CN115236208A - Steel rail health monitoring method based on information enhancement and variable step length sparse expression - Google Patents
Steel rail health monitoring method based on information enhancement and variable step length sparse expression Download PDFInfo
- Publication number
- CN115236208A CN115236208A CN202210735400.1A CN202210735400A CN115236208A CN 115236208 A CN115236208 A CN 115236208A CN 202210735400 A CN202210735400 A CN 202210735400A CN 115236208 A CN115236208 A CN 115236208A
- Authority
- CN
- China
- Prior art keywords
- steel rail
- signal
- matrix
- iteration
- damage
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/14—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object using acoustic emission techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4454—Signal recognition, e.g. specific values or portions, signal events, signatures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4463—Signal correction, e.g. distance amplitude correction [DAC], distance gain size [DGS], noise filtering
Abstract
The invention discloses a steel rail health monitoring method based on information enhancement and variable step length sparse expression, which is characterized in that firstly, a damage information enhancement algorithm with a double-layer Hankerl (Hankerl) matrix and energy compensation is improved based on a singular value decomposition algorithm, and unsteady-state signals and damage signals in acquired acoustic emission signals are separated; then, carrying out self-adaptive adjustment on the iteration step length of the SAMP algorithm by using the inflection point of the residual inner product, and improving the reconstruction precision of the real damage signal; and finally, carrying out self-adaptive weighting on the information entropy of the signal in a unit time window through the short-time energy of the signal, realizing accurate evaluation on the structural health state of the steel rail, and dividing the health stages of the three steel rails by utilizing the constructed structural health indexes. The method can accurately monitor the structural health state of the steel rail in real time, and provides data support and theoretical guidance for maintenance and replacement of the steel rail.
Description
Technical Field
The invention belongs to the field of structural health monitoring of steel rails, relates to a steel rail health monitoring method, and particularly relates to a steel rail structural health acoustic emission monitoring method based on damage information enhancement and variable-step-length sparsity adaptive matching pursuit.
Background
As a new nondestructive testing technology, the acoustic emission technology is introduced into the field of SHM of steel rails with the advantages of high sensitivity, simple equipment, passive monitoring and the like. However, the process of steel rail damage from germination to propagation to fracture is a long term process of energy accumulation and release. Under different stress conditions, acoustic emission signals generated by the movement of the intermolecular dislocation have similar waveform characteristics, and the acoustic emission signals containing non-damaged waveforms exist in the whole steel rail health monitoring process. In the non-damaged signal, two kinds of waveforms are included, namely, a steady state waveform and a non-steady state waveform. A large number of non-damaged waveforms can increase the false alarm rate of the system and reduce the detection accuracy.
Meanwhile, scattering and reflection are generated on a conduction path in the conduction process of the damage signal, so that the original waveform is distorted. And a real damage signal is reconstructed from the complex waveform, so that the accurate evaluation of the structural health state of the steel rail is facilitated. Currently, sparse Representation (SR) is receiving more and more attention in the field of non-destructive inspection due to its advantages of high efficiency and accuracy. Compared with the traditional band-pass filtering method, the sparse representation method reconstructs effective information from the complex signal according to different sparsity degrees of all components. However, the conventional SR method requires a priori information of the signal sparsity, which is not available in the actual SHM of the rail. For example, the sparsity adaptive tracking Matching algorithm (SAMP) needs to set an iteration step, and an improper iteration step will worsen the reconstruction accuracy and calculation efficiency of the ghost algorithm, which requires a great deal of manual experience.
In addition, since the real damage acoustic emission signals are not continuously released and contain a large amount of steady-state signals, the evaluation of the structural health state of the steel rail by means of time-domain waveforms or single characteristics is easily interfered, and the evaluation indexes fluctuate greatly.
For SHMs based on acoustic emission technology, three technical difficulties coexist: firstly, how to enhance damage information from complex waveform components to make the damage information more identifiable; secondly, how to adjust the iteration step length of the SAMP algorithm and improve the reconstruction precision of the real damage signal under the condition of ensuring the calculation efficiency of the algorithm; and thirdly, accurately evaluating the structural health state of the steel rail based on the damage signal, and reducing the fluctuation of the evaluation index caused by the discontinuity of the damage signal.
Disclosure of Invention
In order to realize real-time and accurate monitoring of the health state of the steel rail structure, the invention provides a steel rail health monitoring method based on information enhancement and variable step sparse expression. The method can accurately monitor the structural health state of the steel rail in real time, and provides data support and theoretical guidance for maintenance and replacement of the steel rail.
