A method for extracting vibration characteristics of turnout rail head damage based on VMD-KF

The VMD-KF method addresses the challenge of characterizing non-stationary turnout rail head damage signals by enhancing and extracting vibration features, ensuring structural integrity through improved signal processing.

CN117851812BActive Publication Date: 2024-12-06BENGBU ZHIYUN INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410037211.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-10
Publication Date
2024-12-06
Estimated Expiration
2044-01-10

AI Technical Summary

Technical Problem

Traditional signal processing and analysis methods struggle to characterize the non-stationary and weak impact vibration signals of turnout rail head damage, posing a threat to the structural integrity of wheelsets and turnouts.

Method used

A method utilizing Variable Mode Decomposition (VMD) and Kalman Filter (KF) to extract vibration features from turnout rail head damage, involving signal selection, VMD decomposition, signal enhancement, and feature extraction using Hilbert-Huang Transform (HTT).

Benefits of technology

Effectively monitors and analyzes turnout rail head damage, ensuring the integrity of wheelsets and turnout structures by enhancing weak signals and extracting relevant features from non-stationary vibration data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117851812B_ABST
    Figure CN117851812B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of nondestructive testing technology, and in particular to a method for extracting vibration features of a damaged rail head of a turnout based on VMD-KF, comprising: using the impact at the joint of the rail head and the turnout to preliminarily select a signal segment with multiple impacts for analysis; based on VMD, obtain the stationary intrinsic mode function IMF component of the impact vibration, extract the wheelset information from the low-order IMF component to determine the train speed, and estimate the distribution range of the damage signal; based on KF, extract the vibration signal caused by the damage from the high-order IMF component containing the impact vibration; finally, extract the features of the damaged vibration signal based on the Hilbert-Huang transform HTT. The present invention is used to monitor and analyze the vibration characteristics of the turnout section, which is of great significance to ensuring the integrity of the wheelset and turnout structure; it solves the problem that conventional methods are difficult to extract and analyze when processing non-stationary vibration signals.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of nondestructive testing, and in particular to a method for extracting vibration characteristics of turnout rail head damage based on VMD-KF. Background Art

[0002] Turnouts are an important part of railway infrastructure. They are connecting devices that enable trains to change tracks, and they are also key infrastructure that affects the smooth and safe operation of trains. The braking and starting behaviors of trains when passing through turnouts and the complex wheel-rail contact relationship in the turnout section will lead to large contact stresses. When the contact stress exceeds the yield strength limit of the turnout material, fatigue cracks or peeling of the rail head will occur. When the train is running at high speed, the contact state between the wheelset and the rail head damage changes from rolling friction to non-steady-state contact, resulting in an impact effect. The various indicators of impact vibration can be several times that of conventional rolling friction, accompanied by a large turnout impact acceleration, posing a serious threat to the structural integrity of the turnout rail head and wheelset. In addition, the impact vibration signal characteristics of the wheelset and rail head damage in the turnout section have the problems of weak signals and non-stationary characteristics, which makes it difficult for traditional signal processing and analysis methods to characterize the damage characteristics. Therefore, this application proposes a vibration feature extraction method for turnout rail head damage to monitor and analyze turnout damage and impact vibration characteristics, which is of great significance to ensuring the structural integrity of wheelsets and turnouts. Summary of the invention

[0003] The purpose of the present invention is to solve the shortcomings of the prior art and propose a VMD-KF-based turnout rail head damage vibration feature extraction method to monitor and analyze the turnout interval vibration characteristics, which is of great significance to ensure the integrity of wheelsets and turnout structures.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for extracting vibration features of turnout rail head damage based on VMD-KF, the specific steps are as follows:

[0006] Step S1, preliminarily selecting the original signal: preliminarily selecting the signal for analysis by means of the impact at the joint of the rail head and the turnout;

[0007] Step S2, VMD decomposition: obtaining the accurate speed of the train and the components containing the damage vibration from each order IMF decomposition based on VMD;

[0008] Step S3, signal enhancement: extracting the damage vibration signal from the high-order IMF components containing the impact vibration based on KF;

[0009] Step S4, feature extraction: extracting features of the damage vibration signal based on HTT.

[0010] Preferably, in step S1, it specifically includes:

[0011] Step S101, obtaining the original vibration signal of the train passing through the turnout by vertically pasting the acceleration sensor response to the bottom of the turnout rail;

[0012] Step S102: Considering that a large impact will be generated when a train passes through the joint between the rail head and the turnout, a signal segment with multiple impacts is preliminarily selected from the original signal for analysis.

