Method and system for predicting flexural rigidity of composite material pipe fitting based on deep learning, and terminal

Abstract

The invention relates to the technical field of composite material structural member design, in particular to a composite material pipe fitting bending rigidity prediction method based on deep learning and a system and a terminal thereof. The composite material pipe fitting bending rigidity prediction method comprises the following steps: testing the bending resistance of a layered composite material pipe fitting by using a finite element model under a cantilever beam condition; establishing a basic data set; processing the basic data set to obtain a standard data set; constructing a neural network model, and training the neural network model by using the standard data set to obtain a pipe bending rigidity prediction model; and outputting the flexural rigidity by using the model. The method comprises the following steps: acquiring data by simulating a composite material pipe fitting, dividing the preprocessed data into layer paving data and pipe fitting geometric data as independent variables and flexural rigidity as dependent variables, making a sample data set, constructing a pipe fitting flexural rigidity prediction model, and calculating the flexural rigidity of the pipe fitting through basic data. And low-cost, rapid and accurate design of the composite material pipe fitting is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of structural component design, and in particular to a method for predicting the bending stiffness of composite material pipes based on deep learning, and a system and a terminal thereof. Background Art

[0002] As composite materials have excellent properties such as high specific strength, specific stiffness, and corrosion resistance, more and more traditional metal alloy components are being replaced by composite materials. Accordingly, as the manufacturing industry develops towards intelligence, robotic arms, as the mainstream manufacturing production machinery industry, have begun to develop towards lightweight to improve work efficiency and reduce energy consumption.

[0003] Pipe fittings are commonly used as connecting rods in robotic arms. When they are replaced with composite materials, the anisotropic characteristics of composite materials make their precise design more complicated. Currently, the finite element method is mainly used for simulation design, but it consumes a lot of computing power.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a method for predicting the bending stiffness of composite pipe fittings based on deep learning, and a system and terminal thereof, aiming to solve the problem that the existing method for predicting the bending stiffness of pipe fittings requires a large amount of computing power.

[0006] The technical solution of the present invention is as follows:

[0007] A method for predicting the bending stiffness of composite pipe fittings based on deep learning, comprising the steps of:

[0008] Construct finite element models of laminated composite pipes;

[0009] Using the finite element model under cantilever beam conditions to test the bending resistance of the laminated composite material pipe fitting, and to build a basic data set;

[0010] Performing standardization preprocessing and segmentation processing on the basic data set to obtain a standard data set;

[0011] Constructing a neural network model, and using the standard data set to train the neural network model to obtain a pipe bending stiffness prediction model;

[0012] The parameters of the layered composite material pipe are input into the pipe bending stiffness prediction model, and the bending stiffness of the layered composite material pipe is output.

[0013] The method for predicting the bending stiffness of composite pipes based on deep learning, wherein the step of constructing a finite element model of the layered composite pipes comprises:

[0014] Set the basic properties of single-layer materials in finite element software, build multi-layer material models, and set the angles of each layer;

[0015] The geometric data of the pipe fitting is set, the composite pipe fitting is established by using the multi-layer material model, and the grid units are divided to complete the construction of the finite element model of the layered composite pipe fitting.

[0016] The deep learning-based method for predicting the bending stiffness of composite pipe fittings, wherein the basic properties include Young's modulus, shear modulus, Poisson's ratio, and mass density.

[0017] The method for predicting the bending stiffness of composite pipes based on deep learning, wherein the step of testing the bending performance of the layered composite pipes using the finite element model under cantilever beam conditions and building a basic data set includes:

[0018] The layer laying data and the pipe fitting geometric data are combined, and each parameter interval is scanned with a predetermined step length to obtain multiple sets of independent variable data about the layer laying data and the pipe fitting geometric data;

[0019] A fixed constraint is added to one end of the layered composite material pipe and a static load is applied to the other end. Finite element simulation is performed on each group of the independent variable data, and the maximum deflection of the layered composite material pipe corresponding to each group of independent variable data is recorded. A basic data set including the layer laying data, pipe geometry data and bending stiffness of the layered composite material pipe is established.

