JP2017035973A - Overhead power line state diagnostic device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an overhead power line state diagnostic device capable of detecting an abnormal state of an overhead power line, by using a self-organization map, by utilizing data collected by an electric-track inspection-measurement vehicle.SOLUTION: An overhead power line state diagnostic device comprises an overhead power line state classification map formation device for generating an overhead power line state classification map of partitioning an area collected with every overhead power line state, by forming a self-organization map by adjusting a weight vector of the other node, by successively projecting a map forming input vector on the self-organization map, by generating a multidimensional map forming input vector by determining a value with every vector element on this vibration data, by collecting the map forming vibration data on an overhead power line of a normal state and an abnormal state, and also comprises a diagnostic device for diagnosing the overhead power line state based on a belonging area of a winner node of the overhead power line state classification map, by generating a multidimensional diagnostic input vector on this vibration data, by collecting the diagnostic vibration data in an overhead power line section of a diagnostic object.SELECTED DRAWING: Figure 1

Description

The present invention relates to an overhead line state diagnosis apparatus for diagnosing the presence / absence and contents of an abnormality occurring in an overhead line of an electric railway.

Although the frequency of occurrence of abnormalities in electric railways is not high, it is a facility that is not duplicated. Therefore, once an abnormality occurs, the effect is enormous. Therefore, high-frequency inspections for overhead lines are one of the most important tasks for the safe operation of railways. Examples of preventive maintenance for overhead lines include visual inspections by walking and vehicle patrols, and inspections based on data collected by electric / track inspection vehicles.
On the other hand, the working population has been decreasing regularly in recent years, and maintenance of maintenance quality has become an issue. Therefore, the improvement and simplification of the method of diagnosing the state of the overhead line based on the data measured by the traveling vehicle is attracting attention as a measure for preventing the deterioration of the maintenance quality accompanying the decrease in the working population.

There are two types of overhead wire abnormalities: hanging hanger supporting the trolley wire from the suspended overhead wire, hanger coming off from the pantograph in the straight section There are brace brackets and bent braces that occur due to defective rotating shafts, and brace brackets and bent pullouts when brace brackets and curved brackets are removed. Moreover, there is a change in the span length, etc., which is not abnormal but is different from the normal state.
These abnormalities affect the vibration of the trolley wire that occurs as the pantograph slides, so it may be detected by observing changes in contact force between the overhead wire and the pantograph, acceleration of the pantograph member, etc. .

However, conventionally proposed methods for diagnosing overhead lines using vibration data such as overhead lines and pantographs include methods for detecting shocks with acceleration sensors installed on pantographs and estimating overhead line abnormalities, overhead wire fittings, etc. Although there is a method of detecting an obstacle by acceleration when a pantograph collides with an overhead wire fitting or the like in a situation where it breaks and hangs down, the state that can be grasped is very limited.

A self-organizing map (SOM) is a type of artificial neural network developed by Teuvo Cohonen, a mathematical model that mimics a neural circuit. It is based on pattern recognition of multidimensional data and extracted patterns. Can be considered as one of the multivariate analysis methods. SOM has excellent properties as a data compression means, and can be displayed on a low-dimensional surface while maintaining the phase relationship of multidimensional data.
As described above, the SOM is an effective tool for visualizing high-dimensional data, and the similarity of high-dimensional input data to the SOM is drawn on a two-dimensional or three-dimensional map. Specifically, input data with high similarity are arranged in the vicinity, and as a result, the data is clustered.
In the industrial field, there are many cases where SOM is used for analysis and measurement techniques.

Patent Document 1 discloses an apparatus for diagnosing motor abnormality using a self-organizing map. Motor aging appears in shafts and bearings. In particular, the bearings use ball bearings, and if the bearings are scratched or rusted, a characteristic appears in vibration noise. Therefore, the motor abnormality is detected using the vibration data of the motor collected using the vibration sound sensor manufactured with reference to the auscultation stick used in the field. The invention disclosed in Patent Document 1 converts the output of a vibration sound sensor that contacts a motor and converts vibration into sound into a second sound waveform by wavelet transform. A self-organizing algorithm is applied to the input data extracted from the large amount of second sound waveforms obtained in this way to create a self-organizing map that divides the normal state region and the abnormal state region. The motor diagnosis is to determine whether the vibrating body is abnormal or normal based on the position on the self-organizing map where the input data extracted from the waveform of the second sound acquired by the same method is arranged.

