Pneumatic hammer rivet riveting forming prediction method based on finite element analysis and neural network

Abstract

The invention relates to a pneumatic hammer rivet riveting forming prediction method based on finite element analysis and a neural network, which is realized by constructing a BP neural network model and comprises the following steps: establishing a finite element model of a pneumatic hammer rivet to obtain a riveting deformation model; finite element models of different air pressures, different tightening forces and different striking times are established; verifying the accuracy of the finite element model through a test; rivet upset head data and interference amount data of a finite element model result are extracted to serve as a training set to be stored; and constructing a neural network and training data. By establishing the finite element model of pneumatic hammer riveting, the conditions of the rivet and the riveting plate during each riveting can be monitored in the finite element model; a database of various riveting working conditions of the pneumatic hammer rivet is established through a large amount of finite element modeling calculation, so that the waste of resources is reduced; and finally obtaining a prediction model for riveting forming of the pneumatic hammer rivet through a BP neural network prediction model.

Description

Technical Field

[0001] The invention relates to the field of digital assembly of aircraft, and in particular to a prediction method for pneumatic hammer riveting forming based on finite element analysis and neural network. Background Art

[0002] In the aviation field, assembly and connection are important links in aircraft production, and the labor involved in aircraft assembly accounts for about 45% of the labor involved in aircraft manufacturing, with riveting accounting for about 30%.

[0003] Riveting includes pull riveting, press riveting, spin riveting, self-piercing riveting and pneumatic hammer riveting. Pneumatic hammer riveting has the characteristics of low cost, flexible use and less mutual influence of riveting. In the assembly working environment of some aircraft with complex structure, narrow space and curved surface, the riveting work is often completed by riveting workers holding pneumatic riveting guns. Pneumatic hammer riveting is usually completed by two workers in cooperation, one worker holding a riveting gun and the other worker holding a top iron. The worker holding the riveting gun uses the rivet gun head to hit the placed rivet rod several times until the riveting process is completed by the hitting. Riveting is divided into positive riveting and reverse riveting. The working conditions used for positive riveting and reverse riveting are different. For positive riveting, the rivet gun is pressed against the rivet rod to hit, and for reverse riveting, the rivet gun is pressed against the rivet head to hit.

[0004] The rivet-related parameters after riveting are one of the indicators for evaluating whether the riveting quality is qualified, such as the rivet head height, head diameter and interference of the rivet rod. Now they can meet the forming of the rivet, but the monitoring of each rivet forming has not been realized. The head size is the result after the riveting is completed, and the result cannot reflect the process of pneumatic hammer riveting. It is impossible to identify whether there are errors in the riveting process from the morphology, and it is impossible to know the real-time changes of the head during the pneumatic hammer riveting process. Summary of the invention

[0005] In order to solve the technical problems existing in the prior art, the present invention provides a prediction method for pneumatic hammer riveting forming based on finite element analysis and neural network. By establishing a finite element model of pneumatic hammer riveting and a BP neural network prediction model, a prediction model for pneumatic hammer riveting forming is finally obtained.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A prediction method for pneumatic hammer riveting forming based on finite element analysis and neural network, the prediction method is implemented by constructing a BP neural network model, and the construction of the BP neural network model includes the following steps:

[0008] Step 1, establishing a finite element model of pneumatic hammer riveting to obtain a riveting deformation model;

[0009] Step 2, establishing finite element models with different air pressures, different tightening forces and different striking times;

[0010] Step 3: Verify the accuracy of the finite element model through experiments;

[0011] Step 4, extracting the rivet head data and interference data of the finite element model results and storing them as a training set;

[0012] Step 5: Build a neural network and train the data.

[0013] As a preferred technical solution, in step one, establishing a finite element model of pneumatic hammer riveting includes first establishing a three-dimensional model of hammer riveting, including a top iron, a rivet, a riveting plate, a riveting hammer and a piston, and then importing it into HyperMesh to divide the mesh, and importing the mesh file into ABAQUS software and adding materials, model assembly, load addition and boundary conditions.

[0014] As a preferred technical solution, in step three, a pneumatic hammer riveting test is performed to measure the upsetting head data and interference data of the rivet after riveting to verify the accuracy of the finite element model.

[0015] As a preferred technical solution, in step four, rivet head data and interference data are obtained after analysis using ABAQUS software. The rivet head data include rivet head height and head diameter, and the rivet interference data include head side interference, middle interference and head side interference.

