Engine external pipeline assembly stress prediction method based on machine learning

Abstract

An engine external pipeline assembly stress prediction method based on machine learning comprises the following steps that a data source is obtained, and the data source comprises influence factor data, a strain value of each channel and an external pipeline assembly stress value; performing weighted calculation on the data source to obtain a joint gap comprehensive influence coefficient x1 and other influence coefficients x3; constructing a training model, carrying out polynomial conversion on the influence factor data and the weighted data, and then carrying out training data set fitting to obtain a fitted training model; and on the basis of the training model, giving corresponding data according to a training sample, and obtaining a predicted assembly stress value. The method effectively solves the problem that the result is difficult to judge due to many influence factors of the assembly stress of the conduit, low correlation of variables and no definite mathematical model by adopting a data driving mode, and the prediction method has universality, is simple in operation process and can realize rapid prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction of assembly stress of external pipelines of aero-engines, and particularly relates to a method for predicting assembly stress of external pipelines of an engine based on machine learning. Background Art

[0002] As the main structure for transporting media of an aero-engine, the external pipeline of the aero-engine plays a crucial role in flight safety. During the design process of the external pipeline of the engine, structural compensation of the engine needs to be considered to avoid the problem of large-stress assembly caused by the assembly of the pipeline. The large-stress assembly of the pipeline will not only cause oil leakage after the pipeline is installed, but may also cause the pipeline to break during operation. At present, the commonly adopted measures in the industry are to stipulate the assembly clearance, coaxiality of pipe joints and clamps, and the range that the outer sleeve nut can be manually tightened during the assembly process, etc. However, the same scale is generally adopted for the above control factors at present, and parameters such as the specifications, structures of the pipelines, and the number and structures of the clamps are not considered. This method is bound to cause problems such as difficult assembly, low efficiency or over-strict control of some pipelines.

[0003] At the present stage, in addition to the above control measures for the assembly stress of the external pipeline of the engine, some pipelines will directly measure by pasting strain gauges. Using strain gauges for measurement not only has a large workload, is time-consuming and laborious, but also due to a series of factors such as different pipe joint clearances, centering degrees, different centering and clearances of the clamps, and changes in the springback amount during the placement of the pipeline during each assembly process, the stress difference after each pipeline assembly is large. To obtain the stress data of each assembly, re-measurement is required, which is not conducive to improving the engine assembly efficiency. However, if no measurement is carried out, it is very difficult to ensure that the assembly stress meets the requirements. At the same time, in the prior art, a method using a data acquisition system is adopted to achieve automatic measurement.

[0004] For example, the patent document with the publication number CN118310658A specifically discloses an automatic measurement device and method for the installation stress of the external pipeline of an aero-engine, connecting the strain gauges installed on the pipeline specimen with a strain bridge data acquisition module, and operating in the strain data acquisition software according to the physical parameters of each measuring point strain gauge and the type of installed bridge circuit; at the man-machine interaction interface operation terminal, loading the installation deviation of the pipeline, obtaining the actual loading amount of the installation deviation through the motor rotation position of the man-machine interaction operation terminal, and controlling each servo motor through the installation deviation loading shaft switching button of the man-machine interaction operation terminal to realize the loading of length deviation and lateral deviation of different pipeline specimens; and automatic loading of the installation deviation and measurement of the installation stress under different combinations of clamps and pipeline specimens; after sorting and analyzing the pipeline installation stress test data obtained by the test, forming a test report on automatic loading of the pipeline installation deviation and measurement of the installation stress. However, it uses a large number of devices, the measurement process is complex, there are many influencing factors, the reliability of the measurement process is low, and the universality is low.

[0005] Therefore, there is an urgent need to design a method with few influencing factors, simple operation, high universality, and capable of quickly predicting the assembly stress of the external pipeline of an aeroengine. Summary of the Invention

[0006] To solve the above technical problems, the present application provides a method for predicting the assembly stress of the external pipeline of an engine based on machine learning, including the following steps:

[0007] Obtain data sources; the data sources include: influencing factor data, strain values of each channel, and external pipeline assembly stress values. The influencing factor data includes the conduit diameter x0, the radial clearance m of the joint, the axial clearance n of the joint, the maximum clearance x2 at the clamp, the number of bends x of the conduit, the number of deep-root clamps y, and the number of double-bracket clamps z;

[0008] Perform weighted calculation on the data sources; among them, the comprehensive influence coefficient x1 of the joint clearance is obtained by weighted calculation according to the radial clearance m and the axial clearance n of the joint; other influence coefficients x3 are obtained by weighting according to the number of bends x of the conduit, the number of deep-root clamps y, and the number of double-bracket clamps z;

[0009] Construct a training model; the construction process includes: based on the conduit diameter x0, the comprehensive influence coefficient x1 of the joint clearance, the maximum clearance x2 at the clamp, and other influence coefficients x3, set the assembly stress value as the target value corresponding to x0, x1, x2, and x3, and perform polynomial transformation and then fit the training data set to obtain the fitted training model;

[0010] Based on the training model, given corresponding data according to the training samples, obtain the predicted assembly stress value.