[0013] Preferably, in step S2, it specifically includes:

[0014] Step S201, performing VMD decomposition on the selected signal, setting the initial decomposition layer number, penalty factor α and bandwidth parameter τ;

[0015] Step S202, solving the center frequencies of each order IMF decomposed by VMD;

[0016] Step S203, obtaining the single-sided spectrum of the mode through Hilbert-Huang transform:

[0017]

[0018] Among them, δ(t) is the Dirichlet function, * is the convolution operation, and u k are the decomposed modal components;

[0019] Step S204: By adding exponential Adjust the spectrum of the modal components to baseband:

[0020]

[0021] Among them, δ(t) is the Dirichlet function, u k are the decomposed modal components;

[0022] Step S205: solve the formula in step S204 to obtain the square of the gradient of the frequency-shifted signal L 2 The norm obtains the bandwidth of each mode and establishes the optimal variational model according to the constraints:

[0023]

[0024] Where K is the number of modal components, w k is the decomposed modal component u k The center frequency of , δ(t) is the Dirichlet function;

[0025] Step S206: introduce the Lagrange multiplication operator and the quadratic penalty factor to solve the model of step S205, wherein, in order to ensure the accuracy of signal reconstruction under the Gaussian noise background, α needs to be a sufficiently large positive number, and the expression of the Lagrange multiplication operator is:

[0026]

[0027] Among them, u k are the decomposed modal components, λ(t), f(t) are the input signals, and w k is the decomposed modal component u k The center frequency of

[0028] Step S207: Based on the alternating direction multiplier algorithm, iterate and update {u k (t)}, {w k} and λ(t) to solve the extended Lagrangian expression "saddle point" until the iterative stop condition is met; at this point, the variational model obtains the optimal solution, and the input signal f(t) is decomposed into K finite bandwidth IMF components;

[0029] Step S208: extract and analyze the peak features of the impact signal with a larger amplitude from the low-order IMF components, extract the wheelset information to determine the train speed, and further narrow the distribution range of the damage vibration signal.

[0030] Preferably, in step S3, it specifically includes:

[0031] Step S301, using the KF method to filter the IMF4 component containing noise and vibration high frequency and enhance the weak signal, a pure damage vibration signal can be obtained;

[0032] Step S302: Assume that y(n) is a component containing noise and impact vibration IMF, s(n) is a pure impact vibration signal, and d(n) is a noise signal. The relationship between the three can be expressed as:

[0033] y(n)=s(n)+d(n),0≤n≤N-1

[0034] Where n is the number of data points and N is the frame length;

[0035] Step S303: Perform Fourier transform on both sides of step S302 to obtain:

[0036] Y(ω)=S(ω)+D(ω)

[0037] Step S304: Assuming that s(n) and d(n) are independent signals, then S(ω) and D(ω) are also independent signals, and it can be deduced that:

[0038] E||S(ω)|2 |=E||Y(ω)| 2 |-E||D(ω)| 2 |

[0039] Among them, Y(ω), S(ω) and D(ω) are the short-time power spectra of y(n), s(n) and d(n) respectively;

[0040] Step S305: For the short-term stationary process of the IMF component in a certain frame, there are:

[0041] |S(ω)| 2 =|Y(ω)| 2 -λ d (ω)

[0042] Among them, λ d (ω) is |D(ω)| 2 Statistical average in silent segments;

[0043] Step S306, therefore, the amplitude of the impact vibration signal after the KF method can be expressed as:

[0044]

[0045] Preferably, in step S4, it specifically includes:

[0046] Step S401: Perform Hilbert transform on each IMF component obtained by KF decomposition:

[0047]

[0048] In the formula, Re is a constant, i is the summation value, j is an imaginary unit, and φ i (t) is the phase, α i (t) is the amplitude, ω i (t) is the frequency; φ i (t), α i (t) and ω i (t) are all functions of time;

[0049] In step S402, the Hilbert amplitude spectrum of x(t) in formula S401 is:

[0050]

[0051] Step S403, the boundary spectrum of x(t) is:

[0052] h(ω)=∫ ω H(ω,t)

[0053] Step S404, instantaneous energy density is:

[0054] IE(t)=∫ ω H(ω,t)2dω

[0055] Step S405: Utilize The time-frequency characteristics of the signal can be obtained.

[0056] Preferably, in step S101, the direction of the acceleration signal collected by the acceleration sensor is perpendicular to the pasting position.

[0057] Preferably, in step S208, the wheelset information is the center distance between the wheels.

