Complex structure preform manufacturing yarn let-off tension self-adaptive control method

Abstract

The invention relates to a yarn let-off tension self-adaptive control method for manufacturing a complex-structure preform, which comprises the following steps: according to the curvature of different areas on the surface of a target three-dimensional preform, obtaining the yarn density required by each area, and obtaining the first tension required for weaving the different areas, the first tension being the tension after yarns are wound around a yarn guide bar; the first tension and related parameters under the current working condition are input into an agent model, predicted initial tension is obtained, and the initial tension is the tension before the yarn is wound around the yarn guide bar; controlling the yarn through a tension controller in the device according to the predicted initial tension; the related parameters comprise the contact angle between the yarn and the yarn guide bar, the friction coefficient in the contact process and the structural parameters of the yarn guide bar and the yarn; the proxy model is constructed by training a neural network model through a training set. The initial tension of the yarn can be accurately controlled in real time, and it is ensured that the yarn is always kept in the optimal tension state in the preform weaving process.

Description

Technical Field

[0001] The invention relates to the technical field of braided yarn tension control, in particular to a method for adaptively controlling the warp feeding tension of yarns in manufacturing a preform with a complex structure. Background Art

[0002] Three-dimensional preforms have the characteristics of compact structure, excellent mechanical properties, high strength and high toughness, and have been widely used in high-tech fields such as aerospace. The preparation process of three-dimensional preforms is very complicated, among which the tension of yarn is the key parameter affecting the molding quality of three-dimensional preforms, and directly affects the mechanical properties and structural stability of the final product. Compared with traditional flat fabrics, three-dimensional preforms have significant three-dimensional shapes and complex curved surface structures, and their surfaces show certain bending and undulating shapes, which makes the control of yarn tension during the manufacturing process more stringent. Specifically, different areas of the three-dimensional preform have different requirements for yarn tension. The surface of some areas presents complex curvature due to bending and undulation, and a higher yarn density is required to enhance the strength and bearing capacity of the area; while other areas require a relatively low density. Therefore, the yarn warp tension during the preform manufacturing process needs to be reasonably adjusted according to the different curvatures of the three-dimensional preform surface and the density differences in different areas to ensure the functionality and structural stability of the three-dimensional preform.

[0003] During the weaving process, the yarn is drawn out from the bobbin rack and transferred to the production line of the target three-dimensional preform after changing direction through the yarn guide frame. Since the angle of the yarn changes when it is transferred through the guide frame, the tension of the yarn will also change accordingly, so it is not clear what initial warp tension is required in the actual weaving process. The traditional tension control method determines the setting range of the initial tension of the yarn through continuous debugging based on experience, so the accuracy is not high. In addition, the yarn weaving density in the three-dimensional preform is uneven in space due to the complex local surface curvature. How to reasonably set the initial tension range of the yarn in each area according to the change of the surface curvature of the three-dimensional preform, so that the yarn is always in the best tension state during the weaving process of the entire target three-dimensional preform, and ensure the excellent mechanical properties, structural stability and product quality of the woven three-dimensional preform is a difficult problem that needs to be solved urgently. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for adaptively controlling the warp feeding tension of yarns in manufacturing a complex structure preform, the purpose of which is to ensure that the yarns are always in an optimal tension state during the weaving process of the entire target three-dimensional preform.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The present invention provides a method for adaptively controlling warp letting tension of yarn in manufacturing a complex structure preform, comprising:

[0007] According to the curvature of different areas on the surface of the target three-dimensional preform, the yarn density required for each area is obtained, and the first tension required for weaving different areas is obtained, where the first tension is the tension after the yarn is wound around the yarn guide rod;

[0008] Inputting the first tension and related parameters under the current working condition into the proxy model to obtain a predicted initial tension, wherein the initial tension is the tension of the yarn before it is wound around the yarn guide rod;

[0009] Controlling the yarn according to the predicted initial tension to meet the weaving requirements of the target three-dimensional preform;

[0010] The relevant parameters include the contact angle between the yarn and the yarn guide rod, the friction coefficient during the contact process, and the structural parameters of the yarn guide rod and the yarn;

[0011] The proxy model is constructed by training the neural network model through the training set.

