A Method for Predicting the Assembly and Adjustment Quality of the Inner Bracing Fixture in the Storage Tank Based on Finite Element Simulation and RNN Neural Network

Abstract

A method for predicting the assembly and adjustment quality of an internal support fixture in a storage tank based on finite element simulation and RNN neural network, characterized by the following steps: Step 1: Construct a three-dimensional model of the storage tank cylinder section and the internal support fixture, and conduct finite element analysis on its assembly and adjustment process; Step 2: Construct a dataset of assembly and adjustment quality parameters; conduct preliminary screening and normalization on the data; Step 3: Use the training set to construct a machine learning prediction model; Step 4: Use the test set to test and verify the accuracy of the machine learning prediction model and evaluate it with evaluation indicators. The present invention can reduce the labor intensity of workers, significantly shorten the production cycle, and improve production efficiency.

Description

Technical Field

[0001] The present invention relates to a method for predicting the assembly and adjustment quality of a fixture, in particular to a technology for prediction using finite element and RNN neural network. Specifically, it is a method for predicting the assembly and adjustment quality of an internal support fixture in a storage tank based on finite element simulation and RNN neural network. Background Art

[0002] With the development of welding technology, friction stir welding technology has emerged. Friction stir welding technology is widely used in aerospace, rail transit and other fields. For example, the circumferential seam welding of rocket fuel storage tanks. Rocket storage tanks mostly use aluminum-magnesium alloy and aluminum-lithium alloy materials. The storage tank structure is large in volume, heavy in weight, and weak in rigidity, and is prone to deformation during the processing, with great processing difficulty. Therefore, an internal support fixture is required to ensure the welding quality during the welding process. At present, during the assembly and adjustment process of the internal support fixture for rocket fuel storage tanks, due to the complex structure of the internal support fixture, it completely relies on manual assembly and adjustment and the method of trial and error by experience, resulting in low efficiency, large manual labor, and also lengthening the production cycle of rocket fuel storage tanks. How to speed up the assembly and adjustment speed and improve the assembly and adjustment efficiency is a major problem in the manufacture of rocket storage tanks.

[0003] During the friction stir welding process of a rocket fuel storage tank, since the stirring head of the friction stir welding generates a large downward pressure, an internal support fixture is used for support inside. The internal support fixture generally consists of sixteen telescopic rods and arc-shaped pressing plates on each telescopic rod. In order to ensure the processing quality, when adjusting the fixture, it is necessary to adjust to meet the qualified requirements for roundness, misalignment amount and tire adherence.

[0004] At present, the assembly and adjustment workers all adopt the trial and error method, adjusting the telescopic amount of each rod in turn, and then measuring its roundness, misalignment amount and tire adherence. This method has the problems of low assembly and adjustment efficiency, large work load and unstable assembly and adjustment quality. In order to realize the automatic assembly and adjustment of the internal support fixture for fuel storage tanks, it is urgent to solve the problem of predicting the assembly and adjustment quality of the internal support fixture for storage tanks. Summary of the Invention

[0005] The object of the present invention is to address the problems of large workload and long cycle in the existing assembly and adjustment process of the internal support fixture for storage tanks, and propose a method for predicting the assembly and adjustment quality of the internal support fixture for storage tanks based on finite element simulation and RNN neural network, which can effectively predict the assembly and adjustment quality of the fixture.

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

[0007] A method for predicting the assembly and adjustment quality of the internal support fixture in a storage tank based on finite element simulation and RNN neural network. The quality parameters after assembly and adjustment are obtained by performing finite element rigid-flexible coupling dynamics simulation on the assembly and adjustment process of the internal support fixture in the storage tank. The data obtained from the simulation is initially screened and normalized, and then divided into a training set, a test set, and a cross-validation set. An RNN neural network prediction model is constructed using the training set, and the generalization ability of the RNN neural network prediction model is evaluated using cross-validation. The accuracy of the RNN neural network prediction model is tested using the test set and index evaluation is carried out. The specific steps are as follows:

[0008] Step 1: Use SOLIDWORKS to construct the internal support fixture model for the circumferential weld of the fuel storage tank during friction stir welding, and construct a simplified model of the tank cylinder section.

[0009] Step 2: Import the model using the finite element simulation software ABAQUS, assign material properties to the workpiece and fixture, and perform mesh division. During mesh division, the contact area part is refined to ensure the accuracy of the simulation results.

