Performance prediction method of dual-phase composite pipe based on machine learning

Abstract

The present invention discloses a method for predicting the performance of a two-phase composite pipe made of a material based on machine learning. The method uses a finite element method to establish a composite pipe model, selects the properties of metal and composite materials as part of a database, cleans the data in the database, uses PCA to screen variables, and obtains variables for subsequent model inclusion. A training data set and a test data set are randomly selected to establish a BP neural network model. A ten-fold cross-validation method combined with an R^2 evaluation method is used to determine the final data set selection method and the corresponding weights and thresholds. The final neural network model is established, and the prediction results of the model are evaluated using the test data set. The present invention combines machine learning with known data to predict the energy absorption performance of the composite pipe using a data-driven model. When the composite pipe parameters change, a new model can be obtained to adapt to the new composite pipe simply by changing the finite variables in the model, saving experimental and simulation time.

Description

Technical Field

[0001] The present invention belongs to the technical field of automobile passive safety, and in particular relates to a method for predicting the performance of a dual-phase material composite pipe based on machine learning. Background Art

[0002] With the rapid increase in the number of cars in my country, the importance of passive safety of cars is also increasing day by day. Traditional energy-absorbing structures usually use metal as raw materials, but their performance is not well utilized. On this basis, composite pipes are born.

[0003] Hybrid tubes are made of metal and composite materials and have better energy absorption performance than simple metal tubes. They have good application prospects. However, most of the existing research on the energy absorption performance of hybrid tubes uses quasi-static axial compression to explore the influence of geometric factors and failure modes on energy absorption capacity, such as "Energy absorption and failure pattern of hybrid composite tubes under quasi-static axial compression", Yang, Haiyang et al., Composites Part B: Engineering, Vol. 198, August 2020 and "Buckling behavior of CFRP-aluminum alloy hybrid tubes in axial compression", Peng Feng et al., Engineering Structures, Vol. 132, pp. 624-636, February 2017. While the two aforementioned papers explored some of the factors influencing energy absorption performance, they did not provide a comprehensive system of influencing factors. Furthermore, they required extensive experiments and simulations to obtain a large amount of data. Furthermore, the data obtained was only used to optimize the design of the performance of dual-phase composite tubes, without considering further utilization of the data. When deducing the relationship between different factors and final performance, formulas were generally used. For example, the paper “Buckling behavior of CFRP-aluminum alloy hybrid tubes in axial compression” provides a cylindrical curve design formula for long CFRP-Al hybrid tubes based on finite element analysis and the traditional Perry-Robertson formula. However, this approach not only requires a strong theoretical foundation but also manual selection of parameter values. Furthermore, in practical applications, only structures that have been experimentally or simulated can be applied. If the structural type of the composite tube undergoes significant changes, further simulation or experimentation is required to obtain the energy absorption value, limiting its wide-ranging application and consuming a considerable amount of time. Summary of the Invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and to provide a performance prediction method for dual-phase material composite pipes based on machine learning to solve the problem that the energy absorption performance in common composite pipe applications is difficult to predict. This prediction method can make full use of the obtained data and has a complete set of influencing factors. It can more accurately predict the energy absorption performance of the composite pipe. When deducing the relationship between different factors and the final performance, there is no need to use formulas for deduction. When the geometric parameters or types in the composite pipe structure change, a new model can be obtained to adapt to the new structure by simply changing the finite variables in the model, thereby expanding its scope of application and saving time and processing costs.

[0005] To achieve the above objectives, the present invention provides a method for predicting the performance of a dual-phase composite pipe based on machine learning, comprising the following steps:

[0006] S1: A composite tube model is established using the finite element method, with properties of metals and composite materials selected as part of the database. For the metal part, common mechanical properties of metals are included, including normal elastic modulus, shear elastic modulus, proportional limit, elastic limit, ultimate strength, tensile strength, yield limit, yield strength, and fatigue limit. For the composite part, common mechanical properties of composite materials are included, and different mechanical properties can be selected based on the type of composite material selected.

[0007] S2: Clean the existing data; adjust the attribute parameters in the S1 database through experiments and simulations, and clean and eliminate the data based on the impact on the final performance;

[0008] S3: PCA is used to screen variables and obtain variables to be subsequently included in the model. Variables are screened based on the contribution of each factor in the data obtained in S2. PCA can retain most of the factors with large contributions, and fit several variables with small contributions into new variables with large contributions. The original variables with large contributions that were retained and the newly synthesized variables are used as the total database.