Until now, a lot of vibration data of overhead lines and pantographs have been accumulated by inspections with electric and track inspection vehicles. In addition, the numerical simulation is improved in accuracy so that various phenomena associated with the overhead wire abnormality can be accurately reproduced. In the diagnosis of overhead lines, if SOM can be easily diagnosed by using SOM based on the data collected while traveling with electric and track inspection vehicles, the maintenance quality associated with the decrease in the working population will be improved. Deterioration can be prevented.

However, in order to effectively use SOM for a specific target, it is necessary to appropriately form a data structure according to the case. In order to be able to be used as a tool to detect electrical railway overhead line abnormalities, it is possible to select variables that are closely related to abnormal phenomena, perform processing suitable for applying SOM, and actually detect abnormalities. It is required to develop a technique to do this.

JP 2011-107093 A

Therefore, the present invention provides an overhead line state diagnosis device that can detect an abnormal state of an overhead line by utilizing data collected by an electric / track inspection vehicle and using a self-organizing map (SOM). To do.

The overhead line state diagnosis apparatus of the present invention collects and stores vibration data for forming a map of an overhead line while traveling while sliding the pantograph under the overhead line for a normal state and an abnormal state of the overhead line data collecting apparatus. A map-forming input vector generation device that generates and stores a map-forming input vector by obtaining a value for each vector element by dividing the stored map-forming vibration data into a multi-dimensional vector, Self-organization by repeating the learning to adjust the weight vector so as to reduce the distance from the weight vector of other nodes by sequentially projecting the input vector for map formation for each span onto the winner node of the self-organization map Self-organizing map forming device that forms a map, and an overhead line that divides an area gathered for each overhead line state and generates and stores a diagnostic overhead line state classification map Characterized in that it comprises a status classification map forming apparatus.

Another aspect of the overhead line condition diagnosis apparatus of the present invention is diagnostic data for collecting and storing diagnostic vibration data for overhead lines while traveling while sliding a pantograph under the overhead line in the overhead line section to be diagnosed. A collection device, a diagnostic input vector generation device for generating and storing a diagnostic input vector by classifying the stored diagnostic vibration data for each span, obtaining a value for each vector element, generating a multidimensional vector, and a diameter; A diagnostic apparatus for projecting diagnostic input vectors for each interval onto a winner node of an overhead line state classification map in which a region in which the overhead line state is gathered is divided, and diagnosing the state of the overhead wire based on the region to which the winner node belongs It is characterized by that.

The overhead line state diagnosis apparatus of the present invention creates a state classification map of overhead lines by applying a self-organization algorithm to vibration data of a large amount of overhead lines and pantographs, A self-organization algorithm is applied to the vibration data of the pantograph and projected onto the overhead line state classification map, and the overhead line state at the target location where the data is obtained from the projected position is diagnosed.
As vibration data, data measured on the pantograph side such as the contact force between the running overhead line and the pantograph collected by a sensor installed on the pantograph and the acceleration of the pantograph member can be used. These vibration data reflect changes in span length and changes in span length, such as hanger disconnection, firmness of brace brackets and bent metal fittings, and disconnection of brace brackets and bent metal fittings.

The overhead line condition diagnosis apparatus of the present invention can almost automatically estimate the type and position of overhead line abnormality based on data collected by the electric / track inspection vehicle while traveling. It is possible to reduce the maintenance quality by compensating for the decrease in the working population.

It is a flowchart showing the process performed in the overhead wire state diagnostic apparatus which concerns on one embodiment of this invention. FIG. 1A shows a process for creating a self-organizing map, and FIG. 1B shows a process for diagnosing an overhead wire abnormality. It is a flowchart which shows the creation process of the self-organization map in the overhead wire state diagnostic apparatus based on this embodiment in detail. It is drawing which shows the example of the overhead line state classification | category map in this embodiment. It is a block diagram which shows the structure of the overhead wire state diagnostic apparatus which concerns on one Example of this embodiment. It is a chart which shows the example of the vibration data collected in order to form an overhead line state classification map in this example. It is drawing which shows the example of the overhead line state classification | category map in a present Example. It is drawing which shows the example which detects an abnormal state using the overhead wire state classification | category map in a present Example.

Hereinafter, an embodiment according to the overhead wire state diagnosis apparatus of the present invention will be described in detail with reference to the drawings.