The purpose of the invention is realized by the following technical scheme:
a rail health monitoring method based on information enhancement and variable-step sparse expression is characterized in that firstly, a damage information enhancement algorithm with a double-layer Hankerl (Hankerl) matrix and energy compensation is improved based on a Singular Value Decomposition (SVD) algorithm, and unsteady-state signals and damage signals in collected acoustic emission signals are separated; then, carrying out self-adaptive adjustment on the iteration step length of the SAMP algorithm by using the inflection point of the residual inner product, and improving the reconstruction precision of the real damage signal; and finally, carrying out self-adaptive weighting on the information entropy of the signal in a unit time window through the short-time energy of the signal, realizing accurate evaluation on the structural health state of the steel rail, and dividing the health stages of the three steel rails by utilizing the constructed structural health index. The method comprises the following steps:
the method comprises the following steps: firstly, constructing a two-layer Hankerl matrix based on acoustic emission signals of a steel rail; then, carrying out singular value decomposition on the Hankerl matrix, and setting zero on the first row in the singular value matrix for reconstructing the Hankerl matrix; then, based on the energy compensation coefficient, multiplying the reconstructed Hankerl matrix, and iteratively executing singular value decomposition and energy compensation until the change rate of the singular value ratio is smaller than a threshold value, and stopping iteration; the method comprises the following specific steps:
step one, acoustic emission signal Y to rail o (t)∈R n Taking M sampling points as a segment, and taking Y as a segment o Dividing into N segments with equal length, and constructing a data set Y = [ Y = 1 ,y 2 ,...,y N |y i ∈R M×1 ]A double-layer Hankerl matrix is constructed according to the following structure:
SVD decomposition is performed on H (Y) and a singular value matrix sigma = diag (lambda) 1 ,λ 2 )；
Step two, checking the change rate Delta E of the ratio of the singular values R ＝E R (z-1)/E R (z) is less than threshold δ 1 In which E R ＝λ 1 /λ 2 (ii) a If E R ＜δ 1 Based on the energy compensation coefficient
Performing energy compensation on the reconstructed signal data set to realize damage information enhancement, if E R ≥δ 1 And reconstructing the Hankerl matrix and returning to the step for one-to-one cycle iteration, wherein a specific formula for performing energy compensation on the reconstructed Hankerl matrix is as follows:
wherein U = [ U ] 1 ,u 2 ]Is a left singular matrix, V ∈ R M×N Is a right singular matrix;
step one and three, extraction
Signal segment reconstruction data set in
Step two: first, for the reconstructed lesion data set
Obtaining a matrix M of measured values using a random observation matrix S e Acquiring an initial dictionary D based on a KSVD algorithm; subsequently, M is calculated e The inflection point of the inner product between the inflection point and the sensing matrix gamma = S · D is used for adaptively adjusting the iteration step length of the SAMP algorithm; finally, reconstructing a real damage acoustic emission signal based on the optimal step length; the method comprises the following specific steps:
step two, initializing a measurement value matrix
Wherein S is a random observation matrix; the initial inner product residual is R (1) = M e The method comprises the steps that a sensing matrix gamma = S · D, D is an initial dictionary obtained based on a K-SVD algorithm, the number of initialization iteration times is z =1, and an initial iteration step length L is obtained 1 Initial iteration step size L 1 The specific formula of (A) is as follows:
wherein R is i,j For the jth residual error point in the ith column of residual error vectors, when a plurality of points meeting the conditions exist, only taking the second derivative of the first point as an iteration step length, and when the second derivatives of any residual error inner product points have the same sign, taking 1 as the iteration step length;
step two, when the iteration times z is larger than 1, adaptively adjusting the iteration step length L through an inner product inflection point, wherein the specific formula of the iteration step length L is as follows:
step two and step three, the largest front L in | gamma · R (z-1) | is selected i (z) item sequence number value added to sequence number set B i The serial number in gamma is compared with B i The same item is added to the candidate set C i (z), reconstructing a sensing matrix gamma, wherein the specific formula of the sensing matrix gamma is as follows:
Γ(z)＝C i (z)∪Γ(z-1)；
fourthly, reconstructing a real damaged sparse coefficient matrix through a least square method:
wherein, M e,i Is the ith column in the measurement matrix;
step two and five, selecting
Replacing original sequence number set B by the largest front L (z) item in the sequence i F (z) is reacted with B i The corresponding L (z) term is denoted as Γ' (z), and then the residual is updated, the specific formula of which is as follows:
R i (z)＝M e,i -Γ′(z)(Γ′ T (z)Γ′(z))Γ′ T (z)M e,i ；
wherein M is e,i Is a matrix M e The ith column in (1);
step two, if residual error R (z) is less than or equal to delta 2 Then the reconstruction precision of the real damage signal is shown to meet the system requirement, and the real damage signal is reconstructed according to the system requirement
Reconstructing a real damaged signal set if residual | | | R (z-1) | non-calculation 2 ≤||R(z)|| 2 If yes, returning to the second step for continuous iteration, otherwise, enabling R (z) = R (z-1), and enabling z = z +1, and returning to the second step for continuous iteration;
seventhly, acquiring a real damage signal set Y s Splicing according to columns to obtain reconstructed real damage informationNumber (C)
Step three: firstly, extracting Information Entropy (IE) of a signal based on a reconstructed real impairment signal segment as impairment signal characteristics; then, self-adaptive weighting is carried out on the information entropy by analyzing the change state of the short-time energy of the signal, and a structural health index capable of accurately evaluating the state of the steel rail is constructed; finally, dividing the health state of the steel rail into a safe stage, an extended stage and a non-safe stage based on the index; the method comprises the following specific steps:
step three, taking M sampling points as a section, respectively extracting the information entropy of each section of signals without overlapping, and normalizing the information entropy, wherein the specific formula of the information entropy is as follows:
wherein the content of the first and second substances,
for reconstructed real impairment signal
The number i of the segments in (a),
is composed of
The probability of (a) of (b) being,
is composed of
The probability of the minimum value in the set of values,
is composed of
The probability of the median maximum;
step two, respectively calculating the short-time energy of the real damage signal corresponding to each value in the information entropy, and normalizing the short-time energy, wherein a Hamming window is adopted for calculating the short-time energy, the window length is w, no overlapping exists, and the specific formula of the short-time energy is as follows:
wherein psi (·) is a Hamming window function;
thirdly, self-adaptive weighting is carried out on the information entropy characteristics of the damage signals through short-time energy reconstruction, the structural health index SHI of the steel rail is constructed, meanwhile, the minimum effective weight is added, the change of the health state of the steel rail in the safety stage due to the fact that the short-time energy is too small is prevented, and the structural health index and the minimum effective specific formula are as follows:
wherein ε and η are constants, and ε - η is the minimum effective weight;
step three, dividing the structural health state of the steel rail into three stages based on the structural health index: a secure phase when SHI < 0.1, an extended phase when SHI ∈ [0.1,1), and a non-secure phase when SHI = 1.
