Technical field
[0001]
　The present invention detects the vehicle information represented by the heavy wheel of a wheel comprising railway vehicle traveling on a track or the like, based on the detected vehicle information, railway vehicle determines the vehicle state such as the presence or absence of a railway vehicle abnormality about the state monitoring device. In particular, the present invention relates to a railway vehicle state monitoring apparatus capable of determining easily without requiring much labor for adjustment of parameters for determining the vehicle state.
Background technique
[0002]
　Conventionally, in order to improve the running safety of the railway vehicle, fitted with various sensors on operating the vehicle (rail vehicles in commercial operation), the state of operating the vehicle in the traveling operating line by monitoring by sensors, detecting an abnormality in operating the vehicle during running of the operating vehicle online real-time monitoring being performed (e.g., see Patent documents 1 and 2).
　However, in the above method, when trying to detect the abnormality of the railway vehicle during traveling, it must be fitted with sensors for all railway vehicles, time-consuming maintenance and inspection, etc. of the sensor. Therefore, there is a significant cost problem is required with unable abnormally readily detectable railcar.
[0003]
　To solve the above problems, a wheel load sensor for measuring the wheel loads of the wheels provided on the track, a method of detecting an abnormality of the railway vehicle based on the magnitude of the index represented by wheel loads measured by wheel load sensor It has been proposed (e.g., see Patent Document 3).
　According to such a method, it is possible to easily and inexpensively detect the abnormality of the railway vehicle than in the case of mounting the sensor for each rail vehicle.
[0004]
　However, the method described in Patent Documents 1 to 3 are all compare metrics to a predetermined threshold, a method for detecting an abnormality of a railway vehicle according to the magnitude. Threshold for detecting the abnormality, the structure of the railway vehicle, the loading conditions, appropriate values ​​may vary depending on the running conditions and the like. Therefore, in order to accurately detect the abnormality, it is necessary to determine the number of thresholds for each of these conditions requires tremendous time and effort.
CITATION
Patent Literature
[0005]
Patent Document 1: JP 2009-220815 Patent Publication
Patent Document 2: JP 2011-51518 JP
Patent Document 3: JP 2013-120100 JP
Summary of the Invention
Problems that the Invention is to Solve
[0006]
　The present invention has been made in order to solve the problems of such prior art, easily determined without requiring much labor for adjustment of parameters for determining the vehicle state such as the presence or absence of a railway vehicle abnormality and to provide a railway vehicle state monitoring apparatus capable.
Means for Solving the Problems
[0007]
　To solve the above problems, the railway vehicle state monitoring apparatus according to the present invention includes a detection device for detecting vehicle information railway vehicle is represented by the wheel loads, etc. of the wheels comprising running on a track, the detected vehicle information There are input, and a determination device having a classifier to output a vehicle state such as the presence of abnormalities of the railway vehicle. Then, the classifier using the vehicle information and the vehicle state of the railway vehicle the vehicle condition is known as teacher data, to output the known vehicle state when the vehicle information is input , characterized in that it is produced using a machine learning.
[0008]
　According to the present invention, the vehicle state of outputting the vehicle information and the vehicle state and the teacher data (vehicle information input vehicle condition of the presence or absence of a vehicle abnormality is represented by the wheel loads, etc. of the wheels of the railway vehicle is known used as a combination) of the classifier is generated by machine learning. Then, the vehicle information detected by the detection device, that the detected vehicle information is input to the classifier, the state of the vehicle as the determination result is output from the classifier. According to the present invention, conventional without requiring a great deal of time and effort for determining the threshold value as, for good only generate a classifier by machine learning using the known data, the vehicle state easily it is possible to determine.
[0009]
　Preferably, the vehicle information input on the detected classifier by the detection device, the railway vehicle is represented by the wheel loads of the wheels comprising the classifier, as the vehicle state, the rail vehicle and it outputs the type of presence or absence of abnormality and abnormality.
[0010]
　Specifically, for example, the railway vehicle includes a pair provided before and after the carriage having wheels of left and right two pairs of front and rear vehicle information detected by the detection device has the following formula (1) and (2 a primary spring abnormality index of the primary spring abnormality indicators and rear of the front bogie truck represented respectively), the secondary spring abnormality index of the rail vehicle represented by the following formula (3) There, the vehicle state output from the classifier is an abnormal presence or absence of and abnormal types of the rail vehicle.
　The primary spring abnormality index of the front bogie = (P1 + P4) - (P2 + P3) · · · (1)
　primary spring abnormality index of the rear bogie = (P5 + P8) - ( P6 + P7) ··· (2)
　of the rail vehicle secondary spring abnormality index = (P1 + P3 + P6 + P8) - (P2 + P4 + P5 + P7) ··· (3)
　However, P1 is a wheel load of the front right wheel of the front bogie, P2 is a wheel load of the front left wheel of the front bogie , P3 is a wheel load of the rear right wheel of the front bogie, P4 is a wheel load of the rear left wheel of the front bogie, P5 is a wheel load of the front right wheel of the rear bogie, P6 after wheel wheel load of the front left side of the bogie, P7 is a wheel load of the rear right wheel of the rear bogie, P8 denotes a wheel load of a wheel in the rear left of the rear bogie.
[0011]
　The primary spring abnormality is abnormality of the primary spring provided on the carriage (axial springs), for example, it can be exemplified breakage of the coil spring provided on the carriage.