When there is an abnormality in the overhead line of the electric railway, the vibration mode of the overhead line changes. The fluctuation of the vibration mode is reflected in the contact force fluctuation waveform generated by the pantograph sliding on the trolley line when the vehicle travels, the acceleration change of the pantograph member, and the like. Contact force and acceleration can be continuously measured by a sensor installed on the pantograph side of the electric / track inspection vehicle.
Therefore, in the overhead line state classification map used in the overhead line state diagnosis apparatus of the present embodiment, the pantograph side includes the contact force between the overhead line and the pantograph and the acceleration of the pantograph member as elements constituting the weight vector of the nodes constituting the map. Further, the vibration data measured in (1) and the signal processing results and statistical processing results of these vibration data were used.

In order to generate an overhead line state classification map according to the self-organization algorithm, it is impossible to detect an overhead line state that is not included in the input vector used when creating the self-organization map (SOM) that becomes the overhead line state classification map. The input vector to be used needs to include vibration data about the overhead line in which the abnormality to be included in the classification has occurred. In addition, in order to judge the soundness level of an overhead line, the data regarding a healthy overhead line state are also required.
The input vector can be obtained not only by actual measurement but also by simulation. In order to collect vibration data used to create an overhead line state classification map, it is convenient to use a simulation that can appropriately set an abnormal state to be included in the classification.

Needless to say, when an input vector obtained exclusively by numerical simulation is used to create an overhead line state classification map, it is necessary to use a precise numerical simulation capable of accurately reproducing the current vehicle data.
When the obtained vibration data itself is used as an input vector, the value at the sampling point in the measurement becomes a vector element, and the total number of data becomes the dimension of the input vector. When any processing is performed on the obtained vibration data, the processing result becomes one or more elements of the input vector.

It is effective to use the vibration data itself as an input vector as long as it is vibration data obtained by numerical simulation based on determinism such that the same value is obtained if the overhead line / pantograph conditions are the same.
However, in the case of vibration data obtained by numerical simulation including actual measurement and stochastic processes, even if the overhead line and pantograph conditions are the same, a difference due to noise or measurement error occurs in the result. In this case, when the self-organization algorithm is applied using the vibration data itself as an input vector, there is a possibility that the data is arranged on the self-organization map (SOM) as data based on different conditions.

In order to avoid such misjudgment due to SOM, the vibration data is subjected to some processing such as attenuating noise and emphasizing the characteristics of the overhead line condition, and for example, extracting information such as dominant frequency components and other statistics It is preferably used as a vector element. Since the data processing selected here depends on the characteristics of vibration data or the abnormal state to be detected, it greatly affects the ease and reliability of diagnosis.

When the SOM is created using the collected input vectors, the input vectors obtained in the similar overhead line state are arranged near each other on the map. As a result, the SOM and the overhead line state classification map in which the areas are classified according to the overhead line state, such as “a group area indicating a healthy state”, “a group area indicating a predetermined abnormal state”, and “a region not belonging to any group” on the map Is formed.

FIG. 1 is a flowchart for explaining a diagnosis procedure executed by the overhead line state diagnosis apparatus according to this embodiment.
First, a self-organizing map (SOM) is created using the accumulated vibration data. FIG. 1A is a flowchart showing a self-organizing map creation process executed by the overhead line state diagnosis apparatus of this embodiment.

First, vibration data necessary for mapping is collected and accumulated (S1), and a vibration data database (D1) is generated. The vibration data must include an appropriate number of abnormal vibration data when there is an abnormality that can identify the cause.
The vibration data is acquired individually or in combination by an accelerometer, a laser Doppler vibrometer, a laser displacement meter, a rotary encoder provided at a joint portion of the frame, or a load cell for measuring the axial force of the actuator.
Note that the vibration data to be collected is not limited to actual measurement values, and may be data obtained by simulation. In this case, vibration data is classified and used for each span because it is necessary to specify the occurrence position of the abnormal phenomenon.

Further, for a plurality of indexes that can be involved in a predetermined diagnosis, each vibration data is represented by a multidimensional vector by extracting its components from the vibration data or calculating them (S2). Here, we focus on the maximum and minimum values in the vibration data of spans and the positions indicating them, that is, the distance from the support column, as an indicator that can detect changes in the vibration waveform of overhead lines and pantographs caused by overhead line abnormalities. . In addition to these indices, the standard deviation value, skewness, and kurtosis of the amplitude values sampled for the vibration data are calculated and used as elements of the input vector related to the vibration data between the spans. A dimensional vectorized input vector database (D2) is generated.