[0011] Furthermore, the influencing factor data and the strain value of each channel are measured by the data system by pasting strain gauges on the conduits to be measured, and are obtained through the software corresponding to the data acquisition system on the computer device.

[0012] Furthermore, the method for obtaining the external pipeline assembly stress value is as follows:

[0013] Establish a benchmark in the software and adjust the torque between the outer sleeve nut and the clamp on the conduit joint to the specified torque;

[0014] After the acquisition data displayed in the software is stable, read the strain value of the strain gauge;

[0015] Then, calculate the external pipeline assembly stress value through the calculation formula of strain and stress.

[0016] Further, the weighted calculation of the radial clearance m and axial clearance n of the joint is as follows: the weight value of the radial clearance m of the joint is 0.6, and the weight value of the axial clearance n of the joint is 0.4. The comprehensive influence coefficient x1 of the joint clearance = 0.6m + 0.4n.

[0017] Further, the weighted calculation of the number x of conduit bends, the number y of deep root clamps, and the number z of double clamps is as follows: the weight value of the number x of conduit bends is 0.4, the weight value of the number y of deep root clamps is 0.3, and the weight value of the number z of double clamps is 0.3. The other influence coefficient x3 = (0.4x + 0.3y + 0.3z) / max(x, y, z).

[0018] Further, the polynomial is converted to:

[0019] Taking the conduit diameter x0, the comprehensive influence coefficient x1 of the joint clearance, the maximum clearance x2 at the clamp, and the other influence coefficient x3 as the input values of the training database, and taking the collected strain value as the output value of the training data;

[0020] Using the python programming language to call the sklearn machine learning library, the polynomial is converted into a feature X with a power of 2 i .

[0021] Further, the feature X i has terms of:

[0022] [-x0, x1, x2, x3, x0 2 , x0x1, x0x2, x0x3, x1 2 , x1x2, x1x3, x2 2 , x2x3, x3 2 .

[0023] Further, by using linear regression to predict the feature X i , then constructing the prediction function Yi, and finally using the least squares method to obtain the regression coefficient w1 and the intercept parameter w0, the training model can be obtained. The prediction function Yi is essentially the training model.

[0024] Further, the prediction function Y i = w0 + w1X i , where i is the i-th sample, w0 is called the intercept parameter, and w1 is called the regression coefficient.

[0025] Further, the method for predicting the assembly stress of the engine external pipeline based on machine learning further includes verification after obtaining the predicted assembly stress value. The verification is to use the collected measured value as the verification set to verify the predicted value of the regression model, and the comparison error between the predicted value and the measured value is less than 5%.

[0026] The beneficial effects of the present invention are as follows: By pasting strain gauges on the conduit, the strain gauges are connected to the corresponding test software on the computer device through a data acquisition system to obtain measurement data, perform weighted calculation processing on the data source, construct a training model, and use the measurement data to train the model in a machine learning manner. Finally, the prediction of the assembly stress of the conduit is realized. The method effectively solves the problem that it is difficult to determine the results due to many influencing factors of the conduit assembly stress, low correlation between variables, and no definite mathematical model. Moreover, the prediction method of the present invention has universality, and the operation process is simple, and rapid prediction can be achieved. Brief Description of the Drawings

[0027] Figure 1 is a flowchart of a method for predicting the assembly stress of the external pipeline of an engine based on machine learning provided by the present invention;

[0028] Figure 2 is a schematic diagram of obtaining the assembly stress data of the external pipeline of an engine based on machine learning provided by the present invention. Detailed Embodiments

[0029] The technical solution of the present invention will be further described below, but the scope of protection is not limited thereto.

[0030] An embodiment of the present invention provides a method for predicting the assembly stress of the external pipeline of an engine based on machine learning. First, different conduit specifications, bends, and support conditions are recorded. At the same time, the conduit joints and clamp clearances corresponding to the assembly stress of each conduit are measured. Then, the conduits are classified, and appropriate weights are selected for dimensionality reduction processing. Finally, the experimental data is used to train the model in a machine learning manner, and finally the prediction of the assembly stress of the conduit is realized. By adopting this data-driven method, the problem that it is difficult to determine the results due to many influencing factors of the conduit assembly stress, low correlation between variables, and no definite mathematical model can be effectively solved. Moreover, this method has universality in predicting the assembly stress of the external pipelines of various models of engines and can achieve rapid prediction.

