JP6979898B2 - Condition monitoring device for railroad vehicles - Google Patents

Description

The present invention relates to a condition monitoring device for a railway vehicle that determines the presence or absence of an abnormality in a railway vehicle traveling on a track. In particular, the present invention relates to a railroad vehicle condition monitoring device capable of easily and accurately determining the presence or absence of an abnormality in a railroad vehicle.

From the standpoint of transportation safety, tracks and railroad vehicles are regularly maintained. However, it is not always possible to grasp the condition of railway vehicles. In addition, it is not economical because maintenance is often carried out by inspecting each part or replacing parts regardless of the presence or absence of abnormalities.

For this reason, a method has been proposed in which a wheel load sensor for measuring the wheel load of a wheel is provided on the track, and an abnormality of a railway vehicle is detected based on the magnitude of an index represented by the wheel load measured by the wheel load sensor (). For example, see Patent Document 1).

However, the method described in Patent Document 1 is a method of comparing an index with a predetermined threshold value and detecting an abnormality of a railway vehicle according to the magnitude of the index. The threshold value for detecting this abnormality may change to an appropriate value depending on the structure of the railroad vehicle, loading conditions, running conditions, and the like. Therefore, in order to accurately detect an abnormality in a railroad vehicle, it is necessary to determine a large number of threshold values for each of these conditions, which requires a great deal of time and effort.

Therefore, a railway vehicle condition monitoring device has been proposed that can easily determine the threshold value for determining the presence or absence of an abnormality in a railway vehicle without requiring a great deal of time and effort (see, for example, Patent Document 2). ..
The device described in Patent Document 2 includes a detection device for detecting an index represented by the wheel load of a wheel, and a determination device provided with a classifier for inputting the detected index and outputting the presence or absence of an abnormality in a railroad vehicle. ing. This classifier is generated by machine learning, but machine learning requires teacher data for normal rolling stock and teacher data for abnormal rolling stock.
According to the apparatus described in Patent Document 2, since it is only necessary to generate a classifier by machine learning, it is possible to easily determine the presence or absence of an abnormality in a railway vehicle as compared with the method described in Patent Document 1.

In the apparatus described in Patent Document 2, teacher data about a normal railroad vehicle required for machine learning can be easily prepared. However, it is extremely difficult to prepare a large number of teacher data for the teacher data about the abnormal railroad vehicle because it is difficult to prepare the abnormal railroad vehicle itself. Therefore, there is a problem that the judgment accuracy is deteriorated as a result of insufficient machine learning of the classifier.
Although Patent Document 2 proposes to calculate the teacher data of an abnormal railroad vehicle by numerical simulation using general-purpose mechanism analysis software, it is sufficient because it is not the teacher data of an actual abnormal railroad vehicle. Judgment accuracy cannot be obtained.

Japanese Unexamined Patent Publication No. 2013-120100 Japanese Unexamined Patent Publication No. 2016-113029

The present invention has been made to solve the above-mentioned problems of the prior art, and provides a condition monitoring device for a railway vehicle capable of easily and accurately determining the presence or absence of an abnormality in a railway vehicle. Make it an issue.

As a result of diligent studies in order to solve the above problems, the present inventors have found that when the value of an index (such as a vehicle body twist index) represented by the wheel load of a railroad vehicle is abnormal in the railroad vehicle. It was found that the sudden change was made before and after the maintenance of the track or the railroad vehicle (see Fig. 3 below). The index expressed by the wheel load of a normal railroad vehicle and the maintenance information of the track or railroad vehicle (information about when maintenance was performed on the track or railroad vehicle to be judged) are compared with each other. It can be easily obtained. For this reason, we use teacher data that inputs multiple indicators (time-series data of indicators) detected within a predetermined time for a normal railroad vehicle and outputs the presence or absence of maintenance of the track or railroad vehicle within this predetermined time. , It is relatively easy to generate a classifier by machine learning. If a plurality of indexes detected within a predetermined time are input to the classifier generated by this machine learning for the railway vehicle to be determined for the presence or absence of abnormality, the presence or absence of maintenance of the track or the railway vehicle is output from the classifier. Will be. Then, when the classifier outputs that there is maintenance, the present inventors can collate it with the actual maintenance information, and if the maintenance is actually performed, it can be determined that the railroad vehicle is normal. I came up with the idea that it is possible to determine that an abnormality has occurred in a railroad vehicle if maintenance has not actually been performed.