Next, a self-organizing algorithm is applied to the input vector database (D2) to create a self-organizing map SOM (D3) (S3).

FIG. 2 is a flowchart for explaining a process of creating a self-organizing map by applying a self-organizing algorithm to the map forming input vector.
Here, for generalization, it is assumed that N d-dimensional map forming input vectors are used. The i-th input vector Xi is expressed by the following equation (1), and the maximum value xmaxj and the minimum value xminj of the j-th element are defined by the following equations (2) and (3) for all N input vectors.
(1)
Xi = {xi1, xi2, ... xij, ..., xid} (i = 1, ..., N)
(2)
xmaxj = max (x1j, x2j,..., xNj) (j = 1,..., d)
(3)
xminj = min (x1j, x2j,..., xNj) (j = 1,..., d)

First, the SOM parameter of the self-organizing map is determined (S11).
The SOM parameters include a map shape, a total number M of nodes that determine the map size, a planned learning count T, and a parameter σ that determines the degree of change of the weight vector when the weight vector is updated. These SOM parameters have optimum values depending on the problem of applying the SOM, and more suitable values are obtained by a method such as trial and error.

Next, the map is initialized (S12).
The map is initialized by giving appropriate weight vectors W to all M nodes in the map. The weight vector W has the same dimensions as the input vector. The weight vector Wk given to the kth node is expressed by the following equation (4).
(4)
Wk = {wk1, wk2,..., Wki,..., Wkd} (k = 1,..., M)
The value of the j-th element of each weight vector can be given as a random number generated so as to be uniformly distributed between the minimum value and the maximum value of the j-th element for the input vector with [xminj, xmaxj].

Next, the accumulated vibration data is sequentially introduced into the initialized map, and is transformed into an SOM suitable for the vibration data.
One input vector is input from the previously formed database (D2) of multidimensional input vectors (S13).

The node having the weight vector closest to this input vector is determined in the following manner (S14).
A distance Dik between the i-th input vector Xi and the weight vector Wk assigned to each node is obtained. The distance Dik is given by the following equation (5).
(5)
Dik = | Xi−Wk | (k = 1,..., M)
The weight vector that minimizes the distance Dik is referred to as a winner vector Wl. A node that stores a winner vector is called a winner node.

Next, the weight vector Wk stored in the winner vector and a node located in the vicinity of the winner node is adjusted so as to approach the input vector Xi (S15).
If the weight vector before adjustment is WkOLD and the weight vector after adjustment is WkNEW, the weight vector is adjusted using the parameter σ according to, for example, the following equations (6), (7), and (8): Can do.
(6)
WkNEW = WkOLD + α (Xi-WkOLD)
(7)
α = βexp (−Lk ² / σ)
(8)
β = 1−t / T

Here, Lk is the distance between the k-th node and the winner node, t is the number of learnings at that time, and T is the total number of learnings scheduled.
α is a variable that determines the degree of influence on the correction amount of the weight vector, and the correction amount is increased as it is closer to the winner node and the learning is shallower. Note that β may be a function that decreases as the number of learning increases. Note that the correction calculation of the weight vector may not be performed for nodes that are some distance away from the winner node.
In this way, the weight vector of the node in the vicinity of the winner node is rewritten with a new weight vector approaching the used input vector.

When step S15 is completed, it is checked whether input vector Xi remains (i = N?) (S16), and if there is, step i is incremented, and the process returns to step S13 from the input vector database (D2). The next data is taken in and Steps S14 and S15 are repeated.
If all the input vectors Xi are used, this is regarded as one learning number, and it is confirmed whether the learning number t has reached the predetermined number T (t = T?) (S17). If not, t is incremented, the process returns to step S13, the first data is taken in again from the input vector database (D11), and step S14, step S15 and step S16 are repeated, and learning is repeated.
The input data used in the new learning process may be the same set as that used last time.

When the learning number t reaches the predetermined number T, the weight vector of each node is finally determined. In this way, the multi-dimensional input vector is projected onto a two-dimensional or three-dimensional space by the self-organization algorithm, and the self-organization map SOM (D3) arranged so as to form a group according to the similarity of the input vectors. Is completed.
Therefore, the information related to the state of the overhead line can be classified by including the information related to the state of the overhead line in the input vector.

