A gearbox fault identification and prediction method based on machine learning

Abstract

The present invention provides a gearbox fault identification and prediction method based on machine learning, which relates to the field of gear fault detection. The method includes: obtaining the specification data of the gear and the working condition data under normal operation, performing finite element simulation on the gear, and using a camera module to track and shoot the stress concentration part of the gear tooth root, and obtaining n-frame images of n time nodes of the gear tooth root stress concentration part; importing the boundary calculation strategy to determine the contour boundary of the gear tooth root stress concentration part in the n-frame image; importing the flatness calculation strategy to calculate the flatness of the stress concentration part in the n-frame image; synchronously mixing the flatness with the time node to obtain the flatness time series characteristics of the gear tooth root stress concentration part, inputting the machine learning model to predict the flatness of the gear tooth root stress concentration part at the future time node, and warning the time node where the flatness is greater than the flatness threshold. Improve the work efficiency of managers and help solve system failures.

Description

Technical Field

[0001] The present invention relates to the field of gear fault detection, and in particular to a gearbox fault identification and prediction method based on machine learning. Background Art

[0002] Gearboxes are widely used in mechanical transmission systems, especially in wind power generation, ships, vehicles and other fields. However, long-term operation of gearboxes can lead to gear wear, cracks, broken teeth and other faults, especially the root part of the tooth fault has a greater impact on the reliability of the equipment. Traditional gear fault detection methods include vibration analysis, acoustic emission monitoring and oil analysis, but these methods have certain limitations in real-time, accuracy and robustness. In recent years, with the development of machine learning technology, machine learning-based fault prediction methods have gradually become an important means of gearbox fault diagnosis. Traditional methods such as vibration analysis usually rely on a lot of experience and require professionals to extract and analyze features, resulting in low detection efficiency and susceptibility to noise interference. Although acoustic emission and oil monitoring can provide some early warning information, they often cannot accurately identify the specific location of the fault (such as the root part).

[0003] Tooth root failure usually manifests as crack extension at the stress concentration site. Traditional methods have difficulty distinguishing between stress changes under normal working conditions and characteristic signals at the early stage of failure. Machine learning can capture the signals generated by tiny cracks in the tooth root from a large amount of sensor data by building a complex prediction model and effectively identify them. However, as far as the current existing technology is concerned, it is difficult to label early gear fault data, the sample size is small and difficult to obtain, and traditional processing methods often rely on manual selection of features and require professional knowledge. This method cannot fully adapt to complex working conditions and multiple types of faults. Summary of the invention

[0004] Purpose of the invention: To propose a gearbox fault identification and prediction method based on machine learning to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect of the present invention, a method for identifying and predicting gearbox faults based on machine learning is proposed, comprising the following steps:

[0006] S1. Obtain the specification data of the gears in the target gearbox and the working condition data under normal operation, perform finite element simulation on the gears, and use a camera module to track and shoot the stress concentration part of the gear tooth root, and obtain n frames of images of the stress concentration part of the gear tooth root at n time nodes;

[0007] S2, extracting n frames of images at n time nodes, and determining the contour boundary of the gear tooth root stress concentration location in the n frames of images based on the boundary calculation strategy;

[0008] S3, based on the contour boundary of the gear tooth root stress concentration location in step S2, and based on the flatness calculation strategy, calculate the flatness of the gear tooth root stress concentration location in n frames of images;

[0009] S4. Synchronously mix the flatness of the stress concentration area at the root of the gear tooth in the n-frame image with the time node to obtain the time series characteristics of the flatness of the stress concentration area at the root of the gear tooth, input it into the machine learning model to predict the flatness of the stress concentration area at the root of the gear tooth at future time nodes, and issue an early warning for the time node where the flatness is greater than the flatness threshold.