[0016] As a preferred technical solution, in step five, the obtained ABAQUS data set is input into the BP neural network model for training, and the obtained ABAQUS data set is divided into 70% training set and 30% test set.

[0017] As a preferred technical solution, in step four, multiple evaluation indicators are selected, including mean square error, root mean square error, mean absolute error, and mean absolute percentage error, to evaluate the performance of the neural network model in the prediction task.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] (1) The present invention establishes an accurate pneumatic hammer riveting finite element model through ABAQUS software, and the conditions of the rivet and the riveted plate during each riveting can be monitored in the finite element model.

[0020] (2) The present invention establishes a database of various working conditions of pneumatic hammer riveting through a large amount of finite element modeling calculations, thereby reducing the waste of resources.

[0021] (3) The present invention uses MATLAB software to construct a neural network for pneumatic hammer riveting and riveting forming, and obtains a prediction model for pneumatic hammer riveting and riveting forming. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a finite element model diagram of the pneumatic hammer riveting of the present invention;

[0023] Figure 2 is a finite element analysis result diagram of the present invention;

[0024] Figure 3 It is a data position diagram to be measured for the rivet of the present invention;

[0025] Figure 4 It is a schematic diagram of a single hidden layer BP neural network structure of the present invention;

[0026] Figure 5 It is a schematic diagram of the model training results of the prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network of the present invention.

[0027] In the figure: 1. Top iron; 2. Rivet; 3. Riveting plate; 4. Riveting hammer; 5. Piston. DETAILED DESCRIPTION

[0028] The technical solution of the present invention is further described below in conjunction with specific implementation methods:

[0029] A prediction method for pneumatic hammer riveting forming based on finite element analysis and neural network, the prediction method is implemented by constructing a BP neural network model, and the construction of the BP neural network model includes the following steps:

[0030] Step 1: Use ABAQUS software to establish a finite element model of pneumatic hammer riveting to obtain a riveting deformation model;

[0031] Specifically, Figure 1 As shown, SolidWorks software creates a three-dimensional image of pneumatic hammer riveting, including top iron, rivet, riveting plate, riveting hammer and piston. The three-dimensional image is imported into HyperMesh software for meshing, which is divided into C3D8R type meshes. The mesh file is imported into ABAQUS software for assembly.

[0032] The material properties are set in ABAQUS software. The riveted plate is made of 2060 aluminum alloy, and the rivet is made of 2117 aluminum alloy. The JC constitutive model is used for material assignment.

[0033] Set the analysis step in ABAQUS software and select the dynamic-displayed solver for calculation.

[0034] The load is set in ABAQUS software, and the initial velocity load is given to the piston, so that the piston collides with the riveting hammer, and the riveting hammer hits the rivet until it is deformed.

[0035] The boundary conditions are set in ABAQUS software. The five parts of the pneumatic hammer riveting are all on the same axis, and the fixed support constraints are set for the riveted plate, which is consistent with the actual working conditions.

[0036] Submit the job in ABAQUS software. After setting all the conditions, submit the job and start the calculation. Select 40 threads to improve the calculation efficiency. The riveting results are as follows: Figure 2 shown.

[0037] Step 2: Establish finite element models with different air pressures, different tightening forces, and different striking times to prepare for the subsequent creation of training sets;

[0038] Different air pressures include 0.4MPa, 0.5MPa and 0.6MPa, different strike times include 1-30 strikes, and different tightening forces include 10Kgf, 20Kgf and 30Kgf.

[0039] Different air pressures require fitting the relationship between input air pressure and piston speed.

[0040] Step 3: Verify the accuracy of the finite element model through experiments;

[0041] Specifically, step three requires a pneumatic hammer riveting test bench, a number of rivets and a number of riveted plates.

[0042] The rivet material is 2117 aluminum alloy and the riveted plate material is 2060 aluminum alloy.

[0043] A high-speed force sensor is required during the test to record the force and number of riveting blows.

[0044] like Figure 3 As shown, after the test is completed, the rivet needs to be measured, including the rivet head height, rivet head diameter, rivet head side interference, middle interference and rivet head side interference.

[0045] Rivet measuring tools include height gauge, vernier caliper and micrometer. Use height gauge to measure the head height of rivet, use vernier caliper to measure the head diameter of rivet, and use micrometer to measure the head side interference, middle interference and head side interference of rivet.