Compared with the prior art, the invention has the following advantages:
1. the method utilizes the combination of singular value decomposition and a double-layer Hankerl matrix to decompose the acoustic emission signals of the steel rail, and utilizes the correlation difference among steady-state signals, unstable signals and damage signals in the acoustic emission signals to separate the non-damage components in the acoustic emission signals. Meanwhile, energy compensation is carried out on the reconstructed acoustic emission signals through the constructed energy compensation coefficients. The method improves the amplitude of the signal, enhances the damage information and avoids the problem of difficult subsequent damage detection caused by too low amplitude after reconstruction while ensuring the detection real-time performance. Compared with the traditional filtering method, the method does not depend on the prior information of the signal frequency, and is more suitable for the actual damage monitoring environment.
2. The method is based on the difference of sparsity between the real damage signal and the steady-state signal, the real damage signal is reconstructed by using the improved variable step length SAMP algorithm, and the accuracy of subsequent rail SHM and evaluation is improved. In addition, by utilizing the difference between residual error inner product points in the iterative process of the SAMP algorithm and adaptively adjusting the iterative step length of the algorithm, the optimal sparsity value is searched more quickly and accurately, and the high-precision reconstruction of the real damage acoustic emission signal is realized.
3. The method is based on the reconstructed real damage acoustic emission signal, and self-adaptive weighting is carried out on the information entropy characteristics by using the short-time energy of the signal, so that a structural health index capable of accurately evaluating the health state of the steel rail in real time is constructed. The method can effectively avoid the problem of inaccurate SHM of the steel rail caused by information entropy characteristic fluctuation due to discontinuous damage signals. In addition, by introducing the minimum effective weight, the health index is prevented from being submerged due to too low weight, and the precision of the steel rail SHM is improved.
Drawings
Fig. 1 is a schematic flow chart of a steel rail health monitoring method based on information enhancement and variable step-size sparse expression.
Fig. 2 is a diagram showing the ratio of each signal component in the characteristic value.
Fig. 3 is a diagram of an adjustment process of an iteration step size based on an inner product inflection point.
FIG. 4 is a drawing of the tensile testing apparatus.
Fig. 5 is a diagram of a steady-state signal waveform and a non-steady-state signal waveform.
Fig. 6 is a diagram of steady-state signals and non-steady-state signals after impairment information enhancement.
Fig. 7 is a diagram of real damage acoustic emission signals reconstructed based on the modified variable step-size SAMP algorithm.
FIG. 8 is a graph of kurtosis indicators after different processing methods.
FIG. 9 is an information entropy feature effective value diagram of two steel rail test pieces.
Fig. 10 is a structural health index chart of two steel rail test pieces.
Detailed Description
The technical solution of the present invention is further described below with reference to the accompanying drawings, but the present invention is not limited thereto, and modifications or equivalent substitutions may be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention.
The invention provides a steel rail health monitoring method based on information enhancement and variable step sparse expression, which is used for enhancing damage information and constructing a self-adaptive weighting index to evaluate the structural health state of a steel rail. In the method, singular value decomposition is combined with double-layer Hankerl, so that instable component separation is realized by utilizing the difference of correlation among different components in an acoustic emission signal while the detection real-time performance is ensured, and damage information is enhanced by constructing an energy compensation coefficient. Meanwhile, the step length of the algorithm is adaptively adjusted according to the change of the inner product inflection point in the iterative process of the SAMP algorithm, and a real damage signal is reconstructed more accurately and efficiently according to the difference of sparsity between the damage information and the steady-state information. In addition, a structural health index is constructed based on the information entropy characteristics and the short-time energy weight and is used for evaluating the health state of the steel rail. The structural health index is adaptively weighted by the short-time energy of the segmented calculation signals, so that the phenomenon that the index curve fluctuates too much due to the fact that signals are damaged and discontinuous is avoided, and the false detection rate of a monitoring system is reduced. And the minimum effective weight is introduced, so that the health index under the low weight is not submerged, and the real-time accurate monitoring on the health state of the steel rail structure is realized. As shown in fig. 1, the method comprises the following specific steps:
the method comprises the following steps: acoustic emission signal Y for a section of rail o Taking M sampling points as a segment, and carrying out non-overlapping conversionY o Dividing into a plurality of signal segments and forming an acoustic emission signal set Y = [ Y 1 ,y 2 ,...,y N |y i ∈R M×1 ]. The signal segments with the sequence numbers of 1 to (N-1) and 2 to N are respectively used as two rows of the matrix to form a double-layer Hankerl matrix H (Y). Only one signal segment is different between the upper line and the lower line of the matrix, and the real-time performance of the monitoring system can be guaranteed to the greatest extent. Meanwhile, SVD decomposition is performed on H (Y) to obtain a singular value matrix sigma = diag (λ) 1 ,λ 2 ). Due to the different correlation between the non-stationary signal and the impairment signal, the non-stationary signal component remains at λ 1 Fig. 2 shows the ratio of each signal component to the feature value. Let lambda 1 =0 and reconstructs H (Y) to isolate the non-stationary signal. Continuously iterating the above process until the ratio of singular value to value changes delta E R ＝E R (z-1)/E R (z) whether it is less than a threshold value delta 1 In which E R ＝λ 1 /λ 2 . For enhancing damage information, based on energy compensation coefficient
For the reconstructed acoustic emission signal
Is compensated for. The method comprises the following specific steps:
step one, for the acoustic emission signal Y of the rail o (t) taking M sampling points as a segment, and taking Y as a segment o Dividing into N segments with equal length, and constructing a data set Y = [ Y = 1 ,y 2 ,...,y N |y i ∈R M×1 ]A two-layer Hankerl matrix was constructed according to the following structure:
SVD decomposition is performed on H (Y) and a singular value matrix sigma = diag (lambda) 1 ,λ 2 ). Because there is only one signal segment difference between the upper and lower rows in H (Y)This dual-layer Hankerl matrix guarantees the lowest signal processing delay.
Step two, checking the change rate delta E of the singular value ratio R ＝E R (z-1)/E R (z) is less than threshold δ 1 In which E R ＝λ 1 /λ 2 . If E R ＜δ 1 And then, carrying out energy compensation on the reconstructed signal data set to realize damage information enhancement, wherein the specific formula is as follows:
wherein U = [ U ] 1 ,u 2 ]For the left singular matrix, V ∈ R M×N Is the right singular matrix.
