula 3, using Adam algorithm as the deep learning optimizer, and calculating the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE), and the regression coefficient (R) 2 ) And a peace phase relative error absolute value (RAAE) is used as an assessment index, four neural network models of a multilayer perceptron (MLP), a Recurrent Neural Network (RNN), a long-time memory method (LSTM) and a gated recurrent neural network (GRU) are respectively used for predicting an energy absorption characteristic curve (force-displacement curve) of an energy absorption structure, the prediction accuracy of different models is analyzed by comparison, and FIG. 2 is an ML frame schematic diagram of a prediction model.
From fig. 3 and 4, it can be seen that the fit of the GRU predicted output curve to the real output curve is high, and the loss function is maintained at a small value, which indicates that the accuracy of the GRU model predicted output curve is high. The assessment indexes of the four models in the training period are as R 2 The larger and smaller MAE, RMSE and RAAE indicate that the trained model is more accurate, while GRU is more stationary than the other three models, MAE, RMSE and RAAE are kept at a lower value, R 2 The GRU loss is always kept at a larger value, which indicates that the GRU loss is higher in precision than the other three models.
Wherein n is the number of samples,
predicted output of corresponding neural network for machine learning, y i Is the actual output of the sample.
Step five, through comparative analysis of the prediction effects of the four machine learning network models of MLP, RNN, LSTM and GRU, GRU is used as a prediction model of the energy absorption characteristic curve of the square cone type energy absorption structure of the subway train, and the thickness T of the outer wall is used A Outer wall thickness T B Thickness T of the partition board gb Strength delta of aluminum honeycomb A A And strength delta of aluminum honeycomb B B As optimization variables, the absorption Energy (EA) and the peak force (PCF) of the square cone type energy absorption structure are used as optimization targets, and the established optimization theoretical model is shown as a formula 9; then, resampling the optimization by adopting a Ha Mosi Lei Caiyang method (Hammersley), and calculating corresponding absorption Energy (EA) and peak force (PCF) by utilizing a GRU network model to generate a new DOE; finally, a Global Response Surface Method (GRSM) is adopted to carry out multi-objective optimization, and parameters of a GRSM optimization algorithm are shown in Table 1.
PCF＝max(F(s)) (8)
Wherein s is the compression displacement of the square cone type energy absorption structure; f(s) is the axial force of the square cone type energy absorption structure; t is A And T B The thickness of the outer wall of the square cone type energy absorption structure; t is a unit of gb The thickness of the baffle plate is of a square cone type energy absorption structure; delta A And delta B The strength of the aluminum honeycomb a and the aluminum honeycomb B, respectively.
Wherein EA is the absorption energy of the square cone type energy absorption structure; the PCF is the peak force of the square cone type energy absorption structure;
TABLE 1 parameters of GRSM Algorithm
And 6, obtaining a pareto solution set of an optimization target based on the optimization result of the GRSM, and performing optimal decision on the pareto solution set obtained by optimization by adopting a minimum distance method (shown in a formula 9 below) to obtain an optimal solution. Energy Absorption (EA), specific Energy Absorption (SEA) and average force (F) mean ) And energy absorption efficiency (IFE) is used as an energy absorption evaluation index, and the radar chart is adopted to compare the optimal solution with the experimental result (as shown in figure 4), so that the feasibility of the optimization method is verified
Wherein D is the distance between the knee joint point and the pareto solution point; m is the number of optimization objectives, f i k Is the optimal point k of the ith optimization objective.
Wherein m is the mass of the square cone type energy absorption structure; s is the compression displacement of the square cone type energy absorption structure; f mean Being the average force of a square-cone energy-absorbing structure, PCF being of a square-cone energy-absorbing structurePeak force; IFE is the energy absorption efficiency of the square cone type energy absorption structure.