[0009] S4: randomly selecting a training data set and a test data set for establishing a BP neural network model; using a ten-fold cross-validation method to divide the total database in S3 into ten groups, randomly extracting 80% of the data (i.e., eight of the groups) as the training data set for the BP neural network, and using the remaining 20% ​​of the data (i.e., the remaining two groups) as the test data set for the model; repeating this operation to obtain a total of 45 different training sets and corresponding test sets; the training data set is used to determine the parameters of the BP neural network model and establish the model, and the test data set is used to evaluate the prediction effect of the subsequent model establishment;

[0010] S5: Build BP neural network models using all the training data sets obtained in S4 and their corresponding test data sets, for a total of 45 models. Each model is run 10 times, and the optimal weights and thresholds for each model are determined based on the R^2 results of the model fitting in the 10 times. Use the 10-fold cross-validation method to verify the data set selection method and the corresponding weights and thresholds for the optimal model among the 45 models combined with the R^2 evaluation method.

[0011] S6: Determine network parameters, including the number of network layers, the number of network nodes, the activation function, and select a single hidden layer; the formula for the number of hidden layer nodes is:

[0012]

[0013] Where: N h is the number of hidden layer nodes, N in is the number of input layer nodes. For BP neural network, the number of input layer nodes is the number of input features, N out is the number of output layer nodes, h is a regulation constant ranging from 1 to 10;

[0014] S7: Evaluate the prediction results of the established neural network model using the test data set; apply the neural network model established based on the parameters in S6 to the test data set divided in S4 for model verification, and compare the data in the actual test data set obtained through experiments or simulations with the results predicted by the model. Evaluate the prediction ability of the model through the network training regression in the network model results;

[0015] S8: Use the accuracy evaluation formula to evaluate the model prediction results in S7; the accuracy evaluation formula is:

[0016]

[0017] Where R is the accuracy of the model prediction results, n is the number of correct tests, and N is the total number of tests.

[0018] S9: Appropriately adjust the proportion of the selected random data to test the model results; adjust the proportion of the training data set and the test data set in S4, and randomly select 90% or 70% as the training data set, and use the remaining 10% or 30% as the test data set, and repeat the above S5 to S8 to determine the difference between the model prediction results and the actual results, and evaluate the degree of model dependence on the proportion.

[0019] As a further optimization, the types of composite materials in S1 include polymer composites, ceramic-based composites and metal composites. Since this application is aimed at applications in aspects such as lightweighting of automobiles, polymer composites are selected as the research target. The mechanical performance characteristics of the polymer composites include longitudinal / transverse tensile strength, longitudinal / transverse compressive strength, longitudinal / transverse elastic modulus, principal Poisson's ratio, and shear modulus.

[0020] Advantages and beneficial effects of the present invention

[0021] 1. The present invention combines machine learning with known data and uses a data-driven model to predict the energy absorption performance of the composite pipe, thereby improving data utilization.

[0022] 2. The present invention takes into account comprehensive influencing factors. When establishing a composite pipe model using the finite element method and selecting the properties of metals and composite materials as part of the database, the metal part includes common mechanical properties of metals, such as the metal's positive elastic modulus, shear elastic modulus, proportional limit, elastic limit, strength limit, tensile strength, yield limit, yield strength, fatigue limit, etc. For the composite part, different mechanical properties are also selected according to the type of composite material selected. If a polymer composite material is used, the mechanical performance characteristics including the composite material's longitudinal / transverse tensile strength, longitudinal / transverse compressive strength, longitudinal / transverse elastic modulus, principal Poisson's ratio, shear modulus, etc. are all taken into account, thereby forming a complete system of influencing factors and providing a sufficiently accurate prediction of the energy absorption performance of the composite pipe.