The present invention has been completed based on the above-mentioned findings of the present inventors.
That is, in order to solve the above-mentioned problems, the present invention comprises a detection device for detecting an index represented by the wheel load of a railroad vehicle traveling on a track, and a plurality of detection devices detected by the detection device within a predetermined time. The index is provided with a determination device for determining the presence or absence of an abnormality in the railway vehicle based on the input track or maintenance information of the railway vehicle, and the determination device is input with the plurality of the indexes. It is equipped with a classifier that outputs the presence or absence of maintenance of the track or the railway vehicle, and if the output of the classifier is without maintenance, it is determined that the railway vehicle is normal, and the output of the classifier has maintenance. In the case, based on the input maintenance information, it is determined whether or not the maintenance of the track or the railway vehicle is actually performed within the predetermined time when the plurality of indexes input to the classifier are detected. When the maintenance is actually performed, it is determined that the railway vehicle is normal, and when the maintenance is not actually performed, it is determined that the railway vehicle has an abnormality, and the classifier is normal. For the railway vehicle, when a plurality of the indicators detected within a predetermined time when the maintenance of the track or the railway vehicle is not actually performed is input, no maintenance is output, and maintenance of the track or the railway vehicle is performed. A state monitoring device for a railway vehicle, which is generated by using machine learning so that when a plurality of the indicators detected within a predetermined time actually performed are input, "with maintenance" is output. I will provide a.

According to the railway vehicle condition monitoring device according to the present invention, as teacher data in machine learning to generate a classifier, a normal railway vehicle is detected within a predetermined time when the track or the railway vehicle is not actually maintained. It suffices to prepare a plurality of indicators and a plurality of indicators detected within a predetermined time when the maintenance of the track or the railroad vehicle is actually performed. That is, since an index for an abnormal railroad vehicle is not required as teacher data, a large number of teacher data can be prepared, and as a result of sufficient machine learning of the classifier, it is possible to improve the judgment accuracy. Is.
As the classifier provided in the determination device, various configurations such as an artificial neural network (ANN) and a support vector machine can be adopted as long as they can be generated by using machine learning.

As a preferable index used in the condition monitoring device for a railway vehicle according to the present invention, a vehicle body twist index or a bogie twist index can be exemplified. That is, preferably, the index is a vehicle body twist index represented by the following formula (1) or a bogie twist index represented by the following formula (2) or (3).
Body twist index = (P1 + P3 + P6 + P8)-(P2 + P4 + P5 + P7) ... (1)
Bogie twist index of the front bogie = (P1 + P4)-(P2 + P3) ... (2)
Bogie twist index of the rear bogie = (P5 + P8)-(P6 + P7) ... (3)
However, P1 is the wheel weight of the front right wheel of the front bogie, P2 is the wheel weight of the front left wheel of the front bogie, P3 is the wheel weight of the rear right wheel of the front bogie, and P4 is the front side. P5 is the wheel weight of the front right wheel of the rear bogie, P6 is the wheel weight of the front left wheel of the rear bogie, and P7 is the wheel weight of the front left wheel of the rear bogie. The wheel weight of the wheel on the rear right side of the above is shown, and P8 shows the wheel weight of the wheel on the rear left side of the bogie on the rear side.
The vehicle body twist index, the bogie twist index of the front bogie, and the bogie twist index of the rear bogie can be calculated by detecting the wheel load as is clear from the equations (1) to (3). be. The wheel load can be detected, for example, by installing a wheel load sensor using a strain gauge or a sensor such as a load cell on the track, as described in Patent Document 1. Considering the labor and cost of maintenance and inspection, it is preferable to detect using a sensor provided in the track. However, the present invention is not limited to this, and each index is calculated using a so-called PQ monitoring bogie in which the railroad vehicle (bogie) itself is equipped with a sensor and the wheel load (and lateral pressure) can be measured by the sensor. Is also possible.
Further, when the index used for determining the presence or absence of abnormality in the state monitoring device of the railway vehicle according to the present invention is the bogie twist index, either the bogie twist index of the front bogie or the bogie twist index of the rear bogie. One trolley twist index may be used, or both trolley twist indexes may be used. That is, one of the trolley twist indexes may be input to the classifier provided in the determination device, or both trolley twist indexes may be input to the classifier.