[0012] Further technical solutions are:

[0013] The construction of the training set includes:

[0014] Establish a three-dimensional model of yarn winding around the yarn guide rod;

[0015] Setting calculation conditions, performing numerical simulation on the three-dimensional model, and calculating and obtaining simulation values ​​of the initial tension under different working conditions;

[0016] Performing experimental verification on the simulated value, and correcting the simulated value according to the experimental result;

[0017] The corrected initial tension under each working condition and the corresponding model parameters and structural parameters of the three-dimensional model are used to form sample data, and the sample data corresponding to each working condition are summarized as a first training set;

[0018] The proxy model is constructed by training the neural network model with the first training set;

[0019] The calculation conditions include numerical model parameters and solver analysis conditions.

[0020] The construction of the training set also includes:

[0021] Using the data of the first training set and taking the yarn tension theoretical model as the target fitting function, determining the undetermined coefficients of the target fitting function through nonlinear curve fitting;

[0022] The control variable method is used to calculate the theoretical value of the initial tension under more working conditions through the yarn tension theoretical model;

[0023] The theoretical values ​​and the corresponding variable values ​​constitute a second training set, and together with the first training set constitute a total training set;

[0024] The proxy model is constructed by training the neural network model through the total training set;

[0025] The yarn tension theoretical model is:

[0026]

[0027] In the formula, F b 、F a They represent the tension of the yarn before and after it passes around the guide rod respectively; θ represents the contact angle in arc between the yarn and the guide rod; r is the radius of the guide rod; a is the nonlinear friction coefficient between the yarn and the guide rod, and a and n are the unknown coefficients, both of which are constants related to the materials of the two objects at the junction.

[0028] The numerical model parameters include:

[0029] density, elastic modulus, Poisson’s ratio, and tensile strength of yarns;

[0030] The material of the yarn guide rod;

[0031] The first tension of the yarn after it is wound around the yarn guide rod;

[0032] The contact relationship and friction coefficient between the yarn and the guide wheel;

[0033] The loads and boundary conditions of the three-dimensional model.

[0034] The setting of the solver analysis conditions includes:

[0035] Set two load steps and set the calculation substeps for each load step;

[0036] The gravity is applied to the entire three-dimensional model during the calculation time of the first load step, and the first tension is applied to the yarn during the calculation time of the second load step.

[0037] The setting of the boundary conditions includes:

[0038] Apply fixed support to the two end faces of the yarn guide rod to ensure that it does not move;

[0039] A fixed support is applied to one end of the yarn, and a force is set along the direction of yarn transmission at the other end.

[0040] The step of experimentally verifying the simulated value and correcting the simulated value according to the experimental result includes:

[0041] An experimental platform for winding yarn around a yarn guide rod is built, and experimental tests are carried out using experimental conditions corresponding to the numerical simulation calculation conditions to obtain the test value of the initial tension under the corresponding working conditions;

[0042] The test value is compared with the simulation value. If the error is not within the set range, the simulation value is replaced by the test value, thereby correcting the simulation value.

[0043] The structural parameters of the three-dimensional model include the radius of the yarn, the radius of the yarn guide rod, and the contact angle between the yarn and the yarn guide rod.

[0044] The construction of the proxy model includes:

[0045] Dividing the training set into a training data set and a validation data set;

[0046] Optimizing the neural network model parameters using the training data set;

[0047] Using the validation data set to evaluate the performance of the neural network model;

[0048] The training is completed until the mapping relationship between the corrected initial tension and the corresponding model parameters and the structural parameters of the three-dimensional model meets the set requirements;

[0049] The neural network model includes a convolutional neural network, a recurrent neural network or a multi-layer perceptron.