[0023] 3. The present invention omits the step of manually selecting parameter values ​​in the derivation of the fitting formula, and does not require a strong theoretical basis. Instead, it simply relies on the objective and rational analysis of the data by the BP neural network, thereby reducing the interference of external uncertainties, effectively avoiding the influence of subjective factors caused by humans in the prediction process, and improving the accuracy of the prediction results in the whole process;

[0024] 4. The present invention has strong generalization capabilities. When the structural type of composite tubes undergoes significant changes, especially for those with irregular shapes, the energy absorption performance of these tubes can be predicted by simply changing the limited material, structure-related variables, and network parameters in the model. A new model can be generated to adapt to the new composite tubes, eliminating the need for repeated simulations or experiments to obtain energy absorption values. This expands its scope of application and saves experimental and simulation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0026] Figure 1 is a flow chart of the prediction method of the present invention;

[0027] Figure 2 It is a schematic diagram of the network structure of the present invention;

[0028] Figure 3 Schematic diagram of the principal component contribution rate of the present invention;

[0029] Figure 4 Schematic diagram of the distribution of training set and test set of the present invention;

[0030] Figure 5 This is a schematic diagram of the results after the model training regression of the present invention;

[0031] Figure 6 This is a comparison chart of the energy absorption prediction results of the test set of the present invention and the true value;

[0032] Figure 7 It is a schematic diagram of the model established by the finite element method of the present invention;

[0033] Figure 8 is a top view of the composite pipe of the present invention;

[0034] Figure 9 It is a schematic diagram of the constrained loading of the composite tube of the present invention. DETAILED DESCRIPTION

[0035] The ultimate energy absorption of a composite pipe is influenced by multiple factors: macroscopically, these factors include its external shape, internal structure, selected materials, and the connection between the metal and non-metal. Furthermore, the pipe itself involves both composite and dual-phase metals, and the relationships between material composition, processing, organization, structure, and performance are extremely complex and involve numerous interactions, making it difficult to obtain accurate results through methods such as formula derivation. Therefore, this application was developed to fully utilize existing data, reduce human interference in performance calculations, improve prediction accuracy and computational efficiency, and reduce the time required to predict performance.

[0036] The following example uses a composite tube formed by wrapping an aluminum alloy round tube with carbon fiber as an example. However, all other composite structures formed by combining metal and composite materials are within the scope of the present invention. In this example, the performance indicators are evaluated based on energy absorption (EA).

[0037] like Figure 7-9 As shown, the cross-sectional shape of the composite tube can be other regular shapes or irregular shapes. For irregular shapes, they can be split into relatively regular triangles, circles, rectangles or other regular shapes by fitting approximation, so as to extract feature data in the next step. The specific prediction method is as follows (see Figure 1 ):

[0038] S1: Build a composite tube model using the finite element method, selecting properties of metals and composite materials as part of the database. For the metal part, this example uses aluminum alloy, including common mechanical properties of aluminum alloys, aluminum alloy structure, materials, and parameters related to loading conditions. Common mechanical properties of aluminum alloys are shown in Table 1.

[0039] Table 1

[0040] Common mechanical properties

[0041]

[0042] For the composite material part, including common mechanical properties of composite materials, different mechanical properties can be selected according to the type of composite material selected; the types of composite materials include polymer composite materials (mainly fiber-reinforced polymer materials), ceramic-based composite materials, and metal-based composite materials. Since the present invention is intended for applications such as lightweight automobiles, polymer composite materials are selected as the research target. The composite material in this embodiment takes carbon fiber as an example, including common mechanical properties of carbon fiber, carbon fiber structure, materials, and parameters related to loading conditions. The common mechanical properties of carbon fiber are shown in Table 2:

[0043] Table 2

[0044] Composite material properties

[0045]

[0046] S2: Clean the existing data; adjust the attribute parameters in the S1 database through experiments and simulations, and clean and eliminate the data based on the impact on the final performance. Since there are many types of composite materials that have been used in automobiles, the types can be flexibly selected according to the application requirements, so as to conduct detailed screening of material properties. Since the relationship between the various properties in the two-phase material is not a strict linear relationship, in order to explore the correlation between the variables, it is necessary to construct a nonlinear BP neural network as the machine learning method of the present invention, such as Figure 2As shown in the figure, since the BP neural network with a single hidden layer can approach any continuous nonlinear curve, it can complete a more comprehensive feature extraction of the data. If this step is omitted and the missing data is not cleaned and eliminated, the representativeness and reliability of the network establishment cannot be ensured.