[0020] The method for predicting the bending stiffness of composite pipes based on deep learning, wherein the expression for standardizing and preprocessing the basic data set is:

[0021] Among them, X represents the input data, X mean represents the mean of the data, X std Represents the variance of the data.

[0022] The method for predicting the bending stiffness of composite pipes based on deep learning, wherein the step of constructing a neural network model comprises: constructing a neural network model based on a long short-term memory model combined with a fully connected neural network model, and setting training parameters;

[0023] The training parameters include batch size, training rounds, verification frequency, and learning rate.

[0024] The method for predicting the bending stiffness of composite pipe fittings based on deep learning, wherein the step of training the neural network model using the standard data set to obtain the pipe fitting bending stiffness prediction model includes:

[0025] The input data x of the neural network model is [x1, x2], where x1 corresponds to the ply data of the composite material, and x2 corresponds to the geometric data of the pipe fitting;

[0026] x1 passes through the LSTM layer, then the formula for x1 input processing layer is L(x1)=Lstm(x1,m), where Lstm(x1,m) represents the forward propagation function of the LSTM network, and m represents the number of kernels in the LSTM layer;

[0027] x2 passes through the fully connected layer, then the formula for x2 input processing layer is F1(x2)=Dense(x2;w1,b1), where w1 and b1 are the weight and bias parameters of the fully connected layer respectively;

[0028] Concatenate x1 and x2 according to the feature dimension, the formula is C(x)=Concat(x1, x2), where Concat represents the concatenation processing function;

[0029] The spliced ​​data is input into two layers of fully connected layers for processing to obtain a pipe bending stiffness prediction model; wherein the formula for inputting the spliced ​​data into two layers of fully connected layers for processing is F2(x)=Dense(Dense(x; w2, b2); w3, b3), w2, w3 respectively represent the weights of the two layers of full connection, b2, b3 respectively represent the biases of the two layers of full connection;

[0030] The loss function of the neural network training material is Where y, They are respectively represented as predicted value and true value, and n represents the total number of samples used for training.

[0031] The method for predicting the bending stiffness of composite pipes based on deep learning, wherein the parameters of the layered composite pipes include length, radius, thickness, and layer angle.

[0032] A composite material pipe fitting flexural stiffness prediction system for implementing the composite material pipe fitting flexural stiffness prediction method based on deep learning, comprising:

[0033] A first model building module is used to build a finite element model of a laminated composite pipe;

[0034] A data building module, used to test the bending resistance of the laminated composite material pipe and build a basic data set;

[0035] A data processing module, used for processing the basic data set;

[0036] The second model building module is used to build a neural network model and perform training;

[0037] The data output module is used to output the bending stiffness of laminated composite pipes.

[0038] A terminal comprises a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for predicting the bending stiffness of a composite pipe based on deep learning are implemented.

[0039] Beneficial effect: The present invention provides a method for predicting the bending stiffness of composite pipe fittings based on deep learning, and a system and terminal thereof. The method for predicting the bending stiffness of composite pipe fittings based on deep learning includes the following steps: constructing a finite element model of a layered composite pipe fitting; using the finite element model to test the bending performance of the layered composite pipe fitting under cantilever beam conditions, and building a basic data set; performing standardized preprocessing and segmentation processing on the basic data set to obtain a standard data set; constructing a neural network model, and using the standard data set to train the neural network model to obtain a pipe fitting bending stiffness prediction model; inputting layered composite pipe fitting parameters into the pipe fitting bending stiffness prediction model, and outputting the bending stiffness of the layered composite pipe fitting. The present invention obtains the data through simulation of composite pipe fittings, pre-processes the data, divides the pre-processed data into layer data, pipe geometric data and bending stiffness, takes the layer data and pipe geometric data as independent variables, and bending stiffness as dependent variable, makes a sample data set, and constructs a pipe bending stiffness prediction model, so that the bending stiffness of composite pipe fittings can be inferred from basic data, and the low-cost, fast and accurate design of composite pipe fittings can be achieved. Compared with the traditional pipe fitting reverse design process based on finite element analysis, the neural network model used in the present invention can be reused after only one training, while the finite element must be re-modeled every time the design is designed, and the operator has a certain threshold, and after the neural network model constructed by the present invention is trained, it only needs to be simply called to realize the reverse design of multiple pipe fittings, which greatly improves the design efficiency and reduces the computing power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flow chart of a method for predicting the bending stiffness of composite pipe fittings based on deep learning according to the present invention;