It is preferable that the condition monitoring device for a railway vehicle according to the present invention has a configuration capable of determining not only the presence or absence of an abnormality but also the type of abnormality. According to the findings of the present inventors, for example, after determining the presence or absence of an abnormality using the vehicle body twist index, the type of abnormality can be determined using the bogie twist index.
That is, preferably, the detection device detects the vehicle body twist index and the trolley twist index, and the determination device inputs the plurality of vehicle body twist indexes detected by the detection device within a predetermined time. Based on the track or the maintenance information of the railroad vehicle, the presence or absence of an abnormality in the railroad vehicle is determined, and when it is determined that the railroad vehicle has an abnormality, the plurality of vehicle body twist indexes are detected within a predetermined time. When there is a trolley twist index whose absolute value is equal to or higher than a predetermined threshold among the plurality of trolley twist indexes detected by the detection device, it is determined that the shaft spring included in the railway vehicle is damaged and exists. If not, it is determined that an abnormality related to the secondary spring provided in the railway vehicle has occurred.

In the above preferred configuration, the bogie twist index used to determine the type of abnormality means both the bogie twist index of the front bogie and the bogie twist index of the rear bogie. Further, in the above-mentioned preferable configuration, "when there is a bogie twist index whose absolute value is equal to or higher than a predetermined threshold value" is among the bogie twist index of the front bogie and the bogie twist index of the rear bogie. It means that the absolute value of at least one of them is equal to or more than a predetermined threshold value. Further, in the above preferable configuration, "when there is no bogie twist index whose absolute value is equal to or higher than a predetermined threshold value ... (Omitted) ..." means the bogie twist index of the front bogie and the rear side. It means that the absolute values of both bogie twist indicators of the bogie are less than a predetermined threshold.
In the above preferred configuration, the axle spring is provided for each wheel, and if the axle spring of any of the wheels breaks, the weight of the bogie hanging on the wheel provided with the axle spring will be the weight of that wheel on that bogie. It hangs on the wheels located in the front-rear direction and the left-right direction. For example, if the shaft spring of the front right wheel of the bogie is damaged, the weight of the bogie on the front right wheel is applied to the front left wheel and the rear right wheel. As described above, when the shaft spring is damaged, the influence thereof extends to the wheels located in the front-rear direction and the left-right direction with respect to the wheel in which the shaft spring is damaged.
Therefore, as shown in the above equation (2), the sum of the wheel weights (P1 + P4) of the wheels located on the front right side and the rear left side of the front bogie and the front left side and the rear right side of the front bogie are located. If the absolute value of the bogie twist index (bogie twist index of the front bogie), which is the difference from the total wheel load of each wheel (P2 + P3), is equal to or greater than a predetermined threshold value, the shaft spring of the front bogie is damaged. It can be determined that it is. The same applies to the bogie twist index of the bogie on the rear side represented by the formula (3).
Further, in the above preferable configuration, a pair of left and right secondary springs (air springs) are provided for each bogie, and as an abnormality related to the secondary spring, puncture of the secondary spring and intake / exhaust of the secondary spring are controlled. The failure of the height adjusting valve can be exemplified.