[0047] S3: Regarding the factors affecting the performance of the dual-phase composite pipe, principal component analysis (PCA) is used to screen variables and obtain the variables that are subsequently included in the model. Since there are many factors affecting the performance of the dual-phase composite pipe, and the contribution of each factor is different, experiments and simulations have shown that there is a linear relationship between some of the influencing factors. For most evaluation methods, it is difficult to eliminate the related effects between factors, so it takes a lot of effort to select variables, and PCA can solve this problem. In addition, based on the large number of influencing factors of the dual-phase composite pipe, PCA can screen variables according to the different contributions of each factor in the data obtained in S2, thereby retaining most of the factors with large contributions, and fitting several variables with small contributions into new variables with large contributions, and using the original variables with large contributions and the newly synthesized variables as the total database; such as Figure 3 As shown in the figure, by selecting variables with greater contribution, and then synthesizing the original redundant (multiple variables with smaller contribution) variables as much as possible into new variables that are subsequently incorporated into the model, we finally get four variables with larger contribution. It is not difficult to find in the figure that the sum of the contribution rates of the four principal components is close to 100%, so the replacement effect of the new variables is relatively good; if this step is omitted and the cleaned data is directly substituted into the model, the final result will be overfitted, and the prediction effect within a certain numerical range will not be good, so variable screening is required; and variable screening through PCA can improve the efficiency of the modeling process and ensure the accuracy of the final prediction results.

[0048] S4: The training data set and the test data set for establishing the BP neural network model are selected by random method; since the nonlinear relationship between the parameters of the composite pipe is strong, as much data as possible is needed for fitting, and the processing cost of the dual-phase material composite pipe is high, and the data obtained directly through experiments and simulations are relatively small. In order to make full use of the data, the training data set and the test data set are selected by the ten-fold cross-validation method. When selecting, the ten-fold cross-validation method divides the total database in S3 into ten groups, randomly extracts 80% of the data (i.e., eight groups) as the training data set of the BP neural network, and uses the remaining 20% ​​of the data (i.e., the remaining two groups) as the test data set of the model; repeat this operation to obtain a total of 45 different training sets and their corresponding test sets; the training data set is used to determine the parameters of the BP neural network model and establish its model, and the test data set is used to evaluate the prediction effect of the subsequent model establishment; such as Figure 4As shown in the figure, when the test set is basically covered by the training set, it means that most of the cases in the test set can be well predicted and the prediction effect of the model is very good.

[0049] S5: Build BP neural network models using all the training data sets obtained in S4 and their corresponding test data sets, for a total of 45 models. Each model is run 10 times. This is because for models with random initial points, after more than 10 repetitions, the error will fluctuate around a certain value and it will be difficult to reduce it further. The optimal weights and thresholds for each model are determined using the R^2 results of the model fitting in 10 times. The 10-fold cross-validation method is used to verify the data set selection method and the corresponding weights and thresholds for the optimal model among the 45 models combined with the R^2 evaluation method.

[0050] S6: Determine network parameters, including the number of network layers, the number of network nodes, the activation function, and select a single hidden layer. Here, since the BP neural network with a single hidden layer can approach any continuous nonlinear curve, a single hidden layer is selected. The single hidden layer is selected based on the nonlinear relationship in the bidirectional composite pipe, and the number of hidden layer nodes is determined by the hidden layer node number formula. The hidden layer node number formula is:

[0051]

[0052] Where: N h is the number of hidden layer nodes, N in is the number of input layer nodes. For BP neural network, the number of input layer nodes is the number of input features, N out is the number of nodes in the output layer, and h is a regulation constant ranging from 1 to 10. Different values ​​will affect the training error.

[0053] According to the principle of rounding, the number of nodes and other data need to be integerized during training. Compare the estimation accuracy and generalization ability to determine the optimal number of hidden layer nodes; here the number of nodes is set to 4, the target accuracy is 0.001, the maximum number of iterations is 1000, and the learning rate is 0.01. The final result is as follows Figure 5 shown.

[0054] S7: Evaluate the prediction results of the established neural network model through the test data set; apply the neural network model established based on the parameters in S6 to the test data set divided in S4 for model verification, and compare the data in the actual test data set obtained through experiments or simulations with the results predicted by the model, and evaluate the prediction ability of the model through the network training regression in the network model results, with the error kept within 5%. Figure 6As shown in the figure, the R^2 value, a more intuitive evaluation standard for the model prediction results, has reached 97.999%, indicating that the prediction effect is relatively accurate.