[0041] Figure 2 A detailed flow chart of a method for predicting the bending stiffness of composite pipes based on deep learning according to the present invention;

[0042] Figure 3 Schematic diagram of the finite element model of the CFRP layer-laid pipe in Example 1;

[0043] Figure 4 Schematic diagram of the finite element model of the CFRP pipe under load in Example 1;

[0044] Figure 5 Schematic diagram of the basic neural network model in Example 1;

[0045] Figure 6 A data graph showing the training-validation loss value of the neural network in Example 1;

[0046] Figure 7 This is a graph of the prediction results of the neural network in the test set in Example 1. DETAILED DESCRIPTION

[0047] The present invention provides a method for predicting the bending stiffness of composite pipes based on deep learning, a system and a terminal thereof. To make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0049] like Figure 1 As shown, the present invention provides a method for predicting the bending stiffness of composite pipe fittings based on deep learning, comprising the steps of:

[0050] Step S10: constructing a finite element model of the laminated composite material pipe;

[0051] Step S20: testing the bending resistance of the laminated composite material pipe by using the finite element model under a cantilever beam condition, and building a basic data set;

[0052] Step S30: performing standardization preprocessing and segmentation processing on the basic data set to obtain a standard data set;

[0053] Step S40: constructing a neural network model, and using the standard data set to train the neural network model to obtain a pipe bending stiffness prediction model;

[0054] Step S50: inputting the parameters of the layered composite material pipe into the pipe bending stiffness prediction model, and outputting the bending stiffness of the layered composite material pipe.

[0055] In this embodiment, the data is obtained by simulating composite pipe fittings, and the data is preprocessed. The preprocessed data is divided into layer data, pipe geometry data and bending stiffness. The layer data and pipe geometry data are used as independent variables, and the bending stiffness is used as the dependent variable to make a sample data set, and a pipe bending stiffness prediction model is constructed. In this way, the bending stiffness of composite pipe fittings can be inferred from the basic data, and the low-cost, fast and accurate design of composite pipe fittings can be achieved. Compared with the traditional pipe fitting reverse design process based on finite element analysis, the neural network model used in the present invention can be reused after only one training, while the finite element must be remodeled every time the design is designed, and the operator has a certain threshold. After the neural network model constructed by the present invention is trained, it only needs to be simply called to realize the reverse design of multiple pipe fittings, which greatly improves the design efficiency and reduces the computing power consumption.

[0056] Specifically, the deep learning-based method for predicting the flexural stiffness of composite pipe fittings is suitable for layered pipe fittings constructed of anisotropic composite materials. By properly adjusting the training data set and model parameters, the functions of various types of pipe fitting designs can be better realized, making the deep learning-based method for predicting the flexural stiffness of pipe fittings highly compatible.

[0057] In a preferred embodiment, the deep learning-based composite pipe bending stiffness prediction method is used in the reverse design of CFRP (Carbon Fiber Reforcement Polymer) pipes.

[0058] In some embodiments, in step S10, the step of constructing a finite element model of the laminated composite material pipe comprises:

[0059] Step S11: setting the basic properties of the single-layer material in the finite element software, constructing a multi-layer material model, and setting the angles of each layer;

[0060] Step S12: setting the geometric data of the pipe fitting, establishing the composite material pipe fitting by using the multi-layer material model, and dividing the grid units to complete the construction of the finite element model of the layered composite material pipe fitting.

[0061] In this implementation, first, the basic properties of a single-layer material are set in the finite element software, including the Young's modulus, shear modulus, Poisson's ratio, and mass density of the material. Next, a basic multi-layer material model is established, and the angles of each layer are set. Next, the basic geometric properties of the composite pipe fittings are set, and the composite pipe fittings are established using the multi-layer material, and the grid units are divided to complete the establishment of a finite element model of a basic composite layer-laid pipe fitting.