[0058] By adopting the above technical scheme: the present invention discloses a method for extracting vibration features of turnout rail head damage based on variable mode decomposition (VMD)-Kalman filter (KF), and the steps of the present invention are mainly as follows: with the help of the impact at the joint of the rail head and the turnout, a signal segment with multiple impacts is preliminarily selected for analysis; based on VMD, the stationary intrinsic mode function (IMF) component of the impact vibration is obtained, and the wheelset information is extracted from the low-order IMF component to determine the train speed and estimate the distribution range of the damage signal; based on KF, the vibration signal caused by the damage is extracted from the high-order IMF component containing the impact vibration; finally, the features of the damage vibration signal are extracted based on Hilbert-Huang transform (HTT). Aiming at the problem that the non-stationary characteristics of the wheelset and rail head damage impact vibration signals in the turnout section are difficult to characterize the damage, the present invention proposes a method for extracting turnout impact vibration signals based on VMD-KF, which solves the problem that conventional methods are difficult to extract and analyze when processing non-stationary vibration signals.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The invention is used to monitor and analyze the vibration characteristics of the turnout section, which is of great significance to ensuring the integrity of the wheelset and turnout structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a signal feature extraction flow chart of the present invention;

[0062] Figure 2 It is the original signal diagram of the turnout vibration in the present invention;

[0063] Figure 3 It is a 4-layer VMD decomposition diagram of the original vibration signal in the present invention;

[0064] Figure 4 is the Kalman filter diagram in the present invention;

[0065] Figure 5 It is the impact vibration signal diagram of the damage in the present invention;

[0066] Figure 6 This is the Hilbert-Huang transform result diagram of the impact vibration signal in the present invention. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the protection scope of the present invention. The embodiments described in the present invention are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0068] like Figure 1 As shown, a method for extracting vibration features of turnout rail head damage based on VMD-KF includes the following steps:

[0069] The test conditions of the embodiment are as follows:

[0070] Take the vibration signal of a train passing through a turnout section at a low speed of about 18 km / h as an example, and there is a turnout rail head damage at a distance of 0.3 m from the joint between the turnout and the ordinary rail.

[0071] Step S1, preliminarily selecting the original signal: preliminarily selecting the signal for analysis by means of the impact at the joint of the rail head and the turnout;

[0072] Step S2, VMD decomposition: obtaining the accurate speed of the train and the components containing the damage vibration from each order IMF decomposed by VMD;

[0073] Step S3, signal enhancement: extracting the damage vibration signal from the IMF component containing the impact vibration based on KF;

[0074] Step S4, feature extraction: extracting features of the damage vibration signal based on HTT.

[0075] Specifically, step S1 specifically includes:

[0076] Step S101, obtaining the original vibration signal of the train passing through the turnout by vertically pasting the acceleration sensor response to the bottom of the turnout rail;

[0077] Step S102, considering that a large shock will be generated when the train passes through the joint of the rail head and the turnout, a signal segment with multiple shocks is initially selected from the original signal for analysis. Specifically, since the number of wheels on each bogie is fixed, it can be used to preliminarily determine the time when the wheelset passes the joint. Then, the signal near the time of the large vibration signal is intercepted, and the intercepted original signal is as follows: Figure 2 As shown. Figure 2 It can be found that the initially selected data contains multiple pulse signals, which represent multiple impact phenomena between the wheelset and the turnout joints, damage, etc.

[0078] Specifically, step S2 specifically includes:

[0079] Step S201, performing VMD decomposition on the selected signal, setting the initial decomposition layer number to 2, the penalty factor α and the bandwidth parameter τ to the default parameters of 200 and 0;

[0080] Step S202, solving the center frequency of each order IMF decomposed by VMD;

[0081] Step S203, obtaining the single-sided spectrum of the mode through Hilbert-Huang transform:

[0082]

[0083] Among them, δ(t) is the Dirichlet function, * is the convolution operation, and u k are the decomposed modal components;

[0084] Step S204, by adding exponential Adjust the spectrum of the modal components to baseband:

[0085]

[0086] Among them, δ(t) is the Dirichlet function, u k are the decomposed modal components;

[0087] Step S205, by solving the formula step S204, the square of the gradient of the frequency-shifted signal L 2 The norm obtains the bandwidth of each mode and establishes the optimal variational model according to the constraints:

[0088]

[0089] Where K is the number of modal components, w k is the decomposed modal component u k The center frequency of , δ(t) is the Dirichlet function;

[0090] Step S206, introduce the Lagrange multiplication operator and the quadratic penalty factor to solve the model of step S205. In order to ensure the accuracy of signal reconstruction under Gaussian noise background, α needs to be a sufficiently large positive number. The expression of the Lagrange multiplication operator is:

[0091]