According to the condition monitoring device for a railroad vehicle according to the present invention, a classifier is simply generated by machine learning using known data without requiring a large amount of time and labor for determining a threshold value as in the conventional case. Therefore, it is possible to easily determine whether or not there is an abnormality in the railroad vehicle. Further, according to the condition monitoring device for a railroad vehicle according to the present invention, since an index for an abnormal railroad vehicle is not required as teacher data, a large number of teacher data can be prepared, and machine learning of a classifier is sufficient. As a result, it is possible to improve the determination accuracy. That is, according to the condition monitoring device for a railway vehicle according to the present invention, it is possible to easily and accurately determine the presence or absence of an abnormality in the railway vehicle.

It is a schematic diagram which shows the schematic structure of the state monitoring apparatus which concerns on one Embodiment of this invention. It is a flow chart which shows the schematic operation of the state monitoring apparatus 100 shown in FIG. It is a figure which shows the fluctuation example of the vehicle body twist index. It is explanatory drawing which shows typically the procedure of generating the classifier 21 shown in FIG. 1 by using machine learning. It is explanatory drawing which shows typically the determination operation of the state monitoring apparatus 100 shown in FIG.

Hereinafter, a condition monitoring device for a railway vehicle (hereinafter, appropriately abbreviated as “condition monitoring device”) according to an embodiment of the present invention will be described with reference to the accompanying drawings as appropriate. First, the overall configuration of the condition monitoring device according to the present embodiment will be described.

FIG. 1 is a schematic diagram showing a schematic configuration of a condition monitoring device according to the present embodiment.
As shown in FIG. 1, the state monitoring device 100 according to the present embodiment is a detection device 1 for detecting an index represented by the wheel load of a wheel 31 included in a railway vehicle 3 traveling on a track, and a detection device 1. A plurality of indicators detected within a predetermined time and maintenance information of the track or railroad vehicle 3 (information on the maintenance type, date and time, etc., when maintenance was performed on the track or railroad vehicle 3 to be determined) are input. The railway vehicle 3 is provided with a determination device 2 for determining the presence or absence of an abnormality based on the input plurality of indexes and maintenance information.

The detection device 1 of the present embodiment is attached to the left and right rails R constituting the track, and is connected to the wheel load sensor 11 for measuring the wheel load of the wheels 31 included in the railroad vehicle 3 and the wheel load sensor 11. It is provided with a calculation unit 12. As the wheel load sensor 11, a wheel load sensor using a strain gauge as described in Patent Document 1 described above or a load cell can be used. The calculation unit 12 calculates each index described later based on the wheel load measured by the wheel load sensor 11 and input from the wheel load sensor 11. Specifically, for example, the calculation unit 12 stores calculation formulas for each index (formulas (1) to (3) described later), and the calculation unit 12 stores the wheel weight input from the wheel load sensor 11. It is said to be a personal computer (hereinafter referred to as a PC) in which a program for calculating each index by substituting into the above formula is installed. Then, the calculation unit 12 outputs each calculated index to the determination device 2 (classifier 21 included in the determination device 2).

The determination device 2 of the present embodiment is, for example, a program generated by using machine learning and functioning as a classifier 21 that outputs the presence or absence of maintenance of the track or the railroad vehicle 3 according to a plurality of input indexes. It is said to be an installed PC. As the classifier 21 of this embodiment, an artificial neural network (ANN) is used.
In the condition monitoring device 100 shown in FIG. 1, the calculation unit 12 of the detection device 1 and the determination device 2 are separate bodies, but the present invention is not limited to this, and the calculation unit 12 and the determination device are not limited to this. It is also possible to configure using a single PC on which a program that fulfills both functions of 2 is installed.