[0062] Specifically, constructing the finite element model of the layered composite pipe includes: determining the layer design variables and the pipe geometry variables, where the length of the pipe is set to L [mm], the radius of the middle layer (midpoint in the thickness direction) is R [mm], and the layer angle of each layer is set to θ i , the number of layers is n. When the fiber direction is set to the x-axis direction, the layer angle is 0, and the layer order corresponds to the radial direction from the inner wall to the outer wall of the pipe.

[0063] In some embodiments, the basic properties include but are not limited to Young's modulus, shear modulus, Poisson's ratio, and mass density.

[0064] In some embodiments, in step S20, the step of testing the bending resistance of the laminated composite material pipe using the finite element model under the cantilever beam condition and building a basic data set includes:

[0065] Step S21: combining the layer data and the pipe geometry data, scanning each parameter interval with a predetermined step length, and obtaining multiple sets of independent variable data about the layer data and the pipe geometry data;

[0066] Step S22: Add a fixed constraint to one end of the layered composite material pipe and apply a static load to the other end, perform finite element simulation on each group of the independent variable data, record the maximum deflection of the layered composite material pipe corresponding to each group of independent variable data, and build a basic data set including the layer laying data, pipe geometry data and bending stiffness of the layered composite material pipe.

[0067] Specifically, the finite element model is used to analyze the end displacement of the pipe under the static load in the cantilever beam model, and then the basic design data of the pipe is changed to repeat the finite element analysis, and all the test data are recorded to form a basic data set.

[0068] In some embodiments, a fixed constraint is added to one end of a laminated composite pipe and a static load is applied to the other end to simulate the bending resistance of the rod as a cantilever beam structure. Then, a finite element simulation of the loading is performed and the displacement of the end of the pipe before and after the static load is applied is recorded to obtain the maximum deflection of the composite pipe.

[0069] Specifically, the deflection can be expressed by the formula Converted into bending stiffness; where EI represents the bending stiffness of the pipe, F corresponds to the static load at the end of the pipe, L represents the length of the pipe, and y represents the deflection of the pipe obtained by finite element simulation. This formula can be used to convert the deflection predicted by the model into the bending stiffness of the pipe.

[0070] In some embodiments, the step S21 specifically includes: combining the ply data and the pipe geometry data within a certain design interval, scanning each parameter interval with a certain step size, obtaining multiple sets of independent variable data on the ply data and the pipe geometry characteristics, performing a finite element simulation on each set accordingly, and recording all the data, i.e., changing the basic design and repeating step S22, to form a complete data set including the ply design, geometric design and bending stiffness of the composite pipe.

[0071] In some embodiments, the expression for standardizing the basic data set is: Among them, X represents the input data, X mean represents the mean of the data, X std Represents the variance of the data.

[0072] In some embodiments, the input data X includes layer sequence data, pipe length, pipe radius, pipe thickness, etc., which are all standardized and pre-processed in their respective data spaces.

[0073] In some embodiments, the segmentation process is to divide the basic data set after standardized preprocessing into a training set, a test set and a validation set; specifically, the basic data set after standardized preprocessing is randomly shuffled in order and selected in proportion as the training set, the validation set and the test set.

[0074] In some embodiments, the standard data set may consist of 70% training set, 15% validation set, and 15% test set, or may consist of 80% training set, 10% validation set, and 10% test set, but is not limited thereto.

[0075] In some embodiments, the step of constructing a neural network model includes: constructing a long short-term memory model combined with a fully connected neural network model, and setting training parameters; wherein the training parameters include batch size, training rounds, verification frequency, and learning rate.

[0076] Specifically, a neural network prediction model based on LSTM (Long Short-Term Memory) combined with FCNN (Fully Connected Neural Network) was constructed, and the training parameters were set, including batch size, training rounds, and validation frequency. The optimizer was set to Adam, and the learning rate was set to 0.0001.