It is ensured that the amplitude of the reconstructed signal can be compensated to the level before SVD decomposition. If E R ≥δ 1 And reconstructing a Hankerl matrix and returning to the step for one-cycle iteration.
Step one and three, extraction
Signal segment reconstruction data set in
Step two: for reconstructing a data set
Obtaining a matrix of measured values M using a random observation matrix S e And acquiring an initial dictionary D based on a KSVD algorithm. Subsequently, M is calculated e Inflection points of inner products between the sensing matrix gamma = S · D and the iteration step size of the SAMP algorithm are used for self-adaptively adjusting, and the adjustment process of the iteration step size based on the inflection points of the inner products is shown in FIG. 3. Selecting the largest front L in gamma R (z-1) | i (z) item sequence number value added to sequence number set B i The serial number in gamma is compared with B i Are identical to each otherInto candidate set C i (z), reconstructing the sensing matrix gamma. Then, reconstructing a sparse coefficient matrix of the real damage signal through minimum two-multiplication, and selecting
Replacing original sequence number set B by middle and maximum front L (z) item i In Γ (z) with B i The corresponding L (z) term is recorded as Γ' (z), the residual error matrix is updated, and finally, if the residual error R (z) is less than or equal to δ 2 Then the reconstruction precision of the real damage signal is shown to meet the system requirement, and the real damage signal is reconstructed by
A true impairment signal set is reconstructed. If residual | | | R (z-1) | non-woven phosphor 2 ≤||R(z)|| 2 The step size is readjusted and the iteration continues. Otherwise, let R (z) = R (z-1), z = z +1, and readjust the step size, continue the iteration. The specific flow diagram of the algorithm is shown in fig. 3, and the specific steps are as follows:
step two, firstly, initializing a measurement value matrix
Wherein S is a random observation matrix; the initial inner product residual is R (1) = M e The sensing matrix gamma = S.D, D is an initial dictionary obtained based on a K-SVD algorithm, the number of initialization iterations is z =1, and an initial iteration step length L is obtained 1 The formula is as follows:
wherein R is i,j Is the jth residual point in the ith column of residual vectors. The formula shows that when the sign of the second derivative of a certain point is opposite to that of the second derivative of the next point in the residual inner product, the iteration step of the ith column of residual vectors is the second derivative of the point. When there are a plurality of points that meet the above condition, only the second derivative of the first point is taken as the iteration step. When the second derivatives of any residual inner product points are all of the same sign, the overlapThe substitution step size is taken to be 1. The operation can effectively avoid the problem that the iteration gradient disappears because the residual vector has no inflection point.
Step two, when the iteration times z is larger than 1, adaptively adjusting the iteration step length L through an inner product inflection point, wherein the formula is as follows:
step two and step three, the largest front L in | gamma · R (z-1) | is selected i (z) item sequence number value added to sequence number set B i The serial number in gamma is compared with B i The same item is added to the candidate set C i (z) reconstructing a sensing matrix gamma, wherein the formula is as follows:
Γ(z)＝C i (z)∪Γ(z-1)。
fourthly, reconstructing a real damaged sparse coefficient matrix through a least square method:
wherein M is e,i Is the ith column in the measurement matrix.
Step two and five, selecting
Replacing original sequence number set B by the largest front L (z) item in the sequence i In Γ (z) with B i The corresponding L (z) term is denoted Γ' (z). The residual is then updated, the formula is as follows:
R i (z)＝M e,i -Γ′(z)(Γ′ T (z)Γ′(z))Γ′ T (z)M e,i 。
step two and six, if residual error R (z) is less than or equal to delta 2 Then the reconstruction precision of the real damage signal is shown to meet the system requirement, and the real damage signal is reconstructed according to the system requirement
A true impairment signal set is reconstructed. If residual | | | R (z-1) | non-woven phosphor 2 ≤||R(z)|| 2 And returning to the second step to continue iteration. Otherwise, let R (z) = R (z-1), z = z +1, return to step two and continue iteration.
Seventhly, acquiring a real damage signal set Y s Splicing according to columns to obtain a reconstructed real damage signal
Step three: reconstruction based true impairment signal
The method comprises the steps of taking M sampling points as the length of a window, extracting information entropy characteristics without overlapping, taking M sampling points as a segment, adopting a Hamming window to calculate the short-time energy level of each segment, and calculating the effective value of the information entropy characteristics and the effective value of the short-time energy, so as to avoid the influence on the structural health index of the steel rail due to characteristic fluctuation. Finally, the effective value pair of the short-time energy
The self-adaptive weighting is carried out on the information entropy characteristics, so that a structural health index is constructed and used for acoustic emission monitoring of the steel rail. Based on the structural health index, the structural health state of the steel rail is divided into three stages: when SHI is less than 0.1, the steel rail is in a safe stage, and the performance of the steel rail is good without replacement. When SHI belongs to 0.1,1), the steel rail is in an expansion stage, and at the moment, early cracks and cracks in an expansion state exist in the steel rail, and the steel rail needs to be replaced within a period of time to prevent accidents. When the SHI =1 is an unsafe stage, the internal structure of the rail is damaged, cracks are completely formed, even the rail is in a fracture state, and the rail needs to be replaced to ensure the operation safety of the train. The method comprises the following specific steps:
step three, taking M sampling points as a section, respectively extracting the information entropy of each section of signals without overlapping, and normalizing the information entropy, wherein the formula is as follows:
wherein
Is the i-th segment in the reconstructed real lesion signal.