[0010] A further improvement of the present invention is that step S1 includes the following specific steps:

[0011] S11, obtaining specification data of the gear and working condition data under normal operation; the specification data includes geometric dimensions, density, Young's modulus and Poisson's ratio of the gear;

[0012] S12, taking the camera module as the origin, constructing a finite element simulation model of the gear linkage according to the specification data and the working condition data under normal operation, the material constitutive model is an elastic constitutive model, the mesh division unit type is C3D8R, and the pre-processing of the finite element simulation model is completed;

[0013] S13. Set the analysis type to dynamic implicit, the solver to transient dynamic analysis, the contact type tangential behavior to friction contact and set the friction coefficient, the normal behavior to hard contact, the hard contact surface in the gear linkage as the active surface, the soft contact surface as the driven surface, the contact state to positive contact, the constraint to coupling, set the connector kinematic pair, and simultaneously constrain the displacement and load boundary conditions to complete the solution of the finite element simulation model;

[0014] S14. Output the stress cloud map after solving the finite element model and obtain the stress concentration location of the gear tooth root.

[0015] A further improvement of the present invention is that step S1 also includes: the camera module tracks and shoots the stress concentration part of the gear tooth root after the finite element model is solved according to the rotation speed of the gear, and obtains n frames of images of n time nodes of the stress concentration part of the gear tooth root, wherein the shooting time interval of the n time nodes is consistent with the solution time step in the finite element simulation model.

[0016] A further improvement of the present invention is that the boundary calculation strategy in step S2 includes the following specific steps:

[0017] S21, extracting the i-th frame image and importing it into grayscale processing software, performing grayscale processing on the image to obtain a grayscale image of the i-th frame image;

[0018] S22, dividing the grayscale image into pixels, obtaining stress values ​​of all pixels of the i-th frame image in the finite element model solution, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel below, obtaining the vertical gradient value of the stress value of the pixel, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel on the left, obtaining the horizontal gradient value of the stress value of the pixel;

[0019] S23. Compare the vertical gradient value and the horizontal gradient value of the stress value with the stress gradient threshold value respectively, and extract the pixel points whose vertical gradient value and the horizontal gradient value of the stress value are greater than the stress gradient threshold value, and use the connecting line of these pixel points as the contour boundary of the stress concentration part of the gear tooth root in the image, and repeat the above calculation steps until n frames of images are traversed.

[0020] A further improvement of the present invention is that step S3 includes the following specific steps:

[0021] S31, extracting n frames of images at n time nodes, and at the same time extracting the contour boundary of the gear tooth root stress concentration location in the i-th frame image, and obtaining the number of pixel points within the contour boundary The coordinates of the pixel points within the contour boundary (x, y) and the coordinates of the pixel points within the contour boundary (x, y) correspond to the stress values ​​f(x, y) solved in the finite element simulation model;

[0022] S32, calculating the contour curvature value C(i) of the i-th frame image at the i-th time node, wherein the calculation formula of the contour curvature value C(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed.

[0023] A further improvement of the present invention is that step S3 further comprises:

[0024] S33, extracting n frames of images at n time nodes, and at the same time extracting the gray value of the pixel coordinates (x, y) within the contour boundary of the gear tooth root stress concentration part in the i-th frame image , and extract the mean gray value within the contour boundary of the i-th frame image ;

[0025] S34, calculating the texture difference value W(i) within the contour boundary of the i-th frame image at the i-th time node, wherein the calculation formula of the texture difference value W(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed.

[0026] A further improvement of the present invention is that the calculation formula of the flatness in step S3 is: , where F(i) is the flatness of the stress concentration area at the gear tooth root in the i-th frame image at the i-th time node, a1 and a2 are weight factors, respectively, where a1>a2.

[0027] A further improvement of the present invention is that step S4 includes the following specific steps:

[0028] S41, combining n frames of images and time node n into a feature vector, using a set of n groups of feature vectors as input to a machine learning model, wherein the machine learning model uses the flatness predicted by each group of feature vectors as output, and uses the actual flatness corresponding to each group of feature vectors as a prediction target, and trains the machine learning model until the sum of the prediction accuracies reaches convergence, and stops training;

[0029] S42. Preset a flatness threshold. When the difference between the flatness predicted by the machine learning model and the flatness threshold is less than the difference threshold c, an early warning is triggered. The setting method of the flatness threshold is the minimum flatness value when the gears of the same specification reach the early warning state in the history.

[0030] In a second aspect, a computer-readable storage medium is proposed, on which a computer program is stored. When the computer program is executed, a gearbox fault identification and prediction method based on machine learning disclosed in the above embodiment is implemented.

[0031] In a third aspect, an electronic device is proposed, comprising a memory for storing instructions; and a processor for executing the instructions, so that the electronic device executes a gearbox fault identification and prediction method based on machine learning disclosed in the above embodiment.