[0031] The specific prediction method includes the following steps:

[0032] Obtain the data source;

[0033] The data source includes: influencing factor data, strain values of each channel, and external pipeline assembly stress values. The influencing factor data includes conduit diameter x0, radial clearance m of the joint, axial clearance n of the joint, maximum clearance x2 at the clamp, number of conduit bends x, number of deep root clamps y, and number of double clamps z;

[0034] The influencing factor data and the strain values of each channel are measured by a data acquisition system through pasting strain gauges on the catheter to be measured, and are obtained through software corresponding to the data acquisition system on a computer device. Before pasting the strain gauges, the external pipeline is sandblasted and polished to facilitate pasting the strain gauges and ensure the accuracy of the measurement results. The computer device is connected to the data acquisition system through a data transmission line, and the data acquisition system is connected to the strain gauges through data connection lines.

[0035] The radial clearance m of the joint, the axial clearance n of the joint, and the maximum clearance x2 at the clamp are physical quantities related to the assembly stress of the catheter, and are measured and obtained during the assembly process. The catheter diameter x0, the number of catheter bends x, the number of deep root clamps y, and the number of double clamps z are inherent properties of the catheter and have nothing to do with the assembly, and are obtained by checking the design drawings, so as to establish sample data.

[0036] The method for obtaining the assembly stress value of the external pipeline is as follows:

[0037] Establish a benchmark in the software and adjust the torque between the outer sleeve nut and the clamp on the catheter joint to a specified torque;

[0038] After the acquisition data displayed in the software is stable, read the strain value of the strain gauge;

[0039] Then, through the calculation formula of strain and stress, calculate the assembly stress value of the external pipeline.

[0040] Before measurement, first loosen the outer sleeve nut on the catheter joint to a state where it can be easily rotated. After the benchmark is established, tighten the outer sleeve nut on the catheter joint and the nut on the clamp.

[0041] Perform weighted calculation on the data source, weight the same type of variables and use them as a single variable to reduce influencing factors and improve the efficiency of data training;

[0042] Among them, the comprehensive influence coefficient x1 of the joint clearance is obtained by weighted calculation according to the radial clearance m of the joint and the axial clearance n of the joint; other influence coefficients x3 are obtained by weighting according to the number of catheter bends x, the number of deep root clamps y, and the number of double clamps z;

[0043] The weighted calculation of the radial clearance m of the joint and the axial clearance n of the joint is as follows: the weight value of the radial clearance m of the joint is 0.6, and the weight value of the axial clearance n of the joint is 0.4. The calculation formula of the comprehensive influence coefficient x1 of the joint clearance is:

[0044] x1 = 0.6m + 0.4n (1)

[0045] The weighted calculation of the number of catheter bends \(x\), the number of deep root clamps \(y\), and the number of double - link clamps \(z\) is as follows: the weight of the number of catheter bends \(x\) is \(0.4\), the weight of the number of deep root clamps \(y\) is \(0.3\), and the weight of the number of double - link clamps \(z\) is \(0.3\). The calculation formula for the other influence coefficient \(x3\) is:

[0046] x3=(0.4x + 0.3y + 0.3z) / max(x,y,z) (2)

[0047] Build a training model;

[0048] The building process includes: based on the catheter diameter \(x0\), the comprehensive influence coefficient \(x1\) of the joint clearance, the maximum clearance \(x2\) at the clamp, and the other influence coefficient \(x3\), set the assembly stress value as the target value corresponding to \(x0\), \(x1\), \(x2\), and \(x3\). After performing polynomial transformation, fit the training data set to obtain the trained model after fitting;

[0049] The polynomial transformation is:

[0050] Take the catheter diameter \(x0\), the comprehensive influence coefficient \(x1\) of the joint clearance, the maximum clearance \(x2\) at the clamp, and the other influence coefficient \(x3\) as the input values of the training database, and take the collected strain value as the output value of the training data;

[0051] Use the python programming language to call the sklearn machine learning library to transform the polynomial into a feature \(X\) with a power of 2 i .

[0052] The feature \(X\) i has terms of:

[0053] [-x0,x1,x2,x3,x0 2 ,x0x1,x0x2,x0x3,x1 2 ,x1x2,x1x3,x2 2 ,x2x3,x3 2 (3)

[0054] By predicting using linear regression for the feature \(X\) i , then construct the prediction function \(Yi\) for the \(i\) - th sample,

[0055] The calculation formula for the prediction function \(Yi\) is:

[0056] Y i =w0 + w1X i (4)

[0057] where \(i\) is the \(i\) - th sample, \(w0\) is called the intercept parameter, and \(w1\) is called the regression coefficient.