And step two, respectively calculating the short-time energy of the real damage signal corresponding to each value in the information entropy, and normalizing the short-time energy. Wherein, the calculation of the short-time energy adopts a Hamming window, the window length is w, no overlap exists, and the formula is as follows:
and thirdly, self-adaptive weighting is carried out on the information entropy characteristics of the damage signals through short-time energy of the reconstructed damage signals, and the structural health index SHI of the steel rail is constructed. Meanwhile, the minimum effective weight is added to prevent the change of the health state of the steel rail in the safety stage due to the short-term over-small energy, and the specific formula is as follows:
in the above formula, epsilon-eta ensures that the structural health index does not cause the problem that the rail cannot be accurately evaluated due to too small weight influence in the safety stage of the rail.
And step three, dividing the structural health state of the steel rail into three stages based on the structural health index. When SHI is less than 0.1, the steel rail is in a safe stage, and the performance of the steel rail is good without replacement. When SHI belongs to 0.1,1), the steel rail is in an expansion stage, and at the moment, early cracks and cracks in an expansion state exist in the steel rail, and the steel rail needs to be replaced within a period of time to prevent accidents. When the SHI =1 is an unsafe stage, the internal structure of the steel rail is damaged, cracks are completely formed, even the steel rail is in a fracture state, and the steel rail needs to be replaced to ensure the running safety of the train.
The following describes the embodiments of the present invention with reference to the rail tensile fracture test data:
a standard steel rail test piece is cut from an in-service steel rail and used for a tensile test, the test obtains a complete acoustic emission signal from the intact state to the broken state of the steel rail by applying tensile force to two ends of the steel rail test piece, and the equipment is shown in figure 4. The tensile test equipment comprises a steel rail test piece, a Zwick Z100 tensile breaking machine and a Vallen AE signal acquisition system. The rail test piece is cut from a complete rail and is made of U75V steel, and the detailed dimensions are shown on the left side of FIG. 4. In the test, a stretching machine is used for circularly and continuously applying tension on two ends of a steel rail test piece until the test piece is broken. During the stretching process, the rail gradually changes from the elastic phase to the plastic phase with the generation of AE signals. The VS900-RIC AE sensor and the Vallen AMSY-6 ASIP-2/A AE acquisition system continuously acquire data from the acoustic emission sensor. Empirically, the maximum frequency of the impairment signal is 2.5MHz. Therefore, the sampling frequency of the sensor is set to 5MHz. The safe and unsafe phases of the rail specimen are divided by the yield point of the material, which is determined by previous tensile testing. The safe phase is defined as the material property not reaching the yield point. Otherwise, an unsecured phase is defined.
Executing the step one: based on the tensile test, a section with the length of 1.204 multiplied by 10 is collected 6 Acoustic emission signal Y of individual sampling points o Taking 2048 sampling points as a segment, and carrying out non-overlapping Y conversion o Dividing into a plurality of signal segments and forming an acoustic emission signal set Y = [ Y 1 ,y 2 ,...,y 500 |y i ∈R 2048×500 ]. The acquired acoustic emission signal includes two types of time domain waveforms, wherein a steady state signal waveform and an unsteady state signal waveform are respectively shown in fig. 5 (a) and 5 (b). The signal fragments with the sequence numbers of 1-499 and 2-500 in Y are respectively used as two rows of the matrix to form a double-layer Hankerl matrix H (Y). Only one signal segment is different between the upper line and the lower line of the matrix, and the real-time performance of the monitoring system can be guaranteed to the greatest extent. At the same time, the user can select the desired position,SVD decomposition is performed on H (Y) to obtain a singular value matrix sigma = diag (lambda) 1 ,λ 2 ). Due to the different correlation between the non-stationary signal and the impairment signal, the non-stationary signal component remains at λ 1 The ratio of each signal component in the feature value is shown in fig. 2. Let lambda 1 =0 and reconstructs H (Y) to isolate the non-stationary signal. Continuously iterating the above process until the change rate Delta E of the singular value ratio R ＝E R (z-1)/E R (z) is less than threshold δ 1 ＝1×10 -3 In which E R ＝λ 1 /λ 2 . For enhancing the damage information, based on energy compensation coefficients
For reconstructed acoustic emission signals
Is compensated for. The acoustic emission signal after the enhancement of the damage information is shown in fig. 6. After energy compensation, the amplitude of the unsteady state signal is enhanced by 19.23% at most, and the amplitude of the steady state signal is enhanced by only 9% at most. The result shows that the algorithm can effectively improve the amplitude of the unsteady-state signal and enhance the damage information in the signal.
Executing the step two: for reconstructing a data set
Using a random observation matrix S ∈ R 120×2048 Obtaining a matrix M of measured values e ∈R 120×500 Setting the dictionary atom number to 3000, and obtaining an initial dictionary D e.R based on a KSVD algorithm 2048×3000 . Then, the real damage acoustic emission signals are reconstructed by utilizing the variable step length SAMP algorithm based on the improvement of the inner product inflection point, wherein delta is 2 ＝1×10 -6 . The steady-state signal and the unsteady-state signal in the reconstructed acoustic emission signal are shown in fig. 7 (c) and 7 (f), respectively. Comparing fig. 7 (a), 7 (b) and 7 (c), the amplitude of the non-stationary signal remains at the maximum after enhancement of the impairment information, indicating that the enhanced impairment information in the signal is preserved. On the contraryIn 7 (d), 7 (e), and 7 (f), the maximum amplitude of the steady-state signal is reduced by 18.18%. Meanwhile, as shown by the circled portion in fig. 7, the damage information contained in the enhanced portion is retained. This result occurs because valid damage information is retained in the reconstructed acoustic emission signal. At the same time, based on the difference in sparsity, the steady-state components are separated from the reconstructed acoustic emission signal. Obviously, the reconstructed unsteady signal amplitude has larger fluctuation than that before processing, which is beneficial to the subsequent monitoring of the rail damage.
