A neural network-based measurement point vibration error prediction method

Abstract

This invention discloses a neural network-based method for predicting vibration errors at measurement points. It includes the following steps: constructing geometric models of the fixture and ground within the measurement field; finite element simulation calculations; establishing a neural network data model; and calculating the compensation results for each measurement point under different vibration environments. This method utilizes the transient analysis module in finite element software to solve for the maximum displacement component of the fixture under vibration. Combining information from laser tracker measurements (distance, elevation angle, horizontal angle, equipment position, vibration source amplitude, distance from the vibration source to the measurement point, and distance from the vibration source to the equipment), a neural network-based vibration error prediction model for measurement points is proposed based on the finite element results. This model can calculate the compensation results for each measurement point under different vibration environments, compensating for the measurement results and reducing the time cost of measurement in engineering. It is a fast and effective method for calculating measurement accuracy compensation.

Description

Technical Field

[0001] This invention relates to a method for predicting vibration errors at measurement points based on neural networks, specifically a method for compensating vibration errors in large aircraft assembly tooling based on a neural network model. Background Technology

[0002] In aircraft assembly, the measurement accuracy of laser trackers determines the assembly quality. Boeing has already applied laser trackers to measurement systems for models such as the Boeing 747 and 787. This system can measure the shape of a Boeing 747 fuselage with a diameter of 6 meters with a measurement accuracy of 0.127 mm, and the assembly accuracy of the aircraft reaches 0.25 mm. However, with the development of larger aircraft, there is a need to apply laser trackers to measurement scenarios where the measurement object exceeds 12 meters. The assembly of large-sized aircraft requires the development of large-sized tooling. The measurement of large-sized tooling directly determines the relative positions between various assembly components during the docking process, affecting the adjustment and docking of aircraft components. However, the internal environment of the factory where large tooling is located is generally subject to vibrations caused by milling, drilling, and transport vehicles. Significant vibrations may cause deformation of the tooling, causing the coordinates of the observation points to deviate from the theoretical positions, thereby reducing assembly quality. In addition, vibrations can also reduce the accuracy of the fitting between the theoretical and measured coordinates of the measurement points, reducing assembly stability. Therefore, to address the problem of tooling deformation and reduced assembly accuracy caused by vibration, it is necessary to propose a vibration error analysis method for measurement points of large aircraft assembly tooling. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a neural network-based method for predicting vibration errors at measurement points. The method utilizes the transient analysis module in finite element analysis software to solve for the maximum horizontal displacement component of the fixture under vibration. By combining information from laser tracker measurements of distance, pitch angle, horizontal angle, equipment position, vibration source amplitude, distance from the vibration source to the measurement point, and distance from the vibration source to the equipment, a neural network-based vibration error prediction model is established. This model can calculate the compensation results for each measurement point under vibration conditions, compensate for the measurement results, further improve the measurement accuracy of fixture deformation, and reduce the time cost of measurement in engineering projects.

[0004] The technical solution of this invention is: a method for predicting vibration error at a measurement point based on a neural network, comprising the following steps:

[0005] Step 1: Construct a geometric model of the tooling and ground within the measurement area;

[0006] 1. Establish a precise three-dimensional geometric model of the aircraft tooling, matching the actual object at a 1:1 scale;

[0007] 2. After deleting the bosses, triangles, washers, screws, and bolts from the geometric model, import it into the finite element simulation software.

[0008] Step 2: Finite element simulation calculation;

[0009] 1. Input vibration source information under a specific environmental condition into the finite element analysis software, including vibration amplitude, vibration frequency, and vibration duration;

[0010] 2. Boundary condition setting: The ground is a fixed boundary condition; the interface position between the tooling and the ground is constrained by 6 degrees of freedom;

[0011] 3. Mesh Generation: The maximum dimensions of the mesh (length, width, and height) shall not exceed 1% of the tooling's length, width, and height.

[0012] 4. The ground vibration information, including acceleration, velocity, and displacement, is transmitted to the tooling through the transient response analysis module to obtain the maximum horizontal displacement component U of the tooling under a certain environmental condition. x U y U z .

[0013] Step 3: Establish a neural network data model;

[0014] 1. Input the laser tracker's measurement distance, measurement pitch angle, measurement horizontal angle, equipment position, vibration source amplitude, distance from the vibration source to the point to be measured, and distance from the vibration source to the laser tracker;

[0015] 2. The maximum horizontal displacement component U obtained from finite element calculation x U y U z Establish a neural network data model, construct a measurement point prediction model, and train the neural network according to formula (1):

[0016]

[0017] Where δ is the learning rate in the neural network; n is the number of iterations; and ξ is the coefficient.

