A variable speed wheel pair polygon damage online intelligent diagnosis method, device and medium

Abstract

The application discloses a kind of variable speed wheel pair polygon damage online intelligent diagnosis method, equipment and medium, wherein method includes: based on variable speed working condition rail vehicle wheel polygon vibration signal, according to fault type classification arrangement fault data;Based on mapping conversion from time domain to time-frequency domain, vibration signal is converted into time-frequency domain image;Based on time-frequency domain image drawing rotation angle target frame, mark the image texture reflecting fault characteristics, make fault label file;The time-frequency domain image of fault data and the corresponding fault label file are used as training set;Rotating target recognition neural network is trained based on training set, and wheel polygon fault diagnosis model is obtained;Based on data acquisition equipment, the vibration signal of wheel is obtained and converted into test set image;Wheel polygon fault diagnosis model is used to identify and classify test set image, and wheel polygon fault is diagnosed.The application can be widely applied to variable speed rail vehicle wheel polygon damage online intelligent diagnosis.

Description

Technical Field

[0001] This invention relates to the fields of fault diagnosis and deep learning technology, and in particular to an online diagnostic method, device and medium for polygonal damage to variable speed wheelsets of rail vehicles based on artificial intelligence neural network deep learning technology. Background Technology

[0002] Due to the complex operating environment and harsh stress conditions, rail vehicle wheelsets are prone to sudden damage during operation. Wheel polygonal wear is a major form of wear damage to rail vehicle wheelsets, and the high-frequency vibrations it generates can have a very adverse impact on the safety and stability of train operation. Therefore, conducting online diagnosis of wheel polygonal faults is of great significance for ensuring the safe and stable operation of rail vehicles.

[0003] The current main method for online monitoring of wheelset polygons involves installing accelerometers on the wheel axle boxes to detect vibration signals, calculating evaluation indicators for wheel polygon wear based on instantaneous detection algorithms, or training deep learning networks such as convolutional neural networks and recurrent neural networks to identify vibration signal fault features.

[0004] Existing online diagnostic methods for polygonal damage to wheels have the following problems:

[0005] First, the use of instantaneous index calculation is easily affected by line conditions, which can lead to false alarms. Furthermore, the same polygonal wear pattern on the same wheel can be continuously judged as having a fault, generating a large number of redundant alarm messages.

[0006] Secondly, when the vehicle speed is constant, the fault characteristic frequency is fixed, making it easy to diagnose wheelset polygonal wear. However, the actual operating speed of rail vehicles changes instantaneously, and the wheel polygonal fault characteristic frequency components will evolve into dynamic frequency-changing signals, which will be superimposed with the random vibration noise signals of wheel-rail surface irregularities, posing a significant challenge to the diagnosis of wheelset polygonal faults under variable speed conditions.

[0007] Third, due to the influence of the dynamic characteristics of the vehicle track system, the wheel-rail impact intensity induced by the same wheelset damage varies significantly at different vehicle speeds. Classical statistical regression signal analysis methods are difficult to overcome the influence of time-varying operating conditions on the identification of wheel damage feature information; existing deep learning neural networks mostly use horizontal rectangular boxes as target recognition boxes, which have too much interference background for slanted and narrow targets, making it difficult to learn the target cross-domain invariance and fault diagnosis classification criteria from complex variable speed non-stationary vibration signals. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes an online intelligent diagnostic method, device, and medium for polygonal damage to variable speed wheelsets, which can be widely applied to the diagnosis of polygonal damage to wheels of rail vehicles (high-speed trains, passenger cars, locomotives, freight cars, subways, and other special vehicles).