[0032] Beneficial effects: The present invention constructs a gearbox fault identification and prediction method based on machine learning, which combines the gear tooth root fault detection process with image processing technology to make the detection results more intuitive. The flatness change of the stress concentration part is used as the judgment standard, the amount of data monitoring is small, the pertinence is strong, and the method covers the common fault types of the gear tooth root. Through the prediction of the machine learning model, the gap between the current gear state and the fault state is evaluated, which improves the work efficiency of manual work and helps to further solve the system fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flow chart of a gearbox fault identification and prediction method based on machine learning in an embodiment of the present invention.

[0034] Figure 2 It is a flow topology diagram between various data in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described.

[0036] Combination Figure 1 This embodiment discloses a gearbox fault identification and prediction method based on machine learning, and its internal data flow is shown in Figure 2 The gearbox fault identification and prediction method based on machine learning includes the following specific steps:

[0037] S1. Obtain the specification data of the gears in the target gearbox and the working condition data under normal operation, perform finite element simulation on the gears, and use a camera module to track and shoot the stress concentration part of the gear tooth root, and obtain n frames of images of the stress concentration part of the gear tooth root at n time nodes;

[0038] S2, extracting n frames of images at n time nodes, and determining the contour boundary of the gear tooth root stress concentration location in the n frames of images based on the boundary calculation strategy;

[0039] S3, based on the contour boundary of the gear tooth root stress concentration location in step S2, and based on the flatness calculation strategy, calculate the flatness of the gear tooth root stress concentration location in n frames of images;

[0040] S4. Synchronously mix the flatness of the stress concentration area at the root of the gear tooth in the n-frame image with the time node to obtain the time series characteristics of the flatness of the stress concentration area at the root of the gear tooth, input it into the machine learning model to predict the flatness of the stress concentration area at the root of the gear tooth at future time nodes, and issue an early warning for the time node where the flatness is greater than the flatness threshold.

[0041] In this embodiment, step S1 includes the following specific steps:

[0042] S11. Obtaining specification data of the gear and working condition data under normal operation; the specification data includes geometric dimensions, density, Young's modulus and Poisson's ratio of the gear;

[0043] S12, taking the camera module as the origin, constructing a finite element simulation model of the gear linkage according to the specification data and the working condition data under normal operation, the material constitutive model is an elastic constitutive model, the mesh division unit type is C3D8R, and the pre-processing of the finite element simulation model is completed;

[0044] S13. Set the analysis type to dynamic implicit, the solver to transient dynamic analysis, the contact type tangential behavior to friction contact and set the friction coefficient, the normal behavior to hard contact, the hard contact surface in the gear linkage as the active surface, the soft contact surface as the driven surface, the contact state to positive contact, the constraint to coupling, set the connector kinematic pair, and simultaneously constrain the displacement and load boundary conditions to complete the solution of the finite element simulation model;

[0045] S14. Output the stress cloud map after solving the finite element model and obtain the stress concentration location of the gear tooth root.

[0046] In this embodiment, step S1 also includes: the camera module tracks and shoots the stress concentration part of the gear tooth root after the finite element model is solved according to the rotation speed of the gear, and obtains n frames of images of n time nodes of the stress concentration part of the gear tooth root, wherein the shooting time interval of the n time nodes is consistent with the solution time step in the finite element simulation model.

[0047] In this embodiment, the boundary calculation strategy in step S2 includes the following specific steps:

[0048] S21, extracting the i-th frame image and importing it into grayscale processing software, performing grayscale processing on the image to obtain a grayscale image of the i-th frame image;

[0049] S22, dividing the grayscale image into pixels, obtaining stress values ​​of all pixels of the i-th frame image in the finite element model solution, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel below, obtaining the vertical gradient value of the stress value of the pixel, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel on the left, obtaining the horizontal gradient value of the stress value of the pixel;

[0050] S23. Compare the vertical gradient value and the horizontal gradient value of the stress value with the stress gradient threshold value respectively, and extract the pixel points whose vertical gradient value and the horizontal gradient value of the stress value are greater than the stress gradient threshold value, and use the connecting line of these pixel points as the contour boundary of the stress concentration part of the gear tooth root in the image, and repeat the above calculation steps until n frames of images are traversed.