To further illustrate the efficiency of the optimization process, it is necessary to compare the computation time of the optimization process using the GRU model and the finite element model, respectively. In the model optimization process, the training time of the GRU is 100.5 hours, the single calculation time when the model training is completed is 0.5 seconds, the single calculation time of the finite element model is 1 hour, and a comparison graph of the time required for the GRU and FE models to calculate the DOE without the number of samples is shown in fig. 5. The calculation time of the optimization process of the GRU and FE models is shown in equations 14-16.
T ML ＝t train +n×t ml (14)
Wherein, T ML Total computation time, t, required to compute DOE for machine-learned GRU models train Time required for GRU training, t ml After the model training is successful, the time required for calculating each sample is n is the number of the DOE samples
T FE ＝n×t FE (15)
Wherein, T FE Total computation time, t, required for the computation of DOE for finite element models FE The time required for calculating each sample for the finite element model, n is the number of samples of the DOE.
Wherein, T ML Total computation time required to compute DOE for the machine learning GRU model; t is a unit of FE The total computation time required for the computation of the DOE for the finite element model; the TIEF calculates the time ratio required by the DOE for the machine learning GRU model and the finite element model.
As can be seen from fig. 5, when n =500,t ML And T FE 600.57 and 500, respectively, TIFE 120.12%, and GRU requires more time than finite element model calculation. When n =5000,T ML And T FE 601.19 and 5000, respectively, TIFE is 5.58%, and GRU time is less thanThe finite element model calculates the time. When n is smaller, the calculation time of GRU is longer than that of FE, the calculation of GRU is basically kept unchanged along with the increase of the number of n, the calculation time of finite element model calculation time is greatly increased, and TIFE presents exponential reduction. Therefore, it can be found that when the number of samples sampled in the optimization process is large, the calculation time can be greatly reduced by using the machine learning GRU model, the machine learning GRU model can enable the optimization process to realize large data enabling, and the optimization of the pyramid energy-absorbing structure model is brought into a new field.
The invention is explained and illustrated below with reference to specific examples.
In order to observe the energy absorption characteristics and the behavior mechanism of the end energy absorption structure in the high-speed collision process, a full-size collision test is carried out on the end energy absorption structure on a standard rail. As shown in FIG. 6, the entire experimental system is mainly composed of a launcher for providing a certain speed, a trolley, a force-equalizing plate for impact, a dynamic force sensor installed between the rigid wall and the force-equalizing plate, a speedometer for recording the incident speed of the structure, and a high-speed camera for capturing the impact process. The end energy absorption structure is fixed at the front end of the impact trolley. The trolley is dragged to the far end of the impact point, and the trolley is driven by a motor driving device to impact the sample at the initial speed of 17.9 km/h. The total weight of the trolley is 16.1t.
In order to verify the collision resistance of the square cone type energy absorption structure and verify the correctness of the finite element modeling method, an impact test trolley system is used for performing an impact test. Meanwhile, the numerical value and the theoretical result can be further verified through the experimental result. Fig. 7 (a) shows a comparison of experimental and numerical results in terms of force-displacement curves. The results show that in experiments and simulations, the square-cone energy absorbing structure forms an initial peak force after contact with the rigid wall, and then rapidly decreases. Considering that experimental conditions are more complex than simulation conditions, the force-displacement curves cannot be completely consistent. Nevertheless, it was found that the initial peak force quantities of both were the same, that 12 force peaks were formed, and that the amplitude of the initial peak force was substantially uniform. As shown in fig. 7 (b), the experimental and simulation results are compared in terms of energy-displacement curves. The experimental and simulated energy-displacement curves were consistent throughout the process. For better comparison, the values of the crashworthiness index and the error index are summarized in the table. For EA, IPCF, d, and MCF, the experimental and numerical results were 220.38kJ and 216.86kJ, 549.69kN and 526.67kN, 702.09mm and 711.92mm, 313.89 kN, and 304.61kN, respectively. The data show that the simulation results are in good agreement with the experimental values.