[0009] The technical solution adopted in this invention is as follows:

[0010] A method for online intelligent diagnosis of polygonal damage to variable speed wheelsets includes the following steps:

[0011] Step 1: Based on the polygonal vibration signals of the rail vehicle wheels under variable speed conditions, classify and organize the fault data according to the fault type;

[0012] Step 2: Based on the time-domain to time-frequency domain mapping transformation, the vibration signal is converted into a time-frequency domain image, thereby transforming the fault diagnosis problem into an image classification problem;

[0013] Step 3: Draw a rotated labeled target bounding box based on the time-frequency domain image, and use a closed polygon with a tilt angle to label the target image texture that reflects the fault characteristics;

[0014] Step 4: Create fault label files and use the time-frequency domain images of the fault data and the corresponding fault label files as the training set;

[0015] Step 5: Train the rotating target recognition neural network based on the training set to obtain the wheel polygon fault diagnosis model;

[0016] Step 6: Acquire vibration signals of the target rail vehicle wheels using data acquisition equipment, and convert them into test set images based on time-domain to time-frequency domain mapping transformation;

[0017] Step 7: Based on the wheel polygon fault diagnosis model, identify and classify the test set images to diagnose wheel polygon faults.

[0018] As a preferred approach, in steps 2 and 6, the mapping transformation method applied to the vibration signal from the time domain to the time-frequency domain includes Cohen-type time-frequency transform:

[0019]

[0020] Among them, C x (t,Ω; φ) represents the time-frequency transform representation, where t represents time, Ω represents frequency, and φ and φ(θ,τ) both represent kernel functions; e -j(θt+Ωτ) Let A represent the basis functions for shift and frequency modulation transformations, where j represents the imaginary unit, θ represents the frequency shift, and τ represents the time shift. x (θ,τ) is a fuzzy function of x(t).

[0021] As a preferred embodiment, in step 3, the parameters of the rotated annotation target box include [x] c ,y c [,l,s,θ], where x cIt is the x-coordinate of the center of the rotated annotation target box, y c y is the ordinate of the center of the rotated annotation target box, l is the width of the long side of the rotated annotation target box, s is the width of the short side of the rotated annotation target box, and θ is the angle between the long side of the rotated annotation target box and the x-axis.

[0022] As a preferred method, step 4, the creation of the fault label file includes: writing the fault target category number and the parameters of the rotated label target box in step 3 into a TXT file as the fault label file.

[0023] As a preferred approach, step 5, the process of training the rotating target recognition neural network, includes: adding the rotation angle parameter of the rotating labeled target box to the data loading part, discretizing the angle, and training the rotating target recognition neural network to classify and recognize the angle of the rotating labeled target box.

[0024] As a preferred method, in step 6, the data acquisition device first acquires the vibration signal of the target wheel axle box, and then converts the vibration signal into a test set image.

[0025] A computer device includes a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the above-described online intelligent diagnosis method for polygonal damage to variable speed wheelsets.

[0026] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described online intelligent diagnosis method for polygonal damage to variable speed wheelsets.

[0027] The beneficial effects of this invention are as follows:

[0028] (1) This invention uses artificial intelligence deep learning neural network for fault diagnosis, which avoids the problem that the traditional instantaneous index calculation method is easily affected by line conditions and reduces the interference of a large number of false alarm information to the driver.

[0029] (2) This invention transforms the problem of extracting fault features of complex variable speed vibration signals in the time domain into an image classification problem, enabling the target recognition deep learning neural network to automatically extract low-dimensional features of the data and realize multi-class information transformation and recognition through high-dimensional nonlinear mapping, resulting in stable and accurate detection results.

[0030] (3) The present invention uses a labeled target box with rotation angle parameters, which solves the problems caused by the use of horizontal target boxes in existing deep learning neural networks, such as difficulty in identifying tilted and narrow targets, difficulty in learning the cross-domain invariance of targets and the basis for fault diagnosis classification from complex variable speed non-stationary vibration signals. It realizes intelligent understanding and pattern recognition of wheel polygon damage signals under variable speed conditions, and can train neural network models more accurately.

[0031] (4) The diagnostic results obtained by the method of the present invention are accurate and the diagnostic method has extremely high stability and speed. In addition, the method of the present invention has excellent anti-noise interference ability and is suitable for situations with severe vibration and impact background noise.

[0032] (5) This invention can be widely applied to the diagnosis of polygonal damage faults in the wheels of rail vehicles (trains, passenger cars, locomotives, freight cars, subways and other special vehicles). Attached Figure Description

[0033] Figure 1 This is a flowchart of the online intelligent diagnosis method for polygonal damage to variable speed wheelsets according to the present invention.