[0050] The method controls the yarn according to the predicted initial tension through a yarn letting-off device, wherein the yarn letting-off device comprises:

[0051] A bobbin rack, on which a plurality of bobbins are arranged;

[0052] A guide frame, which is composed of a grid structure of a plurality of yarn guide rods of the same diameter and staggered in a horizontal and vertical direction; after a single yarn is produced from the yarn tube, it passes through a single grid and bypasses the yarn guide rods, thereby being transferred to the production line of the target three-dimensional preform;

[0053] a yarn tension controller, used for adjusting the initial tension of the single yarn to make it consistent with the predicted initial tension;

[0054] The distance between the bobbin frame and the guide frame is adjustable.

[0055] The beneficial effects of the present invention are as follows:

[0056] The present invention realizes the accurate prediction of the tension change before and after the yarn is wound around the yarn guide rod through the construction and application of the proxy model. According to the complex curvature surface and non-uniform yarn density of the three-dimensional preform, the predicted initial tension can be obtained in real time through the proxy model using the measured parameters of the current working conditions, so as to accurately control the initial tension of the yarn, ensure that the yarn is always in the best tension state during the whole weaving process, meet the differentiated tension requirements of different areas of the target three-dimensional preform, and ensure the excellent mechanical properties, structural stability and product quality of the product.

[0057] The numerical simulation calculation results of the present invention have been verified by experiments, and the results are more accurate. The constructed first training set contains sample data under different working conditions, which fully meets the basic requirements for building a proxy model.

[0058] The present invention adopts a combination of numerical simulation and theoretical model calculation to construct a high-quality second training set, expand the working conditions, improve the calculation accuracy and efficiency, and further improve the prediction accuracy of the proxy model.

[0059] The invention is suitable for initial tension control of large-capacity yarns and has a wide application range and strong versatility.

[0060] Other features and advantages of the present invention will be set forth in the following description or may be learned by practicing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic structural diagram of a yarn warp feeding device according to an embodiment of the present invention.

[0062] Figure 2 Schematic diagram of the geometric structure of the yarn tension theoretical model according to an embodiment of the present invention.

[0063] Figure 3 Schematic diagram of the structure of the target three-dimensional preform according to an embodiment of the present invention.

[0064] Figure 4 This is a diagram of numerical simulation results under the working conditions of an embodiment of the present invention.

[0065] Figure 5 This is a diagram showing the calculation results of the yarn tension theoretical model under the working conditions of an embodiment of the present invention.

[0066] Figure 6 It is a comparison chart of the calculation results of the yarn tension theoretical model and the simulation results under the working conditions of the embodiment of the present invention.

[0067] In the figure: 1, yarn tube; 2, yarn tube frame; 3, yarn tension controller; 4, single yarn; 5, guide frame; 51, yarn guide rod. DETAILED DESCRIPTION

[0068] The specific implementation of the present invention is described below with reference to the accompanying drawings.

[0069] This embodiment provides a method for adaptively controlling the yarn letting-off tension in manufacturing a complex structure preform, which can control the operation of the yarn letting-off device, see Figure 1 , the yarn letting-off device comprises:

[0070] A bobbin rack 2, on which a plurality of bobbins 1 are arranged;

[0071] The guide frame 5 is composed of a grid structure of a plurality of yarn guide rods of the same diameter and staggered in a horizontal and vertical direction; after a single yarn 4 is produced from the yarn tube 1, it passes through a single grid and bypasses the yarn guide rods, thereby being transferred to the production line of the target three-dimensional preform;

[0072] The yarn tension controller 3 is used to adjust the initial tension of the single yarn 4.

[0073] As a specific implementation, the distance between the bobbin frame 2 and the guide frame 5 is adjustable.

[0074] As a preferred embodiment, the grid formed by the yarn guide rods staggered horizontally and vertically is evenly arranged in an array.

[0075] As a preferred embodiment, the yarn tubes 1 are evenly distributed on the yarn tube rack 2 in an array.

[0076] As a preferred embodiment, the yarn tension controller 3 is installed on the surface of the yarn creel 2 .

[0077] As a preferred embodiment, the yarn guide rod is made of stainless steel.

