CN103808801A - Real-time detection method for high-speed rail injury based on vibration and audio composite signals - Google Patents

Abstract

The invention discloses a real-time detection method for high-speed railway rail damage based on vibration and audio frequency composite signals. Measure the rail vibration signal and audio signal; 2. According to the actual structure of the track, use the finite element method to construct the rail vibration model and acoustic model, and obtain the vibration and audio signal of typical rail damage; Carry out preprocessing; 4. Fuse vibration and acoustic signals to establish a three-dimensional tensor of vibration, sound frequency and damage type; 5. Use non-negative tensor decomposition to extract the characteristic coefficient of damage type; 6. Use correlation vector machine to establish damage recognition rules and Classify real-time measurement signals to determine damage type. The invention realizes the real-time detection of the rail damage of the high-speed railway and guarantees the safe operation of the high-speed railway.

Description

A kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal

Technical field

The present invention relates to a kind of detection method of rail in high speed railway hurt, especially a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal.

Background technology

Along with the develop rapidly of the construction energetically, particularly high-speed railway of China Railway, the safety of transportation by railroad seems more and more important.Due to the restriction of hurt detection speed, traditional ultrasonic hurt Detection Techniques and pattern cannot meet the flaw detection requirement of high-speed railway, and the research of novel high speed railway inspection technique is extremely urgent.

Rail, in the process using, can fracture, crackle and other hurt form, is rail defects and failures.Rail defects and failures is the main cause of disconnected rail, is the important hidden danger that affects traffic safety, and train derailment accident is mainly produced by rail fracture.Train is in acceleration and braking procedure and while passing through gap, bend and track switch, produce for a long time strong friction, extruding, the bending to rail and impact, rail very easily produces fatigue crack, is just easy to Quick Extended, thereby causes the great serious accidents such as brittle fractures of rail once crackle produces.Friction, extruding, bending and the impact of bullet train to rail etc. effects is more outstanding, and its probability cracking is increased greatly, and speed from crack growth to rail fracture is faster.In order to guarantee the safe operation of high-speed railway, must shorten the sense cycle of high-speed railway.The rate of traffic flow of high-speed railway is large in addition, the speed of a motor vehicle is high, existingly in the time that auxiliary railway track flaw detection pattern detects rail, will take high ferro circuit take large-scale inspection car as main, small-sized defectoscope, serious impact the efficiency of train operation.Meanwhile, small-sized steel rail defectoscope also can only carry out limited detection in subrange.Traditional flaw detection mode is difficult to meet the demand of high-speed railway, must develop as early as possible quick, accurate, real-time high-speed railway inspection technique for this reason.

Summary of the invention

The object of the present invention is to provide a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal, obtain in real time, accurately rail defects and failures information.

The object of the invention is to be achieved through the following technical solutions:

Step 1: vibration transducer and acoustic sensor are installed on high speed railway track rail, combining wireless network node processor is connected and forms the distributed wireless sensor network of a line of high-speed railway, and rail vibration signal and audio signal are carried out to Real-time Collection;

Step 2: according to track practical structures, adopt Finite Element Method to build track vibration model and acoustic model, obtain vibration and the audio signal of typical rail defects and failures;

Step 3: utilize Hilbert-Huang transform to carry out pre-service to the signal collecting, the Hilbert marginal spectrum of picked up signal;

Step 4: merge vibration and acoustic signal and set up vibration, audio frequency and hurt kind three-dimensional tensor;

Step 5: utilize non-negative tensor resolution method to decompose three-dimensional tensor, extract the hurt species characteristic coefficient of typical rail defects and failures;

Step 6: the hurt characteristic coefficient that uses step 5 to extract is trained Method Using Relevance Vector Machine, sets up hurt recognition rule and the vibration measuring in real time and audio signal are classified, and determines hurt type.

The variation that can bring vibration signal and audio signal due to rail defects and failures, in these signals, comprising the information of hurt, the present invention detects vibration and the audio signal of rails hurt in real time by wireless sensor network, utilize Hilbert-Huang transform (Hi lbert-Huang Transform, HHT) carry out Signal Pretreatment, non-negative tensor resolution (Non-negativeTensor FactoriZation, NTF) carry out feature extraction and Method Using Relevance Vector Machine (the Relevance Vector Machine of multidimensional data, RVM) hurt is carried out to discriminator, fast, the real-time information of Obtaining Accurate rail defects and failures, thereby ensure the safe operation of high-speed railway.Compared with prior art, tool has the following advantages in the present invention:

(1) mostly be inspection car and defectoscope for the flaw detection mode of current rail, the present invention, by setting up the track wireless sensor network of detecting a flaw along the line, realizes the Real-Time Monitoring to rail in high speed railway hurt situation;

