CN109649432B - Cloud platform rail integrity monitoring system and method based on guided wave technology - Google Patents

Abstract

The invention discloses a cloud platform rail integrity monitoring system and method based on guided wave technology. The monitoring system includes a front-end monitoring module, a cloud monitoring server and a browsing terminal. The front-end monitoring module includes a guided wave transducer, a temperature sensor and a control cabinet. The control cabinet contains a solar panel module, a charge and discharge control circuit, an energy storage module, and a transmission interface circuit. module, the lower computer control circuit module, the communication interface circuit module, the guided wave transceiver module and the signal conditioning module; first, the characteristic signals and coefficient matrices that characterize the acoustic emission sources at different structures of the rail will be extracted through noise reduction, and then dimensionality reduction and reduction will be carried out. Noise reconstruction obtains the guided wave signal that characterizes the damage of the rail, obtains the damage location, and realizes the integrity monitoring of the rail. The invention can effectively realize the reliable monitoring of the integrity of the rail, greatly improves the cross-regional and real-time monitoring, and has important practical significance and engineering value.

Description

Cloud platform rail integrity monitoring system and method based on guided wave technology

technical field

The invention relates to a rail integrity monitoring system and method, in particular to a cloud platform rail integrity monitoring system and method based on guided wave technology.

Background technique

In recent years, on the one hand, with the rapid development of the national economy, railway transportation, which is the lifeline of the national economy, has achieved unprecedented rapid development; With the substantial improvement, the rail, which is an important part of the track, is bound to be subject to increased loads, impacts, etc., which inevitably leads to an increase in the probability of rail damage, and puts forward stricter requirements for the reliable and safe operation of the track. Therefore, there is an urgent need to propose effective and reliable technologies and methods for the detection and service condition monitoring of in-service railway rails.

Ultrasonic guided wave technology is widely used in non-destructive testing and online monitoring in all walks of life due to its characteristics of long-distance, large-scale, full-section detection, and single-ended transceiver. Railway rails are very suitable for monitoring and evaluation of working conditions using guided wave technology, regardless of material or component type. However, considering that the structural characteristics of the rail are complex and the fasteners are often overlapped, the traditional point-to-point non-destructive testing methods such as ultrasonic testing, magnetic flux leakage testing, machine vision and penetration testing are difficult to meet the requirements of timeliness and cloud online in actual monitoring. , reliability, and strict requirements across a wide range of regions. At the same time, considering the busy operation of railway lines, it has been difficult to use the artificial detection and monitoring technology of the track under the skylight to meet the growing demand for actual detection and monitoring of railways. The relevant early researches with the above-mentioned drawbacks are given here, and the publication number is CN104535652A "A method for detecting damage to a rail", the publication number is CN101398410A "A method and device for detecting rail defects with electromagnetic ultrasonic technology", and the publication number CN104237381A "A method for detecting rail defects" A rail flaw detection method based on image fusion of laser ultrasound and high-speed camera" and the patent "Method and Device for On-Site Ultrasonic Inspection of Railway Tracks" with publication number CN102084245A, etc. With the rapid development of information technology today, especially the rapid development of the Internet, cloud computing, and fifth-generation mobile communication technologies, online, cross-regional, and real-time monitoring technologies based on cloud platforms have increasingly become the focus and frontier of applications in all walks of life. hot spot. The promotion and application of the rail integrity monitoring method and system based on the cloud platform will also help to promote the automation and intelligence of my country's railway track online detection and monitoring technology.

SUMMARY OF THE INVENTION

Aiming at the problems and defects in the above-mentioned background technology, the present invention proposes a cloud platform rail integrity monitoring system and method based on guided wave technology, which can realize real-time online cross-regional monitoring of the service status of railway line network rails.

As shown in Figure 2, the present invention is achieved through the following technical solutions:

1. A cloud platform rail integrity monitoring system based on guided wave technology:

The monitoring system includes a front-end monitoring module, a cloud monitoring server and a browsing terminal. The front-end monitoring module includes a guided wave transducer, a temperature sensor and a control cabinet. The control cabinet includes a solar panel module, a charge and discharge control circuit, and an energy storage module. , transmission interface circuit module, lower computer control circuit module, communication interface circuit module, guided wave transceiver module and signal conditioning module; among them, the guided wave transducer and temperature sensor are installed on the rail to be tested, and the lower computer control circuit module passes through The transmission interface circuit module is electrically connected with the temperature sensor, the charge and discharge control circuit and the guided wave transceiver module respectively, the guided wave transducer is electrically connected with the guided wave transceiver module through the signal conditioning module, the solar panel module is electrically connected with the charge and discharge control circuit, and the charging and discharging control circuit is electrically connected. The discharge control circuit is electrically connected with the energy storage module, and the lower computer control circuit module is connected to the cloud monitoring server through the communication interface circuit module.