[0078] Based on the nonlinear friction theorem, the process of yarn changing direction around an external object can be regarded as the process of yarn going around the surface of a cylinder. Figure 2 The geometric structure of a single yarn 4 passing around the yarn guide rod 51 is shown to establish a theoretical model of yarn tension as follows:

[0079]

[0080] In the formula, F b represents the tension before the yarn passes around the yarn guide rod, i.e. the initial tension, F a represents the tension of the yarn after it passes around the yarn guide rod, that is, the first tension; θ represents the contact angle between the yarn and the yarn guide rod in arc system; r is the radius of the cylindrical yarn guide rod; a is the nonlinear friction coefficient between the yarn and the yarn guide rod, and both a and n are constants related to the materials of the two objects at the junction, and are both unknown coefficients. The specific values ​​can be obtained through nonlinear fitting.

[0081] The adaptive control method comprises the following steps:

[0082] S1. According to the curvature of different regions on the surface of the target three-dimensional preform, the yarn density required for each region is obtained, and the first tension required for weaving different regions is obtained.

[0083] As a specific implementation method, Figure 3 As shown in the figure, A, B, and C are three different areas of the target three-dimensional preform. Due to the specific application fields and scenarios of the target three-dimensional preform, in order to meet specific mechanical performance requirements, such as high load-bearing capacity, high impact resistance, high strength, high wear resistance, etc., the target three-dimensional preform needs to maintain a specific shape and surface curvature, such as the edges, corners and concave parts of the three-dimensional preform need to show a higher curvature; while other areas are close to the plane, so the corresponding curvature values ​​are lower. Therefore, different areas need to show different surface curvatures. The density of the yarn in the local area is positively correlated with the curvature of the surface. For the high-density fabric area, a larger yarn tension is required to prevent the yarn from slipping or deforming during the weaving process, while the distance between the yarns in the low-density area is larger, and a smaller yarn tension is required to prevent the yarn from breaking or the deformation of the fabric. For the area showing a higher curvature, due to the bending and undulation of the surface, the yarns in the area are compressed and arranged more closely, resulting in an increase in the local weaving density coefficient; while the weaving density coefficient of the area with a lower curvature value is smaller. Figure 3 The surface curvatures of the middle regions A, B, and C correspond to the required yarn weaving density coefficients of 0.107-0.493, 0.051-0.222, and 0.061-0.224, respectively. According to the yarn density, the required yarn tension, i.e., the first tension, of the three different regions of the target three-dimensional preform is 1.30N-2.50N, 0.60N-1.20N, and 0.70N-1.25N, respectively.

[0084] S2. Input the first tension and related parameters under the current working condition into the proxy model to obtain the predicted initial tension.

[0085] The related parameters include the contact angle between the yarn and the yarn guide rod, the friction coefficient during the contact process, and the structural parameters of the yarn guide rod and the yarn.

[0086] The proxy model is constructed by training the neural network model through the training set.

[0087] The construction of the training set includes:

[0088] S21. Establish a three-dimensional model of the yarn winding around the yarn guide rod.

[0089] S22, setting calculation conditions, performing numerical simulation on the three-dimensional model, and calculating and obtaining simulation values ​​of the initial tension under different working conditions.

[0090] The calculation conditions include numerical model parameters and solver analysis conditions.

[0091] The numerical model parameters include:

[0092] density, elastic modulus, Poisson’s ratio, and tensile strength of yarns;

[0093] The material of the yarn guide rod;

[0094] The first tension of the yarn after it is wound around the yarn guide rod;

[0095] The contact relationship and friction coefficient between the yarn and the yarn guide rod;

[0096] The loads and boundary conditions of the three-dimensional model.

[0097] S23, performing an experiment to verify the simulation value, and correcting the simulation value according to the experimental result;

[0098] S24, the corrected initial tension under each working condition and the corresponding model parameters and the structural parameters of the three-dimensional model are combined into sample data, and the sample data corresponding to each working condition are summarized into a first training set. Thus, the neural network model can be trained through the first training set to construct the proxy model.

[0099] The structural parameters of the three-dimensional model include the radius of the yarn, the radius of the yarn guide rod, and the contact angle between the yarn and the yarn guide rod.