[0046] The measured data are compared with the simulated data, as shown in Table 1 and Table 2 below:

[0047] Table 1

[0048]

[0049] Table 2

[0050]

[0051]

[0052] The average simulation error of the heading height is 3.78%, the average simulation error of the heading diameter is 2.49%, the average simulation error of the heading side interference is 9.91%, the average simulation error of the middle is 9.88%, and the average simulation error of the nail head side interference is 10.68%, which verifies the accuracy of the finite element model.

[0053] Step 4, extracting the rivet head data and interference data of the finite element model results and storing them as a training set;

[0054] The upsetting head data includes the upsetting head height and the upsetting head diameter, and the interference amount data includes the upsetting head side interference amount, the nail head side interference amount and the upsetting head side interference amount.

[0055] Step 5: Use MATLAB software to build a neural network and train the data. 70% of the data is used as the training set and 30% of the data is used as the test set for training.

[0056] like Figure 4 As shown in the figure, MATLAB software is used to construct a neural network and train data. The BP neural network includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is 3, the hidden layer is set to 8, and the number of neurons in the output layer is 5.

[0057] like Figure 5 As shown in the figure, after the BP neural network is trained, the mean square error MSE, root mean square error RMSE, mean absolute error MAE, and mean absolute percentage error MAPE will be displayed.

[0058] MSE is the expected value of the square of the difference between the predicted value and the actual value, RMSE is the square root of the expected value of the square of the difference between the predicted value and the actual value, and MAE represents the average absolute error between the predicted value and the observed value. The smaller the MAPE value, the better the accuracy of the prediction model.

[0059] The calculation formulas are as follows:

[0060] Mean Square Error (MSE)

[0061] Root mean square error RMSE

[0062] Mean absolute error (MAE)

[0063] Mean absolute percentage error MAPE

[0064] The mean square error (MSE) of the test set is 0.05%.

[0065] The root mean square error RMSE of the test set is 0.52%

[0066] The average absolute error MAE of the test set is 0.67%

[0067] The average absolute percentage error MAPE of the test set is 0.10%

[0068] It can be seen from the simulation and experiment that the experimental data verifies the accuracy of the simulation, and the BP neural network of pneumatic hammer riveting forming is constructed through data simulation. It can be seen from the neural network prediction data that the neural network model has good accuracy.

[0069] This embodiment is only a further explanation of the invention rather than a limitation of the invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as they are within the scope of the claims of the invention, they are protected by the patent law.

Claims

1. A prediction method for pneumatic hammer riveting based on finite element analysis and neural network, characterized in that: The prediction method is implemented by constructing a BP neural network model, and the construction of the BP neural network model includes the following steps: Step 1, establishing a finite element model of pneumatic hammer riveting to obtain a riveting deformation model; Step 2, establishing finite element models with different air pressures, different tightening forces and different striking times; Step 3: Verify the accuracy of the finite element model through experiments; Step 4, extracting the rivet head data and interference data of the finite element model results and storing them as a training set; Step 5: Build a neural network and train the data.

2. The prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network according to claim 1 is characterized in that: In the step one, establishing a finite element model of pneumatic hammer riveting includes first establishing a three-dimensional model of hammer riveting, including a top iron, a rivet, a riveting plate, a riveting hammer and a piston, and then importing it into HyperMesh to divide the mesh, and importing the mesh file into ABAQUS software and adding materials, model assembly, adding loads and applying boundary conditions.

3. The prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network according to claim 1 is characterized in that: In the step three, a pneumatic hammer riveting test is performed to measure the upsetting head data and interference data of the rivet after riveting to verify the accuracy of the finite element model.

4. The prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network according to claim 1 is characterized in that: In the step 4, rivet head data and interference data are obtained after analysis using ABAQUS software. The rivet head data includes rivet head height and head diameter, and the rivet interference data includes head side interference, middle interference and head side interference.

5. The prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network according to claim 4 is characterized in that: In the step five, the obtained ABAQUS data set is input into the BP neural network model for training, and the obtained ABAQUS data set is divided into 70% training set and 30% test set.

6. The prediction method of pneumatic hammer riveting forming based on finite element analysis and neural network according to claim 1 is characterized in that: In the step 4, multiple evaluation indicators are selected, including mean square error, root mean square error, mean absolute error, and mean absolute percentage error, to evaluate the performance of the neural network model in the prediction task.

Images