(2) by setting up rail finite element model to obtain the characteristic signal in various typical hurt situations, for setting up corresponding rail defects and failures sorter, for the hurt classification of measured signal provides hurt judgment criterion;

(3) give full play to HHT and analyze high efficiency non-linear, non-stationary signal, signal is carried out to pre-service, to reduce transmitted data amount, improve node work efficiency;

(4) carry out the extraction of hurt characteristic coefficient in conjunction with vibration and audio frequency signal configuration multi-dimensional signal tensor, can more effective extraction lie in the feature in data with recognition capability with respect to the signal of one type of single use, improve the accuracy of identification.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is system architecture schematic diagram;

Fig. 3 is Hilbert-Huang transform (HHT) process flow diagram;

Fig. 4 is vibration and audio signal structure 2D signal schematic diagram;

Fig. 5 is that vibration signal, audio signal and hurt kind three-dimensional tensor build schematic diagram;

Fig. 6 is non-negative tensor resolution process flow diagram;

Fig. 7 is the rails finite element model of setting up;

Fig. 8 is each rank IMF and the residual error after HHT decomposes.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is further described; but be not limited to this; every technical solution of the present invention is modified or is equal to replacement, and not departing from the spirit and scope of technical solution of the present invention, all should be encompassed in protection scope of the present invention.

Embodiment one: present embodiment provides a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal, as shown in Figure 1, is divided into six steps, and concrete steps are as follows:

Step 1: vibration transducer and acoustic sensor are installed on high speed railway track rail, and combining wireless network node processor forms the distributed wireless sensor network of line of high-speed railway, and rail vibration signal and audio signal are carried out to Real-time Collection.

Distributed wireless sensor network for high-speed railway flaw detection comprises two class sensors: vibration transducer and acoustic sensor.Vibration transducer is mainly used in the collection of track vertical vibration signal, with obtain have or not train through out-of-date track self and train through time the wheel-rail interaction various vibration signals that produced, also comprise the caused vibration characteristics of rail defects and failures simultaneously and change; Acoustic sensor is mainly used in the signals collecting of track train through out-of-date audio frequency, and it had both comprised the caused noise signal of various vibrations, the rail defects and failures information while also comprising train process between wheel track.

Several vibration transducers and acoustics installation of sensors are at rail sidepiece, two kinds of sensors are connected to radio node processor, at train through out-of-date, the vibration transducer vertical track vibration signal that excitation produces to train gathers, the information that acoustic sensor collection comprises wheel track audio features, wireless network node processor carries out pre-service to the vibration signal collecting and audio signal, and send to information processing centre or patrol and examine inspection car by wireless sensor network, the wireless network node processor working state control of node when being responsible for having or not train also simultaneously, so that the saving energy, extend the serviceable life of node.Information processing centre or patrol and examine inspection car the signal receiving is classified by hurt recognition rule method, judges the kind of hurt.Its whole system structure as shown in Figure 2.

Step 2: according to track practical structures, adopt Finite Element Method to build track vibration model and acoustic model, obtain vibration and the audio signal of typical rail defects and failures.

Adopt finite element modeling method to set up rail three-dimensional finite element model (Fig. 7), obtain vibration and the audio signal of typical hurt, for the foundation of rail defects and failures recognition classifier provides sample data, for the hurt classification of measured signal provides hurt judgment criterion.Finite element model need to be selected suitable temporal resolution and spatial resolution, and its time interval and element length formula are as follows:

The time interval: $\mathrm{\Δt}=\frac{1}{20f_{\max }}---\left(1\right),$

Element length: $l_e=\frac{{\mathrm{\λ}}_{\min }}{20}---\left(2\right),$

In formula: Δ t is the time interval, f _maxfor the highest frequency of interested signal, l _efor selected finite element element length, λ _minfor the minimum wavelength of interested signal.

Step 3: utilize Hilbert-Huang transform to carry out pre-service to the signal collecting, the Hilbert marginal spectrum of picked up signal.