[0051] In this embodiment, step S3 includes the following specific steps:

[0052] S31, extracting n frames of images at n time nodes, and at the same time extracting the contour boundary of the gear tooth root stress concentration location in the i-th frame image, and obtaining the number of pixel points within the contour boundary The coordinates of the pixel points within the contour boundary (x, y) and the coordinates of the pixel points within the contour boundary (x, y) correspond to the stress values ​​f(x, y) solved in the finite element simulation model;

[0053] S32, calculating the contour curvature value C(i) of the i-th frame image at the i-th time node, wherein the calculation formula of the contour curvature value C(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed.

[0054] In this embodiment, step S3 further includes:

[0055] S33, extracting n frames of images at n time nodes, and at the same time extracting the gray value of the pixel coordinates (x, y) within the contour boundary of the gear tooth root stress concentration part in the i-th frame image , and extract the mean gray value within the contour boundary of the i-th frame image ;

[0056] S34, calculating the texture difference value W(i) within the contour boundary of the i-th frame image at the i-th time node, wherein the calculation formula of the texture difference value W(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed.

[0057] In this embodiment, the calculation formula of flatness in step S3 is: , where F(i) is the flatness of the stress concentration area at the gear tooth root in the i-th frame image at the i-th time node, a1 and a2 are weight factors, respectively, where a1>a2.

[0058] In this embodiment, step S4 includes the following specific steps:

[0059] S41, combining n frames of images and time node n into a feature vector, taking a set of n groups of feature vectors as input of a machine learning model, the machine learning model taking the flatness predicted by each group of feature vectors as output, taking the actual flatness corresponding to each group of feature vectors as a prediction target, training the machine learning model until the sum of the prediction accuracies reaches convergence, and stopping the training;

[0060] S42. Preset a flatness threshold. When the difference between the flatness predicted by the machine learning model and the flatness threshold is less than the difference threshold c, an early warning is triggered. The flatness threshold is set according to the minimum flatness value when the gears of the same specification reach the early warning state in the past.

[0061] In this embodiment, the machine learning model can be any one of a convolutional neural network, a random forest, and a deep neural network. The above machine learning models are all open source models.

[0062] The gearbox fault identification and prediction method based on machine learning disclosed in the above embodiment can be written into a set of executable instructions in the actual application process, and the executable instructions are written into a storage medium and run on an electronic device. More specific examples of the computer-readable storage medium mentioned in this embodiment may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier, which carries a readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which can send, propagate or transmit a program for use by or in combination with an instruction execution system, device or device. When the above electronic device is running, the complete gearbox fault identification and prediction method based on machine learning disclosed in any of the above embodiments can be automatically executed, and the specific method is not repeated here.

[0063] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A gearbox fault identification and prediction method based on machine learning, characterized in that: The steps include: S1. Obtain the specification data of the gears in the target gearbox and the working condition data under normal operation, perform finite element simulation on the gears, and use a camera module to track and shoot the stress concentration part of the gear tooth root, and obtain n frames of images of the stress concentration part of the gear tooth root at n time nodes; S2, extracting n frames of images at n time nodes, and determining the contour boundary of the gear tooth root stress concentration location in the n frames of images based on the boundary calculation strategy; S3, based on the contour boundary of the gear tooth root stress concentration location in step S2, and based on the flatness calculation strategy, calculate the flatness of the gear tooth root stress concentration location in n frames of images; S31, extracting n frames of images at n time nodes, and at the same time extracting the contour boundary of the gear tooth root stress concentration location in the i-th frame image, and obtaining the number of pixel points N within the contour boundary M, the coordinates of the pixel point within the contour boundary (x, y), the coordinates of the pixel point within the contour boundary (x, y) correspond to the stress value f(x, y) solved in the finite element simulation model; S32, calculating the contour curvature value C(i) of the i-th frame image at the i-th time node, wherein the calculation formula of the contour curvature value C(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed; S33, extracting n frames of images at n time nodes, and at the same time extracting the gray value of the pixel coordinates (x, y) within the contour boundary of the gear tooth root stress concentration part in the i-th frame image , and extract the mean gray value within the contour boundary of the i-th frame image ; S34, calculating the texture difference value W(i) within the contour boundary of the i-th frame image at the i-th time node, wherein the calculation formula of the texture difference value W(i) is: ; Repeat the above calculation until n frames of images at n time nodes are traversed; The calculation formula for flatness is: F(i)=a1 C(i)+a2 W(i), where F(i) is the flatness of the stress concentration area at the gear tooth root in the i-th frame image at the i-th time node, and a1 and a2 are weight factors, where a1>a2; S4. Synchronously mix the flatness of the stress concentration area at the root of the gear tooth in the n-frame image with the time node to obtain the time series characteristics of the flatness of the stress concentration area at the root of the gear tooth, input it into the machine learning model to predict the flatness of the stress concentration area at the root of the gear tooth at future time nodes, and issue an early warning for the time node where the flatness is greater than the flatness threshold.