[0077] In some embodiments, in step S40, the step of training the neural network model using the standard data set to obtain a pipe bending stiffness prediction model includes:

[0078] The input data x of the neural network model is [x1, x2], where x1 corresponds to the ply data of the composite material, and x2 corresponds to the geometric data of the pipe fitting;

[0079] x1 passes through the LSTM layer, then the formula for x1 input processing layer is L(x1)=Lstm(x1,m), where Lstm(x1,m) represents the forward propagation function of the LSTM network, and m represents the number of kernels in the LSTM layer;

[0080] x2 passes through the fully connected layer, then the formula for x2 input processing layer is F1(x2)=Dense(x2;w1,b1), where w1 and b1 are the weight and bias parameters of the fully connected layer respectively;

[0081] Concatenate x1 and x2 according to the feature dimension, the formula is C(x)=Concat(x1, x2), where Concat represents the concatenation processing function;

[0082] The spliced ​​data is input into two layers of fully connected layers for processing to obtain a pipe bending stiffness prediction model; wherein the formula for inputting the spliced ​​data into two layers of fully connected layers for processing is F2(x)=Dense(Dense(x; w2, b2); w3, b3), w2, w3 respectively represent the weights of the two layers of full connection, b2, b3 respectively represent the biases of the two layers of full connection;

[0083] The loss function of the neural network training material is Where y, They are respectively represented as predicted value and true value, and n represents the total number of samples used for training.

[0084] In this embodiment, compared with the traditional pipe fitting reverse design process based on finite element analysis, the neural network model used in the present invention can be reused after only one training, while the finite element must be remodeled for each design and the operator has a certain threshold. After the neural network model constructed by the present invention is trained, it only needs to be simply called to realize multiple pipe fitting reverse designs, which greatly improves the design efficiency and reduces computing power consumption.

[0085] In some embodiments, after obtaining the pipe bending stiffness prediction model, it also includes evaluating the pipe bending stiffness prediction model; the step of evaluating the pipe bending stiffness prediction model includes: using the mean square error, root mean square error and mean absolute error of the model prediction results and actual results in the test set to evaluate the quality of the pipe bending stiffness prediction model.

[0086] Specifically, the test set is input into the pipe bending stiffness prediction model to calculate the model predicted bending stiffness value, and then the mean square error, root mean square error, and mean absolute error of the two are solved to evaluate the quality of the model, and the training and validation loss value graphs during the training process are checked to check whether the training is completed well. In addition, the hyperparameters such as optimizer, training rounds, and loss function can be adjusted to optimize the network and improve the fitting ability of the network.

[0087] In some embodiments, the deflection value and the laying data of the layered composite material pipe and the geometric data of the pipe can be used to form a basic data set. At this time, the output of the pipe bending stiffness prediction model is the deflection value, and then the deflection value is converted into bending stiffness using the relationship between bending stiffness, deflection and geometric parameters to finally obtain the bending stiffness value; the deflection can also be converted first using the relationship between bending stiffness, deflection and geometric parameters, and the layer laying data of the layered composite material pipe, the geometric data of the pipe and the bending stiffness are used as the basic data set.

[0088] In some embodiments, the parameters of the laminated composite material pipe include length, radius, thickness, and ply angle.

[0089] Specifically, the length, radius, thickness and ply angle of the pipe are used as input data, and then the mean value X of the original data stored in the standardization preprocessing process is used. mean and variance X std , the input data is standardized, and then the standardized data is input into the pipe bending stiffness prediction model, the model obtains a standardized output value, and then the output value is inversely standardized. Among them, the inverse standardization process is specifically: x*X sts +X mean , where x is the value to be inversely normalized, and the predicted value y of the pipe deflection is obtained. Since the test is a cantilever beam, the relationship between its bending stiffness, deflection and geometric parameters is:

[0090]

[0091] Among them, EI represents the bending stiffness of the pipe fitting, F corresponds to the static load at the end of the pipe fitting, and L represents the length of the pipe fitting. This formula can be used to convert the deflection predicted by the model into the bending stiffness of the pipe fitting.