Because the data volume of vibrating and audio signal comprises is larger, and the transmittability of wireless network node is limited, need to carry out pre-service to signal, in radio node processor, adopt HHT method respectively vibration and audio signal to be carried out to essential characteristic analysis, arrive to obtain the Hilbert marginal spectrum of signal, reduce the data volume that signal comprises, improved the transfer efficiency of wireless network node.Its flow process as shown in Figure 3, specifically describes as follows:

Step 1), find out vibration or maximum value and the minimal value of audio signal x (t);

Step 2), generate respectively local maximum and the local minimum envelope of signal by cubic spline interpolation method;

Step 3), the local maximum envelope of signal and local minimum envelope are added and are averaging, obtain local envelope average m (t);

Step 4), from signal, deduct local envelope average: h (t)=x (t)-m (t);

Step 5), judge whether h (t) meets two conditions of IMF: condition one, in whole function, the number of extreme point equates with the number that passes through zero point, or only differs 1; Condition two, is at any time zero by the defined envelope average of extreme value envelope; If meet, obtain first IMF c ₁(t), define residual error item: r simultaneously ₁(t)=x (t)-c ₁(t), execution step 6), if do not met, by signal h (t) execution step 1) to step 4);

Step 6), continue the above-mentioned screening process of application using residual error as signal to be decomposed and decomposite each IMFc _i(t), i=1 ..., n, when decomposing to r and be normal function or monotonic quantity or only having the function of an extreme point, decomposes and finishes; Obtain the decomposition of the following form of x (t):

, wherein, residual error item r has represented the basic trend of signal x (t), the number that rear IMF is decomposed in n representative;

Step 7), x obtained above (t) decomposed form is carried out to Hilbert transform, structure analytic signal, be expressed as polar form, and get real part, obtain hilbert spectrum H (w, t): wherein a _iand w (t) _i(t) be all the function of time t, a _i(t) be amplitude function, w _i(t) be phase function, Re is realistic computing, has reacted the relation between time t, phase place w and H (w, t);

Step 8), combination

from 0 to T (T original signal length) integration

the h (w) obtaining is Hilbert marginal spectrum, the amplitude h (w) that has reflected signal on whole frequency band with the situation of change (Fig. 8) of frequency.

Step 4: merge vibration and acoustic signal and set up vibration, audio frequency and hurt kind three-dimensional tensor.

Because tensor has retained multidimensional information and the relation between them, therefore adopt tensor algorithm to construct multidimensional tensor signal and can more effective extraction lie in the characteristic information with recognition capability in data, further improve the accuracy of identification.Utilize the Hilbert marginal spectrum of the pretreated vibration of HHT and audio signal, and set up three-dimensional tensor in conjunction with hurt kind, its building process, as shown in Fig. 4-5, specifically describes as follows:

Step 1), vibration transducer measures hurt vibration signal on rail, be designated as: x ₁, and it is carried out to HHT conversion obtain Hilbert marginal spectrum, be designated as: f ₁; Acoustic sensor is measured hurt audio signal on rail, is designated as: x ₂, and it is carried out to HHT conversion obtain Hilbert marginal spectrum, be designated as: f ₂;

Step 2), each f _i, (i=1,2) are all the signals of one dimension, these 2 one-dimensional signals of utilization vibration and audio frequency construct a 2D signal and are designated as: x ^{vibration × acoustic}, wherein vibration represents that vibration signal value, acoustic represent audio signal value;

Step 3), according to different hurt situations, hurt kind builds the third dimension of signal, is designated as: x ^{vibration × acoustic × class}, wherein vibration represents that vibration signal value, acoustic represent that audio signal value, class represent hurt kind.

Step 5: utilize non-negative tensor resolution to extract the hurt species characteristic coefficient of typical rail defects and failures.

The three-dimensional tensor signal indication of setting up in step 4 is: x ^{vibration × acoustic × class}=G × ₁a ⁽¹⁾× ₂a ⁽²⁾× ₃a ⁽³⁾, wherein G, A ⁽¹⁾, A ⁽²⁾, A ⁽³⁾>=0, carry out non-negative tensor resolution flow process as shown in Figure 6, specifically describe as follows:

Step 1), random initializtion A ⁽ⁿ⁾(n=1,2,3), the least mean-square error C of computational minimization _new=|| X _(n)-A ⁽ⁿ⁾z ^{(n) T}|| ², wherein: X _(n)for original signal x ^{vibration × acoustic × class}, $Z^{\left(n\right)}=A^{\left(N\right)}|\⊗|A^{(N-1)}|\⊗|...|\⊗|A^{(n+1)}|\⊗|A^{(n-1)}|\⊗|...|\⊗|A^{\left(1\right)},$ N=1,2,3, N is order of a tensor number, T represents matrix transpose;

Step 2), iterative utilize the A solving in this step ⁽ⁿ⁾the C that calculating makes new advances _new=|| X _(n)-A ⁽ⁿ⁾z ^{(n) T}|| ², n=1,2,3; Simultaneously by step 1) the middle C calculating _newvalue assignment is to C _old;

Step 3), judgement

δ is error precision requirement, sets up by step 2) in the C that calculates _newvalue assignment is to C _old, continue execution step 2) in calculating C _newwith step 3); Be false and perform step 4);

Step 4), be met the A of requirement ⁽¹⁾, A ⁽²⁾, A ⁽³⁾matrix; Wherein: A ⁽¹⁾, A ⁽²⁾for the matrix producing in decomposable process, as new substrate, A ⁽³⁾for the characteristic coefficient matrix of hurt classification, wherein every a line is a proper vector, corresponding a kind of hurt situation.