[0058] It is represented in matrix form as:

[0059]

[0060] Finally, the least squares method is used to obtain the regression coefficients w1 and the intercept parameter w0, and then the training model can be obtained.

[0061] The task of linear regression is to construct a prediction function to map the linear relationship between the input feature matrix X and the label value Y. The prediction function Yi is essentially the training model, and the core of constructing the prediction function Yi is to find the parameter vector of the model.

[0062] Based on the training model, corresponding data is given according to the training samples to obtain the predicted assembly stress value.

[0063] In this embodiment, the method for predicting the assembly stress of the engine external pipeline based on machine learning further includes verification after obtaining the predicted assembly stress value. The verification is to use the collected measured values as a verification set to verify the predicted values of the regression model, and the error between the predicted values and the measured values is less than 5%.

[0064] The specific embodiments of the present invention disclosed above are only for illustration. However, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for predicting the assembly stress of the external pipeline of an engine based on machine learning, characterized in that, It includes the following steps: Obtain the data source; The data source includes: influencing factor data, the strain value of each channel, and the external pipeline assembly stress value. The influencing factor data includes the conduit diameter x0, the radial clearance m of the joint, the axial clearance n of the joint, the maximum clearance x2 at the clamp, the number of conduit bends x, the number of deep root clamps y, and the number of double clamps z; Perform weighted calculation on the data source; among them, the comprehensive influence coefficient x1 of the joint clearance is obtained by weighted calculation according to the radial clearance m and the axial clearance n of the joint; other influence coefficients x3 are obtained by weighting according to the number of conduit bends x, the number of deep root clamps y, and the number of double clamps z; Construct a training model; the construction process includes: based on the conduit diameter x0, the comprehensive influence coefficient x1 of the joint clearance, the maximum clearance x2 at the clamp, and other influence coefficients x3, set the assembly stress value as the target value corresponding to x0, x1, x2, and x3, and perform training dataset fitting after polynomial transformation to obtain the trained model; Based on the training model, given corresponding data according to the training samples, obtain the predicted assembly stress value.

2. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 1, wherein The influencing factor data and the strain value of each channel are measured by the data system by pasting strain gauges on the conduit to be measured, and are obtained through the software corresponding to the data acquisition system on the computer device.

3. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 2, wherein The method for obtaining the external pipeline assembly stress value is as follows: Establish a benchmark in the software and adjust the torque between the outer sleeve nut and the clamp on the conduit joint to the specified torque; After the acquisition data displayed in the software is stable, read the strain value of the strain gauge; Then calculate the external pipeline assembly stress value through the calculation formula of strain and stress.

4. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 3, characterized in that, The weighted calculation of the radial clearance m and the axial clearance n of the joint is: the weight of the radial clearance m of the joint is 0.6, the weight of the axial clearance n of the joint is 0.4, and the comprehensive influence coefficient x1 of the joint clearance = 0.6m + 0.4n.

5. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 4, wherein The weighted calculation of the number of conduit bends x, the number of deep root clamps y, and the number of double clamps z is: the weight of the number of conduit bends x is 0.4, the weight of the number of deep root clamps y is 0.3, and the weight of the number of double clamps z is 0.

3. The other influence coefficient x3 = (0.4x + 0.3y + 0.3z) / max(x, y, z).

6. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 5, wherein The polynomial transformation is: Take the conduit diameter x0, the comprehensive influence coefficient x1 of the joint clearance, the maximum clearance x2 at the clamp, and other influence coefficients x3 as the input values of the training database, and take the collected strain value as the output value of the training data; Use the Python programming language to call the scikit-learn machine learning library to convert the polynomial into a feature X with a second power i .

7. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 6, wherein The said feature X i The items of: [-x0,x1,x2,x3,x0 2 ,x0x1,x0x2,x0x3,x1 2 ,x1x2,x1x3,x2 2 ,x2x3,x3 2 。 8. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 7, wherein By taking the feature X i perform prediction using linear regression, and then construct the prediction function Y i , and finally obtain the regression coefficient w1 and the intercept parameter w0 by using the least squares method, then the training model can be obtained, and the prediction function Yi is essentially the training model.

9. The method for predicting the assembly stress of the external pipeline of the engine based on machine learning according to claim 8, characterized in that, The prediction function Y i = w0 + w1X i , where i is the i-th sample, w0 is called the intercept parameter, and w1 is called the regression coefficient.

10. The method for predicting the assembly stress of the engine external pipeline based on machine learning according to claim 9, wherein The method for predicting the external pipeline assembly stress of the engine based on machine learning further includes verification after obtaining the predicted assembly stress value. The verification is to use the collected measured value as the verification set to verify the predicted value of the regression model, and the error between the predicted value and the measured value is less than 5%.

Images