[0010] Step 3: Use finite element simulation to perform rigid-flexible coupling implicit dynamics analysis, and simulate the deformation of the cylinder section during the assembly and adjustment of the internal support fixture in the storage tank. The internal support fixture is set as a rigid body model, and the cylinder section is set as a flexible model. The adjustment parameters are the 16 expansion amounts of the 16 rods of the internal support fixture. After obtaining the shape of the workpiece after simulation, extract the features and calculate its assembly and adjustment quality parameters: roundness, tire-fitting degree, misalignment amount, and the position corresponding to the maximum value, and construct an assembly and adjustment quality parameter data set.

[0011] Step 4: Initially screen and normalize the data, and divide the data in the data set into a training set, a cross-validation set, and a test set according to the ratio of 8:1:1.

[0012] Step 5: Use the training set to construct a machine learning prediction model. The initial shape parameters of the un-welded cylinder section and the 16 expansion amounts of the internal support fixture are used as input parameters. Through the RNN neural network machine learning prediction model, calculate the roundness, tire-fitting degree, misalignment amount of the circumferential weld after assembly and adjustment, and the position corresponding to the maximum value of each quality index, and evaluate the accuracy of the RNN neural network prediction model with evaluation indexes. Use cross-validation to evaluate the generalization ability of the RNN neural network prediction model.

[0013] Step 6: Use the test set to test and verify the accuracy of the machine learning prediction model and evaluate it with evaluation indexes. When the accuracy meets the standard, output the RNN neural network prediction model. When the accuracy does not meet the standard, repeat Step 5 until the accuracy meets the standard. After the accuracy meets the standard, output the RNN neural network prediction model, which is the prediction model for the assembly and adjustment quality of the internal support fixture in the storage tank.

[0014] The finite element analysis method for the internal support fixture in the storage tank includes the following steps:

[0015] Step 1, key part of pre - processing: Using the model simplification method. Since the cylindrical section of the storage tank belongs to a thin - walled circular part and its thickness remains basically unchanged during the welding process, a group of discrete points with unequal radial distances can be connected in sequence to form a polygon to replace the circular cylindrical section, thus simplifying the three - dimensional simulation into a two - dimensional simulation.

[0016] Step 2, finite element analysis. When setting the boundary conditions, set the outward - convex points as fixed constraints; set the telescopic amount to zero when the internal support fixture just touches the cylindrical section; conduct a dynamic implicit analysis on the process of adjusting the telescopic amount to clamp the workpiece, and two adjacent cylindrical sections are clamped simultaneously.

[0017] Step 3, post - processing stage. All output quantities are related to displacement. Just set the output displacement field. Calculate the roundness, misalignment amount, tire - fitting degree, and the corresponding position of the maximum value from the output displacement field; extract the displacement change situation of the cross - sections of two adjacent cylindrical sections, and take the roundness of the two cross - section shapes with the largest roundness as the roundness; extract the maximum value of the distance between two discrete points at the corresponding positions of the two cross - sections as the misalignment amount; extract the maximum value of the distance from each discrete point on the two cross - sections to the pressure plate of the internal support fixture as the tire - fitting degree; at the same time, obtain the positions of the discrete points corresponding to each maximum value.

[0018] The described data set consists of the following parameters: The internal support fixture has 16 telescopic rods, and the un - welded cylindrical section has 6 initial - state parameters: roundness, roughness, misalignment amount, and the corresponding positions of the maximum values, a total of 22 input parameters. By adjusting the telescopic amounts of the 16 telescopic rods, the roundness, roughness, misalignment amount after adjustment, and the discrete - point numbers corresponding to the maximum values for reflecting the positions to be adjusted are obtained, a total of 6 output parameters.

[0019] When constructing the machine - learning prediction model using the training set, the machine - learning method selects the RNN neural network; the telescopic amounts of the 16 telescopic rods affect each other pairwise. The smaller the misalignment amount and the tire - fitting degree, the smaller the roundness will be. There is an interaction relationship among the input parameters, and the RNN conforms to this feature; the input layer is 16 + 6, the output layer is 6, and the hidden layer is set to 5; at the same time, in order to improve the problem of small RNN gradients, the GEU unit is adopted to change the RNN hidden layer so that it can better capture the deep connections.

[0020] The beneficial effects of the present invention are:

[0021] The present invention breaks the traditional manual debugging method, reduces the labor intensity of workers, reduces costs, significantly shortens the production cycle, and improves production efficiency. The present invention combines the advanced technologies of finite element and neural network in machine learning, making the fixture adjustment process gradually intelligent. Using this method, similar problems can be solved. A large number of complex fixtures are required in the aerospace field, and this method can be used to achieve the adjustment of the fixtures. This method is not only limited to the adjustment of the internal support fixture of the fuel storage tank, but can also be applied to the adjustment of other complex fixtures. Brief Description of the Drawings

[0022] Figure 1 Schematic diagram of the barrel section structure of the rocket fuel tank according to an embodiment of the present invention.