As an index to be input to the classifier 21, for example, a vehicle body twist index represented by the following formula (1) or a bogie twist index represented by the following formula (2) or (3) can be used. .. Further, as an index to be input to the classifier 21, a vehicle body roll index represented by the following equation (4) can be used. In this embodiment, a vehicle body twist index is used as an index to be input to the classifier 21. Further, in the present embodiment, in order to determine the type of abnormality, a bogie twist index is used as an index to be input to the determination device 2 (not input to the classifier 21).
Body twist index = (P1 + P3 + P6 + P8)-(P2 + P4 + P5 + P7) ... (1)
Bogie twist index of the front bogie = (P1 + P4)-(P2 + P3) ... (2)
Bogie twist index of the rear bogie = (P5 + P8)-(P6 + P7) ... (3)
Body roll index = (P1 + P3 + P5 + P7)-(P2 + P4 + P6 + P8) ... (4)
However, P1 is the wheel weight of the front right wheel of the front bogie, P2 is the wheel weight of the front left wheel of the front bogie, P3 is the wheel weight of the rear right wheel of the front bogie, and P4 is the front side. P5 is the wheel weight of the front right wheel of the rear bogie, P6 is the wheel weight of the front left wheel of the rear bogie, and P7 is the wheel weight of the front left wheel of the rear bogie. The wheel weight of the wheel on the rear right side of the above is shown, and P8 shows the wheel weight of the wheel on the rear left side of the bogie on the rear side.

Hereinafter, the operation of the condition monitoring device 100 according to the present embodiment will be specifically described.
FIG. 2 is a flow chart showing a schematic operation of the condition monitoring device 100 according to the present embodiment.
As shown in FIG. 2, the condition monitoring device 100 first determines whether or not to execute machine learning of the classifier 21 (S1 in FIG. 2). If the machine learning of the classifier 21 has never been executed, the machine learning is naturally executed (“Yes” in S1 of FIG. 2, S2 of FIG. 2). Further, if the machine learning of the classifier 21 has already been executed, for example, immediately after the maintenance of the track or the railroad vehicle 3 is performed, or after a predetermined predetermined time has elapsed, the condition monitoring device 100 automatically performs the machine learning. It may be configured to execute machine learning. Further, for example, the condition monitoring device 100 can be configured to execute machine learning by a manual instruction of an operator. By performing machine learning, the classifier 21 is generated. In other cases (“No” in S1 of FIG. 2), the condition monitoring device 100 executes the next operation (S3 of FIG. 2) using the existing classifier 21 without executing new machine learning. ..

FIG. 3 is a diagram showing an example of fluctuation of the vehicle body twist index. Specifically, FIG. 3 shows that for the same normal railroad vehicle 3 traveling on the same track, the wheel load of the wheel 31 is measured by the wheel load sensor 11 for about 3 years, and the vehicle body twist index is calculated by the calculation unit 12. This is the result of calculation and storage. Note that FIG. 3 includes a period in which the vehicle body twist index could not be calculated due to the suspension of the railway vehicle 3 in addition to the period in which the vehicle body twist index could not be calculated due to maintenance of the track or the railway vehicle.
As shown in FIG. 3, the vehicle body twist index suddenly changes before and after the maintenance of the track or the maintenance of the railroad vehicle 3. Although not shown, the vehicle body twist index changes abruptly even when an abnormality occurs in the railway vehicle 3 as compared with the normal case. The condition monitoring device 100 executes machine learning by utilizing the phenomenon that the index represented by the wheel load of the wheel 31 included in the railway vehicle 3 such as the vehicle body twist index suddenly changes, and generates the classifier 21.

FIG. 4 is an explanatory diagram schematically showing a procedure for generating the classifier 21 by using machine learning. As shown in FIG. 4A, during machine learning, as teacher data (combination of input / output to the classifier 21), the track or the railway vehicle 3 is not actually maintained for the normal railway vehicle 3. The combination data of a plurality of vehicle body twist indexes T1 detected within a predetermined time and no maintenance is used. In the example shown in FIG. 4A, n (for example, 40) consecutive T1 (1), T1 (2), T1 (3) ... T1 (n) are input data to the classifier 21. The vehicle body twist index T1 detected in the above is used, and "0" (a numerical value indicating no maintenance) is used as output data from the classifier 21, and machine learning is executed.