[0092] Specifically, Figure 2As shown, the prediction method of the stiffness strength of pipe fittings based on the neural network model specifically includes the following steps: establishing a finite element model of CFRP pipe fittings, determining the layer design variables and the geometric variables of the pipe fittings; applying a fixed constraint at one end of the pipe fittings and a static load at the other end to obtain the displacement of the end of the rod; selecting several groups of different layer data and geometric data of the rod, repeating the previous step, and obtaining several groups of CFRP pipe fittings bending simulation data, i.e., basic data sets; preprocessing the original data set, classifying the input data into layer sequence data and pipe fittings geometric data, inputting the neural network model and training, and obtaining the bending stiffness predictor of the pipe fittings; using the mean square error, root mean square error, and mean absolute error of the model prediction results and actual results in the test set to evaluate the quality of the model; given the length, radius, thickness, and layer angle of the pipe fitting, using the trained neural network to predict the bending stiffness of the pipe fitting.

[0093] In addition, the present invention also provides a composite material pipe fitting flexural stiffness prediction system for realizing a composite material pipe fitting flexural stiffness prediction method based on deep learning, comprising:

[0094] A first model building module is used to build a finite element model of a laminated composite pipe;

[0095] A data building module, used to test the bending resistance of the laminated composite material pipe and build a basic data set;

[0096] A data processing module, used for processing the basic data set;

[0097] The second model building module is used to build a neural network model and perform training;

[0098] The data output module is used to output the bending stiffness of laminated composite pipes.

[0099] In addition, the present invention also provides a terminal, which includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of a method for predicting the bending stiffness of composite pipes based on deep learning are implemented.

[0100] In some embodiments, the memory may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the memory may also include both an internal storage unit of the terminal and an external storage device. The memory is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal, etc. The memory may also be used to temporarily store data that has been output or is to be output. In one embodiment, a composite pipe bending stiffness prediction program is stored on the memory, and the composite pipe bending stiffness prediction program can be executed by a processor, thereby realizing the composite pipe bending stiffness prediction method of the present invention.

[0101] In some embodiments, the processor may be a central processing unit (CPU), a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory, such as executing the composite pipe bending stiffness prediction method.

[0102] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer-readable storage medium stores a composite material pipe fitting bending stiffness prediction program, and when the composite material pipe fitting bending stiffness prediction program is executed by a processor, the steps of the composite material pipe fitting bending stiffness prediction method are implemented.

[0103] The following examples are further given to illustrate the present invention in detail. It should also be understood that the following examples are only used to further illustrate the present invention and cannot be understood as limiting the scope of protection of the present invention. Some non-essential improvements and adjustments made by those skilled in the art based on the above content of the present invention belong to the scope of protection of the present invention.

[0104] Example 1

[0105] This embodiment first establishes a basic model of finite element analysis, then uses finite element analysis to construct a data set, pre-processes the data set, and then constructs a neural network model, which specifically includes the following steps:

[0106] Step 1: Create a pipe made of multiple layers of unidirectional carbon fiber reinforced composite prepreg (hereinafter referred to as CFRP pipe), such as Figure 3 As shown, the thickness of each layer of composite material is 0.2 mm, the length of the tube is L [mm], and the middle radius is R.

[0107] The mechanical properties of single-layer CFRP materials include:

[0108] Young's modulus: (E1, E2, E3) = (121 GPa, 8.6 GPa, 8.6 GPa);

[0109] Shear modulus: (G1, G2, G3) = (4.7 GPa, 3.1 GPa, 4.7 GPa);

[0110] Poisson's ratio: (ν1, ν2, ν3) = (0.27, 0.4, 0.27), where E1, E2, E3 represent the Young's modulus in the x, y, z directions respectively; G1, G2, G3 represent the shear modulus in the x, y, z directions respectively; ν1, ν2, ν3 represent the Poisson's ratio in the x, y, z directions respectively, and the mass density of the composite material is 1490 g / cm 3 .

[0111] Step 2: Apply a fixed constraint to the left end of the pipe and a static load F of 100N to the right end. Perform loading analysis and record the displacement ω in the z direction of the end of the pipe, as shown in Figure 4 shown.

[0112] Step 3: Change the basic design data of the pipe fitting, repeat step 2, select the pipe fitting length L∈[300mm,1200mm], the pipe fitting radius R∈[30mm,50mm], set the step size of the two variables to 20mm and 2mm respectively, and combine them to construct 503 sets of pipe fitting geometric data, and further select multiple different layup combinations.