[0034] Figure 2 This is a schematic diagram of the polygonal wear of the wheel represented by polar coordinates in Example 2.

[0035] Figure 3 This is a circumferential unfolded diagram of the polygonal wear of the wheel under acceleration conditions in Example 2.

[0036] Figure 4 This is a mixed vibration signal diagram of the wheel-rail impact background noise and the polygonal wear vibration of the wheel under the acceleration condition in Example 2.

[0037] Figure 5 In Example 2 Figure 3 The diagram shows the Cohen-type time-frequency mapping transformation of the vibration signal.

[0038] Figure 6 In Example 2 Figure 4 The diagram shows the Cohen-type time-frequency mapping transformation of the vibration signal.

[0039] Figure 7 This is a schematic diagram of how the neural network correctly diagnosed the polygonal wear of the wheel under weak noise deceleration conditions in Example 2.

[0040] Figure 8 This is a schematic diagram of how the neural network correctly diagnosed the polygonal wear of the wheel under strong noise acceleration conditions in Example 2. Detailed Implementation

[0041] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments are now described. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0042] Example 1

[0043] This embodiment provides an online intelligent diagnosis method for polygonal damage to variable speed wheelsets, such as... Figure 1 As shown, it includes the following steps:

[0044] Step 1: Based on the polygonal vibration signals of the rail vehicle wheels under variable speed conditions, classify and organize the fault data according to the fault type;

[0045] Step 2: Based on the time-domain to time-frequency domain mapping transformation, the vibration signal is converted into a time-frequency domain image, thereby transforming the fault diagnosis problem into an image classification problem;

[0046] Step 3: Draw a rotated labeled target bounding box based on the time-frequency domain image, and use closed polygons with tilt angles to annotate the target image texture that reflects the fault characteristics;

[0047] Step 4: Create fault label files and use the time-frequency domain images of the fault data and the corresponding fault label files as the training set;

[0048] Step 5: Train the rotating target recognition neural network based on the training set to obtain the wheel polygon fault diagnosis model;

[0049] Step 6: Acquire vibration signals of the target rail vehicle wheels using data acquisition equipment, and convert them into test set images based on time-domain to time-frequency domain mapping transformation;

[0050] Step 7: Based on the wheel polygon fault diagnosis model, identify and classify the test set images to diagnose wheel polygon faults.

[0051] As a preferred approach, in steps 2 and 6, the time-domain to time-frequency domain mapping transformation is a Cohen-type time-frequency transform:

[0052]

[0053] Among them, C x (t,Ω; φ) represents the time-frequency transform, where t represents time, Ω represents frequency, and φ and φ(θ,τ) both represent kernel functions; e -j(θt+Ωτ) Let A represent the basis functions for shift and frequency modulation transformations, where j represents the imaginary unit, θ represents the frequency shift, and τ represents the time shift. x (θ,τ) is a fuzzy function of x(t).

[0054] As a preferred approach, in step 3, the parameters for rotating the labeled target box include [x] c ,y c [,l,s,θ], where x c It is the x-coordinate of the center of the rotated annotation target box, y cy is the ordinate of the center of the rotated annotation target box, l is the width of the long side of the rotated annotation target box, s is the width of the short side of the rotated annotation target box, and θ is the angle between the long side of the rotated annotation target box and the x-axis.

[0055] As a preferred method, step 4, creating the fault label file includes: writing the fault target category number and the parameters of the rotated label target box from step 3 into a TXT file as the fault label file.

[0056] Because the traditional YOLO V5 algorithm fault tag file format is [x c ,y c [,w,h], where x c It is the x-coordinate of the center of the bounding box, y c y is the ordinate of the center of the target bounding box, w is the width of the target bounding box, and h is the height of the target bounding box. Therefore, the traditional YOLO V5 algorithm can only be trained and predicted using horizontal rectangular bounding boxes. For oblique and narrow targets, there is too much background interference, making it difficult to learn the target's cross-domain invariance and fault diagnosis classification criteria from complex variable-speed non-stationary vibration signals.

[0057] As a preferred approach, this embodiment implements a rotating target recognition neural network based on an improved YOLO V5 algorithm: the rotation angle parameter of the rotating labeled target box is added to the data loading part, the angle is discretized, and the rotating target recognition neural network is trained to classify and recognize the angle of the rotating labeled target box.