[0023] Figure 2 Simplified diagram of the internal support fixture in the rocket fuel tank according to an embodiment of the present invention.

[0024] Figure 3 Flow chart of the prediction method according to the present invention.

[0025] Figure 4 Structural diagram of the RNN neural network according to the present invention. Detailed Embodiments

[0026] The present invention will be further described below with reference to the drawings and embodiments.

[0027] As Figures 1-4 shown.

[0028] The general idea of the present invention is to use finite element simulation to construct a fixture assembly quality data set. After preprocessing the data in the data set, it is divided into a training set, a cross-validation set, and a test set according to a certain ratio through data normalization. The RNN neural network machine learning method is used to construct a prediction model for the assembly quality of the internal support fixture of the fuel tank. The generalization ability of the model is evaluated by the method of cross-validation to prevent overfitting. Then, the test set is used to test the reliability of the model. Numerical indicators such as the correlation coefficient R, the mean absolute error MAE, and the relative error RAE are used to evaluate the accuracy of the model, and finally the best prediction model is obtained. This prediction model can predict the assembly quality of the internal support fixture of the fuel tank during the friction stir welding process. Using this method to predict the fixture assembly quality can improve the development efficiency and shorten the production cycle.

[0029] A method for predicting the assembly quality of the internal support fixture of the fuel tank based on finite element simulation and RNN neural network, taking the assembly of the 3350mm internal support fixture of the fuel tank shown in the processing Figure 1 as an example, the specific implementation steps are as follows (as Figure 3 shown):

[0030] 1. Use SOLIDWORKS to construct a 3350mm internal support fixture model for the circumferential weld of the rocket fuel tank during friction stir welding, and construct a simplified model of the actual rocket fuel tank barrel section, with a roundness of 1.8mm, an offset of 0.8mm, a tire adherence of 0.41mm, the maximum roundness corresponding to 150 degrees, and the minimum roundness corresponding to 265 degrees.

[0031] 2. Use finite element simulation software ABAQUS to import the model, assign material properties to the workpiece and fixture, and perform mesh division. When dividing the mesh, refine the mesh in the contact area to ensure the accuracy of the simulation results. The barrel material is: aluminum-lithium alloy 2090, with a Young's modulus of 80Gpa and a density of 2.59g / cm -3 , Poisson's ratio is 0.3. The material of the internal support fixture is set to structural steel (such as Figure 2 As shown), Young's modulus is 206 / Gpa, and Poisson's ratio is 0.28.

[0032] 3. Use finite element simulation to perform rigid-flexible coupling implicit dynamics analysis, set the dynamic implicit analysis step, and simulate the deformation of the cylinder segment when the tank internal support fixture is installed. The internal support fixture is set as a rigid body model, and the cylinder segment is set as a flexible model. The parameters are adjusted to the 16 expansion and contraction amounts of the 16 rods of the internal support fixture, and the displacement cloud map of the simulated workpiece is obtained. After extracting the features, the installation quality is calculated: roundness, tire adhesion, misalignment, and the corresponding position of the maximum value. Construct an installation quality parameter data set.

[0033] 4. Perform preliminary screening and normalization on the data, remove the data with a displacement greater than 2 mm, normalize the remaining data according to the following formula, and divide the data in the data set into a training set, a cross-validation set, and a test set in a ratio of 8:1:1;

[0034] The normalization formula is:

[0035]

[0036] In the formula, y i is the normalized data, x i is the original data, x min is the minimum value of each dimension in the original data, x max is the maximum value of each dimension in the original data.

[0037] 5. Use the training set to build a machine learning prediction model, such as Figure 4 As shown. The initial shape parameters of the unprocessed barrel section (roundness, adhesion, misalignment and the corresponding position of the maximum value) and the 16 expansion and contraction amounts of the inner support fixture are used as input parameters. The RNN neural network prediction model is used to calculate the annular seam roundness, adhesion, misalignment and the corresponding position of the maximum value of each quality index after adjustment. The GEU unit is used to change the RNN hidden layer so that it can better capture the deep connection. The accuracy of the RNN neural network prediction model is evaluated by evaluation indicators, and the generalization ability of the RNN neural network prediction model is evaluated by cross-validation.

[0038] The evaluation metrics are one or more of the correlation coefficient R, mean absolute error MAE, and relative error RAE. Preferably, the calculation formula for the correlation coefficient R is:

[0039]

[0040] The calculation formula for the mean absolute error MAE is:

[0041]

[0042] The calculation formula for the relative error RAE is:

[0043]

[0044] In the formula, N is the total number of samples, y ai and y pi represent the true value and the predicted value respectively, represents the average value of all values of the simulation value, represents the average value of all values of the predicted value.