[0113] The following ply combinations were tested in this example:

[0114] [0,45,0,90,0,45,0,90,0,45]、[0,90,0,90,0,90,0,90,0,90]、[-45,45,0,-45,45,0,-45,45,0,0]、[30,60,30,60,30,60,30,60,30,60]、[90,90,0 ,90,90,90,90,0,90,90]、[10,20,30,40,50,50,40,30,20,10]、[0,90,90,0,0,90,90,0,0,90,90,0]、[45,90,-45,90,45,90,-45,90,45,90,-45,90 ]、[-45,45,-45,45,-45,45,-45,45,-45,45,-45,45]、[45,0,-45,0,90,0,45,0,-45,0,90,0]、[90,45,0,45,90,0,90,45,0,45,90,0]、[30,45,60,90,60,45,30,45,60,90,60,45]、[90,45,-45,90,90,45,-45,90,90,45,-45,90,90], where the unit is the angle value, which represents the ply angle of the composite prepreg from the first layer to the last layer. A total of 6072 sets of simulation data are obtained, which constitute the basic data set.

[0115] Step 4: Preprocess the data, standardize the geometric parameters, displacement parameters and layer angle in this embodiment, and set the dimension of all layer angle data to 12. Then save the standardized intermediate quantities, that is, the mean and variance of each parameter, for use in the subsequent de-standardization process.

[0116] Step 5: Build a neural network model based on LSTM and FCNN architecture (such as Figure 5 As shown), the data set in step 4 is divided into 80% training set, 10% validation set, and 10% test set. The loss function is set to Mse, the optimization solver is Adam, and 200 epochs are trained. The learning rate is set to 0.0001. The training results are shown in Figure 6 As shown. The mean square error is 0.000186, the root mean square error is 0.013623, and the mean absolute error is 0.010115, indicating that the model training results fit the data very well. The prediction effect is tested on the test set, and the prediction results are plotted as shown below. Figure 7 shown.

[0117] Step 6: Use the trained model to predict the bending stiffness of the pipe. The ply angle of the pipe is [0, 90, 60, 90, 0, 0, 90, 60, 90, 0], the length of the pipe is 1100 mm, and the radius is 35 mm. Use the mean and variance of the parameters saved in step 4 to unify the above design parameters, then input the model to obtain the prediction results and perform de-normalization to complete the prediction of the bending stiffness of the ply composite pipe.

[0118] In summary, the present invention provides a method for predicting the bending stiffness of composite pipe fittings based on deep learning, and a system and terminal thereof. The method for predicting the bending stiffness of pipe fittings based on deep learning includes the steps of: constructing a finite element model of a layered composite pipe fitting; using the finite element model to test the bending performance of the layered composite pipe fitting under cantilever beam conditions, and building a basic data set; performing standardized preprocessing and segmentation processing on the basic data set to obtain a standard data set; constructing a neural network model, and using the standard data set to train the neural network model to obtain a pipe fitting bending stiffness prediction model; inputting layered composite pipe fitting parameters into the pipe fitting bending stiffness prediction model, and outputting the bending stiffness of the layered composite pipe fitting. The present invention obtains the data through simulation of composite pipe fittings, pre-processes the data, divides the pre-processed data into layer data, pipe geometric data and bending stiffness, takes the layer data and pipe geometric data as independent variables, and bending stiffness as dependent variable, makes a sample data set, and constructs a pipe bending stiffness prediction model, so that the bending stiffness of composite pipe fittings can be inferred from basic data, and the low-cost, fast and accurate design of composite pipe fittings can be achieved. Compared with the traditional pipe fitting reverse design process based on finite element analysis, the neural network model used in the present invention can be reused after only one training, while the finite element must be re-modeled every time the design is designed, and the operator has a certain threshold, and after the neural network model constructed by the present invention is trained, it only needs to be simply called to realize the reverse design of multiple pipe fittings, which greatly improves the design efficiency and reduces the computing power consumption.