[0100] As a specific implementation, steps S21 and S22 specifically include the following steps:

[0101] (1) 3D model construction:

[0102] According to the actual working conditions of the yarn, SolidWorks 3D modeling software is used to draw an arbitrary circle with the origin as the center on the front view reference plane, and the diameter of the circle is set to 40mm using the smart dimension function. Then the stretching depth is set to 40mm to complete the construction of a 3D model of a cylinder with a diameter and height of 40mm. Then a new reference plane is created that coincides with the middle plane of the cylinder, and an arc is drawn on this reference plane with the center of the end face of the cylinder as the center of the circle, and the curvature radius of the arc segment is specified as 20.5mm, indicating that the cross section of the yarn is a circle with a diameter of 1mm. Then, with the outer surface of the cylinder as the reference object, two path lines are drawn on the new reference plane, and the endpoints of the two path lines are made to coincide with the two endpoints of the arc.

[0103] A tangent matching relationship is added to the created path line and arc line, and the contact angle (90°~180°) between the yarn and the cylinder is set through smart dimensions to make it conform to the actual path. After checking that the path relationship between the cylinder and the yarn is correct, the constructed 3D model is saved in step file format, and finally a 3D model file that can be imported into the Ansys simulation tool is generated.

[0104] (2) Calculation condition settings:

[0105] Import the constructed 3D model file into the Workbench module in the Ansys simulation software, select the transient structure module, and set the yarn density to 1800kg / m according to the actual working conditions. 3 , elastic modulus of 80 Gpa, Poisson's ratio of 0.25 and tensile strength of 3530 MPa, give the cylinder a stainless steel material, set the mesh unit type of the yarn to cable unit, define the contact relationship between the yarn and the cylinder as friction contact, and specify the friction coefficient as 0.2.

[0106] Boundary condition setting: Fixed supports are applied to the two end faces of the cylinder to ensure that it does not move, and one end of the yarn

[0107] A fixed support was applied, and a force was set at the other end along the direction of yarn transmission, simulating the first tension of 1.35N after the frictional contact between the yarn and the external cylinder;

[0108] Load condition setting: Standard gravity is applied to the entire system at the same time to approach the stress state in the actual production environment.

[0109] Meshing: The yarn is finely meshed with a unit size of 0.5 mm to ensure the accuracy of the force results, and the cylinder is meshed with a unit size of 2 mm to shorten the calculation time.

[0110] Solver analysis condition setting: set two load steps in the solver analysis setting, apply gravity to the entire system during the calculation time of the first load step, apply initial tension to the yarn during the calculation time of the second load step, set the number of calculation sub-steps of all load steps to 200 to ensure the convergence of the solution process, and the entire simulation process lasts 4s.

[0111] This specific implementation only sets the relevant parameters and friction coefficient of the yarn according to specific production conditions, and sets the load conditions and analysis conditions. It can be understood that the relevant parameters of the yarn, the friction coefficient between the yarn and the cylinder, and the first tension of the yarn can be adjusted according to actual working conditions.

[0112] (3) Data acquisition and analysis:

[0113] After completing the simulation load conditions and analysis conditions, the solution calculation is performed to output the equivalent stress results before and after the yarn passes around the cylinder, and the initial tension of the yarn is calculated based on the cross-sectional area of ​​the yarn. The numerical simulation calculation results are shown in Figure 2. Figure 4 shown.

[0114] This specific implementation method constructs a three-dimensional model of a circular yarn with a diameter of 1 mm in contact with a cylinder with a radius of 20 mm based on actual working conditions, and sets the contact angle between the yarn and the cylinder to 90° to 180°. Taking the contact angle of 90° as the reference, a simulation is performed every 10° until 180°, and a series of data is obtained, and a data set of yarn tension changes under current conditions is obtained.

[0115] The above steps are used to combine yarns with different cross-sectional sizes and cylinders with different radii, and a series of friction coefficients and first tensions are set to simulate the initial tension of the yarn under various working conditions to obtain a large amount of simulation data.