Therefore, returning to step S4 in FIG. 1A, when a set of input vectors is applied to the self-organizing map SOM (D3), each of the input vectors is arranged at a node having a weight vector that minimizes the distance. At this time, since input vectors having high similarity are arranged in the vicinity, an overhead line state classification map (D4) in which a region is defined for each overhead line state is formed on the map by finding an overhead line state common to similar groups. (S4).

FIG. 3 is a chart showing an example of the overhead line state classification map in the present embodiment.
The illustrated map has a shape in which 631 regular hexagonal tiles are spread in a regular hexagon. A tile represents a node, and an appropriate weight vector was assigned to each tile in the first stage. Extracting predetermined characteristic values from vibration data acquired for various overhead states including abnormal states and converting them into multidimensional vectors as elements, processing them according to a self-organization algorithm to obtain node weight vectors By repeating the learning to be adjusted, as shown in FIG. 3, in an area gathered for each abnormality type such as “healthy” “a certain abnormality has occurred” or “an unexpected abnormality has occurred” The partitioned overhead line state classification map is completed.

When the vibration data obtained in the inspection target section is projected onto the overhead line state classification map thus formed, the overhead line state can be diagnosed based on the region to which the projected position belongs.
In the diagnosis of the overhead line state, an input vector having the same characteristic value as that used in the overhead line state classification map is generated from the vibration data acquired in the field, and this is used as a node having the most approximate weight vector on the SOM. Project. Depending on the area to which the node thus projected belongs, the overhead line state of the section from which the data was acquired is either “sound”, “an abnormality has occurred”, or “an unexpected abnormality has occurred” Can be estimated, and an overhead wire abnormality can be diagnosed.

FIG. 1B is a flowchart showing an overhead wire abnormality diagnosis process.
The vibration data for diagnosis is acquired by running the electric / track inspection vehicle in the section to be diagnosed (S5). For the acquired vibration data for diagnosis (D5), the components of the same elements as the weight vectors of the overhead line state classification map are calculated for each span and converted into multidimensional vectors to obtain diagnostic input vector data (D6) ( S6). For the input vector data for diagnosis (D6), a node having a weight vector closest to the input vector is selected (S7). By knowing the region to which this node belongs in the overhead line state classification map (D3), the overhead line state for each span in the diagnosis target section is diagnosed (S8).

Vibration data obtained by running an electric / track inspection vehicle by previously forming an overhead line state classification map in which regions are classified according to overhead line states based on a large amount of accumulated data by the overhead line state diagnosis device of the present invention. By applying a self-organization algorithm to a multidimensional vector and projecting it onto the overhead line state classification map, the overhead line state in the target section can be diagnosed. Therefore, even if the inspection density is maintained or increased, the frequency of the patrol inspection at the site can be reduced, and the deterioration of the maintenance quality of the overhead wire can be prevented by supplementing the decrease in the working population.

Hereinafter, an embodiment of the overhead wire state diagnosis apparatus according to the present embodiment will be described with reference to the drawings.
FIG. 4 is a block diagram showing the configuration of the overhead wire state diagnosis apparatus in this embodiment.
In this example, in view of the ease of experimentation, an overhead line state classification map was created based on vibration data obtained by reproducing overhead line conditions using overhead line / pantograph system simulation software.

The overhead line state diagnosis apparatus according to the present embodiment collects vibration data for map creation, creates a self-organizing map SOM using the vibration data, and creates an overhead line state classification map that divides an area where data whose properties are approximated. A map creation unit 100 to be created, and a state determination unit 200 that collects actual vibration data, multi-dimensionally vectorizes them, projects them on the overhead line state classification map, and diagnoses the overhead line state based on the region to which the projection position belongs.

The map creation unit 100 is a personal computer equipped with a simulation program, and generates a map creation vibration data D11. The map creation unit 100 extracts a characteristic value of the data from the vibration data and multi-dimensionalizes it, and creates a multi-dimensional input vector D12 for map creation. An information processing device 12 for generating a self-organizing map (SOP) D13 by applying a self-organizing algorithm to a multidimensional input vector, and an input vector projected on the self-organizing map SOP A map forming device 14 that forms and stores an overhead line state classification map D14 by dividing a normal region, an abnormal region, and an intermediate region on the basis of characteristics common to these groups.