The guided wave transducer sends an ultrasonic guided wave to the rail, and the ultrasonic guided wave propagates along the rail and encounters a defect reflection to generate an echo signal, and the echo signal is received by the guided wave transducer as a guided wave signal for monitoring the rail, and Send to the lower computer control circuit module; the temperature sensor detects the real-time temperature of the steel rail near the guided wave transducer, and sends it to the lower computer control circuit module.

The solar panel module collects solar energy and converts it into electrical energy, and then charges the energy storage module through the charge and discharge control circuit, and then the energy storage module supplies power to the guided wave transducer, the temperature sensor and the entire control cabinet.

The browsing terminal includes a computer/mobile phone terminal.

2. A cloud platform rail integrity monitoring method based on guided wave technology, using the above system, and the method is as follows:

S1. The first part of the monitoring method will extract the characteristic signals and coefficient matrices of the acoustic emission sources at different structures of the rail through noise reduction. The steps are as follows:

S1.1: According to the geometry and physical properties of the rail to be measured, pre-set the mode and frequency of the ultrasonic guided wave used by the guided wave transducer, and pre-set the acquisition and monitoring parameters of the guided wave transducer. The energy device and temperature sensor obtain the guided wave signal and temperature signal of the monitoring rail;

The acquisition parameters of the guided wave transducer include acquisition time duration A, data storage capacity B, acquisition times C, acquisition period Ts, monitoring temperature D, monitoring time interval E, distance threshold F, total number of iterations N, amplitude threshold Z, Sampling frequency Fs and ultrasonic guided wave velocity V.

The acquisition parameters of the guided wave transducer further include a data storage capacity B, and the data storage capacity B is the data capacity of the stored guided wave signal and temperature signal.

The collection time duration A is greater than the collection period TS.

S1.2: The number of acquisitions C is carried out in the acquisition time duration A, the guided wave signal X, the guided wave signal X and the temperature signal collected at the same time in each acquisition period T _S are transmitted to the cloud monitoring server, according to C The data _format ^Y =(X ₁ , X ₂ , .

The time interval between adjacent collection time durations A is the monitoring time interval E.

The guided wave signal is essentially decomposed to represent the linear superposition of different sound source signals: Y=MR+N(t). In order to obtain the weight matrix M and the sound source signal data R, N(t) represents the noise signal data. The present invention does not need to collect and process redundant guided wave signals in different complex environments when the rail structure is complete enough, and can directly extract and obtain sound source signal data R from any guided wave signals containing damage and defects.

S1.3: For the two groups of guided wave monitoring data before and after (cross processing) acquired under different acquisition time duration A, the signal data dimensions of the two groups of guided wave monitoring data are q and w. The row direction superposition together forms a new higher-dimensional data to be analyzed Z', and then the data to be analyzed Z' is subjected to feature scaling and sparse standardization processing according to the following formula to obtain a scale-normalized multi-dimensional data with a mean value of 0 and a variance of 1 Data to be analyzed Z:

Among them, E(z) and σ are the mean and standard deviation of the data to be analyzed Z' respectively, Z _i represents the i-th group of signals in the multi-dimensional data to be analyzed Z, Z _i ' represents the i-th group of the data to be analyzed Z'Signal;

S1.4: Carry out target optimization: perform blind source separation on the multi-dimensional data to be analyzed Z, and obtain the sound source signal data representing rail damage:

First, the following formula is used to iteratively solve to obtain the weight coefficient of the sound source signal in the guided wave monitoring data:

S1.4.1: Initialize a weight coefficient W ₀ with a two-norm of 1 and an iteration count n=1;

S1.4.2: Iteratively solve according to the following formula:

W _n =E{Z(W ^T _n-1 Z) ³ }-3W _n-1 , n=n+1

Among them, W _n represents the weight coefficient vector obtained by the nth iteration, Z represents the multi-dimensional data to be analyzed, E{} represents the expectation function; δ _1n ~δ _Ln respectively represent the coefficient value in the nth weight coefficient vector W _n , L Represents the total number of strongly correlated weight coefficients;

S1.4.3: After each iterative solution, normalize the weight coefficient vector W _n , and then judge:

If |W ^T _n W _n-1 | does not converge to 1, perform step S1.4.2 again;

If |W ^T _n W _n-1 | converges to 1 and satisfies the condition that the number of iterations n is less than the total number of iterations N, the weight coefficient vector W _n under the current iteration number is output as a strongly correlated weight coefficient, and added to the coefficient In the matrix W* and continue to step S1.4.2, until the number of iterations ordinal n is equal to the total number of iterations N, the iteration is terminated, and a total of q+w groups of strongly correlated weight coefficients are obtained;

S1.4.4: Finally obtain the coefficient matrix W*, W*=(W ₁ , W ₂ ,..., W _q+w ) ^T , according to the formula R=W*×Z to obtain the sound sources representing different sound sources in the rail to be tested Signal data R, R=(r ₁ , r ₂ ,...,r _L ) ^T , r ₁ , r ₂ ,..., r _L represents each guided wave signal corresponding to the guided wave monitoring data in the sound source signal data R The sound source sub-signal of ;

S1.5: Solve the generalized inverse matrix of the coefficient matrix W* to obtain the weight matrix M, which is expressed as:

In the formula, β ₁₁ ~ β _qL respectively represent the q rows of coefficient values of the sound source signal corresponding to the guided wave monitoring data with dimension q, as a set of coefficient groups; α ₁₁ ~α _wL represent the guided wave monitoring data with dimension w The w lines of coefficient values corresponding to the sound source signal are used as another set of coefficient groups; the q lines of coefficient values and the w lines of coefficient values correspond to the signal data dimensions q and w of the two groups of guided wave monitoring data before and after respectively;

Construct the following reference matrix K:

Among them, the number of elements -1 in each column is q and the number of elements 1 is w;

S1.6: Perform similarity measurement, and calculate the similarity between a column _pi of the weight matrix M and a column _ki corresponding to the reference matrix K respectively:

For each column vector pi of the weight matrix M and each column vector _ki of the reference matrix K, the normalization process is carried out in the same way as in step S1.3 to obtain the normalized respective row and column vectors p ^* _i and k ^* _{i ,} and then according to The similarity distance is obtained by the following formula:

θ _i = |1-p ^* _i ×k ^* _i |

In the formula, θ _i represents the similarity distance between the normalized column vector p ^* _i of the weight matrix M and the normalized column vector k ^* _i corresponding to the reference matrix K;

Then, all similarity distance results are formed into a similarity distance set ξ={θ ₁ ,θ ₂ ,...,θ _L };

S2. Perform dimensionality reduction and noise reduction reconstruction to obtain a guided wave signal representing rail damage, and obtain the damage location:

S2.1: Perform feature extraction and dimension reduction processing on the weight matrix M to obtain a strongly correlated sound source signal:

According to the distance threshold F, each column vector in the weight matrix M corresponding to the similarity distance θ _i and each sound source sub-signal in the sound source signal data R are extracted from the similarity distance set ξ satisfying the inequality relationship θ _i <F, The following damage matrix P _(q+w)×H is composed of the extracted column vectors, and the following strongly correlated sound source matrix is composed of the extracted sound source sub-signals

Among them, H represents the number of strongly correlated sound source signals, as a strongly correlated sound source component;

Through the threshold value F, the present invention can automatically extract the required number of reconstructed sound source signals to obtain damage signals. The shorter the similarity distance, the more relevant the required sound source signals and the damage are.

S2.2: In the two sets of coefficient groups of the damage matrix P _(q+w)×H , the coefficient groups with relatively small matrix coefficient values are discarded, and the coefficient groups with relatively large matrix coefficients are retained, so that the damage matrix P _{( q+w)×H} extraction to obtain the construction coefficient matrix P _w×H ;

The following takes the coefficient group correspondence of w coefficient values as an example, that is, the coefficient values are relatively large:

S2.3: Reconstruct the ultrasonic guided wave damage signal Y ^defect containing the characteristic information of the track damage according to the structure coefficient matrix and the strongly correlated sound source matrix;

For the construction coefficient matrix P _w×H , the coefficient values of each column are averaged to remove interference according to the following formula, and a new one-dimensional row vector P _1×H ^NEW is obtained, so as to avoid mixing noise and abnormal interference of calculation results:

The ultrasonic guided wave damage signal Y ^defect containing the characteristic information of the track damage is further obtained according to the following formula:

S2.4: Obtain the envelope information according to the damage signal Y ^defect processing,

If the envelope amplitude in the envelope information is greater than the amplitude threshold Z, it is considered that the rail to be tested is damaged;

If the envelope amplitude in the envelope information is not greater than the amplitude threshold Z, it is considered that there is no damage to the rail to be tested;

In the case of damage to the rail to be tested, the ultrasonic guided wave damage signal Y ^defect is processed by the following formula at the acquisition time t corresponding to the damage of the rail to be tested to obtain the location of the rail damage:

s=t×V/2

t=1/Fs×i, i=0,1,…,num, num=Ts×Fs

Among them, Ts represents the acquisition period of the guided wave transducer, Fs represents the sampling frequency of the guided wave transducer, V represents the ultrasonic guided wave velocity emitted by the guided wave transducer, i represents the ordinal number of the ith acquisition, and num represents all the The total number of acquisitions, s represents the distance from the location of the damage on the rail to the guided wave transducer;

The rail integrity monitoring of the present invention takes the rail damage and its location as an indication of integrity.

The discontinuous and continuous acquisition is the acquisition of the guided wave signal and temperature with a data length of Ts×Fs and a continuous number of times C under each acquisition interval duration A according to monitoring requirements.

When the data of the guided wave signal and temperature signal stored in the control circuit module of the lower computer reaches the data storage capacity B, the big data of the ultrasonic guided wave obtained from the monitoring will be mined, and the following set cross-processing analysis method will be used to reduce the false alarm rate Find the damage:

A: The guided wave signals at the same monitoring temperature D but at different times are processed and analyzed according to the steps to obtain the location of the rail damage;

B: Under the fixed monitoring time interval E, the two groups of guided wave monitoring data whose time difference is the monitoring time interval E are processed and analyzed according to the steps to obtain the location of the rail damage;

C: The two groups of guided wave monitoring data in the stored data are randomly selected and processed and analyzed according to the steps to obtain the location of the rail damage.

The beneficial effects of the present invention are:

The invention has the characteristics of online real-time, cross-region, low cost and high robustness, which not only greatly improves the deficiencies and problems existing in the traditional detection and monitoring means and technologies, but also improves the service state evaluation and monitoring of the railway line network. The level of automation and intelligence; the inherent advantages of ultrasonic guided wave technology will greatly improve the detection and monitoring efficiency, scope and system reliability of railway tracks; the development and application of cloud-based monitoring algorithms will greatly improve the cross-platform nature of rail online monitoring and remote convenience.

The invention can not only realize the monitoring and processing of cracks and broken lines, but also can realize the monitoring trend evaluation of the corrosion-type defect information in the slow transformation. In particular, the adopted cross-processing method greatly improves the identification and tracking of corrosion-type damage , improve the monitoring accuracy, and avoid the problems that the monitoring accuracy in the corrosion growth process is insufficient and difficult to find in the traditional monitoring method.

The monitoring system of the invention has a high degree of intelligence, and can realize the tracking monitoring and real-time perception of the corrosion growth information through the cross-processing analysis method, which is beneficial to the real-time long-term monitoring of the whole life cycle of the rail.

Through algorithm optimization, the invention obtains de-redundant and mutually independent sound source signals through iteration, realizes noise reduction, greatly improves the signal-to-noise ratio of the signal, and thus improves the accuracy of the monitoring algorithm; , for the damage reconstruction, the present invention performs triple dimension reduction processing, which greatly reduces the computational complexity, reduces the monitoring time, and improves the efficiency and reliability of the monitoring algorithm. significant. One is the iterative extraction of the sound source signal by the original guided wave signal algorithm, which realizes one-fold dimensionality reduction; the other is the double noise reduction for a large number of obtained sound source signals according to the relationship of similarity distance proposed by the present invention. , and further obtain the signal of strong correlation damage, which reduces the redundancy of the traditional monitoring method; thirdly, in the damage signal reconstruction step, according to the relative magnitude relationship of the two sets of coefficient values in the damage matrix, the smaller contribution is further removed. The coefficient value is extracted to obtain a lower-dimensional construction coefficient matrix to achieve triple noise reduction. In the embodiment of the present invention, the fourth dimension reduction is further mentioned, that is, extracting a certain row vector for the finally obtained construction coefficient matrix to simplify the operation of mean value de-interference. Therefore, the computational complexity is reduced, and the efficiency and reliability of the monitoring algorithm are greatly improved, which is of great significance to the analysis of the big data obtained from the rail guided wave monitoring.

The invention considers and combines the mining problem of the large data of the rail guided waves obtained by long-term monitoring, reduces the damage false alarm rate through cross-processing analysis, and improves the monitoring accuracy. At the same time, an innovation is made in the algorithm. In the selection of the two sets of monitoring data, compared with the traditional method, the damage comparison can only be carried out with the reference database, that is, the guided wave database under the complete and healthy working state of the rail. The present invention can realize different monitoring. The processing and analysis of the guided wave data at the time, ingeniously constructing the reference matrix, realizing the measurement and extraction of the iteratively obtained sound source signal, and processing the coefficient values of the constructed coefficient matrix through the set averaging method to remove the interference and reduce the accidental factors. It can reduce the false alarm of rail damage and improve the accuracy.

At the same time, the invention does not need to establish an over-redundant reference library in the early stage of service of the rail, and can monitor as needed, and start from the current moment of monitoring to carry out multi-dimensional monitoring and analysis, so as to realize the evaluation and determination of multiple defects and damage location, using the reference library When there are multiple damages in the detection signal, it is difficult to judge the damage of the reconstructed signal in practice. In actual monitoring, the potential damage of the rail is not only many, but also in different forms, so the traditional benchmark library method will be difficult to achieve. .

It can be seen that the present invention can effectively realize the reliable monitoring of the integrity of the rail, greatly improve the cross-regional and real-time monitoring, and has important practical significance and engineering value.

Description of drawings

Fig. 1 is a system block diagram of the system of the present invention.

Figure 2 is a flow chart of the method of the present invention.

FIG. 3 is a correlation waveform diagram of an embodiment of the present invention.

FIG. 4 is a waveform diagram of a damaged signal according to an embodiment of the present invention.

FIG. 5 is a damage signal and an envelope diagram of an embodiment of the present invention.

FIG. 6 is an envelope diagram of an embodiment of the present invention.

FIG. 7 is a damage location diagram of an embodiment of the present invention.

FIG. 8 is an impairment signal according to an embodiment of the present invention.

FIG. 9 is a damage location diagram of an embodiment of the present invention.

FIG. 10 is a damage signal diagram of an embodiment of the present invention.

FIG. 11 is an actual rail monitoring signal diagram of the embodiment of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments.

As shown in Figure 1, the monitoring system includes a front-end monitoring module, a cloud monitoring server and a browsing terminal. The front-end monitoring module includes a guided wave transducer, a temperature sensor and a control cabinet. The control cabinet contains a solar panel module, a charge and discharge control circuit , energy storage module, transmission interface circuit module, lower computer control circuit module, communication interface circuit module, guided wave transceiver module and signal conditioning module; among them, the guided wave transducer and temperature sensor are installed on the rail to be tested, and the lower computer The control circuit module is electrically connected to the temperature sensor, the charge and discharge control circuit and the guided wave transceiver module through the transmission interface circuit module, respectively. The guided wave transducer is electrically connected to the guided wave transceiver module through the signal conditioning module, and the solar panel module is connected to the charge and discharge control circuit. Electrical connection, the charging and discharging control circuit is electrically connected with the energy storage module, and the lower computer control circuit module is connected to the cloud monitoring server through the communication interface circuit module.

The solar panel module collects solar energy and converts it into electrical energy, and then charges the energy storage module through the charge and discharge control circuit, and then the energy storage module supplies power to the guided wave transducer, the temperature sensor and the entire control cabinet. The guided wave transducer sends an ultrasonic guided wave to the rail. After the ultrasonic guided wave propagates along the rail, it encounters defects and reflects and generates an echo signal. The conditioning module adjusts, and the guided wave transceiver module receives and sends it to the lower computer control circuit module, and the lower computer control circuit module transmits the received data to the cloud monitoring server for storage; the temperature sensor detects the real-time temperature of the rail near the guided wave transducer, Send to the lower computer control circuit module. The browsing terminal includes a computer/mobile terminal. The computer/mobile terminal is connected to access the cloud monitoring server, and receives alarm and message information.

As shown in Fig. 2, the embodiment implemented according to the complete method of the present invention is as follows:

Example 1:

As shown in Figure 3 to Figure 7, the cloud platform performs monitoring processing and analysis according to the guided wave data obtained by the monitoring front-end monitoring.

Here, a large amount of guided wave monitoring data, that is, guided wave big data, is further mined and analyzed through cross processing. Here, two sets of track waist monitoring data ^{Yi and Y j with} an interval of E are taken for monitoring and analysis, where i=22 , j=22 monitoring data, standardize according to step S1.3, after processing according to step S1.4.1 to step S1.4.3 in claim 5, the dimension of the obtained coefficient matrix W* is 44×23, for 44 For the guided wave signal of 3 dimensions, after being processed in step S1.4.4 of claim 5, a total of 23 characteristic sound source information can be obtained. After programming in MATLAB, each group of coefficient data of the characteristic signal can be visualized, a total of 23 groups .

In order to obtain θ _i = |1-p ^* _i ×k ^* _i |, the data needs to be normalized here, so as to avoid inaccurate results due to the irregular size of the data itself. Here we take covariance calculation as an example to illustrate, let Y ^* _i = k×Y _i , Y ^* _j = k×Y _j , obviously cov(Y ^* _i , Y ^* _j )=k2cov(Y _i , Y _j ), it can be seen that the data itself is not uniform enough, resulting in the same characteristics The data have mixed results. That is, in order to avoid the influence of the numerical value of _pi and ki on the similarity, it is necessary to process _pi and _ki according to the standardized formula in step S1.3 to obtain the processed data p ^* _i and _k ^* _i . Here, it needs to be processed according to the standardized formula in step S1.3 in claim 5, to obtain new data p ^* _i and k ^* _i .

Further according to step S1.6, the similarity distance set ξ={θ ₁ , θ ₂ ,...,θ _L } is obtained, and the trend curve of the similarity is obtained, which is displayed according to the correlation distance from low to high, as shown in Figure 3 . The black dotted line in Figure 3 is the distance threshold F set in step S2.1, and the parameter θ _i is also drawn in Figure 3. It can be intuitively seen that the part of the coefficient that satisfies the setting relationship θ _i <F, the smaller the corresponding weight matrix. The coefficients and sound source components in M characterize the damage better. Next, the damage signal Y ^defect is obtained according to steps S2.1 to 2.3, as shown in FIG. 4 .

Go to step 2.4 to obtain a clear waveform diagram of the defect position, as shown in Figure 5 and Figure 6, according to the set amplitude threshold Z, determine that the rail is damaged, and give the damage position, as shown in Figure 7, to complete the positioning evaluation.

In further actual detection, for step S2.3, in order to improve the monitoring efficiency, it can also be implemented in this way, the damage signal can be constructed by a certain row of coefficient values of the construction coefficient matrix P _{w × H} , and the coefficient group of w coefficient values corresponds to For example, it is expressed as follows:

可以求得含有轨道损伤特征信息的超声导波损伤信号Y^defect，进一步步骤S2.4再由Y^defect处理获得包络信息，通过设置阈值得到伤损位置信息，完成监测。可以降低运算的计算复杂度，提高计算速度。Here, the mean value de-interference operation can be simplified to construct damage by extracting any row vector. Here, taking the i-th row of the construction coefficient matrix as an example, a new one-dimensional row vector P _1×H ^NEW =[α _i1 α _ii … α _iH ], then according to

The ultrasonic guided wave damage signal Y ^defect containing the characteristic information of the track damage can be obtained, and in further step S2.4, the envelope information is obtained by Y ^defect processing, and the damage position information is obtained by setting a threshold to complete the monitoring. It can reduce the computational complexity of the operation and improve the computational speed.

Example 2:

Figure 11 shows the actual rail waist monitoring signal, the signal-to-noise ratio is low, and it is difficult to identify the damage.

The present invention can also monitor, evaluate and locate multiple damages. Here, for example, the actual conditions of the rails containing 4 damages and growing into 5 damages are detected, analyzed and compared based on the present invention and the traditional benchmark library method, that is, damage growth. The process is monitored by cross-processing analysis. In actual monitoring, the guided wave acquisition mode is set as excitation and emission by the transducer at one end of the rail, and the transducer at the other end receives the rail guided wave signal to obtain the collected data, which is transmitted from the front-end monitoring module to the cloud monitoring server.

The specific implementation of the present invention adopts the cross processing method, and four damaged guided wave signals and five damaged guided wave signals can be selected for processing and analysis. In fact, this situation can be equivalent to 0 damage and 1 damage. After the method is processed, the feature sound source signal and coefficient extraction are completed according to steps S1.3 to S1.6, the reconstructed signal in Fig. 8 is obtained in steps S2.1 to S2.3, and the signal in Fig. If the damage is determined and located, the newly added signal of the fifth damage can be obtained, so a better processing result can be obtained.

By analogy, when there are more damages, such as nine, ten, etc., this method can better distinguish by cross-processing analysis and selecting different analysis data. However, when using the traditional analysis method based on the reference library, since the guided wave signal of the reference library is complete and does not contain damage, the guided wave monitoring signal of the rail is equivalent to 0 when the analysis contains five damages. In the case of one damage and five damages, it is necessary to find out five kinds of damage information at the same time. Considering that in actual propagation, multiple back-and-forth reflections, superposition cancellation and conversion of the guided wave signal (different acoustic emission sources) lead to the reconstruction of the five features. , the position of each damage cannot be obtained from the obtained guided wave signal containing five features, as shown in Figure 10 Y ^defect , all signals are aliased together, superimposed and canceled each other, and it is difficult to determine the damage position.

By analogy, the traditional monitoring method based on the benchmark library or the sample library can complete the damage assessment when there are few damages. In practical applications, as the monitoring continues, the amount of data increases and the damage grows and increases. Reliable monitoring and analysis of rail damage will be difficult to accomplish.

At the same time, in the present invention, the defined steps S1.5 to S1.6 use the similarity distance θ _i = |1-|p ^* _i ×k ^* _i || to measure the similarity between the source signal and the rail damage, by taking the absolute The value method makes the two sets of guided wave data used in step S1.3 have no sequence and signal source requirements. The traditional method requires these two sets of data to have sequence and source requirements, because the traditional method is based on the reference library or sample library. That is, the first group of data needs to be the data in the benchmark library or the sample library, and the second group is the data from damage monitoring, that is, the traditional method defaults that the first group of data is non-damaged, and the second group of data needs to be analyzed. The signal is damaged, which limits the application scope of the traditional method, but the present invention does not need to consider it, and the present invention has no special requirements for the sequence of the two sets of data.

It can be concluded that the monitoring data is cross-processed. According to the preset settings, a horizontal comparison is made at a given monitoring temperature D, and the rail guided wave signals in different monitoring periods are processed and analyzed according to the monitoring steps to obtain the rail monitoring status, so as to realize different preset monitoring. During the period of time, under the same temperature conditions, the detection and positioning of the service state; it is also possible to monitor and analyze the two sets of monitoring data Y _i and Y _j with the interval E under the monitoring time interval E according to the preset setting, and obtain the service state of the rail. Analysis results; can also randomly select two groups of data Y _i and Y _j in the stored data according to the preset and perform cross processing according to the monitoring processing method to obtain the service status of the monitored rails.

The above-mentioned specific embodiments are used to explain the present invention, rather than limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims all fall into the protection scope of the present invention.

Claims (3)

1. A cloud platform steel rail integrity monitoring method based on a guided wave technology adopts a cloud platform steel rail integrity monitoring system based on the guided wave technology, the monitoring system comprises a front end monitoring module, a cloud monitoring server and a browsing terminal, the front end monitoring module comprises a guided wave transducer, a temperature sensor and a control cabinet, and a solar panel module, a charging and discharging control circuit, an energy storage module, a transmission interface circuit module, a lower computer control circuit module, a communication interface circuit module, a guided wave receiving and sending module and a signal conditioning module are arranged in the control cabinet; the system comprises a power supply, a transmission interface circuit module, a lower computer control circuit module, a temperature sensor, a charge-discharge control circuit, a guided wave transceiving module, a solar panel module, an energy storage module, a transmission interface circuit module, a solar panel module, a power supply module and a cloud monitoring server, wherein the guided wave transducer and the temperature sensor are both arranged on a steel rail to be detected;

the guided wave transducer sends ultrasonic guided waves to the steel rail, the ultrasonic guided waves are transmitted along the steel rail and then encounter defects to be reflected to generate echo signals, and the echo signals are received by the guided wave transducer to serve as guided wave signals for monitoring the steel rail and sent to the lower computer control circuit module; the temperature sensor detects the real-time temperature of the steel rail near the guided wave transducer and sends the real-time temperature to the lower computer control circuit module;

the method is characterized in that:

s1, the steps are as follows:

s1.1: according to the geometric and physical properties of the steel rail to be detected, the mode and frequency of ultrasonic guided waves adopted by a guided wave transducer are preset, the acquisition monitoring parameters of the guided wave transducer are preset, and guided wave signals and temperature signals of the monitored steel rail are obtained by discontinuously and continuously acquiring the guided wave transducer and a temperature sensor;

s1.2: the acquisition times C are carried out in the acquisition time duration A, and each acquisition period T_SInterior guided wave signal X, the temperature signal that guided wave signal X and same time were gathered transmit to high in the clouds monitoring server, and data format Y according to C signal data dimension is (X ═₁，X₂，…，X_c)^TStoring to form guided wave monitoring data;

s1.3: for front and rear groups of guided wave monitoring data acquired under different acquisition time durations A, the signal data dimensions of the two groups of guided wave monitoring data are q and w, the two groups of guided wave monitoring data are overlapped together according to the row direction to form new higher-dimensional data Z 'to be analyzed, and then the data Z' to be analyzed is subjected to standardized processing of feature scaling and sparse processing according to the following formula to obtain scale-normalized multidimensional data Z to be analyzed, wherein the mean value of the data Z to be analyzed is 0, and the variance of the data Z to be analyzed is 1:

wherein E (Z) and σ are the mean and standard deviation of the data Z' to be analyzed, Z_iRepresenting the i-th set of signals in the multi-dimensional data Z to be analyzed, Z_i'represents the ith set of signals in the data Z' to be analyzed;

s1.4: carrying out target optimization: blind source separation is carried out on the multidimensional data Z to be analyzed, and sound source signal data representing steel rail damage are obtained:

firstly, the following formula is adopted to iteratively solve and obtain the weight coefficient of the sound source signal in the guided wave monitoring data:

s1.4.1: initializing a weight factor W with a two-norm 1₀And an iteration count n ═ 1;

s1.4.2: the iterative solution is performed according to the following formula:

wherein, W_nRepresenting a weight coefficient vector obtained by nth iteration, Z representing multidimensional data to be analyzed, and E { } representing an expectation function; delta_1n～δ_LnRespectively represent the nth weight coefficient vector W_nL represents the total number of strongly correlated weight coefficients;

s1.4.3: after each iteration solution, weighting coefficient vector W is calculated_nCarrying out normalization processing, and then judging:

if W^T _nW_n-1If | does not converge to 1, then step S1.4.2 is re-executed;

if W^T _nW_n-1If | converges to 1 and satisfies the condition that the iteration number N is less than the total iteration number N, outputting the weight coefficient vector W under the current iteration number_nAdding the strong correlation weight coefficients into the coefficient matrix W and continuing to carry out step S1.4.2 until the iteration number N is equal to the total iteration number N, and terminating the iteration to obtain q + W groups of strong correlation weight coefficients;

s1.4.4: finally, a coefficient matrix W, W ═ W (W) is obtained₁，W₂，…，W_q+w)^TObtaining sound source signal data R representing different sound sources in the steel rail to be measured according to a formula R-W- × Z, wherein R-W (R-R)₁,r₂,...,r_L)^T，r₁,r₂,...,r_LA source subsignal representing each guided wave signal corresponding to the guided wave monitoring data in the source signal data R;

s1.5: solving the generalized inverse matrix of the coefficient matrix W to obtain a weight matrix M, which is expressed as:

in the formula, β₁₁～β_qLThe guided wave monitoring data respectively representing the dimension q corresponds to q rows of coefficient values of the acoustic source signal as a set of coefficient groups α₁₁～α_wLThe guided wave monitoring data with the dimension w corresponds to the w rows of coefficient values of the sound source signal as another group of coefficient groups; the q rows of coefficient values and the w rows of coefficient values respectively correspond to signal data dimensions q and w of the front and rear groups of guided wave monitoring data;

the following reference matrix K is constructed:

wherein the number of the element-1 in each column is q and the number of the element-1 in each column is w;

s1.6: performing similarity measurement, and respectively calculating a column p of the weight matrix M_iOne column K corresponding to the reference matrix K_iSimilarity between:

for each column vector p of the weight matrix M_iAnd each column vector K of the reference matrix K_iBy normalizing to obtain normalized respective row-column vectors p^* _iAnd k^* _iThen, the similarity distance is obtained according to the following formula:

θ_i＝|1-p^* _i×k^* _i|

in the formula, theta_iA normalized column of vectors p representing a weight matrix M^* _iNormalized column of vectors K corresponding to reference matrix K^* _iSimilarity distance between them;

all similarity distance results are then grouped into a similarity distance set ξ ═ θ₁,θ₂,...,θ_L}；

S2, reducing the dimensions and denoising, reconstructing to obtain guided wave signals representing the damage of the steel rail, and obtaining the damage position:

s2.1: and (3) performing feature extraction and dimension reduction processing on the weight matrix M to obtain a strongly-correlated sound source signal:

extracting the similarity distance set ξ satisfying the inequality relation theta according to the distance threshold value F_i<Distance of similarity of F θ_iThe column vectors in the corresponding weight matrix M and the sound source sub-signals in the sound source signal data R form the following impairment matrix P from the extracted column vectors_(q+w)×HFrom the extracted source sub-signals, the following strongly correlated source matrix is formed

Wherein H represents the number of strongly correlated sound source signals as strongly correlated sound source components;

s2.2: damage matrix P_(q+w)×HThe two coefficient groups of the matrix are abandoned, and the coefficient group with relatively larger matrix coefficient is reserved, so that the matrix P is damaged_(q+w)×HExtracting to obtain a structural coefficient matrix P_w×H；

The following takes as an example a coefficient group correspondence of w coefficient values, i.e. coefficient values which are relatively large:

s2.3: reconstructing ultrasonic guided wave damage signal Y containing track damage characteristic information according to construction coefficient matrix and strongly-correlated sound source matrix^defect；

For the constructed coefficient matrix P_w×HTaking the mean value to remove the interference to the coefficient value of each column according to the following formula to obtain a new one-dimensional row vector P_1×H ^NEW：

Further, an ultrasonic guided wave damage signal Y containing track damage characteristic information is obtained according to the following formula^defect：

S2.4: according to the damage signal Y^defectThe processing obtains the envelope information and the envelope information,

if the part of the envelope information, of which the envelope amplitude is greater than the amplitude threshold value Z, determining that the steel rail to be detected is damaged;

if the part of the envelope information, of which the envelope amplitude is not greater than the amplitude threshold value Z, determining that the steel rail to be detected has no damage;

under the condition that the steel rail to be detected is damaged, the ultrasonic guided wave damage signal Y^defectAnd (3) processing the acquisition time t corresponding to the damage of the steel rail to be detected by adopting the following formula to obtain the positioning of the damage of the steel rail:

s＝t×V/2

t＝1/Fs×i，i＝0,1,…,num，num＝Ts×Fs

wherein Ts represents the acquisition period of the guided wave transducer, Fs represents the sampling frequency of the guided wave transducer, V represents the ultrasonic guided wave speed sent by the guided wave transducer, i represents the ordinal number of the ith acquisition, num represents the total number of all the acquisitions, and s represents the distance from the position of the damage existing in the steel rail to the guided wave transducer.

2. The method and the system for monitoring the integrity of the cloud platform steel rail based on the guided wave technology are characterized in that the discontinuous continuous acquisition is the acquisition of guided wave signals and temperature of which the times C of continuous development and the data length are Ts × Fs at each acquisition interval time A according to monitoring requirements.

3. The method and the system for monitoring integrity of the steel rail of the cloud platform based on the guided wave technology are characterized in that: when the data of the guided wave signals and the temperature signals stored by the lower computer control circuit module reach a data storage amount B, the following set cross processing analysis method is adopted to reduce the missing report rate and find out the damage:

a: processing and analyzing the guided wave signals at the same monitoring temperature D but different time according to the steps to obtain the location of the steel rail damage;

b: under a fixed monitoring time interval E, processing and analyzing two groups of guided wave monitoring data with the time difference of the monitoring time interval E according to the steps to obtain the location of the damage of the steel rail;

c: and randomly extracting two groups of guided wave monitoring data in the stored data, and processing and analyzing the two groups of guided wave monitoring data according to the steps to obtain the positioning of the steel rail damage.

Images