Furthermore, the proposed theoretical model predicts a theoretical dynamic MCF of the end energy absorbing structure of 296.22kN. The dynamic steady-state force predicted by theory is well matched with an impact test, the relative error is 5.63%, and the requirement on engineering calculation precision is met. Therefore, the proposed theoretical model and the constructed finite element model have sufficient accuracy to study the energy absorption characteristics of the end energy absorbing structure.
TABLE 2 comparison of experimental and simulation results
Accurate prediction of the impact response is not the only criterion and requires efficient prediction of the deformation mode. The experimental and simulated deformation process of the pyramid-type energy absorbing structure is shown in fig. 8. It can be seen that the deformation mode of the pyramid-type energy-absorbing structure in the simulation is basically consistent with the experimental result in the collision process. Whether experimental or simulated, the deformation is stable and orderly from the collision end to the rear end, and finally a regular shape is formed. Due to the presence of the membrane, 12 folds occurred during both the simulation and the experimental deformation.
In conclusion, the force-displacement, energy-displacement and structural deformation modes in the numerical simulation are well matched with the impact experiment, and the simulation model has good precision and can be used for follow-up research.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. The machine learning-based multi-objective optimization method for the collision resistance of the square cone type energy absorption structure is characterized by comprising the following steps of:
establishing a finite element simulation model of a square cone type energy absorption structure of the subway train;
extracting the structural parameters and the energy absorption characteristic curve of the subway train energy absorption structure based on a method of combining the established finite element simulation model with experimental design; according to the method, a Latin hypercube method is adopted for sampling, virtual experiment design is carried out, an optimal energy absorption characteristic curve prediction model of a subway train pyramid type energy absorption structure is determined, and variables and an optimization target are optimized;
establishing an optimization theoretical model according to the optimal energy absorption characteristic curve prediction model, the optimization variables and the optimization target;
resampling the optimization theoretical model by adopting a Ha Mosi Lei Caiyang method, and calculating corresponding absorption energy and peak force by utilizing an optimal energy absorption characteristic curve prediction model to generate a new DOE; performing multi-objective optimization by adopting a global response surface method to obtain an optimization result;
and obtaining a pareto solution set of an optimization target based on an optimization result, and performing optimal decision on the pareto solution set obtained by optimization by adopting a minimum distance method to obtain an optimal solution.
2. The machine learning-based multi-objective optimization method for collision resistance of the square cone type energy absorption structure according to claim 1, wherein a full-size dynamic impact experiment of the square cone type energy absorption structure is performed based on the same boundary conditions of finite element simulation, and the accuracy of the established finite element model is verified by comparing a force-displacement curve, a displacement-energy curve and a deformation sequence pattern between the experiment and the simulation in a mode of combining the experiment and the simulation.
3. The machine learning-based multi-objective optimization method for the collision resistance of the pyramid type energy absorption structure according to claim 1, characterized in that structural parameters and energy absorption characteristic curves of the subway train energy absorption structure are extracted based on a method combining an established finite element simulation model and experimental design; according to the method, a Latin hypercube method is adopted for sampling, virtual experiment design is carried out, an optimal energy absorption characteristic curve prediction model of a subway train pyramid type energy absorption structure is determined, and variables and an optimization target are optimized, and the method specifically comprises the following steps:
taking five input variables of the DOE as input features of network training in machine learning, and taking the output of the DOE as actual output of the network training in the machine learning;
randomly dividing the input and the actual output of the network training into a training set and a test set;
normalizing the input data and the actual output data;
mapping an input geometric figure of the energy absorption structure into a Machine Learning (ML) system structure of a required output response, using a normalized training set and a normalized test set as samples, respectively using four machine learning network models of MLP, RNN, LSTM and GRU to predict an energy absorption characteristic curve of the energy absorption structure, adopting comparison analysis to predict the accuracy of different models, and using the network model with the highest accuracy of the predicted output curve as an optimal energy absorption characteristic curve prediction model of the square cone type energy absorption structure of the subway train.