An appropriate evaluation index is crucial for evaluating the performance of the reconstructed real impairment signal. Due to the lack of a priori information about the frequency of the signal, its performance cannot be evaluated with commonly used indicators such as signal-to-noise ratio. Fortunately, the Kurtosis, the fourth-order center-to-center distance (Kurtosis), can effectively reflect the impulse component in the signal, and is commonly used for mechanical fault detection. Therefore, the invention introduces the average kurtosis difference as an evaluation index, and the calculation method is as follows:
in general, a higher average kurtosis value indicates that there are more impact components in the reconstructed signal, indicating that the impairment information is more significant in the monitoring process. For comparison, three-layer wavelets, low-pass filtering with 25% cutoff frequency, conventional SVD and improved variable-step SAMP algorithm reconstruction proposed by the present invention are used herein
As shown in fig. 8.
The results in fig. 8 indicate that the proposed method is about 6 times higher than the other methods in the kurtosis index. The method provided by the invention has the advantages that the reconstructed signal contains more impact components, has more obvious damage information and is beneficial to the evaluation of the steel rail.
And (5) executing the third step: based on the acoustic emission signals acquired by the steel rail test pieces, the real damage acoustic emission signals are reconstructed based on the method provided by the invention. And taking w =2048 sampling points as the length of the window, and extracting information entropy characteristics without overlapping. The information entropy characteristic valid values of the unprocessed signal and the reconstructed signal are compared, wherein epsilon =1 and eta =0.1, and the result is shown in fig. 9. In the first 160 sets of signals, the rail test piece is in a safe stage. In fig. 9 (a) and 9 (b), the method proposed by the present invention significantly reduces the effective value of the information entropy, 61.82% and 94.15% respectively, compared to the original signal. Meanwhile, after the method provided by the invention is adopted for processing, the fluctuation of the effective value of the information entropy in the safety stage is small, and the accuracy of the steel rail SHM is favorably improved. In a word, based on the method provided by the invention, the information entropy of the acoustic emission signal in the safety stage is reduced, and the whole acoustic emission signal is more stable, which proves that the method provided by the invention can effectively enhance the damage information and separate the non-damage components.
And then weighting the information entropy characteristics by using the short-time energy of the corresponding segment to construct the structural health index of the steel rail. Finally, based on the structural health indexes of the two steel rail test pieces, the structural health state of the steel rail is divided into a safe stage, an extended stage and an unsafe stage, and the result is shown in fig. 10. As can be seen from fig. 10, the structural health index constructed by the method of the present invention can clearly distinguish three different rail structural health phases.
Claims (10)
1. A steel rail health monitoring method based on information enhancement and variable step sparse expression is characterized by comprising the following steps:
the method comprises the following steps: firstly, constructing a two-layer Hankerl matrix based on acoustic emission signals of a steel rail; then, carrying out singular value decomposition on the Hankerl matrix, and setting zero on the first row in the singular value matrix for reconstructing the Hankerl matrix; then, based on an energy compensation coefficient, multiplying the reconstructed Hankerl matrix, and iteratively executing singular value decomposition and energy compensation until the change rate of a singular value ratio is smaller than a threshold value, and stopping iteration;
step two: first, for the reconstructed lesion data set
Obtaining a matrix of measured values M using a random observation matrix S e Acquiring an initial dictionary D based on a KSVD algorithm; subsequently, M is calculated e The inflection point of the inner product between the inflection point and the sensing matrix gamma = S · D is used for adaptively adjusting the iteration step length of the SAMP algorithm; finally, reconstructing a real damage acoustic emission signal based on the optimal step length;
step three: firstly, extracting the information entropy of a signal based on a reconstructed real damage signal in a segmentation way to be used as the characteristics of the damage signal; then, self-adaptive weighting is carried out on the information entropy by analyzing the change state of the short-time energy of the signal, and a structural health index capable of accurately evaluating the state of the steel rail is constructed; and finally, dividing the health state of the steel rail into a safe stage, an extended stage and an unsafe stage based on the index.
2. The steel rail health monitoring method based on information enhancement and variable step size sparse representation according to claim 1, characterized in that the specific steps of the first step are as follows:
step one, for the acoustic emission signal Y of the rail o (t)∈R n Taking M sampling points as a segment, and taking Y as a segment o Dividing into N segments with equal length, and constructing a data set Y = [ Y = 1 ,y 2 ,...,y N |y i ∈R M×1 ]A two-layer Hankerl matrix was constructed according to the following structure:
SVD decomposition is performed on H (Y) and a singular value matrix sigma = diag (lambda) 1 ,λ 2 )；
Step two, checking the change rate delta E of the singular value ratio R ＝E R (z-1)/E R (z) is less than threshold δ 1 In which E R ＝λ 1 /λ 2 (ii) a If E R ＜δ 1 Based on the energy compensation coefficient
Performing energy compensation on the reconstructed signal data set to realize damage information enhancement, if E R ≥δ 1 Reconstructing a Hankerl matrix and returning to the step for one-to-one cycle iteration;
step one and three, extraction
Signal segment reconstruction data set in
3. The steel rail health monitoring method based on information enhancement and variable-step sparse expression as claimed in claim 2, wherein in the second step, a specific formula for performing energy compensation on the reconstructed Hankerl matrix is as follows:
wherein U = [ U ] 1 ,u 2 ]Is a left singular matrix, V ∈ R M×N Is the right singular matrix.