Similarly, as shown in FIG. 4 (b), during machine learning, as another teacher data, a plurality of detections of a normal railroad vehicle within a predetermined time when the track or the maintenance of the railroad vehicle 3 is actually performed. The combination data of the vehicle body twist index T1 (including the vehicle body twist index T1 before the maintenance execution and the vehicle body twist index T1 after the maintenance execution within this predetermined time) and the maintenance is used. In the example shown in FIG. 4B, n (for example, 40) consecutive T1 (1), T1 (2), T1 (3) ... T1 (n) are input data to the classifier 21. The vehicle body twist index T1 detected in the above is used, and "1" (a numerical value indicating that maintenance is present) is used as output data from the classifier 21, and machine learning is executed.
When the maintenance is actually performed, the vehicle body twist index T1 is not detected during the maintenance. Therefore, the i vehicle body twist index T1 before the maintenance and the n−i vehicle body twist after the maintenance are executed. A total of n vehicle body twist indexes T1 are used in combination with the index T1.

By executing the above machine learning, the classifier 21 inputs a plurality of vehicle body twist indexes T1 detected within a predetermined time when the track or the railway vehicle 3 is not actually maintained for the normal railway vehicle 3. When this is done, no maintenance (0) is output, and when a plurality of vehicle body twist indexes T1 detected within a predetermined time when the track or the railway vehicle 3 is actually maintained are input, maintenance is performed (1). It will be output.
As for the more detailed contents of machine learning, for example, known machine learning procedures used for artificial neural networks can be applied, and therefore the detailed description thereof will be omitted here.
The classifier 21 is preferably generated for each railroad vehicle 3 to be determined for the presence or absence of abnormality. Further, it is preferable that the classifier 21 is generated for each orbit (orbit to which the detection device 1 is attached) to be determined.

Using the classifier 21 generated as described above, the condition monitoring device 100 determines the presence / absence of an abnormality in the railway vehicle 3 which is the target for determining the presence / absence of an abnormality.
FIG. 5 is an explanatory diagram schematically showing the determination operation of the state monitoring device 100. Hereinafter, a specific description will be given with reference to FIG. 5 in addition to FIG.
First, in the classifier 21 generated by machine learning, a plurality of railroad vehicles 3 (calculated by the calculation unit 12) detected by the detection device 1 for the railroad vehicle 3 to be determined for the presence or absence of an abnormality (same as when machine learning is executed). The vehicle body twist indexes T1 (1), T1 (2), T1 (3) ... T1 (n) are input (S3 in FIG. 2, FIG. 5).
Then, when the output of the classifier 21 is no maintenance (0) (when "Yes" in S4 of FIG. 2), the determination device 2 determines that the railroad vehicle 3 is normal (S5 of FIG. 2, Figure 5), the result is output.

On the other hand, when the output of the classifier 21 has maintenance (1) (when "No" in S4 of FIG. 2), the maintenance information of the track or the railroad vehicle 3 (the track or the railroad to be determined) is sent to the determination device 2. Information on the type of maintenance, the date and time, etc., such as when the maintenance was performed on the vehicle 3) is input (S6 in FIG. 2, FIG. 5). Specifically, the determination device 2 refers to the track and the railroad vehicle 3 to be determined with respect to a predetermined management device (not shown) that collectively manages maintenance information for the plurality of tracks and the plurality of railroad vehicles 3. Requests the transmission of maintenance information, and thereby receives maintenance information from the management device. In the present embodiment, the case where the maintenance information is input to the determination device 2 for the first time when the output of the classifier 21 has maintenance (1) is illustrated, but the present invention is not limited to this, and the classification is not limited to this. It is also possible to adopt a configuration in which maintenance information is input to the determination device 2 in advance regardless of the output of the device 21.