The state determination unit 200 is mounted on the panda graph side of the electric / trajectory inspection vehicle traveling in the target area, collects the vibration signal for diagnosis in real time, and collects the vibration signal for diagnosis in real time. A data logger 22 that is recorded as diagnostic vibration data D22 that can be called and used, an information processing device 23 that extracts characteristic values of the vibration data cut out for each span to obtain a multidimensional input vector D23 for diagnosis, An arithmetic processing unit 24 for finding a winner node of the overhead line state classification map D14 corresponding to the vibration data for each interval and a determination device 25 for diagnosing the overhead line state for each span based on the position of the winner node are provided.
The information processing device 23, the arithmetic processing device 24, and the determination device 25 are configured by a computer or the like separately provided in a test room or the like, so that vibration data is acquired from the data logger 22 that has collected the data, and the arithmetic processing is performed. Good.

FIG. 5 is a chart showing an example of vibration data collected to form an overhead line state classification map in the present embodiment. The data shown in FIG. 5 is obtained by reproducing the normal overhead line situation.

Here, in the self-organizing map, the SOM size is 631 nodes, and the shape is a two-dimensional regular hexagon in which 631 tiles are spread, but the number of nodes and the shape can be changed. First, a multidimensional weight vector having values randomly generated from 0 to 1 for each element is set in each node. Here, since a 9-dimensional input vector is used as will be described later, the weight vector is also 9-dimensional accordingly.
The vibration data was replaced by the contact force response waveform between the overhead line and the pantograph detected by the contact force sensor installed on the pantograph side. The contact force has a high correlation with the vibration of the overhead wire, and a large amount of measured data has been accumulated conventionally.

Taking the main specifications of simulated overhead line, 150 mm ² cross-sectional area in the trolley wire, 116 mm ² in catenary, 1.334kg / m linear density of trolley lines, with catenary 1.05 kg / m, the Young's modulus Is 120 GPa for the trolley wire, 84.7 GPa for the suspension wire, tension is 26,000 N for the trolley wire, and 20,000 N for the suspension wire.
In the simulation, the overhead wire is composed of 31 diameters supported by 32 support points. The standard span length was 63 m except for the vicinity of both ends of the overhead line between the first support point to the fifth support point and between the 28th support point and the 32nd support point. The overhead line height was 5.08 m at the hanger position and the suspension line height was 6.48 m at the support point except for both ends of the overhead line in the vicinity of the holding. Moreover, the overhead lines were displaced by 0.2 m by sandwiching the center of the line in the horizontal direction alternately for each support point by the support device except for the vicinity of both ends of the overhead line.

According to the simulation, it is assumed that the pantograph is in contact between the sixth support point and the 24th support point except for the vicinity of both ends of the overhead line, and from the eleventh support point corresponding to the central part of the travel section that is not affected by the boundary retaining point. Contact force response waveforms at the 21st support point (11th to 20th spans) were acquired and used as vibration data to create an overhead line state classification map.

For example, in the simulation, a reference state with no abnormalities in the overhead wire, a case where the pulling angle of the anti-rest fitting is shifted from the reference as an abnormal state, a case where the anti-rest furniture is astringent, and a case where the anti-rest fitting is missing are set. And executed. The overhead line abnormality is assumed to exist at one support point in the measurement section, and the pantograph is assumed to run at a constant speed of 300 km.

In addition, it is thought that the abnormality of the firmness of a bent metal fitting or a defect | deletion is detected as the same situation as the firmness or defect | deletion of a bracing metal fitting.
For an AC 25 kV overhead line, these overhead line states were expressed in the simulation, and vibration data was obtained when one pantograph traveled at 300 km per hour under the overhead line.

4 cases in which the brace fitting at a certain support point (any one of the thirteenth support point to the sixteenth support point) is solid, and a brace at a certain support point (any one of the thirteenth support point to the sixteenth support point) A total of 28 cases including 4 cases with missing metal fittings and 16 cases with the pulling angle of the brace fitting at a certain support point (any one of the 13th support point to the 16th support point) being 0 degree, 20 degrees, 30 degrees and 40 degrees. A case was prepared.

Changes in the sag of the overhead wire due to tension changes and changes in the span length are in the low frequency region of the contact force response waveform, hangers are abnormal in the intermediate frequency region, and phenomena such as shocks are in the high frequency region. It is estimated that it can be detected as fluctuations in
Further, the change in span length affects the dominant frequency component of the vibration of the overhead wire, and the abnormality of the hanger or the clasp has an effect on the vibration frequency, the peak value of the vibration, or the peak position.
It is expected that the accidental vibration waveform deviation that is not considered to be an abnormal overhead line is canceled by statistical processing of the waveform.