4. The machine learning-based multi-objective impact resistance optimization method for the pyramid energy absorption structure of the subway train as claimed in claim 1, wherein the optimal energy absorption characteristic curve prediction model of the pyramid energy absorption structure of the subway train is a gated recurrent neural network model, and the outer wall thickness T is used A Outer wall thickness T B Thickness T of the partition board gb Strength delta of aluminum honeycomb A A And strength delta of aluminum honeycomb B B As optimization variables, the absorption energy and the peak force of the square cone type energy absorption structure are optimization targets.
5. The machine learning-based multi-objective optimization method for crash resistance of the pyramid energy absorption structure according to claim 1, wherein the established optimization theoretical model is as follows:
PCF＝max(F(s))
s is the compression displacement of the square cone type energy absorption structure; f(s) is the axial force of the square cone type energy absorption structure; t is A And T B The thickness of the outer wall of the square cone type energy absorption structure; t is gb The thickness of the baffle plate is of a square cone type energy absorption structure; delta A And delta B The strength of the aluminum honeycomb A and the strength of the aluminum honeycomb B are respectively; EA is the energy absorption of the square cone type energy absorption structure; the PCF is the peak force of the square cone type energy absorption structure.
6. The machine learning-based multi-objective optimization method for crash resistance of the pyramid energy absorption structure according to claim 1, wherein a pareto solution set of an optimization objective is obtained based on an optimization result of GRSM, and an optimal decision is performed on the pareto solution set obtained by optimization by using a minimum distance method to obtain an optimal solution, specifically:
absorbing energy EA, specific energy SEA and average force F mean And the energy absorption efficiency IFE is used as an energy absorption evaluation index, and the radar map is adopted to compare the optimal solution with the experimental result, so that the feasibility of the optimization method is verified:
wherein D is the distance between the knee joint point and the pareto solution point; m is the number of optimization objectives, f i k An optimal point k of the ith optimization target;
wherein m is the mass of the square cone type energy absorption structure; s is the compression displacement of the square cone type energy absorption structure; f mean The PCF is the peak force of the square cone type energy absorption structure; IFE is the energy absorption efficiency of the square cone type energy absorption structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211151523.7A CN115481488B (en) | 2022-09-21 | 2022-09-21 | Fang Zhui type energy-absorbing structure crashworthiness multi-objective optimization method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211151523.7A CN115481488B (en) | 2022-09-21 | 2022-09-21 | Fang Zhui type energy-absorbing structure crashworthiness multi-objective optimization method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115481488A true CN115481488A (en) | 2022-12-16 |
| CN115481488B CN115481488B (en) | 2024-03-01 |
Family
ID=84423642
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211151523.7A Active CN115481488B (en) | 2022-09-21 | 2022-09-21 | Fang Zhui type energy-absorbing structure crashworthiness multi-objective optimization method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115481488B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116911144A (en) * | 2023-09-11 | 2023-10-20 | 西南交通大学 | Train collision energy management optimization method based on machine learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106081126A (en) * | 2016-06-13 | 2016-11-09 | 王晨 | Bionical cellular active safety escape compartment embeds application and the design of aviation aircraft |
| CN109697309A (en) * | 2018-12-05 | 2019-04-30 | 西北机电工程研究所 | A kind of bullet high speed impact gets into resistance fast acquiring method |
| CN110020466A (en) * | 2019-03-19 | 2019-07-16 | 南京理工大学 | Negative poisson's ratio structure energy-absorption box optimization design method based on agent model |
| CN110309601A (en) * | 2019-07-04 | 2019-10-08 | 中南大学 | A kind of optimization method and system for long volume EMU impact energy management |
| WO2022007409A1 (en) * | 2020-07-08 | 2022-01-13 | 大连理工大学 | Curve reinforced structure layout intelligent design method based on image feature learning |