[0119] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for predicting the bending stiffness of composite pipes based on deep learning, characterized in that: Includes steps: Construct finite element models of laminated composite pipes; Using the finite element model under cantilever beam conditions to test the bending resistance of the laminated composite material pipe fittings, and to build a basic data set; Performing standardization preprocessing and segmentation processing on the basic data set to obtain a standard data set; Constructing a neural network model, and using the standard data set to train the neural network model to obtain a pipe bending stiffness prediction model; The parameters of the layered composite material pipe are input into the pipe bending stiffness prediction model, and the bending stiffness of the layered composite material pipe is output.

2. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The step of constructing a finite element model of a laminated composite material pipe comprises: Set the basic properties of single-layer materials in finite element software, build multi-layer material models, and set the angles of each layer; The geometric data of the pipe fitting is set, the composite pipe fitting is established by using the multi-layer material model, and the grid units are divided to complete the construction of the finite element model of the layered composite pipe fitting.

3. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 2, characterized in that: The basic properties include Young's modulus, shear modulus, Poisson's ratio, and mass density.

4. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The step of testing the bending resistance of the laminated composite material pipe by using the finite element model under the cantilever beam condition and building a basic data set comprises: The layer laying data and the pipe fitting geometric data are combined, and each parameter interval is scanned with a predetermined step length to obtain multiple sets of independent variable data about the layer laying data and the pipe fitting geometric data; A fixed constraint is added to one end of the layered composite material pipe and a static load is applied to the other end. Finite element simulation is performed on each group of the independent variable data, and the maximum deflection of the layered composite material pipe corresponding to each group of independent variable data is recorded. A basic data set including the layer laying data, pipe geometry data and bending stiffness of the layered composite material pipe is established.

5. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The expression for standardizing and preprocessing the basic data set is: Among them, X represents the input data, X meam represents the mean of the data, X std Represents the variance of the data.

6. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The step of constructing a neural network model includes: constructing a neural network model based on a long short-term memory model combined with a fully connected neural network model, and setting training parameters; The training parameters include batch size, training rounds, verification frequency, and learning rate.

7. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The step of training the neural network model using the standard data set to obtain a pipe bending stiffness prediction model includes: The input data x of the neural network model is [x1, x2], where x1 corresponds to the ply data of the composite material, and x2 corresponds to the geometric data of the pipe; x1 passes through the LSTM layer, then the formula for x1 input processing layer is L(x1)=Lstm(x1,m), where Lstm(x1,m) represents the forward propagation function of the LSTM network, and m represents the number of kernels in the LSTM layer; x2 passes through the fully connected layer, then the formula for x2 input processing layer is F1(x2)=Dense(x2;w1,b1), where w1 and b1 are the weight and bias parameters of the fully connected layer respectively; Concatenate x1 and x2 according to the feature dimension, the formula is C(x) = Concat(x1, x2), where Concat represents the concatenation processing function; The spliced ​​data is input into two layers of fully connected layers for processing to obtain a pipe bending stiffness prediction model; wherein the formula for inputting the spliced ​​data into two layers of fully connected layers for processing is F2(x)=Dense(Dense(x; w2, b2); w3, b3), w2, w3 respectively represent the weights of the two layers of full connection, and b2, b3 respectively represent the biases of the two layers of full connection; The loss function of the neural network training material is Where y, They are respectively represented as predicted value and true value, and n represents the total number of samples used for training.

8. The method for predicting the bending stiffness of composite pipes based on deep learning according to claim 1, characterized in that: The parameters of the laminated composite material pipe include length, radius, thickness and layer angle.

9. A composite pipe flexural stiffness prediction system for implementing the composite pipe flexural stiffness prediction method based on deep learning as claimed in any one of claims 1 to 8, characterized in that: include: A first model building module is used to build a finite element model of a laminated composite pipe; A data building module, used to test the bending resistance of the laminated composite material pipe and build a basic data set; A data processing module, used for processing the basic data set; The second model building module is used to build a neural network model and perform training; The data output module is used to output the bending stiffness of laminated composite pipes.

10. A terminal, characterized in that: The terminal includes a memory, a processor, and a program stored in the memory and executable on the processor. When the program is executed by the processor, the steps of the method for predicting the bending stiffness of a composite pipe based on deep learning as described in any one of claims 1 to 8 are implemented.

Images