4. A steel rail health monitoring method based on information enhancement and variable-step sparse representation according to claim 1, characterized in that the specific steps of the second step are as follows:
step two, initializing a measurement value matrix
Wherein S is a random observation matrix; the initial inner product residual error is R (1) = M e The sensing matrix gamma = S.D, D is an initial dictionary obtained based on a K-SVD algorithm, the number of initialization iterations is z =1, and an initial iteration step length L is obtained 1 When there are a plurality of points satisfying the above condition, only the second derivative of the first point is taken as the iteration stepWhen the second derivatives of any residual inner product points have the same sign, the iteration step length is 1;
step two, when the iteration times z is larger than 1, adaptively adjusting the iteration step length L through the inner product inflection point;
step two and step three, the largest front L in | gamma · R (z-1) | is selected i (z) item sequence number value added to sequence number set B i The serial number in gamma is compared with B i The same item is added to the candidate set C i (z), reconstructing a sensing matrix gamma;
fourthly, reconstructing a real damaged sparse coefficient matrix through a least square method:
wherein M is e,i Is the ith column in the measurement matrix;
step two and five, selecting
Replacing original sequence number set B by the largest front L (z) item in the sequence i In Γ (z) with B i The corresponding L (z) term is denoted Γ' (z), and the residual is then updated;
step two, if residual error R (z) is less than or equal to delta 2 Then the reconstruction precision of the real damage signal is shown to meet the system requirement, and the real damage signal is reconstructed by
Reconstructing a real damaged signal set if residual | | | R (z-1) | non-calculation 2 ≤||R(z)|| 2 If yes, returning to the second step for continuous iteration, otherwise, making R (z) = R (z-1), z = z +1, and returning to the second step for continuous iteration;
seventhly, collecting the obtained real damage signal set Y s Splicing according to columns to obtain reconstructed real damage signals
5. The steel rail health monitoring method based on information enhancement and variable-step sparse representation according to claim 4, wherein in the first step, an initial iteration step L is adopted 1 The specific formula of (2) is as follows:
wherein R is i,j A jth residual error point in the ith column of residual error vectors;
in the second step, the specific formula of the iteration step length L is as follows:
6. the steel rail health monitoring method based on information enhancement and variable step size sparse representation according to claim 4, wherein in the second step and the third step, a specific formula of a sensing matrix Γ is as follows:
Γ(z)＝C i (z)∪Γ(z-1)；
in the second and fifth steps, the specific formula of the residual error is as follows:
R i (z)＝M e,i -Γ′(z)(Γ′ T (z)Γ′(z))Γ′ T (z)M e,i 。
7. a steel rail health monitoring method based on information enhancement and variable-step sparse representation according to claim 4, characterized in that the specific steps of the third step are as follows:
step three, taking M sampling points as one section, respectively extracting the information entropy of each section of signal without overlapping, and normalizing the information entropy;
step two, respectively calculating the short-time energy of the real damage signal corresponding to each value in the information entropy, and normalizing the short-time energy, wherein a Hamming window is adopted for calculating the short-time energy, the window length is w, and no overlap exists;
thirdly, self-adaptive weighting is carried out on the information entropy characteristics of the damage signals through short-time energy reconstruction, the structural health index SHI of the steel rail is constructed, meanwhile, the minimum effective weight is added, and the change of the health state of the steel rail in the safety stage due to the fact that the short-time energy is too small is prevented;
step three, dividing the structural health state of the steel rail into three stages based on the structural health index: a secure phase when SHI < 0.1, an extended phase when SHI ∈ [0.1,1), and a non-secure phase when SHI = 1.
8. A steel rail health monitoring method based on information enhancement and variable step size sparse representation according to claim 7, wherein in the step three, a specific formula of information entropy is as follows:
wherein the content of the first and second substances,
for the i-th segment in the reconstructed real impairment signal,
is composed of
The probability of (a) of (b) being,
is composed of
The probability of the minimum value in the set of values,
is composed of
The probability of the median maximum.
9. A steel rail health monitoring method based on information enhancement and variable step size sparse representation according to claim 7, characterized in that the specific formula of short-time energy in the third two is as follows:
where ψ (-) is a Hamming window function.
10. A steel rail health monitoring method based on information enhancement and variable-step sparse representation according to claim 7, wherein in the third step, the specific formulas of the structural health index and the minimum effective weight are as follows:
where ε and η are constants and ε - η is the minimum effective weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210735400.1A CN115236208B (en) | 2022-06-27 | 2022-06-27 | Steel rail health monitoring method based on information enhancement and variable step length sparse expression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210735400.1A CN115236208B (en) | 2022-06-27 | 2022-06-27 | Steel rail health monitoring method based on information enhancement and variable step length sparse expression |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115236208A true CN115236208A (en) | 2022-10-25 |
| CN115236208B CN115236208B (en) | 2023-04-07 |
Family
ID=83671617
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210735400.1A Active CN115236208B (en) | 2022-06-27 | 2022-06-27 | Steel rail health monitoring method based on information enhancement and variable step length sparse expression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115236208B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110222781A1 (en) * | 2010-03-15 | 2011-09-15 | U.S. Government As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| US20160162577A1 (en) * | 2014-12-08 | 2016-06-09 | Mitsubishi Electric Research Laboratories, Inc. | Method for Segmenting and Tracking Content in Videos Using Low-Dimensional Subspaces and Sparse Vectors |
| CN106500735A (en) * | 2016-11-03 | 2017-03-15 | 重庆邮电大学 | A kind of FBG signal adaptive restorative procedures based on compressed sensing |