[0092] Among them, u k are the decomposed modal components, λ(t), f(t) are the input signals, and w k is the decomposed modal component u k The center frequency of

[0093] Step S207, based on the alternating direction multiplier algorithm, iterate and update {u k (t)}, {w k} and λ(t) to solve the extended Lagrangian expression "saddle point" until the iterative stop condition is met. At this point, the variational model obtains the optimal solution, and the input signal f(t) is decomposed into K finite bandwidth IMF components;

[0094] Step S208, extract and analyze the peak features of the impact signal with a large amplitude from the low-order IMF components, extract the wheel set information to determine the train speed, and further narrow the distribution range of the damage vibration signal. Specifically, in this embodiment, the decomposition level is finally determined to be 4, and the decomposition result is as follows: Figure 3 As shown. The wheel information can be found from the low-frequency IMF1 layer, and the true wave speed is determined to be 18km / h. From the high-frequency IMF4 layer, it can be found that there is a small pulse before the wheel information, which may be a damage signal, but the amplitude of the damage signal is low and the signal-to-noise ratio is low.

[0095] Specifically, step S3 specifically includes:

[0096] Step S301, using the KF method to filter the IMF4 component containing noise and vibration high frequency and enhance the weak signal, a pure damage vibration signal can be obtained;

[0097] In step S302, assuming that y(n) is a component containing noise and impact vibration IMF, s(n) is a pure impact vibration signal, and d(n) is a noise signal, the relationship between the three can be expressed as:

[0098] y(n)=s(n)+d(n),0≤n≤N-1

[0099] Where n is the number of data points and N is the frame length;

[0100] Step S303, perform Fourier transform on both sides of step S302, and obtain:

[0101] Y(ω)=S(ω)+D(ω)

[0102] In step S304, assuming that s(n) and d(n) are independent signals, then S(ω) and D(ω) are also independent signals, and it can be deduced that:

[0103] E||S(ω)| 2 |=E||Y(ω)| 2 |-E||D(ω)| 2 |

[0104] Among them, Y(ω), S(ω) and D(ω) are the short-time power spectra of y(n), s(n) and d(n) respectively;

[0105] Step S305, for the short-term stationary process of the IMF component in a certain frame, there are:

[0106] |S(ω)| 2 =|Y(ω)| 2 -λ d (ω)

[0107] Among them, λ d (ω) is |D(ω)| 2 Statistical average in silent segments;

[0108] Step S306, therefore, the amplitude of the impact vibration signal after the KF method can be expressed as:

[0109]

[0110] In this embodiment, the Kalman filter diagram is as follows: Figure 4 As shown in Figure 1, the pulse signal caused by damage can be significantly observed. By extracting the impact vibration signal of the damage, the signal is as follows Figure 5 shown.

[0111] Specifically, step S4 specifically includes:

[0112] Step S401, perform Hilbert transform on each IMF component obtained by KF decomposition:

[0113]

[0114] In the formula, Re represents a constant, i represents the summation value, j is an imaginary unit, φ i (t) is the phase, α i (t) is the amplitude, ω i (t) is the frequency; α i (t), ai(t) and ω i (t) are all functions of time;

[0115] Step S402, the Hilbert amplitude spectrum of x(t) in formula S401 is:

[0116]

[0117] Step S403, the boundary spectrum of x(t) is:

[0118] h(ω)=∫ ω H(ω,t)

[0119] Step S404, the instantaneous energy density is:

[0120] IE(t)=∫ ω H(ω,t)2dω

[0121] Step S405, using The time-frequency characteristics of the signal can be obtained.

[0122] In this embodiment, the Hilbert-Huang transform result of the damaged impact vibration signal is as follows: Figure 6 As shown in the figure, it can be found that the wheel has multiple impacts on the damage, and the frequency of the impact is mostly distributed in 3500-5000Hz. By extracting and analyzing each impact signal, it is of great significance for the status monitoring of the turnout.