[0116] As a specific implementation, step S23 specifically includes:

[0117] An experimental platform for winding yarn around a yarn guide rod is built, and experimental tests are carried out using experimental conditions corresponding to the numerical simulation calculation conditions to obtain the test value of the initial tension under the corresponding working conditions;

[0118] The test value is compared with the simulation value. If the error is not within the set range, the simulation value is replaced by the test value, thereby correcting the simulation value.

[0119] Since the numerical simulation calculation time is long and the simulation value must be verified by experiments to ensure accuracy, the simulation data obtained in actual operation is limited, that is, the amount of data in the first training set is limited. To this end, this embodiment further optimizes the construction of the training set in the following manner: The construction of the training set also includes:

[0120] S25, using the data of the first training set, the yarn tension theoretical model described above is a target fitting function, and the undetermined coefficients a and n of the target fitting function are determined by nonlinear curve fitting;

[0121] S26, using a control variable method and the yarn tension theoretical model to calculate theoretical values ​​of initial tension under more working conditions;

[0122] S27: The theoretical values ​​and the corresponding variable values ​​constitute a second training set, and together with the first training set constitute a total training set.

[0123] As a specific implementation, step S25 includes:

[0124] Using the nonlinear fitting function of MATLAB software, the contact angle θ, the radius r of the cylindrical yarn guide rod, and the tension F before the yarn passes around the yarn guide rod under the corresponding working conditions in the first training set are used. b and the tension F of the yarn after it passes around the yarn guide rod a , nonlinear curve fitting is performed on a and n to obtain the specific values ​​of a and n, which are substituted into the target fitting function to obtain the nonlinear yarn tension theoretical model for a cylinder with a friction coefficient of 0.2, a yarn cross-section of 1 mm in diameter, and a radius of 20 mm.

[0125] As a specific implementation, step S26 includes:

[0126] According to the yarn tension theoretical model, the contact angle θ between the yarn and the yarn guide rod, the radius r of the cylindrical yarn guide rod, and the friction coefficient a between the yarn and the yarn guide rod are changed in turn through the control variable method, and the corresponding results, i.e., the theoretical values, are calculated to form the second training set.

[0127] This embodiment can analyze the variation of yarn tension with related parameters based on a large amount of data obtained by combining numerical simulation with theoretical models, and provide a basis for research and application in related fields.

[0128] The results calculated using the yarn tension theory model are as follows: Figure 5 shown.

[0129] Comparison between the calculation results of the yarn tension theoretical model and the numerical simulation results Figure 6 As shown in Figure 2, it can be seen that the fitting results of the two are good, which verifies the accuracy of the calculation results of this implementation.

[0130] The data of the second training set is directly obtained by calculating the theoretical model, which not only has high accuracy but also large data volume. Figure 5 It can be seen that the tension under any contact angle condition can be calculated theoretically, and the number of conditions covered is huge. Therefore, the accuracy of the proxy model can be further improved by training the neural network model with the total training set.

[0131] The construction of the proxy model includes taking the contact angle, friction coefficient, radius of the cylinder, first tension after the yarn transfer angle changes as input, and taking the initial tension before the yarn transfer angle changes as output, to train and optimize the neural network model. As a specific implementation, it specifically includes the following steps:

[0132] (1) Selecting a neural network model: Analyze the size and characteristics of the training set and determine a suitable neural network model. Common model types include convolutional neural networks (CNNs), recurrent neural networks (RNNs), and multi-layer perceptrons (MLPs). Since the input parameters of the training set usually do not have significant temporal or spatial characteristics, this embodiment prefers an MLP model that is simple in structure and suitable for processing scalar inputs.

[0133] In addition, the structural parameters of the neural network model, such as the number of layers and the number of neurons in each layer, are adjusted according to the data characteristics of the training set to take into account both the expressive power and training efficiency of the model.

[0134] (2) Training set division: The training set is divided into a training data set and a validation data set in a ratio of 7:3; the training data set is used to optimize the parameters of the neural network model. The validation data set is used to evaluate the performance of the neural network model. During the division process, the consistency of data distribution is ensured to avoid the impact of data bias on the model training effect.

