A method for detecting stress in aircraft piping assembly based on displacement characteristic data

Abstract

This invention belongs to the field of aircraft piping assembly stress detection technology, and relates to a method for aircraft piping assembly stress detection based on displacement feature data. This invention determines displacement feature data points through optimization methods, constructs a lightweight proxy model using deep learning neural network technology, and then combines it with a stress field inversion algorithm to complete the three-dimensional reconstruction and cloud map presentation of the stress field, achieving the purpose of on-site stress detection in aircraft piping assembly. This invention can detect the stress field of aircraft piping assembly using a small amount of easily measurable displacement feature data. Compared with traditional strain gauge or manual detection methods, it improves detection accuracy and efficiency, achieving the effect of reconstructing a complete aircraft piping stress field from a small amount of easily measurable displacement feature data, thus achieving the goal of on-site inspectability, measurability, and visualization of the aircraft piping assembly stress field. This invention can improve the detection speed and accuracy of aircraft piping assembly stress, solving the problem of inefficient stress detection in aircraft production sites.

Description

Technical Field

[0001] This invention belongs to the field of aircraft piping assembly stress detection technology, and relates to an aircraft piping assembly stress detection method based on displacement characteristic data. Background Technology

[0002] Modern aircraft extensively utilize hydraulic lines. These lines operate at high pressures, have complex shapes, and numerous joints, making them prone to leaks, fractures, and fatigue. Statistics show that piping problems account for 71% of manufacturing issues in domestic aircraft development, with hydraulic line problems being particularly prominent, accounting for 82.3% of all piping problems. Research has found that initial material defects, manufacturing deviations, assembly stress, and vibration loads are the main causes of piping failures. Vibration loads are normal operating loads and cannot be avoided; therefore, control of piping problems during development primarily focuses on three aspects: initial material defects, manufacturing deviations, and assembly stress. Currently, detection methods for initial material defects and manufacturing deviations are relatively comprehensive, while methods for detecting assembly stress are quite limited. Existing methods for testing assembly stress include modal analysis, X-ray analysis, strain gauge testing, and inductive coating testing. These methods generally suffer from drawbacks such as long testing cycles, narrow applicability, and low efficiency. They are only suitable for laboratory testing scenarios and not for testing the assembly stress of aircraft piping in the complex assembly environment of aircraft production sites. Currently, the testing of aircraft piping assembly stress in aircraft production sites still relies mainly on manual visual inspection and touch, which results in long testing cycles and makes it difficult to provide quantitative references for assembly stress control. There is an urgent need to develop a new method for testing the assembly stress of aircraft piping. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of current methods for detecting stress fields in aircraft piping assemblies, including a lack of methods, low accuracy, and long testing cycles. This invention provides a method for detecting stress in aircraft piping assemblies based on displacement feature data. First, displacement feature data points are determined through optimization methods. Then, a lightweight proxy model is constructed using deep learning neural network technology. Finally, combined with a stress field inversion algorithm, a three-dimensional reconstruction and cloud map presentation of the stress field are completed, achieving the goal of on-site stress detection in aircraft piping assemblies. This invention can detect the stress field of aircraft piping assemblies using a small amount of easily measurable displacement feature data, improving detection accuracy and efficiency compared to traditional strain gauge or manual detection methods.

[0004] The technical solution of this invention is: to provide a method for detecting stress in aircraft piping assembly based on displacement characteristic data, comprising:

[0005] Step 1: Establishment of the finite element model of the aircraft piping: Based on the three-dimensional digital model of a certain aircraft piping to be tested, construct a finite element model of the aircraft piping for numerical simulation.

[0006] Step 2, Displacement Feature Point Identification: On the finite element model of the aircraft piping, determine the displacement feature points;

[0007] Step 3, 3D→2D spatial mapping: The 3D to 2D data dimensionality is reduced from three-dimensional to two-dimensional using the spatial mapping method;

[0008] Step 4: Determine the input and output physical quantities: Determine the input physical quantity as the displacement data at the displacement feature points, where the displacement data is a set of displacement values ​​at the displacement feature points; determine the output physical quantity as a set of three-dimensional stress data of the aircraft piping finite element model.

[0009] Step 5: Assembly stress sample generation and expansion: Select 100-200 sets of typical displacement characteristic data, including the maximum and minimum displacement boundaries, from the displacement database measured on the production site. Through finite element calculation, generate the corresponding aircraft pipeline assembly stress field to form an assembly stress seed sample library. Then, expand the sample library to generate the assembly stress sample library.