[0018] 3. Determine the number of neurons N in the hidden layer according to formula (2). n And begin training the neural network;

[0019]

[0020] Where, N in N represents the number of nodes in the input layer of the neural network. out is the number of nodes in the output layer of the neural network, and Const is a coefficient.

[0021] 4. For the maximum horizontal displacement component U obtained from finite element calculation xU y U z 40% of the results are used in steps 2 and 3 for neural network learning, 40% are used for data validation, and 20% are used for neural network data validation, ultimately obtaining U. x U y U z The compensation result.

[0022] Step 4: Calculate the compensation results for each measurement point under different vibration environments.

[0023] 1. Repeat steps 3 and 4 to obtain the maximum horizontal displacement component U at each measurement point in a large tooling measurement field under vibration conditions. x U y U z The compensation result;

[0024] 2. Repeat steps 3 and 4 to establish and calculate a database of compensation results for all measurement points under different vibration amplitudes, vibration frequencies, and vibration durations in different environments.

[0025] Beneficial effects of the invention: To prevent the decrease in assembly accuracy due to vibration during assembly, which leads to time-consuming and costly processes, this invention proposes a neural network-based method for predicting vibration errors at measurement points. This method constructs a geometric model of the tooling and ground within the measurement field. By inputting the vibration source location, vibration amplitude, vibration frequency, and vibration duration into finite element software, the maximum horizontal displacement component of the tooling under vibration is solved using the transient analysis module. Combining the input measurement distance, measurement pitch angle, measurement horizontal angle, equipment position, vibration source amplitude, distance from the vibration source to the measurement point, and distance from the vibration source to the equipment, a neural network-based model for predicting vibration errors at measurement points is proposed based on the finite element results. This model can calculate the compensation results for each measurement point under vibration conditions, compensating for the measurement results and reducing the time cost of measurement in engineering. It is a fast and effective method for calculating measurement accuracy compensation.

[0026] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

[0027] The parts not covered in this invention are the same as or can be implemented using existing technologies. Attached Figure Description

[0028] Figure 1 The tooling finite element model in step one of this invention

[0029] Figure 2 The vibration information input in step two of this invention

[0030] Figure 3 This is the neural network model calculation process in step three of the present invention.

[0031] Figure 4 This is the database for calculating the compensation results of each measurement point under vibration conditions in step four of this invention.

[0032] Numbering in the diagram: 1. Tooling; 2. Ground where the tooling is located. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0034] A method for deformation analysis of a measurement field in a non-uniform temperature field for a large tool 1 includes the following steps:

[0035] Step 1: Construction of the geometric model of the site fixtures and ground:

[0036] 1-1 A precise 3D geometric model of the aircraft tooling, at a 1:1 scale, was created using CATIA R2018 software, such as... Figure 1 As shown;

[0037] 1-2 After deleting the bosses, triangles, washers, screws, and bolts from the geometric model, import it into the finite element simulation software.

[0038] Step 2: Finite element simulation calculation

[0039] 2-1 In the ABAQUS 6.14 finite element analysis software, input the vibration source information, including the distance from the vibration source depth to the fixture being 5m. Assume the vibration source is a sinusoidal wave along the height direction, with a vibration amplitude of 0.1mm, a vibration frequency of 2Hz, and a vibration duration of 10s. Figure 2 As shown;

[0040] 2-2 Boundary Condition Setting: The ground is a fixed boundary condition; the interface position between the tooling and the ground is constrained by 6 degrees of freedom;

[0041] 2-3 Mesh Generation: The maximum dimensions of the mesh (length, width, and height) shall not exceed 1% of the tooling's length, width, and height;

[0042] 2-4 The ground vibration information (including acceleration, velocity, and displacement) is transmitted to the fixture through the transient response analysis module to obtain the maximum horizontal displacement component U of the fixture under vibration. x =0.0029mm, U y =0.0021mm, U z =0.0032mm;

[0043] Step 3: Establish the neural network data model:

[0044] 3-1 Input the actual laser tracker measurement distance as follows: 5.2m, elevation angle as follows: 0.349rad, horizontal angle as follows: 1.902rad, equipment position as follows: 2.011m, 1.179m, vibration source amplitude as follows: 0.1mm, distance from vibration source to measurement point as follows: 5m, distance from vibration source to equipment as follows: 7.21m.