[0058] As a preferred method, in step 6, the data acquisition device first acquires the vibration signal of the wheel axle box of the target rail vehicle, and then converts the vibration signal into a test set image.

[0059] Example 2

[0060] This embodiment is based on embodiment 1:

[0061] This embodiment provides an online intelligent diagnosis method for polygonal damage in transmission wheelsets, applied to the diagnosis of polygonal faults in 20th-order wheels, such as... Figure 1 As shown, it includes the following steps:

[0062] Step 1: Classify and organize the polygonal vibration signals of rail vehicle wheels under variable speed conditions according to the fault type.

[0063] like Figure 2 In this embodiment, the wear of a 20th-order polygonal wheel is represented by polar coordinates, such as... Figure 3 This is the accelerated operating condition in this embodiment. Figure 2 The diagram shown is a circumferential unfolded view of the polygonal wear of the wheel. Figure 4This is a mixed vibration signal diagram of the wheel-rail impact background noise and the polygonal wear vibration of the wheel under the acceleration condition in this embodiment.

[0064] Step 2: Apply a mapping transformation from the time domain to the time-frequency domain to the vibration signal, converting the vibration signal into a time-frequency domain image.

[0065] As a preferred approach, the time-domain to time-frequency domain mapping transformation applied to the vibration signal is a Cohen-type time-frequency transform:

[0066]

[0067] Among them, C x (t,Ω; φ) represents the time-frequency transform representation, where t represents time, Ω represents frequency, and φ and φ(θ,τ) both represent kernel functions; e -j(θt+Ωτ) Let A represent the basis functions for shift and frequency modulation transformations, where j represents the imaginary unit, θ represents the frequency shift, and τ represents the time shift. x (θ,τ) is a fuzzy function of x(t).

[0068] like Figure 5 This is in this embodiment Figure 3 The Cohen-type time-frequency mapping transformation diagram of the vibration signal is shown below. Figure 6 This is in this embodiment Figure 4 The diagram shows the Cohen-type time-frequency mapping transformation of the vibration signal.

[0069] Step 3: Treat the fault diagnosis problem as an image classification problem. Draw rotated labeled target boxes on the time-frequency domain image, and use closed polygons with tilt angles to label image texture targets that reflect fault characteristics.

[0070] As a preferred approach, the parameters of the rotated annotation target box include [x c ,y c [,l,s,θ], where x c It is the x-coordinate of the center of the bounding box, y c θ is the ordinate of the center of the target box, l is the width of the long side of the target box, s is the width of the short side of the target box, and θ is the angle between the long side of the rotated annotation target box and the x-axis.

[0071] Step 4: Create fault label files. The time-frequency domain images of the fault data and the corresponding fault label files constitute the training set of the rotating target recognition neural network.

[0072] As a preferred method, the fault label file is created by writing the fault target category number and the rotation annotation target box parameters from step 3 into a TXT file.

[0073] Step 5: Train the rotating target recognition neural network using the training set to obtain the wheel polygon fault diagnosis model.

[0074] As a preferred approach, the rotating target recognition neural network employs an improved YOLO V5 algorithm. Specifically, a rotation angle parameter for the target bounding box is added to the data loading section, the angle is discretized, and the rotating target recognition neural network is trained to classify and recognize the angle of the rotated labeled target bounding box.

[0075] Step 6: Convert the vibration signal acquired by the data acquisition device into a test set image through a mapping transformation from the time domain to the time-frequency domain.

[0076] As a preferred method, the data acquisition device acquires the vibration acceleration signal of the wheel axle box.

[0077] As a preferred approach, the time-domain to time-frequency domain mapping transformation applied to the vibration signal is a Cohen-type time-frequency transform:

[0078]

[0079] Among them, C x (t,Ω; φ) represents the time-frequency transform representation, where t represents time, Ω represents frequency, and φ and φ(θ,τ) both represent kernel functions; e -j(θt+Ωτ) Let A represent the basis functions for shift and frequency modulation transformations, where j represents the imaginary unit, θ represents the frequency shift, and τ represents the time shift. x (θ,τ) is a fuzzy function of x(t).