[0010] Step 6, Neural Network Model Construction: A fully connected neural network model is used, with input parameters X. 1k Output parameter Y ′ 1m Where k is the number of input parameters and m is the number of output parameters, the number of hidden layers is determined to be 2 layers, divided into input hidden layers and output hidden layers. The number of neurons in the input hidden layer is n, and the number of neurons in the output hidden layer is m. The response functions of the hidden layers all adopt the form of linear equations, and the coefficient of the linear term of the response function of the input hidden layer is... The constant term is The coefficients of the first term of the response function of the output hidden layer are The constant term is The final expression for the neural network model is:

[0011]

[0012] Step 7, Neural Network Model Training: Input the assembly stress sample library data generated in Step 5 through amplification into the neural network model constructed in Step 6 to train the neural network model;

[0013] Step 8, Proxy Model Generation: After all the model training in Step 7 is completed, a proxy model will be generated. A proxy model with parameters, whose proxy input parameters Agent output parameters The mathematical expression for this proxy model is:

[0014]

[0015] Step 9: Measurement of displacement data of feature points in the reference state: The displacement data of the pipeline displacement feature points in the unassembled reference state are measured by 3D scanning method to obtain the displacement feature data in the reference state.

[0016] Step 10: Measurement of displacement data of feature points after assembly: The displacement data of the pipeline displacement feature points in the assembled state are measured by 3D scanning method to obtain the displacement feature data in the assembled state.

[0017] Step 11: Stress field inversion based on feature data: ... and Subtraction yields the relative displacement data at the characteristic points of the pipeline displacement. Will Input into the proxy model generated in step 8, and calculate the result with... Corresponding pipeline assembly stress field data Y 1m ;

[0018] Step 12: 2D→3D Spatial Inverse Mapping: First, the pipeline assembly stress field data Y... 1m Convert the data into a 2D data matrix, and then, based on the mapping relationship formed in step 3, back-map the stress field data in the 2D data matrix to the grid nodes in 3D space to generate grid node stress field data.

[0019] Step 13: Aircraft piping stress field reconstruction and cloud map generation: Load the mesh node stress field data generated in Step 12 onto the aircraft piping finite element model, and generate a stress field cloud map by rendering.

[0020] Step 1 specifically involves setting the simulation parameters of the aircraft piping finite element model according to the actual situation, selecting 2D elements to establish the aircraft piping finite element model.

[0021] The simulation parameters include material, properties, load, constraints, solution parameters, and output parameters.

[0022] The displacement feature points in step 2 are a small number of easily measurable points selected from the center line of the pipeline. Their number and location are determined by the minimum correlation optimization method. These are the measurement points used to determine the displacement feature data, which is the displacement data of the displacement feature points.

[0023] The displacement data is a set of displacement values ​​at a set of displacement feature points.

[0024] The method for data dimensionality reduction in step 3 is as follows: a spatial mapping method of projection aggregation compression is used to map the 3D space mesh nodes of the aircraft pipeline finite element model to a 2D space data point matrix, while recording the mapping relationship between the mesh node number and the position of the data point matrix element.

[0025] The data dot matrix is ​​a two-dimensional array used to store the output data of the neural network model.

[0026] The assembly stress seed samples in the assembly stress seed sample library in step 5 consist of displacement feature data and corresponding assembly stress field data.

[0027] Step 5, amplification, involves grouping the assembly stress seed samples, randomly selecting assembly stress seed sample A and assembly stress seed sample B, and randomly combining them. The amplification amount is 100 times the number of assembly stress seed samples, resulting in assembly stress samples.

[0028] Step 7 specifically involves using the displacement feature data from the assembly stress samples as the input parameter X of the neural network model. 1k The assembly stress field data in the assembly stress sample is used as the target output parameter Y of the neural network model. 1m The neural network model is trained iteratively, with the number of iterations equal to the number of samples, Sum. The neural network model is trained using an iterative method, where the number of iterations equals the number of samples, Sum. The input is fed into a neural network model, and through calculation, a set of output parameters is obtained. With the target output parameters in this sample By comparison, the parameters of the hidden layers of the neural network model are analyzed using the minimum mean error method. Update and generate Complete one training iteration of the neural network model; in the next training iteration, the next sample will be used... The input is fed into the neural network model, while also using the data obtained from the previous calculation. Generate calculation output parameters With the target output parameters in the next sample By making a comparison, the minimum mean error method is used to... Update the sample; where s represents the number of iterations, s = 1, 2, 3, ..., Sum. Repeat the above training method until every sample has been used.