In view of the above, the vibration waveform is divided for each span, and using vibration measurement values cut out at every sampling time, the maximum value and minimum value, and the position where the maximum value and minimum value are expressed (from the support point) ), The average value μ of the sampled vibration measurement values, the standard deviation σ, the skewness (skewness) μs, the kurtosis (kultosis) μk, and the value obtained based on the wavelet transform are extracted as characteristic values. This is an input vector element. However, it goes without saying that the number of dimensions of the input vector can be selected as appropriate.
As the values of the elements of the input vector, values obtained by normalizing the respective elements with an appropriate range including the maximum value and the minimum value appearing in the collected vibration data as 1 were used.

In this example, the sampled vibration measurement values were wavelet transformed using a 6 wave number Morlet wavelet function. As is well known, the wavelet transform is defined by convolution of a signal with a wavelet function, which is a small vibration, subjected to translation and expansion / contraction operations. The wavelet transform is an excellent method for detecting the fluctuation characteristics of a non-stationary signal because it can measure the position of a signal in time and frequency.
In this embodiment, two-dimensional data obtained by performing wavelet transform on vibration data for each span, the span average for an appropriate section is calculated, and the wavelet transform value and span average for each span are calculated. The correlation coefficient was calculated as one element of the input vector. This makes it possible to accurately detect the occurrence of the unsteady state and the generated span position by using values obtained based on the wavelet transform.

Here, the skewness μs is an index representing the asymmetry of the distribution, and is expressed by the following equation (9). Ri is an amplitude value, and μ is an average value of vibration measurement values between spans.
(9)
μs = (1 / N) Σ (Ri−μ) ³ / (((1 / N) Σ (Ri−μ) ² ) ^1/2 ) ³
Further, the kurtosis μk is an index representing the sharpness of the distribution and is expressed by the equation (10).
(10)
μk = (1 / N) Σ (Ri−μ) ⁴ / (((1 / N) Σ (Ri−μ) ² ) ^1/2 ) ⁴

The self-organizing map SOM is formed by repeatedly using the input vectors of the 11th to 20th spans formed for at least 64 section data including the normal state and the abnormal state.
When the number of learning repeated when forming the SOM is set to 100 and the weight vector assigned to each node is corrected according to the procedure shown in FIG. 1A, a final SOM is obtained.

By plotting a set of input vectors on the SOM and comparing the abnormal contents of the input vectors, the input vectors that indicate the abnormalities of the hanger, the steadyness of the steady bracket, the lack of the steady bracket hanger, and the respective abnormalities are nearby. It can be seen that the regions are distinguished. Moreover, although it is not an abnormal state, it turns out that it forms a group also when the span length changes. In this way, an overhead line state classification map is formed on the map that divides the area according to the overhead line state, such as “group area indicating a healthy state”, “group area indicating a predetermined abnormal state”, and “area not belonging to any group”. The

FIG. 6 is a diagram showing an example of the overhead line state classification map generated in this embodiment, and FIG. 7 shows an example in which the determination accuracy is verified by plotting the input vector generated from the vibration data on the overhead line state classification map of FIG. It is a drawing.
This figure shows the angle abnormality of the bent metal fittings for convenience of verification.
The angle of the bent metal fitting attached to the starting end of the span is 10.4 degrees as standard, but vibration data is generated by the simulator when there are abnormal angles of 0 degrees, 20 degrees, 30 degrees, and 40 degrees. The vibration data of and the normal vibration data are formed together.

Looking at the completed SOM, the input vector when the angle of the bent metal fitting is large is concentrated near the edge of the lower left part of the map, and it is scattered slightly upward and scattered upward, and at the upper left edge part of the map The distribution then disappears.
The region surrounded by the dotted line where the vibration data in the abnormal state is plotted is the region to which the plot of the input vector of vibration data when the overhead line in the span has an abnormal state should belong.
In this way, when the area where the overhead line states are classified is entered in the self-organizing map, the overhead line state classification map is obtained.

FIG. 7 is a plot of vibration data obtained in the target section on the overhead line state classification map thus obtained. The vibration data 1 to 4 in the abnormal state all belong to the abnormal region, and there was no mistake of detection omission. The vibration data in the normal state was plotted in the normal region in many cases, but there were only a few examples 5 to 7 that were determined to be abnormal although they were not abnormal.
It can be said that the determination accuracy is generally good.