-
2022
- 2022-09-21 CN CN202211151523.7A patent/CN115481488B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106081126A (en) * | 2016-06-13 | 2016-11-09 | 王晨 | Bionical cellular active safety escape compartment embeds application and the design of aviation aircraft |
| CN109697309A (en) * | 2018-12-05 | 2019-04-30 | 西北机电工程研究所 | A kind of bullet high speed impact gets into resistance fast acquiring method |
| CN110020466A (en) * | 2019-03-19 | 2019-07-16 | 南京理工大学 | Negative poisson's ratio structure energy-absorption box optimization design method based on agent model |
| CN110309601A (en) * | 2019-07-04 | 2019-10-08 | 中南大学 | A kind of optimization method and system for long volume EMU impact energy management |
| WO2022007409A1 (en) * | 2020-07-08 | 2022-01-13 | 大连理工大学 | Curve reinforced structure layout intelligent design method based on image feature learning |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116911144A (en) * | 2023-09-11 | 2023-10-20 | 西南交通大学 | Train collision energy management optimization method based on machine learning |
| CN116911144B (en) * | 2023-09-11 | 2023-11-21 | 西南交通大学 | Train collision energy management optimization method based on machine learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115481488B (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Marzbanrad et al. | Multi-objective optimization of aluminum hollow tubes for vehicle crash energy absorption using a genetic algorithm and neural networks | |
| Zhang et al. | Vehicle crash accident reconstruction based on the analysis 3D deformation of the auto-body | |
| Karimi et al. | Signal reconstruction, modeling and simulation of a vehicle full-scale crash test based on Morlet wavelets | |
| Pawlus et al. | Investigation of vehicle crash modeling techniques: theory and application | |
| CN108595898A (en) | Finite element modeling method based on vehicle collision simulation and system | |
| CN115577436B (en) | Combined deep learning method for solving wind-induced vibration response of uncertain structure | |
| CN115481488A (en) | Machine learning-based multi-target optimization method for collision resistance of square cone type energy absorption structure | |
| CN114997027B (en) | Method for intelligently solving random signals of axle system | |
| CN106056147A (en) | System and method for establishing target division remote damage assessment of different vehicle types based artificial intelligence radial basis function neural network method | |
| Koch et al. | Machine learning for predicting the damaged parts of a low speed vehicle crash | |
| CN117669389B (en) | Random vibration analysis method for earthquake-vehicle-bridge system based on deep learning | |
| Yessoufou et al. | One-class convolutional neural network (OC-CNN) model for rapid bridge damage detection using bridge response data | |
| Guo et al. | Machine learning-based crashworthiness optimization for the square cone energy-absorbing structure of the subway vehicle | |
| Qiu et al. | Recurrent neural network based driving cycle development for light duty vehicles in Beijing | |
| Liang et al. | 1d convolutional neural networks for fault diagnosis of high-speed train bogie | |
| Wang et al. | Bridge weigh-in-motion through bidirectional Recurrent Neural Network with long short-term memory and attention mechanism | |
| CN117313248A (en) | Vehicle body structure optimization method and system considering vehicle body performance and collision injury | |
| Xiang et al. | Vibration-based structural damage identification using P-CNN and time-frequency hybrid index under the conditions of uncertainties and incomplete measurements | |
| Shi et al. | Train-induced vibration response reconstruction for bridge damage detection with a deep learning methodology | |
| Fu et al. | Prediction of automobile aerodynamic drag coefficient for SUV cars based on a novel XGBoost model | |
| Chen et al. | A comparative study of vehicle frame load spectrum extraction using SA-1DCNN deep learning and virtual iteration | |
| CN106067038A (en) | Point long-range loss assessment system of vehicle and a method is set up based on artificial intelligence's supervised learning Nae Bayesianmethod | |
| Pawlus et al. | Performance evaluation of feed forward neural networks for modeling a vehicle to pole central collision | |
| Borse et al. | Development of a machine learning-based design optimization method for crashworthiness analysis. | |
| CN102999672A (en) | Parallel support vector machine approximate model optimization method based on automobile crashworthiness |
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 |