When the output of the classifier 21 has maintenance (1), the determination device 2 detects a plurality of (n) vehicle body twist indexes T1 input to the classifier 21 based on the input maintenance information within a predetermined time. In (S7, FIG. 5 of FIG. 2), it is determined whether or not the maintenance of the track or the railroad vehicle 3 is actually performed. Specifically, the date and time within the predetermined time when the plurality of vehicle body twist indexes T1 input to the classifier 21 are detected are stored in the determination device 2, and the determination device 2 is configured to store the stored predetermined time. If the date and time in the above are compared with the date and time when the maintenance included in the maintenance information was actually performed and they match (the earliest date and time and the latest date and time within the predetermined time when multiple vehicle body twist indexes T1 are detected). (If the date and time when the maintenance was actually performed is included), it is determined that the maintenance of the track or the railroad vehicle 3 was actually performed, and if they do not match, the maintenance of the track or the railroad vehicle 3 is performed. Judge that it is not actually done.

When the track or railroad vehicle 3 is actually maintained within a predetermined time when the determination device 2 detects a plurality of vehicle body twist indexes T1 input to the classifier 21 (“Yes” in S7 of FIG. 2). In the case of), it is determined that the railroad vehicle 3 is normal (S8 in FIG. 2, FIG. 5), and the result is output.
On the other hand, when the determination device 2 actually does not actually perform maintenance on the track or the railroad vehicle 3 within a predetermined time when a plurality of vehicle body twist indexes T1 input to the classifier 21 are detected (FIG. 2). In the case of "No" in S7), it is determined that an abnormality has occurred in the railway vehicle 3 (S9 in FIG. 2, FIG. 5), and the result is output.

As a preferred embodiment, the condition monitoring device 100 according to the present embodiment has a configuration capable of determining not only the presence or absence of an abnormality but also the type of the abnormality.
Specifically, when the determination device 2 included in the state monitoring device 100 determines that an abnormality has occurred in the railroad vehicle 3 (S9, FIG. 5 in FIG. 2), a plurality of (n) vehicle body twist indexes T1 are set. Multiple (n) bogie twist indexes continuously detected by the detection device 1 within the detected predetermined time (bogie twist index of the front bogie and bogie twist index of the rear bogie) T2 (1), T2 (2) ), T2 (3) ... T2 (n), it is determined whether or not there is a bogie twist index T2 whose absolute value is equal to or greater than a predetermined threshold value (S10, FIG. 5 in FIG. 2).
As this threshold value, it is preferable to use a threshold value that fluctuates at an appropriate timing, instead of using the same fixed threshold value that was initially set. For example, the average value of a plurality of bogie twist indexes T2 detected immediately after maintenance of the track or the railroad vehicle 3 is performed (a state in which it is very likely that no abnormality has occurred in the railroad vehicle 3) is calculated, and this average is calculated. It is possible to set a threshold for a value that exceeds the value.

According to the findings of the present inventors, the shaft spring included in the railcar 3 is damaged when the bogie twist index T2 whose absolute value is equal to or higher than a predetermined threshold value is present, and the railcar 3 is provided when the absolute value is not present. In most cases, an abnormality related to the secondary spring has occurred.
Therefore, the determination device 2 determines that the shaft spring included in the railcar 3 is damaged when the bogie twist index T2 whose absolute value is equal to or higher than a predetermined threshold value is present among the plurality of bogie twist indexes T2. However, when it does not exist (S11 in FIG. 2 and FIG. 5), an abnormality related to the secondary spring provided in the railcar 3 (a puncture of the secondary spring or a failure of the height adjusting valve that controls the intake / exhaust of the secondary spring) occurs. It is determined that the operation is performed (S12 in FIG. 2, FIG. 5), and the result is output.