[0045] 3-2 Based on the maximum horizontal displacement component obtained from finite element analysis, a neural network data model is established to construct a measurement point prediction model, such as... Figure 3 As shown. The neural network is trained according to formula (1):

[0046]

[0047] Where δ is the learning rate in the neural network, ranging from 0.001 to 0.1; n is the number of iterations, with a recommended value of 1.0 × 10⁻⁶. 5 ~1.0×10 6 ξ is a coefficient, and a value of 0.15 to 0.25 is recommended.

[0048] 3-3 Determine the number of neurons N in the hidden layer according to formula (2) n And begin training the neural network;

[0049]

[0050] Where, N in N represents the number of nodes in the input layer of the neural network. out N represents the number of nodes in the output layer of the neural network, and Const is a coefficient. In this invention, N... in =7, N out =3, Const=1. Determine the number of neurons N in the hidden layer. n =10, start neural network training using the Train function in MATLAB 2018 software;

[0051] 3-4 For the maximum horizontal displacement component U obtained from finite element calculation x U y U z 40% of the results are used for neural network learning through steps 3-2 to 3-3, 40% are used for data validation, and 20% are used for neural network data validation, ultimately obtaining U. x U y U z The compensation result.

[0052] Step 4: Calculate the compensation results for each measurement point under vibration conditions.

[0053] 4-1 Repeat steps 3 and 4 to obtain the three maximum horizontal displacement components U at the measurement point under the large tooling measurement field in a vibration environment. x U y U z The compensation results are 0.00236mm, -0.04205mm, and 0.00785mm;

[0054] 4-2 Repeat steps 3 and 4 to establish and calculate a database of compensation results for all measurement points under different vibration amplitudes, frequencies, and durations in different environments, such as... Figure 4 As shown.

[0055] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

[0056] The parts not covered in this invention are the same as or can be implemented using existing technologies.

Claims

1. A method for predicting vibration error at measurement points based on neural networks, characterized in that, Includes the following steps: Step 1 involves constructing a geometric model of the site's tooling and ground. The specific process is as follows: 1-1 Establish a precise three-dimensional geometric model of the aircraft tooling at a 1:1 scale with the actual object; 1-2 After deleting the bosses, triangles, washers, screws, and bolts from the geometric model, import it into the finite element simulation software; Step two, finite element simulation calculation, is as follows: 2-1 Input vibration source information under a specific environmental condition into the finite element analysis software, including vibration amplitude, vibration frequency, and vibration duration; 2-2 Boundary Condition Setting: The ground is a fixed boundary condition; the interface position between the tooling and the ground is constrained by 6 degrees of freedom; 2-3 Mesh Generation: The maximum dimensions of the mesh (length, width, and height) shall not exceed 1% of the tooling's length, width, and height; 2-4 The transient response analysis module transmits ground vibration information, including acceleration, velocity, and displacement, to the fixture to obtain the maximum displacement component of the fixture under a certain environmental condition. U x , U y , U z ; Step 3 involves establishing a neural network data model, the specific process of which is as follows: 3-1 Input the following information for the laser tracker: measurement distance, measurement pitch angle, measurement horizontal angle, equipment position, vibration source amplitude, distance from the vibration source to the point to be measured, and distance from the vibration source to the laser tracker; 3-2 Maximum displacement components obtained from finite element calculation U x , U y , U z Establish a neural network data model, construct a measurement point prediction model, and train the neural network according to Formula 1: （1） in, The learning rate in a neural network; n For the number of iterations, For coefficients; 3-3 Determine the number of neurons in the hidden layer according to Formula 2 N n And begin training the neural network; （2） in, This represents the number of nodes in the input layer of the neural network. This represents the number of nodes in the output layer of the neural network. For coefficients; 3-4 For the maximum displacement component obtained from finite element calculation U x , U y , U z 40% of the results are used for neural network learning through steps 3-2 to 3-3, 40% are used for data validation, and 20% are used for neural network data validation, ultimately yielding... U x , U y , U z The compensation result; Step four calculates the compensation results for each measurement point under different vibration environments. The specific process is as follows: 4-1 Obtaining the maximum displacement component at each measurement point in a large tooling measurement field under vibration conditions U x , U y , U z The compensation result; 4-2 Establish and calculate a database of compensation results for all measurement points under different vibration amplitudes, vibration frequencies, and vibration durations in different environments.

Images