[0135] (3) Neural network model training: Take X in the training data set as input and the corresponding Y as output to train the neural network. The training set contains sample pairs (X 1 ,Y 1 )、(X 2 ,Y 2 )…(Xs,Ys). X represents a number of input parameter vectors, including but not limited to contact angle, friction coefficient and radius of the cylinder; Y represents the initial tension of the yarn corresponding to the input parameter. The subscripts 1, 2…, s represent the sample data numbers corresponding to different working conditions.

[0136] During the training process, select appropriate loss functions (such as mean square error MSE) and optimization algorithms (such as Adam or SGD), and improve training efficiency and results by setting hyperparameters such as learning rate and batch size. At the same time, early stopping mechanism can be used to prevent overfitting of the model.

[0137] (4) Evaluation of the neural network model: Input X in the validation data set into the trained model and calculate the error indicators between the model output and Y in the validation data set, such as mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R 2 ). These indicators are used to comprehensively evaluate the model's fit and prediction accuracy to determine whether it meets the performance requirements.

[0138] (5) Optimization of the neural network model: Optimize the model according to the evaluation results. Optimization methods include but are not limited to the following: 1. Adjust the network structure, such as increasing the number of hidden layers or the number of neurons in each layer to improve the learning ability of the model; 2. Adjust hyperparameters, such as optimizing the learning rate, regularization coefficient, etc., to balance the training speed and generalization ability of the model; 3. Increase the amount of data or introduce data enhancement technology to expand the diversity of training data; 4. Introduce cross-validation methods to further verify the stability and robustness of the model. The optimized model needs to repeat steps (3) and (4) to ensure that the mapping relationship between the corrected initial tension and the corresponding model parameters and the structural parameters of the three-dimensional model meets the requirements, ensuring that the model meets expectations and completes the training.

[0139] The trained neural network model is used as a proxy model in the actual scene. By inputting the first tension and related parameter data X measured on the spot, the proxy model can quickly output the predicted initial tension value Y.

[0140] S3. Control the initial tension of the yarn through the yarn tension controller 3 to make it consistent with the initial tension value Y, so as to meet the weaving requirements of the target three-dimensional preform.

[0141] In summary, the present invention obtains the first tension required in different areas according to the surface curvature required in different areas of the target three-dimensional preform, and uses three-dimensional simulation or three-dimensional simulation combined with theoretical models to obtain the change law of the initial tension with the first tension, contact angle, friction coefficient, yarn and yarn guide rod structural parameters, and obtains the data under different working conditions to form a training set. Then the neural network model is trained and optimized to obtain a proxy model that can accurately predict the yarn tension under different working conditions. The measured related parameters are input through the proxy model to obtain the predicted initial tension of the yarn in real time. Finally, the yarn is regulated by the yarn tension controller according to the predicted initial tension of the yarn, so that it is transmitted to the production line of the target three-dimensional preform in the optimal tension state. The weaving requirements of the target three-dimensional preform are met in a targeted manner, and the yarn is always in the optimal tension state during the entire weaving process, ensuring the excellent mechanical properties, structural stability and product quality of the target three-dimensional preform. The present invention is suitable for the initial tension control of large-capacity yarns.

[0142] Those skilled in the art can understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for adaptively controlling the yarn let-off tension in manufacturing a complex structure preform, characterized in that: include: According to the curvature of different areas on the surface of the target three-dimensional preform, the yarn density required for each area is obtained, and the first tension required for weaving different areas is obtained, where the first tension is the tension after the yarn is wound around the yarn guide rod; Inputting the first tension and related parameters under the current working condition into the proxy model to obtain a predicted initial tension, wherein the initial tension is the tension of the yarn before it is wound around the yarn guide rod; Controlling the yarn according to the predicted initial tension to meet the weaving requirements of the target three-dimensional preform; The relevant parameters include the contact angle between the yarn and the yarn guide rod, the friction coefficient during the contact process, and the structural parameters of the yarn guide rod and the yarn; The proxy model is constructed by training the neural network model through the training set.