[0123] The description and practice disclosed in the present invention are easy to think and understand for ordinary technicians in the technical field, and several improvements and modifications can be made without departing from the principles of the present invention. Therefore, modifications or improvements made without departing from the spirit of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for extracting vibration characteristics of turnout rail head damage based on VMD-KF, characterized in that: The specific steps are as follows: Step S1, preliminarily selecting the original signal: preliminarily selecting the signal for analysis by means of the impact at the joint of the rail head and the turnout; Step S2, VMD decomposition: obtaining the accurate speed of the train and the components containing the damage vibration from each order IMF decomposition based on VMD; Step S3, signal enhancement: extracting the damage vibration signal from the high-order IMF components containing the impact vibration based on KF; Step S4, feature extraction: extracting features of the damage vibration signal based on HTT; In step S2, it specifically includes: Step S201, performing VMD decomposition on the selected signal, setting the initial decomposition layer number, penalty factor α and bandwidth parameter τ; Step S202, solving the center frequencies of each order IMF decomposed by VMD; Step S203, obtaining the single-sided spectrum of the mode through Hilbert-Huang transform: Among them, δ(t) is the Dirichlet function, * is the convolution operation, and u k are the decomposed modal components; Step S204: By adding exponential Adjust the spectrum of the modal components to baseband: Among them, δ(t) is the Dirichlet function, u k are the decomposed modal components; Step S205: solve the formula in step S204 to obtain the square of the gradient of the frequency-shifted signal L 2 The norm obtains the bandwidth of each mode and establishes the optimal variational model according to the constraints: Where K is the number of modal components, w k is the decomposed modal component u k The center frequency of , δ(t) is the Dirichlet function; Step S206: Introduce the Lagrange multiplication operator and the quadratic penalty factor to solve the model of step S205. In order to ensure the accuracy of signal reconstruction under the Gaussian noise background, α needs to be a sufficiently large positive number. The expression of the Lagrange multiplication operator is: Among them, u k are the decomposed modal components, λ(t), f(t) are the input signals, and w k is the decomposed modal component u k The center frequency of Step S207: Based on the alternating direction multiplier algorithm, iteratively update {u k (t)}, {w k } and λ(t) to solve the extended Lagrangian expression "saddle point" until the iterative stop condition is met; at this point, the variational model obtains the optimal solution, and the input signal f(t) is decomposed into K finite bandwidth IMF components; Step S208, extracting and analyzing the peak features of the impact signal with a large amplitude from the low-order IMF components, extracting the wheelset information to determine the train speed, and further narrowing the distribution range of the damage vibration signal; In step S3, it specifically includes: Step S301, using the KF method to filter the IMF4 component containing noise and vibration high frequency and enhance the weak signal, a pure damage vibration signal can be obtained; Step S302: Assume that y(n) is a component containing noise and impact vibration IMF, s(n) is a pure impact vibration signal, and d(n) is a noise signal. The relationship between the three can be expressed as: y(n)=s(n)+d(n),0≤n≤N-1 Where n is the number of data points and N is the frame length; Step S303: Perform Fourier transform on both sides of step S302 to obtain: Y(ω)=S(ω)+D(ω) Step S304: Assuming that s(n) and d(n) are independent signals, then S(ω) and D(ω) are also independent signals, and it can be deduced that: E||S(ω)| 2 |=E||Y(ω)| 2 |-E||D(ω)| 2 | Among them, Y(ω), S(ω) and D(ω) are the short-time power spectra of y(n), s(n) and d(n) respectively; Step S305: For the short-term stationary process of the IMF component in a certain frame, there are: |S(ω)| 2 =|Y(ω)| 2 -l d (oh) Among them, λ d (ω) is |D(ω)| 2 Statistical average in silent segments; Step S306, therefore, the amplitude of the impact vibration signal after the KF method can be expressed as: In step S4, it specifically includes: Step S401: Perform Hilbert transform on each IMF component obtained by KF decomposition: In the formula, Re is a constant, i is the summation value, j is an imaginary unit, and φ i (t) is the phase, α i (t) is the amplitude, ω i (t) is the frequency; φ i (t), α i (t) and ω i (t) are all functions of time; In step S402, the Hilbert amplitude spectrum of x(t) in formula S401 is: Step S403, the boundary spectrum of x(t) is: Step S404, instantaneous energy density is: IE(t)=∫ ω H(ω,t)2dω Step S405: Utilize The time-frequency characteristics of the signal can be obtained.

2. The method for extracting vibration characteristics of turnout rail head damage based on VMD-KF according to claim 1 is characterized in that: In step S1, it specifically includes: Step S101, obtaining the original vibration signal of the train passing through the turnout by vertically pasting the acceleration sensor response to the bottom of the turnout rail; Step S102: Considering that a large impact will be generated when a train passes through the joint between the rail head and the turnout, a signal segment with multiple impacts is preliminarily selected from the original signal for analysis.

3. The method for extracting vibration characteristics of turnout rail head damage based on VMD-KF according to claim 2 is characterized in that: In step S101 , the direction of the acceleration signal collected by the acceleration sensor is perpendicular to the pasting position.

4. The method for extracting vibration characteristics of turnout rail head damage based on VMD-KF according to claim 3 is characterized in that: In step S208, the wheelset information is the center distance between the wheels.