[0029] The beneficial effect of this invention is that it enables the reconstruction of a complete aircraft piping stress field from a small amount of easily measurable displacement characteristic data, achieving the goal of on-site inspection, measurement, and visualization of the aircraft piping assembly stress field. This invention can improve the detection speed and accuracy of aircraft piping assembly stress, solve the problem of inefficient piping assembly stress detection on aircraft production sites, and can be used to develop a new type of on-site aircraft piping assembly stress detection equipment. Attached Figure Description

[0030] Figure 1 This is a flowchart of the detection method described in this invention.

[0031] Figure 2 This is a schematic diagram of step 1 of the present invention.

[0032] Figure 3 This is a schematic diagram of step 2 of the present invention.

[0033] Figure 4 This is a schematic diagram of step 3 of the present invention.

[0034] Figure 5 This is a schematic diagram of step 4 of the present invention.

[0035] Figure 6 This is a schematic diagram of step 5 of the present invention.

[0036] Figure 7 This is a schematic diagram of step 6 of the present invention.

[0037] Figure 8 This is a schematic diagram of step 7 of the present invention.

[0038] Figure 9 This is a schematic diagram of step 8 of the present invention.

[0039] Figure 10 This is a schematic diagram of step 11 of the present invention.

[0040] Figure 11 This is a schematic diagram of step 12 of the present invention.

[0041] Figure 12 This is a schematic diagram of step 13 of the present invention.

[0042] In the diagram: 101, 3D digital model; 102, finite element model; 201, displacement feature points; 301, mesh nodes; 302, data lattice; 401, displacement data; 402, 3D stress field; 500, assembly stress seed sample library; 501, assembly stress seed sample A; 502, assembly stress seed sample B; 550, assembly stress sample library; 551, assembly stress sample; 600, neural network model; 601, input parameter X 1k 602. Output parameter Y′ 1m 680. Proxy Model; 681. Proxy Input Parameters 682. Agent Output Parameters 701. Relative Displacement Data 702. 2D data matrix; 703. Stress field data; 704. Stress field cloud map. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0044] like Figure 1 As shown, the specific implementation steps of the aircraft piping assembly stress detection method based on displacement feature data described in this invention are as follows:

[0045] Step 1: Establish the finite element model of the aircraft piping. For example... Figure 2 As shown, based on a three-dimensional digital model 101 of an aircraft pipeline to be inspected, a finite element model 102 for numerical simulation of the aircraft pipeline is constructed. The mesh type of the finite element model 102 is selected as 2D element. The simulation parameters of the finite element model 102 are set according to the actual situation. The simulation parameters include material, properties, load, constraints, solution parameters, and output parameters. The finite element model 102 is used for subsequent identification of displacement feature points and generation of assembly stress samples.

[0046] Step 2: Identification of displacement feature points. For example... Figure 3 As shown, displacement feature points 201 are determined on the finite element model 102 of the aircraft piping system. These displacement feature points 201 are selected from the centerline of the piping system, and their number and location are determined using a minimum correlation optimization method. These displacement feature points are used to determine the measurement points for displacement feature data, which is the displacement data of these feature points and is used as data input for the aircraft piping assembly stress detection method.

[0047] Step 3, as follows Figure 4 As shown, 3D to 2D spatial mapping is performed. The finite element model 102 of the aircraft piping system is reduced from three-dimensional to two-dimensional using a spatial mapping method. A projection-aggregation compression spatial mapping method is employed to map the 3D space mesh nodes 301 to the 2D space data point matrix 302, while simultaneously recording the mapping relationship between the mesh node numbers and the positions of the data point matrix elements. The mesh nodes 301 are the mesh nodes of the finite element model. The data point matrix 302 is a two-dimensional array used to subsequently store the output data of the neural network model 600.

[0048] Step 4, as follows Figure 5 As shown, the input and output physical quantities are determined. The input physical quantity is determined to be the displacement data 401 on the displacement feature point 201, which is a set of displacement values ​​on the displacement feature point 201. The output physical quantity is determined to be the three-dimensional stress field 402 of the aircraft piping finite element model 102, which is a set of stress values ​​of the aircraft piping finite element model 102.

[0049] Step 5, as follows Figure 6As shown, the assembly stress sample generation and amplification are performed. From a displacement database measured on the production site, 100-200 sets of typical displacement characteristic data are selected as load conditions for the finite element model. Through finite element calculation, the corresponding aircraft piping assembly stress field is generated, forming an assembly stress seed sample library 500. This is then amplified to generate an assembly stress sample library 550. The assembly stress seed samples in the assembly stress seed sample library 500 consist of displacement characteristic data and corresponding assembly stress field data, used for assembly stress sample amplification. The amplification is achieved by randomly combining assembly stress seed samples A501 and B502, with the amplification amount generally being 100 times the number of assembly stress seed samples, resulting in assembly stress sample 551.