2. The method for identifying and predicting gearbox faults based on machine learning according to claim 1, characterized in that: Step S1 includes the following specific steps: S11, obtaining specification data of the gear and working condition data under normal operation; the specification data includes geometric dimensions, density, Young's modulus and Poisson's ratio of the gear; S12, taking the camera module as the origin, constructing a finite element simulation model of the gear linkage according to the specification data and the working condition data under normal operation, the material constitutive model is an elastic constitutive model, the mesh division unit type is C3D8R, and the pre-processing of the finite element simulation model is completed; S13. Set the analysis type to dynamic implicit, the solver to transient dynamic analysis, the contact type tangential behavior to friction contact and set the friction coefficient, the normal behavior to hard contact, the hard contact surface in the gear linkage as the active surface, the soft contact surface as the driven surface, the contact state to positive contact, the constraint to coupling, set the connector kinematic pair, and simultaneously constrain the displacement and load boundary conditions to complete the solution of the finite element simulation model; S14. Output the stress cloud map after solving the finite element model and obtain the stress concentration location of the gear tooth root.

3. The method for identifying and predicting gearbox faults based on machine learning according to claim 2, characterized in that: Step S1 also includes: the camera module tracks and shoots the stress concentration part of the gear tooth root after the finite element model is solved according to the rotation speed of the gear, and obtains n frames of images of n time nodes of the stress concentration part of the gear tooth root, wherein the shooting time interval of the n time nodes is consistent with the solution time step in the finite element simulation model.

4. The method for identifying and predicting gearbox faults based on machine learning according to claim 3 is characterized in that: The boundary calculation strategy in step S2 includes the following specific steps: S21, extracting the i-th frame image and importing it into grayscale processing software, performing grayscale processing on the image to obtain a grayscale image of the i-th frame image; S22, dividing the grayscale image into pixels, obtaining stress values ​​of all pixels of the i-th frame image in the finite element model solution, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel below, obtaining the vertical gradient value of the stress value of the pixel, calculating the difference between the stress value corresponding to each pixel and the stress value corresponding to the adjacent pixel on the left, obtaining the horizontal gradient value of the stress value of the pixel; S23. Compare the vertical gradient value and the horizontal gradient value of the stress value with the stress gradient threshold value respectively, and extract the pixel points whose vertical gradient value and the horizontal gradient value of the stress value are greater than the stress gradient threshold value, and use the connecting line of these pixel points as the contour boundary of the stress concentration part of the gear tooth root in the image, and repeat the above calculation steps until n frames of images are traversed.

5. The method for identifying and predicting gearbox faults based on machine learning according to claim 1, characterized in that: Step S4 includes the following specific steps: S41, combining n frames of images and time node n into a feature vector, using a set of n groups of feature vectors as input to a machine learning model, wherein the machine learning model uses the flatness predicted by each group of feature vectors as output, and uses the actual flatness corresponding to each group of feature vectors as a prediction target, and trains the machine learning model until the sum of the prediction accuracies reaches convergence, and stops training; S42. Preset a flatness threshold. When the difference between the flatness predicted by the machine learning model and the flatness threshold is less than the difference threshold c, an early warning is triggered. The setting method of the flatness threshold is the minimum flatness value when the gears of the same specification reach the early warning state in the history.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a gearbox fault identification and prediction method based on machine learning as described in any one of claims 1 to 5 is implemented.

7. An electronic device, characterized in that: It includes a memory for storing instructions; a processor for executing the instructions, so that the device implements a gearbox fault identification and prediction method based on machine learning as described in any one of claims 1 to 5.

Images