[0045] 6. Use the test set to test and verify the accuracy of the machine learning prediction model and evaluate it with the evaluation metrics. When the accuracy meets the standard, output the RNN neural network prediction model. When the accuracy does not meet the standard, repeat step 5 until the accuracy meets the standard. After the accuracy meets the standard, output the RNN neural network prediction model, which is the prediction model for the assembly and adjustment quality of the internal support fixture of the rocket storage tank.

[0046] The parts not involved in the present invention are the same as or can be implemented by the prior art.

Claims

1. A method for predicting the assembly and adjustment quality of the internal support fixture in a storage tank based on finite element simulation and RNN neural network, characterized in that It includes the following steps: Step 1: Construct the 3D models of the storage tank cylinder section and the internal support fixture, and conduct finite element analysis on their assembly and adjustment process to obtain the shape of the cylinder workpiece after simulation under different expansion amounts of the internal support fixture. After extracting features, calculate its assembly and adjustment quality parameters: roundness, tire fitting degree, and offset amount; Step 2: Construct a dataset of assembly and adjustment quality parameters; conduct preliminary screening and normalization on the data, and divide the data in the dataset into a training set, a cross-validation set, and a test set according to the ratio of 8:1:1; Step 3: Use the training set to construct a machine learning prediction model. The initial shape parameters of the un-welded cylinder section and the 16 expansion amounts of the internal support fixture are used as input parameters. Through the RNN neural network prediction model, calculate the roundness of the circumferential weld, tire fitting degree, offset amount after assembly and adjustment, and the positions corresponding to the maximum values of each quality index, and evaluate the accuracy of the RNN neural network prediction model with evaluation indicators, and adopt cross-validation to evaluate the generalization ability of the RNN neural network prediction model; Step 4: Use the test set to test and verify the accuracy of the machine learning prediction model and evaluate it with evaluation indicators. When the accuracy meets the standard, output the RNN neural network prediction model. When the accuracy does not meet the standard, repeat Step 3 until the accuracy meets the standard; after the accuracy meets the standard, output the RNN neural network prediction model, which is the prediction model for the assembly and adjustment quality of the internal support fixture of the storage tank; The finite element analysis method of the internal support fixture of the storage tank includes the following steps: Step 1, the key part of preprocessing: Use the model simplification method. Since the storage tank cylinder section belongs to a thin-walled circular part and the thickness remains basically unchanged during the welding process, a group of discrete points with unequal radial distances can be connected in sequence to form a polygon to replace the circular cylinder section, and the 3D simulation can be simplified to a 2D simulation; Step 2, finite element analysis. When setting the boundary conditions, set the outward convex points as fixed constraints; set the expansion amount to zero when the internal support fixture just contacts the cylinder section; conduct dynamic implicit analysis on the process of adjusting the expansion amount to press the workpiece, and two adjacent cylinder sections are pressed simultaneously; Step 3, in the post-processing stage, all output quantities are related to displacement. Just set the output displacement field. Calculate the roundness, offset amount, tire fitting degree, and the corresponding positions of the maximum values from the output displacement field; extract the displacement change situation of the cross-sections of two adjacent cylinder sections, and take the roundness with the largest roundness in the two cross-section shapes as the roundness; extract the maximum value of the distance between two discrete points at the corresponding positions of the two cross-sections as the offset amount; extract the maximum value of the distance from each discrete point on the two cross-sections to the pressure plate of the internal support fixture as the tire fitting degree; at the same time, obtain the positions of the discrete points corresponding to each maximum value; The dataset consists of the following parameters: The internal support fixture has 16 telescopic rods, and the un-welded cylinder section has 6 initial state parameters: roundness, roughness, offset amount, and the corresponding positions of the maximum values, a total of 22 input parameters. By changing the expansion amounts of the 16 telescopic rods, the roundness, roughness, offset amount after assembly and adjustment, and the discrete point numbers corresponding to the maximum values and used to reflect the positions that need to be adjusted are obtained, a total of 6 output parameters; When constructing a machine learning prediction model using a training set, the machine learning method selects an RNN neural network; the telescopic amounts of the 16 telescopic rods influence each other pairwise. The smaller the offset and the tire attachment degree, the smaller the roundness will necessarily be. There is an interaction relationship between the input parameters, and the RNN conforms to this characteristic; the input layer is 16 + 6, the output layer is 6, and the hidden layer is set to 5; at the same time, in order to improve the problem of small RNN gradients, the GEU unit is adopted to change the RNN hidden layer so that it can better capture deep connections.

Images