[0080] Step 7: Use the trained wheel polygon fault diagnosis model to classify and identify the test set images. Diagnose and identify wheel polygon faults based on the neural network output. Some results are shown below:

[0081] like Figure 7 As shown, the trained wheel polygon fault diagnosis model correctly diagnosed the wheel polygon wear under deceleration conditions, with a fault label of 0.

[0082] like Figure 8 As shown, the trained wheel polygon fault diagnosis model correctly diagnosed the wheel polygon wear under acceleration conditions, and the fault label was 1.

[0083] Example 3

[0084] This embodiment is based on embodiment 1:

[0085] This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the online intelligent diagnosis method for polygonal damage of variable speed wheelsets in Embodiment 1. The computer program can be in the form of source code, object code, executable file, or some intermediate form.

[0086] Example 4

[0087] This embodiment is based on embodiment 1:

[0088] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the online intelligent diagnosis method for polygonal damage to variable speed wheelsets in Embodiment 1. The computer program can be in the form of source code, object code, executable file, or some intermediate form. The storage medium includes any entity or device capable of carrying computer program code, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content contained in the storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. In some jurisdictions, according to legislation and patent practice, the storage medium does not include electrical carrier signals and telecommunication signals.

[0089] It should be noted that, for the sake of simplicity, the foregoing method embodiments are described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. An online intelligent diagnosis method for polygonal damage to variable speed wheelsets, characterized in that, Includes the following steps: Step 1: Based on the polygonal vibration signals of the rail vehicle wheels under variable speed conditions, classify and organize the fault data according to the fault type; Step 2: Based on the time-domain to time-frequency domain mapping transformation, the vibration signal is converted into a time-frequency domain image, thereby transforming the fault diagnosis problem into an image classification problem; Step 3: Draw a rotated labeled target bounding box based on the time-frequency domain image, and use a closed polygon with a tilt angle to label the target image texture that reflects the fault characteristics; Step 4: Create fault label files and use the time-frequency domain images of the fault data and the corresponding fault label files as the training set; Step 5: Train the rotating target recognition neural network based on the training set to obtain the wheel polygon fault diagnosis model; Step 6: Acquire vibration signals of the target rail vehicle wheels using data acquisition equipment, and convert them into test set images based on time-domain to time-frequency domain mapping transformation; Step 7: Based on the wheel polygon fault diagnosis model, identify and classify the test set images to diagnose wheel polygon faults; In steps 2 and 6, the mapping transformation method applied to the vibration signal from the time domain to the time-frequency domain includes a Cohen-type time-frequency transform, which is as follows: in, This represents the time-frequency transform representation. Indicates time, Indicates frequency, and Both represent kernel functions; Represents the basis functions for shift and frequency modulation transforms. Represents the imaginary unit. Indicates frequency shift, Indicates time shift; yes fuzzy functions; In step 3, the parameters of the rotated annotation target box include ,in, It is the x-coordinate of the center of the rotated annotation target box. It is the ordinate of the center of the rotated annotation target box. It is the width of the longer side of the rotated annotation target box. It is the width of the shorter side of the rotated annotation target box. It is the sum of the long side of the rotated annotation target box. The angle between the axes; In step 5, the process of training the rotating target recognition neural network includes: adding the rotation angle parameter of the rotating labeled target box to the data loading part, discretizing the angle, and training the rotating target recognition neural network to classify and recognize the angle of the rotating labeled target box.

2. The online intelligent diagnosis method for polygonal damage to variable speed wheelsets according to claim 1, characterized in that, In step 4, creating the fault label file includes writing the fault target category number and the parameters of the rotated label target box from step 3 into the fault label file.

3. The online intelligent diagnosis method for polygonal damage to variable speed wheelsets according to claim 1, characterized in that, In step 6, the data acquisition device first acquires the vibration signal of the target wheel axle box, and then converts the vibration signal into a test set image.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the online intelligent diagnosis method for polygonal damage of variable speed wheelsets as described in any one of claims 1-3.

5. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the online intelligent diagnosis method for polygonal damage of variable speed wheelsets as described in any one of claims 1-3.

Images