By using the overhead line state classification map of this embodiment, the vibration data obtained from the electric / track inspection vehicle traveling in the inspection section is divided into spans to generate input vectors and plotted on the overhead line state classification map The presence / absence and type of abnormality can be estimated based on the region to which the position belongs.
These abnormality diagnoses can be mechanically performed using a data logging device, a signal processing device, and an arithmetic processing device. In addition, even when the occurrence location of an abnormality is inspected based on the result of diagnosis, the state and position of the abnormality are clarified, so that the field work can be performed simply and accurately.

In the present embodiment, a two-dimensional hexagonal map having 631 nodes is selected, but it goes without saying that the number of nodes and the shape can be changed depending on the situation. In addition, it is effective to use a cylindrical map or a spherical map having no boundary, particularly in order to prevent a problem that the distance between vectors with adjacent nodes jumps at the boundary part of the map.
Furthermore, in this embodiment, the vibration data of the overhead line is collected by using the contact force sensor between the overhead line and the pantograph of the electric / track inspection vehicle. However, even if the sensor is mounted on an ordinary vehicle, the contact force sensor is used instead. You may measure using the displacement sensor and acceleration sensor of a pantograph member.

Further, although the input vector has been made seven-dimensional, it may be formed with a larger number of dimensions, or it may be of lower dimensions if an abnormal state distribution region appears.
Furthermore, the maximum amplitude width of vibration data, the peak frequency obtained by Fourier transform, and the like can be used as elements of the input vector. Further, a correlation coefficient obtained by comparing the waveform after Fourier transform with a standard waveform reflecting the normal state can be used as an element.

DESCRIPTION OF SYMBOLS 100 Map preparation part 11 Simulator 12 Information processing apparatus 13 Arithmetic processing apparatus 14 Map formation apparatus 200 State determination part 21 Vibration sensor 22 Data logger 23 Information processing apparatus 24 Arithmetic processing apparatus 25 Determination apparatus

Claims (6)

A map forming data collecting device that collects and stores vibration data for forming a map of an overhead line while traveling while sliding a pantograph under the overhead line for a normal state and an abnormal state overhead line,
A map forming input vector generating device that generates a map forming input vector by storing the map forming vibration data for each span and obtaining a value for each vector element to generate a map forming input vector.
The map formation input vector for each span is projected onto the winner node of the self-organizing map sequentially, and learning for adjusting the weight vector so as to reduce the distance between the weight vectors of other nodes is repeated and self-organization is performed. A self-organizing map forming device for forming an integrated map;
An overhead line state diagnosis apparatus comprising: an overhead line state classification map forming apparatus that divides a region gathered for each overhead line state and generates and stores a diagnostic overhead line state classification map. A diagnostic data collection device that collects and stores diagnostic vibration data of the overhead line while traveling while sliding the pantograph under the overhead line in the overhead line section to be diagnosed;
A diagnostic input vector generation device that generates and stores a diagnostic input vector by dividing the span of the stored diagnostic vibration data for each span and obtaining a value for each vector element to generate a multidimensional vector; and
The state of the overhead line based on the region to which the winner node belongs is projected onto the winner node of the overhead line state classification map according to claim 1, wherein a region in which the diagnostic input vector for each span is gathered for each overhead line state is partitioned. An overhead line state diagnosis device comprising: a diagnosis device for diagnosing 2. The overhead line condition diagnosis apparatus according to claim 1, wherein the map forming data collection device is a simulator for estimating vibration data of an overhead line. 4. The overhead line state diagnosis apparatus according to claim 1, wherein the vibration data for map formation or the vibration data for diagnosis is data of contact force between overhead lines and pantographs. 5. The map-forming vibration data or the diagnostic vibration data includes an accelerometer that measures displacement of the hull, a laser Doppler vibrometer, a laser displacement meter, a rotary encoder provided at the joint of the frame, or an axial force of the actuator. The overhead line state diagnosis apparatus according to any one of claims 1 to 4, wherein the overhead line state diagnosis apparatus is measured by a load cell. The vector element includes the maximum value and minimum value in the vibration waveform, the position where the maximum value and the minimum value are expressed, the maximum amplitude width, the average value μ of the sampled vibration measurement values, the standard deviation, the skewness, the cusp. The overhead line state diagnosis apparatus according to any one of claims 1 to 5, including any one of a degree (Cultosis), a value obtained based on wavelet transformation, and a peak frequency obtained by Fourier transformation.

Images