[0055] S8: Use the accuracy evaluation formula to evaluate the prediction results of the model in S7. The accuracy of a single prediction cannot accurately reflect the prediction performance of the model. Multiple predictions must be examined. The accuracy evaluation formula is:

[0056]

[0057] Among them, R is the accuracy of the model prediction results, n is the number of correct tests, and the evaluation standard for the number of correct tests is that if the difference between the value predicted by the model and the true value is within 5%, the model prediction result is considered correct. N is the total number of tests. The accuracy evaluation formula is used to evaluate the model prediction results, and the accuracy rate reaches more than 95%.

[0058] S9: Appropriately adjust the proportion of the selected random data to test the model results; when the data is proportionally distributed, adjust the proportion of the training data set and the test data set in S4, and randomly select 90% or 70% as the training data set, and use the remaining 10% or 30% as the test data set, and repeat the above S5 to S8 to perform repeated operations. By comparing the difference between the results predicted by the model and the actual results after the proportion is changed, the degree of dependence of the model on the proportion is evaluated. The model obtained by the 80% proportion selected in this embodiment has the best fitting effect.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the embodiments of the present invention have been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the embodiments of the present invention.

Claims

1. A method for predicting the performance of a dual-phase composite pipe based on machine learning, characterized by: The following steps are involved: S1: Establish a composite tube model using the finite element method, selecting the properties of metals and composite materials as part of the database. For the metal part, common mechanical properties of metals are included; for the composite part, common mechanical properties of composite materials are included. Different mechanical properties can be selected based on the type of composite material selected. S2: Clean the existing data; adjust the attribute parameters in the S1 database through experiments and simulations, and clean and eliminate the data based on the impact on the final performance; S3: PCA is used to screen variables and obtain variables to be subsequently included in the model. Variables are screened based on the contribution of each factor in the data obtained in S2. PCA can retain most of the factors with large contributions, and fit several variables with small contributions into new variables with large contributions. The original variables with large contributions that were retained and the newly synthesized variables are used as the total database. S4: The training data set and test data set for establishing the BP neural network model are selected by random method; The total database in S3 was divided into ten groups using the ten-fold cross-validation method. 80% of the data was randomly selected as the training data set for the BP neural network, and the remaining 20% ​​of the data was used as the test data set for the model. The training data set is used to determine the parameters of the BP neural network model and to establish the model, and the test data set is used to evaluate the prediction effect of the subsequent model establishment; S5: Establish multiple BP neural network models using all the training data sets obtained in S4 and their corresponding test data sets. Each neural network model is run 10 times repeatedly, and the optimal weights and thresholds of each model are determined based on the R^2 results of the model fitting in the 10 times. The 10-fold cross-validation method was used to verify the data set selection method and corresponding weights and thresholds of the optimal model in multiple models combined with the R^2 evaluation method; S6: Determine network parameters, including the number of network layers, the number of network nodes, the activation function, and select a single hidden layer; the formula for the number of hidden layer nodes is: Where: N h is the number of hidden layer nodes, N in is the number of input layer nodes. For BP neural network, the number of input layer nodes is the number of input features, N out is the number of output layer nodes, h is a regulation constant ranging from 1 to 10; S7: Evaluate the prediction results of the established neural network model using the test data set; apply the neural network model established based on the parameters in S6 to the test data set divided in S4 for model verification, and compare the data in the actual test data set obtained through experiments or simulations with the results predicted by the model. Evaluate the prediction ability of the model through the network training regression in the network model results; S8: Use the accuracy evaluation formula to evaluate the model prediction results in S7; the accuracy evaluation formula is: Where R is the accuracy of the model prediction results, n is the number of correct tests, and N is the total number of tests; S9: Appropriately adjust the proportion of the selected random data to test the model results; adjust the proportion of the training data set and the test data set in S4, and randomly select 90% or 70% as the training data set, and use the remaining 10% or 30% as the test data set, and repeat the above S5 to S8 to determine the difference between the model prediction results and the actual results, and evaluate the degree of model dependence on the proportion.

2. The method for predicting the performance of a dual-phase composite pipe based on machine learning according to claim 1, characterized in that: The types of composite materials in S1 include polymer composite materials, ceramic matrix composite materials and metal composite materials.

Images