| CN107093184A (en) * | 2017-04-11 | 2017-08-25 | 湖北理工学院 | A kind of ultrasonic image sequence dividing method based on sparse features and shape correlation |
| CN109039960A (en) * | 2018-06-26 | 2018-12-18 | 东南大学 | A kind of underwater sound condition of sparse channel estimation variable step degree of rarefication Adaptive matching method for tracing |
| CN109670465A (en) * | 2018-12-25 | 2019-04-23 | 重庆师范大学 | The signal reconfiguring method of compressed sensing based adaptive determining degree of rarefication |
| CN109859223A (en) * | 2019-01-14 | 2019-06-07 | 江苏大学 | A kind of compression sampling nonlinear iteration optimal reconfiguration method based on target plant |
| CN111582137A (en) * | 2020-04-30 | 2020-08-25 | 燕山大学 | Rolling bearing signal reconstruction method and system |
| CN111735518A (en) * | 2020-06-19 | 2020-10-02 | 山东省特种设备检验研究院有限公司 | Intelligent detection system for accumulated liquid in oil and gas pipeline |
| US20210150705A1 (en) * | 2019-11-19 | 2021-05-20 | Uih America, Inc. | Systems and methods for determining plasma input function used in positron emission tomography imaging |
| US20210341436A1 (en) * | 2018-10-08 | 2021-11-04 | Ecole Polytechnique Federale De Lausanne (Epfl) | Image reconstruction method based on a trained non-linear mapping |
| CN114330455A (en) * | 2022-01-05 | 2022-04-12 | 哈尔滨工业大学 | Steel rail acoustic emission signal rapid high-precision reconstruction method based on compressed sensing |
-
2022
- 2022-06-27 CN CN202210735400.1A patent/CN115236208B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110222781A1 (en) * | 2010-03-15 | 2011-09-15 | U.S. Government As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| US20160162577A1 (en) * | 2014-12-08 | 2016-06-09 | Mitsubishi Electric Research Laboratories, Inc. | Method for Segmenting and Tracking Content in Videos Using Low-Dimensional Subspaces and Sparse Vectors |
| CN106500735A (en) * | 2016-11-03 | 2017-03-15 | 重庆邮电大学 | A kind of FBG signal adaptive restorative procedures based on compressed sensing |
| CN107093184A (en) * | 2017-04-11 | 2017-08-25 | 湖北理工学院 | A kind of ultrasonic image sequence dividing method based on sparse features and shape correlation |
| CN109039960A (en) * | 2018-06-26 | 2018-12-18 | 东南大学 | A kind of underwater sound condition of sparse channel estimation variable step degree of rarefication Adaptive matching method for tracing |
| US20210341436A1 (en) * | 2018-10-08 | 2021-11-04 | Ecole Polytechnique Federale De Lausanne (Epfl) | Image reconstruction method based on a trained non-linear mapping |
| CN109670465A (en) * | 2018-12-25 | 2019-04-23 | 重庆师范大学 | The signal reconfiguring method of compressed sensing based adaptive determining degree of rarefication |
| CN109859223A (en) * | 2019-01-14 | 2019-06-07 | 江苏大学 | A kind of compression sampling nonlinear iteration optimal reconfiguration method based on target plant |
| US20210150705A1 (en) * | 2019-11-19 | 2021-05-20 | Uih America, Inc. | Systems and methods for determining plasma input function used in positron emission tomography imaging |
| CN111582137A (en) * | 2020-04-30 | 2020-08-25 | 燕山大学 | Rolling bearing signal reconstruction method and system |
| CN111735518A (en) * | 2020-06-19 | 2020-10-02 | 山东省特种设备检验研究院有限公司 | Intelligent detection system for accumulated liquid in oil and gas pipeline |
| CN114330455A (en) * | 2022-01-05 | 2022-04-12 | 哈尔滨工业大学 | Steel rail acoustic emission signal rapid high-precision reconstruction method based on compressed sensing |
Non-Patent Citations (3)
| Title |
|---|
| Y.KONG ET AL.: "Discriminative dictionary iearning based sparse representation classification for intelligent fault identification of planet bearings in wind turbine", 《RENEWABLE ENERGY》 * |
| YONGQI CHANG ET AL.: "An efficient method for wheel-flattened defects detection based on acoustic emission technique", 《IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS AND FREQUENCY CONTROL》 * |
| 于志 等: "有源噪声主动控制中一种改进的变步长算法研究", 《工程控制与应用》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115236208B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109084186B (en) | Pipeline leakage signal identification method based on improved ELMD (ensemble empirical mode decomposition) multi-scale entropy | |
| US9534993B2 (en) | Entropy-based impact load identification | |
| Chen et al. | Compressed sensing based on dictionary learning for extracting impulse components | |
| Croxford et al. | Efficient temperature compensation strategies for guided wave structural health monitoring | |
| Sugiyama et al. | Firmness measurement of muskmelons by acoustic impulse transmission | |
| Chen et al. | An adaptive Morlet wavelet filter for time-of-flight estimation in ultrasonic damage assessment | |
| Tang et al. | A correlated empirical mode decomposition method for partial discharge signal denoising | |
| CN110286155B (en) | Damage detection method and system for multilayer composite material | |
| CN109507304B (en) | Defect detection method based on ultrasonic flaw detection | |
| Wang et al. | Prediction model of natural gas pipeline crack evolution based on optimized DCNN-LSTM | |
| CN109632973A (en) | A kind of ultrasound echo signal extracting method based on Based on Multiscale Matching tracking | |
| Hou et al. | Prediction of firmness and pH for “golden delicious” apple based on elasticity index from modal analysis | |
| Song et al. | An improved structural health monitoring method utilizing sparse representation for acoustic emission signals in rails | |
| Dettmer et al. | Joint time/frequency-domain inversion of reflection data for seabed geoacoustic profiles and uncertainties | |
| CN107037136B (en) | Array type ultrasonic detection method for heat exchange dirt of pipeline | |
| CN110487499B (en) | Bolt pretightening force identification method based on singular value decomposition and support vector machine | |
| CN109682892B (en) | Signal denoising method based on time-frequency analysis | |
| CN115236208B (en) | Steel rail health monitoring method based on information enhancement and variable step length sparse expression | |
| CN108415880B (en) | Line loss characteristic analysis method based on sample entropy and wavelet transformation | |
| CN111505709B (en) | Attenuation qualitative analysis method based on sparse spectral decomposition | |
| CN116522080A (en) | Partial discharge signal noise reduction method | |
| CN114330455B (en) | Steel rail acoustic emission signal rapid high-precision reconstruction method based on compressed sensing | |
| CN111611533B (en) | Mill load characteristic extraction method based on self-adaptive VMD and improved power spectrum | |
| Wang et al. | Early sub-surface fault detection in rolling element bearing using acoustic emission signal based on a hybrid parameter of energy entropy and deep autoencoder | |
| Ortega et al. | Hilbert–Huang transform analysis of storm waves |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |