CN110991828A - Steel rail fastener state detection method based on information entropy theory - Google Patents

Abstract

The invention provides a steel rail fastener state detection method based on an information entropy theory, and relates to the technical field of steel rail state detection. According to the invention, the acceleration sensor is used for collecting the acceleration of the rail in the vertical direction at a set frequency and uploading the acceleration to the cloud server. And for the acquired data, after removing a trend term, measuring the chaos or dispersion degree of the steel rail system by adopting a CSAE concept, and quantitatively representing the uniformity of the energy distribution of the system. Judging the state of the steel rail fastener according to the CSAE value: the CSAE value is smaller when the steel rail fastener is loosened or falls off than when the steel rail fastener is in a normal screwing state. The method can greatly reduce the consumption of human resources, and has the advantages of timely information feedback and low hardware cost.

Description

Steel rail fastener state detection method based on information entropy theory

Technical Field

The invention relates to the technical field of steel rail state detection, in particular to a steel rail fastener state detection method based on an information entropy theory.

Background

The detection of the state of the railway fastener is important to the railway safety, but the method for manually detecting the state of the railway fastener cannot complete heavy railway maintenance tasks and cannot follow the trend of high-speed development of railway technology. There are many image-based fastener inspection methods available today. Yang and Feng et al provide image processing and pattern recognition techniques for detecting rail fastener defects. Mazzeo et al describe a vision-based system that automatically detects the absence or presence of tie bolts that secure the rails to the crossties. These methods, however, do not detect the extent of fastener loosening because they take a view of the entire fastener from the top. Some researchers have focused on the method of structured light, with the camera acquisition producing a pattern of structured light on the target. Garcia et al propose a structured light based track measurement system for assessing gauge and detecting missing track fasteners. Aytekin et al have constructed a real-time railway fastener detection system using a structured light sensor to detect missing hex-head fasteners. None of the above methods based on structured light sensors accurately detect the absence of a rail fastener, and none identify a partially worn or loose fastener. Mao et al propose a method for strictly detecting fasteners of a high-speed railway based on a structured light sensor. And acquiring accurate and dense fastener point clouds from the structured light sensor, and detecting not only missing fasteners but also partially worn or loosened fasteners by using the decision tree classifier. The defect is that the precision is influenced by the external environment, and the sensitivity to the loosening degree of the fastener is lower. Gibert et al propose a new railway fastener detection algorithm, and explore the advantages of a deep volume and a neural network in classification and identification.

The structural damage identification based on the vibration data has wide application prospect in structural health monitoring. The vibration characteristics of an object can be affected by changes in the physical properties of the structure. In recent years, many scholars have analyzed the dynamic response of a wheel-rail system in the event of a failure of an under-rail support (fastener or tie) by establishing different dynamic models, and have made many advances in the detection of fasteners. Xu et al suggest that failure of a rail fastener enhances the interaction between the wheel and rail, and that this interaction increases significantly as the speed of train operation increases. Wang et al propose a method for identifying rail fastener looseness that uses a self power density method to analyze vibration signals of four acceleration sensors mounted on the head of a rail. Zhou and Wei et al propose a wavelet-based energy spectrum method for wavelet decomposition of vibration signals to identify frequency domain energy changes before and after fastener loosening. This method can identify loose fasteners, but the accuracy is related to the selection of the wavelet. Zhao et al have designed a wireless transmission's data acquisition system to strain voltage is the evaluation index, can detect the rail fastener of disappearance or fracture. However, the system needs to be installed at the bottom of the rail, and installation and movement are difficult.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a steel rail fastener state detection method based on an information entropy theory aiming at the defects of the prior art, so as to realize the detection of the state of a railway fastener.

In order to solve the technical problems, the technical scheme adopted by the invention is as follows:

the invention provides a steel rail fastener state detection method based on an information entropy theory, which comprises the following steps of:

step 1: installing lambda acceleration sensors at rail web positions above the rail fasteners and between the rail fasteners, collecting acceleration of the rail in the vertical direction at a set frequency by using the acceleration sensors, taking the acceleration of the rail in the vertical direction as original data, and uploading the original data to a cloud server;

step 2: removing trend items from the acquired original data, measuring the chaos or dispersion degree of the steel rail system by adopting a CSAE concept, and further quantitatively representing the uniformity of the energy distribution of the steel rail system to obtain a CSAE value;

step 2.1: obtaining data after removing the trend item by using an R language;

step 2.2: from the data obtained in step 2.1, count data { θ } ═ θ } of the acceleration signal is obtained by a two-dimensional division method₁,θ₂,…,θ_n}；

Step 2.3: calculating the frequency corresponding to the divided counting data;

the probability of the signal { theta } is pi respectively₁，…，π_nSimultaneously has pi_i> 0 and sigma pi_iIf 1 is true, the solution method is:

wherein, pi_iIs the frequency corresponding to the ith count data, n is the number of count data, θ_iAnd theta_jRespectively representing the ith and jth count data;

step 2.4: calculating a CSAE value;

step 2.4.1: the Horvitz-Thompson estimator is described as:

wherein A is_iIndicating that the sample contains the ith unit of event, I (A)_i) Is an index function, i.e. when event A_iIs true time I (A)_i) 1, otherwise I (a)_i)＝0；H^HTA value calculated for the Horvitz-Thompson estimator;

step 2.4.2: the Horvitz-Thompson estimator is combined with the traditional information entropy to obtain the amplitude spectrum entropy CSAE, and the following formula is shown:

wherein H^CSIs a CSAE (carrier-assisted laser ablation imaging system),

is the corrected frequency;

and step 3: judging the state of the steel rail fastener according to the CSAE value: when the CSAE value is larger than the CSAE value of the same position in a normal screwing state, judging that the steel rail fastener at the current moment is loosened or falls off;

the specific steps of the step 2.1 are as follows:

step 2.1.1: in the R language, loading a pragma packet, setting an order value interval, namely setting a polynomial order m, and solving each order coefficient of the polynomial by using a polyfit function according to original acceleration data acc and time t;

step 2.1.2: using a polyfal fitting polynomial according to each order coefficient of the polynomial and time t to obtain acceleration data p _ acc after polynomial fitting; and finally obtaining acceleration data acc _1 after removing the trend term, wherein the formula is as follows:

acc_1＝acc-p_acc。

the specific steps of step 2.2 are as follows:

step 2.2.1: determining the resonant frequency f of the fastener and the rail_k；

Step 2.2.2: the vibration signal from which the trend term is removed is set to { X } ═ X₁,…,x_kAt 1/f on the time axis_kSetting an amplitude axis interval P for interval size division, and dividing a time domain plane where the X is located into a plurality of equal-area parts according to the amplitude and the frequency of the signal;

step 2.2.3: by theta_iThe number of X components in each region is expressed, and count data of the vibration signal is obtained in the form of { θ } -, where θ is expressed₁,θ₂,…,θ_n}。

Adopt the produced beneficial effect of above-mentioned technical scheme to lie in: according to the steel rail fastener state detection method based on the information entropy theory, a sensor collects vibration generated by a steel rail, and data are analyzed by utilizing the Chao-Shen information entropy theory, so that state information of a rail fastener is obtained; the method can greatly reduce the consumption of human resources, and has the advantages of timely information feedback and low hardware cost.

Drawings

FIG. 1 is a flow chart of a method provided by an embodiment of the present invention;

FIG. 2 is a schematic diagram of a track mounted sensor provided by an embodiment of the present invention;

FIG. 3 is a three-dimensional view of a rail mounting location provided by an embodiment of the invention

FIG. 4 is a schematic diagram of processing data by a two-dimensional segmentation method according to an embodiment of the present invention;

FIG. 5 is a diagram illustrating the increased robustness of CASE according to an embodiment of the present invention, wherein a is a data graph with singularities, and b is a time-domain segmentation graph of a signal;

FIG. 6 is a comparison graph of measured data provided by an embodiment of the present invention; wherein, a is a schematic diagram of measured data of an excitation position 1, and b is a schematic diagram of measured data of an excitation position 2;

where 1-fastener, k 1-actuated position 1, k 2-actuated position 2.

Detailed Description

The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

As shown in fig. 1, the method of the present embodiment is as follows.

The invention provides a steel rail fastener state detection method based on an information entropy theory, which comprises the following steps of:

step 1: installing lambda acceleration sensors at rail web positions above the rail fasteners and between the rail fasteners, collecting acceleration of the rail in the vertical direction at a set frequency by using the acceleration sensors, taking the acceleration of the rail in the vertical direction as original data, and uploading the original data to a cloud server; the whole device is small in size and can be conveniently and flexibly arranged at a plurality of positions of the rail.

In this embodiment, an acceleration sensor (MMA7361) is deployed at a rail web (MMA7361 is a low-power-consumption and low-profile capacitive digital acceleration sensor chip, and a three-axis micro-mechanical accelerometer is adopted), as shown in fig. 2 to 3, the sensor collects vibration data of a rail in a vertical direction (Z direction of the sensor) at a certain sampling frequency. And uploading to the cloud. Since the maximum frequency of rail vibrations occurs at around 1300Hz, the data frequency that can be analyzed should reach around 1500Hz, from which the sampling frequency of the plant arrangement is set to 3300 Hz.

Using a digital torque wrench to adjust the torque of the target fastener to the magnitude shown in table 1 while adjusting the tightening torque of the other fasteners to about 120N · m, where 140N · m corresponds to a fully tightened state of the rail fastener; 0N · m corresponds to a state in which the rail fastener is completely detached. Under each condition, each excitation point k1 and k2 was subjected to at least 5 excitations, respectively, and experimental data were collected.

TABLE 1 test conditions

Working conditions	moment/N.m	Degree of loosening/%)
			1	0	100
2	20	85
			3	55	60
4	85	40
			5	110	20
6	140	0

Step 2: removing trend items from the acquired original data, measuring the chaos or dispersion degree of the steel rail system by adopting a CSAE concept, and further quantitatively representing the uniformity of the energy distribution of the steel rail system to obtain a CSAE value; csae is the content about "Chao-Shen information entropy"; e is entcopy a is amplitude;

step 2.1: when processing data, firstly removing a nonlinear trend generated by temperature and the like, and obtaining data after removing a trend term by using an R language; the de-trending process may eliminate the effect of offsets generated by the sensors while acquiring data on the post-calculations. Raw data is collected from the rails and removing trends from the data can focus the analysis on fluctuations in the data trends themselves. The specific method comprises the following steps:

step 2.1.1: in the R language, loading a pragma packet, setting an order value interval, namely setting a polynomial order m, and solving each order coefficient of the polynomial by using a polyfit function according to original acceleration data acc and time t; the statements are as follows:

p＝polyfit(t,acc,m)

wherein t is time, acc is original acceleration data, and m is polynomial order;

step 2.1.2: obtaining polynomial fitted acceleration data p _ acc by using a polyfal fitting polynomial according to each order coefficient of the polynomial and time t, wherein the polynomial fitted acceleration data p _ acc is described as follows:

p_acc＝polyval(p,t)

wherein, p _ acc is acceleration data after polynomial fitting, and p is the coefficient of each order of the polynomial;

and finally obtaining acceleration data acc _1 after removing the trend term, wherein the formula is as follows:

acc_1＝acc-p_acc

step 2.2: changes in the physical structure result in changes in the vibration signal that are reflected in the energy distribution of the acceleration signal in the time domain. The original vibration data has the characteristic of high precision and can reflect the stability of a vibration system most, but the original vibration data has the defect that a negative value is difficult to directly apply an information entropy theory. In order to apply the shannon entropy theory to the vibration data, it is first necessary to obtain count data reflecting the distribution of the acceleration signal. From the data obtained in step 2.1, count data { θ } ═ θ } of the acceleration signal is obtained by a two-dimensional division method₁,θ₂,…,θ_n}；

Step 2.2.1: determining the resonant frequency f of the fastener and the rail_k；

Step 2.2.2: the vibration signal from which the trend term is removed is set to { X } ═ X₁,…,x_kAt 1/f on the time axis_kSetting an amplitude axis interval P for interval size division, selecting a proper interval by P according to the size of P, and dividing a time domain plane where X is located into a plurality of equal-area parts according to the amplitude and the frequency of a signal; in this study, the acceleration amplitude is divided equally into 8 sections, the time range is divided equally into 35 sections, and finally the time domain is divided into 8 × 35 subfields. As shown in fig. 4;

step 2.2.3: by theta_iThe number of X components in each region is expressed, and count data of the vibration signal is obtained in the form of { θ } -, where θ is expressed₁,θ₂,…,θ_n}；

Step 2.3: calculating the frequency corresponding to the divided counting data;

the probability of the signal { theta } is pi respectively₁,…,π_nSimultaneously has pi_i> 0 and sigma pi_iIf 1 is true, the solution method is:

wherein, pi_iIs the frequency corresponding to the ith count data, n is the number of count data, θ_iAnd theta_jRespectively representing the ith and jth count data;

the shannon entropy in natural units is then given by:

wherein H (U) is information entropy,. pi_iThe frequency corresponding to the ith count data.

Since the count data is sparse data with many 0 s, the sparse data is corrected by the following formula:

wherein f is₁Is the number of counts in the data other than 0, i.e. θ_iThe number of the non-equal to 0,

denotes the corrected frequency, π_iThe frequency corresponding to the ith count data.

Step 2.4: calculating a CSAE value;

step 2.4.1: the signal with singular points has great influence on the distribution of the signal in the time domain interval, and the information entropy is a method for quantizing the time domain distribution, so that the singular data can have great influence on the result if the singular data are not processed well. The CSAE method obtains the counting data of the original data by carrying out unit division on the original data in a time domain plane according to the amplitude and the frequency of the original data, not only keeps the distribution condition of the data, but also can eliminate the influence caused by signal singularity. Therefore, the CSAE estimation method is invented, and combines a Horvitz-Thompson estimator with the traditional information entropy. The Horvitz-Thompson estimator is described as:

wherein A is_iIndicating that the sample contains the ith unit of event, I (A)_i) Is a normal index function, i.e. when event A_iIs true time I (A)_i) 1, otherwise I (a)_i)＝0；H^HTA value calculated for the Horvitz-Thompson estimator;

chao and Shen propose a new estimation method for Shannon entropy. The method combines the Horvitz-Thompson estimator with Good-Turing empirical unit probability correction. The Horvitz-Thompson estimator can reduce the influence caused by serious uneven distribution of data;

the citation paper Nonparametric evaluation of Shannon's index of diversity where area unseen speciales in sample;

step 2.4.2: the Horvitz-Thompson estimator is combined with the traditional information entropy to obtain the amplitude spectrum entropy CSAE, and the following formula is shown:

wherein H^CSIs a CSAE (carrier-assisted laser ablation imaging system),

is the corrected frequency;

and step 3: judging the state of the steel rail fastener according to the CSAE value: when the CSAE value is larger than the CSAE value of the same position in a normal screwing state, judging that the steel rail fastener at the current moment is loosened or falls off; because the CSAE values of different types of rails and different positions of the same type of rails are different, the comparison of the relative sizes of the CSAEs of the same signal source is more meaningful. If the fastener loosens slightly, this results in only a small change in CSAE, indicating that slight loosening has little effect on the system architecture. Meanwhile, because CSAE may fluctuate significantly when the fastener conditions are normal, it is difficult for the CSAE to accurately identify fastener conditions with a loosening coefficient of less than 60%. When the fastener is severely loosened and even falls off, the CSAE value is significantly increased, which can be used for accurate identification. Especially, when the fastener is nearly stripped, the system structure is seriously damaged, and the CSAE change amplitude is large.

Measuring the distribution uniformity of the signal in the time domain according to the CSAE, and reflecting the time complexity of the signal energy distribution; the vibration signals obtained by train wheels passing through a steel rail are distributed on a time domain plane and depend on two parameters of the amplitude and the attenuation speed of the vibration signals; when the fastener is normally locked, the vibration amplitude of the steel rail is small, the attenuation speed is high, the time domain energy distribution is concentrated, and the CSAE value is small; on the contrary, when the fastener is loosened, the vertical constraint borne by the steel rail is reduced, the amplitude is increased, and the attenuation rate is reduced, so that the distribution of vibration signals on the time domain is more dispersed, which causes the low stability of a system formed by the steel rail and the fastener for fixing the steel rail and generates relatively high entropy; as the stiffness scaling factor decreases, the failure factor becomes greater, and the CSAE value gradually increases, as shown in fig. 6; this demonstrates that the CSAE method has good recognition capability for spring fasteners of different stiffness and damping.

For the original signals obtained under each set of test conditions, a Chao-Shen entropy method was used to obtain CSAE values, and the calculated CSAE mean values were used as final results for comparison, with the results shown in fig. 6. It can be observed that when the fastener is severely loosened and even comes off, the CSAE value increases significantly, which can be used for accurate identification. Particularly, when the fastener is close to falling off, the system structure is seriously damaged, and the CSAE change amplitude is large; the CSAE values obtained at the positions of the special observation points (the sensor S1-the sensor S5) where the rails have no crosstie support are larger than the CSAE values obtained at the positions where the rails have the crosstie support. The reason is that the constraint force at the supporting point is large, the vibration response is weak, and the CSAE is small; this vibration-based structural health monitoring method can detect not only missing fasteners, but also partially loose fasteners.

The Chao-Shen information entropy theory is applied to the state recognition of the rail fastener. Experimental results show that the time domain attenuation speed of the rail vibration is reduced due to the loosening of the fastening piece, and a large entropy value is generated. By adopting an analysis method of the amplitude spectrum entropy, the loosening state of the fastening piece can be accurately identified, and the intelligent fastening piece monitoring system is greatly helpful for maintaining the railway.

The CSAE method obtains the counting data of the original data by carrying out unit division on the original data in a time domain plane according to the amplitude and the frequency of the original data, not only keeps the distribution condition of the data, but also can eliminate the influence caused by signal singularity. As shown in fig. 5, the singular points in the time domain are only divided into one unit (assuming that the index is i), and the count data is 1, the influence thereof is further weakened when calculating the frequency (pi i is 1/n, n is the total). For the data shown in fig. 5(b) and fig. 5(b), the amplitude entropy calculated by using the CSAE method is 5.5152 and 5.5286, respectively, which shows that the CSAE method has strong robustness.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions and scope of the present invention as defined in the appended claims.

Claims (3)

1. A steel rail fastener state detection method based on an information entropy theory is characterized in that: the method comprises the following steps:

step 1: installing lambda acceleration sensors at rail web positions above the rail fasteners and between the rail fasteners, collecting acceleration of the rail in the vertical direction at a set frequency by using the acceleration sensors, taking the acceleration of the rail in the vertical direction as original data, and uploading the original data to a cloud server;

step 2: removing trend items from the acquired original data, measuring the chaos or dispersion degree of the steel rail system by adopting a CSAE concept, and further quantitatively representing the uniformity of the energy distribution of the steel rail system to obtain a CSAE value;

step 2.1: obtaining data after removing the trend item by using an R language;

step 2.2: from the data obtained in step 2.1, count data { θ } ═ θ } of the acceleration signal is obtained by a two-dimensional division method₁,θ₂,…,θ_n}；

Step 2.3: calculating the frequency corresponding to the divided counting data;

the probability of the signal { theta } is pi respectively₁,…,π_nSimultaneously has pi_i> 0 and sigma pi_iIf 1 is true, the solution method is:

wherein, pi_iIs the frequency corresponding to the ith count data, n is the number of count data, θ_iAnd theta_jRespectively representing the ith and jth count data;

step 2.4: calculating a CSAE value;

step 2.4.1: the Horvitz-Thompson estimator is described as:

wherein A is_iIndicating that the sample contains the ith unit of event, I (A)_i) Is an index function, i.e. when event A_iIs true time I (A)_i) 1, otherwise I (a)_i)＝0；H^HTA value calculated for the Horvitz-Thompson estimator;

step 2.4.2: the Horvitz-Thompson estimator is combined with the traditional information entropy to obtain the amplitude spectrum entropy CSAE, and the following formula is shown:

wherein H^CSIs a CSAE (carrier-assisted laser ablation imaging system),

is the corrected frequency;

and step 3: judging the state of the steel rail fastener according to the CSAE value: and when the CSAE value is larger than the CSAE value in the normal screwing state at the same position, judging that the steel rail fastener at the current moment is loosened or falls off.

2. A steel rail fastener state detection method based on an information entropy theory according to claim 1, characterized in that: the specific steps of the step 2.1 are as follows:

step 2.1.1: in the R language, loading a pragma packet, setting an order value interval, namely setting a polynomial order m, and solving each order coefficient of the polynomial by using a polyfit function according to original acceleration data acc and time t;

step 2.1.2: using a polyfal fitting polynomial according to each order coefficient of the polynomial and time t to obtain acceleration data p _ acc after polynomial fitting; and finally obtaining acceleration data acc _1 after removing the trend term, wherein the formula is as follows:

acc_1＝acc-p_acc。

3. a steel rail fastener state detection method based on an information entropy theory according to claim 1, characterized in that: the specific steps of step 2.2 are as follows:

step 2.2.1: determining the resonant frequency f of the fastener and the rail_k；

Step 2.2.2: the vibration signal from which the trend term is removed is set to { X } ═ X₁,…,x_kAt 1/f on the time axis_kSetting an amplitude axis interval P for interval size division, and dividing a time domain plane where the X is located into a plurality of equal-area parts according to the amplitude and the frequency of the signal;

step 2.2.3: by theta_iThe number of X components in each region is expressed, and count data of the vibration signal is obtained in the form of { θ } -, where θ is expressed₁,θ₂,…,θ_n}。

Images