2. The method according to claim 1, characterized in that The construction of the training set includes: Establish a three-dimensional model of yarn winding around the yarn guide rod; Setting calculation conditions, performing numerical simulation on the three-dimensional model, and calculating and obtaining simulation values ​​of the initial tension under different working conditions; Performing experimental verification on the simulated value, and correcting the simulated value according to the experimental result; The corrected initial tension under each working condition and the corresponding model parameters and structural parameters of the three-dimensional model are used to form sample data, and the sample data corresponding to each working condition are summarized as a first training set; The proxy model is constructed by training the neural network model with the first training set; The calculation conditions include numerical model parameters and solver analysis conditions.

3. The method according to claim 2, characterized in that The construction of the training set also includes: Using the data of the first training set and taking the yarn tension theoretical model as the target fitting function, determining the undetermined coefficients of the target fitting function through nonlinear curve fitting; The control variable method is adopted to calculate the theoretical value of the initial tension under more working conditions through the yarn tension theoretical model; The theoretical values ​​and the corresponding variable values ​​constitute a second training set, and together with the first training set constitute a total training set; The proxy model is constructed by training the neural network model through the total training set; The yarn tension theoretical model is: In the formula, F b 、F a They represent the tension of the yarn before and after it passes around the guide rod respectively; θ represents the contact angle in arc between the yarn and the guide rod; r is the radius of the guide rod; a is the nonlinear friction coefficient between the yarn and the guide rod, and a and n are the unknown coefficients, both of which are constants related to the materials of the two objects at the junction.

4. The method according to claim 2, characterized in that: The numerical model parameters include: density, elastic modulus, Poisson’s ratio, and tensile strength of yarns; The material of the yarn guide rod; The first tension of the yarn after it is wound around the yarn guide rod; The contact relationship and friction coefficient between the yarn and the guide wheel; The loads and boundary conditions of the three-dimensional model.

5. The method according to claim 4, characterized in that The setting of the solver analysis conditions includes: Set two load steps and set the calculation substeps for each load step; The gravity is applied to the entire three-dimensional model during the calculation time of the first load step, and the first tension is applied to the yarn during the calculation time of the second load step.

6. The method according to claim 4, characterized in that The setting of the boundary conditions includes: Apply fixed support to the two end faces of the yarn guide rod to ensure that it does not move; A fixed support is applied to one end of the yarn, and a force is set along the direction of yarn transmission at the other end.

7. The method according to claim 2, characterized in that: The step of experimentally verifying the simulated value and correcting the simulated value according to the experimental result includes: An experimental platform for winding yarn around a yarn guide rod is built, and experimental tests are carried out using experimental conditions corresponding to the numerical simulation calculation conditions to obtain the test value of the initial tension under the corresponding working conditions; The test value is compared with the simulation value. If the error is not within the set range, the simulation value is replaced by the test value, thereby correcting the simulation value.

8. The method according to claim 2, characterized in that: The structural parameters of the three-dimensional model include the radius of the yarn, the radius of the yarn guide rod, and the contact angle between the yarn and the yarn guide rod.

9. The method according to claim 2, characterized in that: The construction of the proxy model includes: Dividing the training set into a training data set and a validation data set; Optimizing the neural network model parameters using the training data set; Using the validation data set to evaluate the performance of the neural network model; The training is completed until the mapping relationship between the corrected initial tension and the corresponding model parameters and the structural parameters of the three-dimensional model meets the set requirements; The neural network model includes a convolutional neural network, a recurrent neural network or a multi-layer perceptron.

10. The method according to claim 1, characterized in that The method controls the yarn according to the predicted initial tension through a yarn letting-off device, wherein the yarn letting-off device comprises: A bobbin rack (2) on which a plurality of bobbins (1) are arranged; A guide frame (5) is composed of a plurality of yarn guide rods of the same diameter and staggered in a horizontal and vertical direction to form a grid structure; after a single yarn (4) is produced from the yarn tube (1), it passes through a single grid and bypasses the yarn guide rods, thereby being transferred to a production line of a target three-dimensional preform; A yarn tension controller (3) for adjusting the initial tension of the single yarn (4) to make it consistent with the predicted initial tension; The distance between the bobbin frame (2) and the guide frame (5) is adjustable.

Images