[0050] Step 6, as follows Figure 7 As shown, the neural network model is constructed. A fully connected neural network model 600 is used, with input parameters X. 1k 601, Output parameter Y′ 1m 602. The number of hidden layers is determined to be two, consisting of an input hidden layer and an output hidden layer. The number of neurons in the input hidden layer is n, and the number of neurons in the output hidden layer is m. The response functions of the hidden layers are all in the form of linear equations. The coefficient of the linear term of the response function of the input hidden layer is... The constant term is The coefficients of the first term of the response function of the output hidden layer are The constant term is The final expression for the neural network model is:

[0051]

[0052] Step 7, as follows Figure 8 As shown, the neural network model is trained. The assembly stress sample library generated in step 5 through amplification is input into the neural network model constructed in step 6 for neural network model training. The displacement feature data in the assembly stress samples are used as the input parameter X of the neural network model. 1k The assembly stress field data in the assembly stress sample is used as the target output parameter Y of the neural network model. 1m The neural network model is trained iteratively, taking a sample... The input is fed into a neural network model, and through calculation, a set of output parameters is obtained. With the target output parameters in this sample By comparison, the parameters of the hidden layers of the neural network model are analyzed using the minimum mean error method. Update and generate Complete one training iteration of the neural network model. In the next training iteration, the next sample will be used... The input is fed into the neural network model, while also using the data obtained from the previous calculation. Generate calculation output parameters With the target output parameters in the next sample By making a comparison, the minimum mean error method is used to... Update the training. Repeat the above training method until every sample has been used.

[0053] Step 8, as follows Figure 9 As shown, the proxy model is generated. Once the model training in step 7 is complete, a proxy model will be generated. The surrogate model 680 is a parameter, and its surrogate input parameter X 1k 681, Agent output parameter Y 1m 682, the mathematical expression for the proxy model 680 is:

[0054]

[0055] Step 9: Measurement of displacement data of feature points in the reference state. The displacement data of the pipeline's displacement feature points in the unassembled reference state are measured using a 3D scanning method to obtain the displacement feature data in the reference state.

[0056] Step 10: Measurement of displacement data of feature points after assembly. The displacement data of pipeline displacement feature points in the assembled state are measured using a 3D scanning method to obtain the displacement feature data in the assembled state.

[0057] Step 11: Stress field inversion based on feature data. and Subtraction yields the relative displacement data at the characteristic points of the pipeline displacement. 701, will Input 701 into the proxy model 680 generated in step 8, and calculate the result with... The corresponding pipeline assembly stress field data Y for 701 1m .

[0058] Step 12: As Figure 10-11 As shown, this is a 2D to 3D spatial inverse mapping. First, the pipeline assembly stress field data Y... 1m The data is converted into a 2D data matrix 702. Then, based on the mapping relationship formed in step 3, the stress field data in the 2D data matrix 702 is back-mapped onto the grid nodes in 3D space to generate grid node stress field data 703.

[0059] Step 13: As Figure 12As shown, the stress field of the aircraft piping is reconstructed and a cloud map is generated. The stress field data 703 of the mesh nodes generated in step 12 is loaded onto the finite element model of the aircraft piping, and a stress field cloud map 704 is generated by rendering.