As described above, according to the state monitoring device 100 according to the present embodiment, maintenance of the track or the railroad vehicle 3 is actually performed for the normal railroad vehicle 3 as the teacher data in the machine learning for generating the classifier 21. A plurality of indicators detected within a predetermined time not performed (for example, vehicle body twist index T1) and a plurality of indicators detected within a predetermined time when maintenance of the track or railroad vehicle 3 is actually performed (for example, vehicle body twist index T1). It is only necessary to prepare the index T1). That is, since the index for the abnormal railroad vehicle 3 is not required as the teacher data, a large number of teacher data can be prepared, and the machine learning of the classifier 21 can be sufficiently performed, and as a result, the determination accuracy is improved. Is possible.
Further, according to the condition monitoring device 100 according to the present embodiment, not only the presence or absence of an abnormality but also the type of the abnormality can be determined, so that appropriate measures can be taken according to the type of the abnormality.

1 ... Detection device 2 ... Judgment device 3 ... Railway vehicle 11 ... Wheel load sensor 12 ... Calculation unit 21 ... Classifier 31 ... Wheel 100 ... State of railway vehicle Monitoring device

Claims (3)

A detection device that detects an index represented by the wheel load of a railroad vehicle running on a track, and a detection device.
A determination device for determining the presence or absence of an abnormality in the railway vehicle based on the plurality of indicators detected by the detection device within a predetermined time and the input maintenance information of the track or the railway vehicle is provided.
The determination device is
A classifier for inputting the plurality of indicators and outputting the presence or absence of maintenance of the track or the railway vehicle is provided.
If the output of the classifier is without maintenance, it is determined that the railroad vehicle is normal.
When the output of the classifier has maintenance, the maintenance of the track or the railroad vehicle is actually performed within a predetermined time when a plurality of the indexes input to the classifier are detected based on the input maintenance information. It is determined whether or not the maintenance has been performed, and when the maintenance is actually performed, it is determined that the railroad vehicle is normal, and when the maintenance is not actually performed, it is determined that the railroad vehicle has an abnormality. death,
The classifier outputs no maintenance for the normal railroad vehicle when a plurality of the indicators detected within a predetermined time when the track or the railroad vehicle is not actually maintained is input. Generated using machine learning to output with maintenance when a plurality of the indicators detected within a predetermined time when the track or the rolling stock is actually maintained are input.
A condition monitoring device for railroad vehicles. The index is a vehicle body twist index represented by the following formula (1), or a bogie twist index represented by the following formula (2) or (3).
The condition monitoring device for a railroad vehicle according to claim 1.
Body twist index = (P1 + P3 + P6 + P8)-(P2 + P4 + P5 + P7) ... (1)
Bogie twist index of the front bogie = (P1 + P4)-(P2 + P3) ... (2)
Bogie twist index of the rear bogie = (P5 + P8)-(P6 + P7) ... (3)
However, P1 is the wheel weight of the front right wheel of the front bogie, P2 is the wheel weight of the front left wheel of the front bogie, P3 is the wheel weight of the rear right wheel of the front bogie, and P4 is the front side. P5 is the wheel weight of the front right wheel of the rear bogie, P6 is the wheel weight of the front left wheel of the rear bogie, and P7 is the wheel weight of the front left wheel of the rear bogie. The wheel weight of the wheel on the rear right side of the above is shown, and P8 shows the wheel weight of the wheel on the rear left side of the bogie on the rear side. The detection device detects the vehicle body twist index and the bogie twist index, and
The determination device determines whether or not there is an abnormality in the railway vehicle based on the plurality of vehicle body twist indexes detected by the detection device within a predetermined time and the input maintenance information of the track or the railway vehicle. When it is determined that an abnormality has occurred in the railroad vehicle, the absolute value of the plurality of trolley twist indexes detected by the detection device within the predetermined time when the plurality of vehicle body twist indexes are detected is a predetermined threshold. When the trolley twist index equal to or greater than the value is present, it is determined that the shaft spring provided in the railroad vehicle is damaged, and when it is not present, it is determined that an abnormality related to the secondary spring provided in the railroad vehicle has occurred.
The condition monitoring device for a railroad vehicle according to claim 2.

Images