Step 6: utilize Method Using Relevance Vector Machine set up hurt recognition rule and real-time measuring-signal is classified, determine hurt type.

Based on Method Using Relevance Vector Machine sorting technique, the proper vector coefficient of the typical hurt situation to rail defects and failures vibration signal and audio signal is trained, and obtains corresponding Method Using Relevance Vector Machine, sets up hurt recognition rule, realize the online of rail vibration and audio signal and detect in real time, specifically describe as follows:

Step 1), utilize vibration and the audio signal of the typical hurt that step 2 obtains, and the signal processing method of integrating step three, step 4 and step 5 obtains proper vector coefficient;

Step 2), utilize step 1) proper vector coefficient carry out Method Using Relevance Vector Machine training, obtain having the Method Using Relevance Vector Machine of hurt discriminator, set up hurt recognition rule;

Step 3), by the distributed wireless sensor network of track, measure train excitation lower rail vibration signal and audio signal, and these signals utilized to step 3, step 4 and step 5 calculating signal characteristic coefficient;

Step 4), utilize step 2) in obtain Method Using Relevance Vector Machine to step 3) in the signal characteristic coefficient that obtains classify, identify the hurt type of this real-time measuring-signal.

Claims (7)

1. the rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal, is characterized in that described method step is as follows:

Step 1: vibration transducer and acoustic sensor are installed on high speed railway track rail, combining wireless network node processor is connected and forms the distributed wireless sensor network of a line of high-speed railway, and rail vibration signal and audio signal are carried out to Real-time Collection;

Step 2: according to track practical structures, adopt Finite Element Method to build track vibration model and acoustic model, obtain vibration and the audio signal of typical rail defects and failures;

Step 3: utilize Hilbert-Huang transform to carry out pre-service to the signal collecting, the Hilbert marginal spectrum of picked up signal;

Step 4: merge vibration and acoustic signal and set up vibration, audio frequency and hurt kind three-dimensional tensor;

Step 5: utilize non-negative tensor resolution method to decompose three-dimensional tensor, extract the hurt species characteristic coefficient of typical rail defects and failures;

Step 6: the hurt characteristic coefficient that uses step 5 to extract is trained Method Using Relevance Vector Machine, sets up hurt recognition rule and the vibration measuring in real time and audio signal are classified, and determines hurt type.

2. a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal according to claim 1, is characterized in that the concrete steps of described step 1 are as follows:

Several vibration transducers and acoustics installation of sensors are at high speed railway track rail sidepiece, and two kinds of sensors are connected to radio node processor form the distributed wireless sensor network of a line of high-speed railway, at train through out-of-date, the vibration transducer vertical track vibration signal that excitation produces to train gathers, acoustic sensor gathers the information of wheel track audio features, and will collect signal and carry out being sent to information processing centre or being patrolled and examined inspection car by wireless sensor network after pre-service.

3. a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal according to claim 1, it is characterized in that in described step 2, adopt finite element modeling method to set up rail three-dimensional finite element model, obtain vibration and the audio signal of typical hurt.

4. a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal according to claim 1, is characterized in that the concrete steps of described step 3 are as follows:

Step 1), find out vibration or maximum value and the minimal value of audio signal x (t);

Step 2), generate respectively local maximum and the local minimum envelope of signal by cubic spline interpolation method;

Step 3), the local maximum envelope of signal and local minimum envelope are added and are averaging, obtain local envelope average m (t);

Step 4), from signal, deduct local envelope average: h (t)=x (t)-m (t);

Step 5), judge whether h (t) meets the condition of IMF: condition one, in whole function, the number of extreme point equates with the number that passes through zero point, or only differs 1; Condition two, is at any time zero by the defined envelope average of extreme value envelope; If meet, obtain first IMF c ₁(t), define residual error item: r simultaneously ₁(t)=x (t)-c ₁(t), execution step 6), if do not met, by signal h (t) execution step 1) to step 4);

Step 6), continue the above-mentioned screening process of application using residual error as signal to be decomposed and decomposite each IMFc _i(t), i=1 ..., n, when decomposing to r and be normal function, monotonic quantity or only having the function of an extreme point, decomposes and finishes, and obtains the decomposition of the following form of x (t):

wherein, residual error item r has represented the basic trend of signal x (t);