Claims

1. A method for detecting stress in aircraft piping assembly based on displacement characteristic data, characterized in that, include: Step 1: Establishment of the finite element model of the aircraft piping: Based on the three-dimensional digital model of a certain aircraft piping to be tested, construct a finite element model of the aircraft piping for numerical simulation. Step 2, Displacement Feature Point Identification: On the finite element model of the aircraft pipeline, determine the displacement feature points; the displacement feature points are a small number of easily measurable points selected from the center line of the pipeline, and their number and location are determined by the minimum correlation optimization method. Step 3, 3D→2D spatial mapping: The 3D to 2D data dimensionality is reduced from three-dimensional to two-dimensional using the spatial mapping method; Step 4: Determine the input and output physical quantities: Determine the input physical quantity as the displacement data at the displacement feature points, where the displacement data is a set of displacement values ​​at the displacement feature points; determine the output physical quantity as a set of three-dimensional stress data of the aircraft piping finite element model. Step 5: Assembly stress sample generation and expansion: Select 100-200 sets of typical displacement characteristic data, including the maximum and minimum displacement boundaries, from the displacement database measured on the production site. Through finite element calculation, generate the corresponding aircraft pipeline assembly stress field to form an assembly stress seed sample library. Then, expand the sample library to generate the assembly stress sample library. Step 6, Neural Network Model Construction: A fully connected neural network model is used, with input parameters X. 1k Output parameter Y ′ 1m Where k is the number of input parameters and m is the number of output parameters, the number of hidden layers is determined to be 2 layers, divided into input hidden layers and output hidden layers. The number of neurons in the input hidden layer is n, and the number of neurons in the output hidden layer is m. The response functions of the hidden layers all adopt the form of linear equations, and the coefficient of the linear term of the response function of the input hidden layer is... The constant term is The coefficients of the first term of the response function of the output hidden layer are The constant term is The final expression for the neural network model is: Step 7, Neural Network Model Training: Input the assembly stress sample library data generated in Step 5 through amplification into the neural network model constructed in Step 6 to train the neural network model; Step 8, Proxy Model Generation: After all the model training in Step 7 is completed, a proxy model will be generated. A proxy model with parameters, whose proxy input parameters Agent output parameters The mathematical expression for this proxy model is: Step 9: Measurement of displacement data of feature points in the reference state: The displacement data of the pipeline displacement feature points in the unassembled reference state are measured by 3D scanning method to obtain the displacement feature data in the reference state. Step 10: Measurement of displacement data of feature points after assembly: The displacement data of the pipeline displacement feature points in the assembled state are measured by 3D scanning method to obtain the displacement feature data in the assembled state. Step 11: Stress field inversion based on feature data: ... and Subtraction yields the relative displacement data at the characteristic points of the pipeline displacement. Will Input into the proxy model generated in step 8, and calculate the result with... Corresponding pipeline assembly stress field data Y 1m ; Step 12: 2D→3D Spatial Inverse Mapping: First, the pipeline assembly stress field data Y... 1m Convert the data into a 2D data matrix, and then, based on the mapping relationship formed in step 3, back-map the stress field data in the 2D data matrix to the grid nodes in 3D space to generate grid node stress field data. Step 13: Aircraft piping stress field reconstruction and cloud map generation: Load the mesh node stress field data generated in Step 12 onto the aircraft piping finite element model, and generate a stress field cloud map by rendering.

2. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 1, characterized in that, Step 1 specifically involves setting the simulation parameters of the aircraft piping finite element model according to the actual situation, selecting 2D elements to establish the aircraft piping finite element model.

3. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 2, characterized in that, The simulation parameters include material, properties, load, constraints, solution parameters, and output parameters.

4. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 1, characterized in that, The displacement feature point in step 2 is a measurement point used to determine the displacement feature data, and the displacement feature data is the displacement data of the displacement feature point.

5. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 4, characterized in that, The displacement data is a set of displacement values ​​at a set of displacement feature points.

6. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 1, characterized in that, The method for data dimensionality reduction in step 3 is as follows: a spatial mapping method of projection aggregation compression is used to map the 3D space mesh nodes of the aircraft pipeline finite element model to a 2D space data point matrix, while recording the mapping relationship between the mesh node number and the position of the data point matrix element.

7. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 6, characterized in that, The data dot matrix is ​​a two-dimensional array used to store the output data of the neural network model.

8. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 1, characterized in that, The assembly stress seed samples in the assembly stress seed sample library in step 5 consist of displacement feature data and corresponding assembly stress field data.

9. The aircraft piping assembly stress detection method based on displacement characteristic data according to claim 1, characterized in that, Step 5, amplification, involves grouping the assembly stress seed samples, randomly selecting assembly stress seed sample A and assembly stress seed sample B, and randomly combining them. The amplification amount is 100 times the number of assembly stress seed samples, resulting in assembly stress samples.

10. The aircraft piping assembly stress detection method based on displacement feature data according to claim 1, characterized in that, Step 7 specifically involves using the displacement feature data from the assembly stress samples as the input parameter X of the neural network model. 1k The assembly stress field data in the assembly stress sample is used as the target output parameter Y of the neural network model. 1m The neural network model is trained iteratively, with the number of iterations equal to the number of samples, Sum. The neural network model is trained using an iterative method, where the number of iterations equals the number of samples, Sum. The input is fed into a neural network model, and through calculation, a set of output parameters is obtained. With the target output parameters in this sample By comparison, the parameters of the hidden layers of the neural network model are analyzed using the minimum mean error method. Update and generate Complete one training iteration of the neural network model; in the next training iteration, the next sample will be used... The input is fed into the neural network model, while also using the data obtained from the previous calculation. Generate calculation output parameters With the target output parameters in the next sample By making a comparison, the minimum mean error method is used to... Update the sample; where s represents the number of iterations, s = 1, 2, 3, ..., Sum; repeat the above training method until every sample has been used.

Images