Step mule 7), x obtained above (t) decomposed form is carried out to Hilbert transform, structure analytic signal, be expressed as polar form, and get real part, obtain hilbert spectrum H (w, t):

wherein a _iand w (t) _i(t) be all the function of time t, a _i(t) be amplitude function, w _i(t) be phase function, Re is realistic computing, has reacted the relation between time t, phase place w and H (w, t):

Step 8), combination $H(w,t)=\mathrm{Re}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{a}_i\left(t\right)e^{j\∫w_i\left(t\right)\mathrm{dt}}$ From 0 to T integration $h\left(w\right)={\∫}_0^{T}H(w,t)\mathrm{dt},$ T is original signal length, and the h (w) obtaining is Hilbert marginal spectrum.

5. a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal according to claim 1, is characterized in that the concrete steps of described step 4 are as follows:

Step 1), vibration transducer measures hurt vibration signal on rail, be designated as: x ₁, and it is carried out to HHT conversion obtain Hilbert marginal spectrum, be designated as: f ₁; Acoustic sensor is measured hurt audio signal on rail, is designated as: x ₂, and it is carried out to HHT conversion obtain Hilbert marginal spectrum, be designated as: f ₂;

Step 2), each f _i, (i=1,2) are all the signals of one dimension, these 2 one-dimensional signals of utilization vibration and audio frequency construct a 2D signal and are designated as: x ^{vibration × acoustic}, wherein vibration represents that vibration signal value, acoustic represent audio signal value;

Step 3), according to different hurt situations, hurt kind builds the third dimension of signal, is designated as: x ^{vibration × acoustic × class}, wherein vibration represents that vibration signal value, acoustic represent that audio signal value, class represent hurt kind.

6. a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal according to claim 1, is characterized in that the concrete steps of described step 5 are as follows:

Three-dimensional tensor signal indication is: x ^{vibration × acoustic × class}=G × ₁a ⁽¹⁾× 2A ⁽²⁾× 3A ⁽³⁾, wherein G, A ⁽¹⁾, A ⁽²⁾, A ⁽³⁾>=0, carry out non-negative tensor resolution, its step is as follows:

Step 1), random initializtion A ⁽ⁿ⁾(n=1,2,3), the least mean-square error C of computational minimization _new=|| X _(n)-A ⁽ⁿ⁾z ^{(n) T}|| ², wherein: X _(n)for original signal x ^{vibration × acoustic × class}, $Z^{\left(n\right)}=A^{\left(N\right)}|\⊗|A^{(N-1)}|\⊗|...|\⊗|A^{(n+1)}|\⊗|A^{(n-1)}|\⊗|...|\⊗|A^{\left(1\right)},$ N=1,2,3, n is order of a tensor number, T represents matrix transpose;

Step 2), iterative

utilize the A solving in this step ⁽ⁿ⁾the C that calculating makes new advances _new=|| X _(n)-A ⁽ⁿ⁾z ^{(n) T}|| ², n=1,2,3; Simultaneously by step 1) the middle C calculating _newvalue assignment is to C _old;

Step 3), judgement

δ is error precision requirement, sets up by step 2) in the C that calculates _newvalue assignment is to C _old, continue execution step 2) in calculating C _newwith step 3); Be false and perform step 4);

Step 4), be met the A of requirement ⁽¹⁾, A ⁽²⁾, A ⁽³⁾matrix; Wherein: A ⁽¹⁾, A ⁽²⁾for the matrix producing in decomposable process, as new substrate, A ⁽³⁾for the characteristic coefficient matrix of hurt classification, wherein every a line is a proper vector, corresponding a kind of hurt situation.

7. according to a kind of rail in high speed railway hurt real-time detection method based on vibration and audio frequency composite signal described in claim 1,4,5 or 6, it is characterized in that the concrete steps of described step 6 are as follows:

Step 1), utilize vibration and the audio signal of the typical hurt that step 2 obtains, and the signal processing method of integrating step three, step 4 and step 5 obtains proper vector coefficient;

Step 2), utilize step 1) proper vector coefficient carry out Method Using Relevance Vector Machine training, obtain having the Method Using Relevance Vector Machine of hurt discriminator, set up hurt recognition rule;

Step 3), by the distributed wireless sensor network of track, measure train excitation lower rail vibration signal and audio signal, and these signals utilized to step 3, step 4 and step 5 calculating signal characteristic coefficient;

Step 4), utilize step 2) in obtain Method Using Relevance Vector Machine to step 3) in the signal characteristic coefficient that obtains classify, identify the hurt type of this real-time measuring-signal.

Images