　The primary spring is provided for each wheel, the primary spring of one of the wheels is abnormal, the weight of the bogie to which the primary spring was hanging on wheels provided, before and after the wheels in the bogie It applied to the wheels located on the direction and the lateral direction. For example, when the front right wheel of the primary spring of the bogie becomes abnormal, the weight of the truck had hanging on the front right wheel is applied to the front left wheel and rear right wheels. Thus, when the primary spring abnormality occurs, the effect would extend to the wheels located in the front-rear direction and the left-right direction relative to the wheel in which the primary spring abnormality occurs.
　Therefore, the sum of the respective wheel load of the wheel located on the front right side and rear left side of the front bogie and (P1 + P4), the sum of the respective wheel load of the wheel located on the front left side and rear right side of the front bogie (P2 + P3) by evaluating the difference between, it would be able to detect the primary spring abnormality of the front bogie. Therefore, in the preferred configuration described above, as the vehicle information detected by the detecting device (vehicle information to be input to the classifier) uses a primary spring abnormality index of the front bogie of the formula (1). The same reasons also the uses a primary spring abnormality reference side of the carriage after the formula (2).
[0012]
　On the other hand, the secondary spring abnormality is abnormality in the secondary spring provided on the carriage, for example, can be exemplified an abnormality in supply and exhaust air spring provided on the carriage.
　The secondary spring left and right of the front and rear bogie, is provided on the front, rear, left and right of a railway vehicle in other words, the secondary spring one of the bogie can not be support the weight of the vehicle body becomes abnormal, the carriage body weight that was applied to the wheels of the secondary spring is provided side would take to the wheel in the vicinity of the secondary spring which is positioned in the longitudinal direction and the lateral direction of the secondary spring. For example, when the right side of the secondary spring of the front bogie becomes abnormal body weight that was applied to the wheel on the right side of the front bogie (front and rear pair of wheels) is located on the left side of the front bogie wheel It applied to the wheels (front and rear pair of wheels) positioned on the right side of the truck (front and rear pair of wheels) and the rear side. In this way, the secondary spring abnormality occurs, the effect would extend to the wheel in the vicinity of the secondary spring which is positioned in the longitudinal direction or lateral direction relative to the secondary spring abnormality occurs.
　Therefore, the sum of the respective wheel load of the wheel located on the left of the right and rear of the front bogie truck and (P1 + P3 + P6 + P8 ), each of the wheels of the wheel located on the right side of the left and rear of the carriage of the front bogie by evaluating the difference between the sum of the weight (P2 + P4 + P5 + P7 ), it is considered possible to detect the secondary spring abnormality railcar. Therefore, in the preferred configuration described above, as the vehicle information detected by the detecting device (vehicle information to be input to the classifier), it is used secondary spring abnormality reference railcar of formula (3).
[0013]
　As described above, in the preferred configuration described above, as the vehicle information to be detected (vehicle information to be input to the classifier), primary spring abnormality index of the front bogie, the rear bogie primary spring abnormality indicators and railcar for use secondary spring abnormality index, it can be expected that the accuracy of determining the presence or absence of and abnormal types of vehicle to be output from the classifier as the determination result abnormality (primary spring and secondary spring abnormalities associated) is increased.
　Incidentally, the primary spring abnormality index of the front bogie, the secondary spring abnormality index of the primary spring abnormality indicators and rail vehicle of the rear of the truck, as is clear from equation (1) to (3), wheel wheel It can be calculated by detecting the weight. Wheel load of the wheel, for example, as described in Patent Document 3, can be detected by installing a wheel load sensor and the load cell with strain gauge track.
[0014]
　Here, the secondary spring abnormality index of the primary spring abnormality index and railcar of the front and rear bogie described above, becomes an index for detecting abnormalities related to the primary spring and secondary spring simultaneously, railcar It can also be considered to be the indicator of the individual differences. That is also conceivable to use one of the rail vehicle indicators to identify the other railway vehicle, as an index for determining the vehicle knitting other words.
　In this case, similarly to the above-mentioned preferable configuration, as the vehicle information to be detected (vehicle information to be input to the classifier) primary spring abnormality index of the front bogie, the rear bogie primary spring abnormality indicators and railcar while using the secondary spring abnormality index, comprising a vehicle state is determined result to the vehicle knitting.
　In other words, preferably, the rail vehicle pair provided before and after the carriage having wheels of left and right two pairs of front and rear vehicle information detected by the detection device, in the above equations (1) and (2) a primary spring abnormality index of the primary spring abnormality indicators and rear of the front bogie truck respectively represented a secondary spring abnormality index of the rail vehicle represented by the above formula (3), vehicle state output from the classifier is a vehicle organization of the railway vehicle.
[0015]
　According to such a preferred configuration, the vehicle information relating to the organization without receiving from the outside, the primary springs of the primary spring abnormality indicators and rear of the front bogie of the vehicle information (railway vehicle detected by the detector carriage abnormality it is possible to determine the vehicle knitting by using the index, only secondary spring abnormality index) of a railway vehicle. Thus, for example, as in the preferred arrangement described above, the classifier that outputs an abnormality presence or absence of and abnormal types of railway vehicles (first classifier), as in this preferred configuration, the classification of outputting the vehicle knitting vessels by (second classifier) ​​with both to produce the determination device of advance provided inputs respectively vehicle information detected by the detecting device to both the first classifier and the second classifier , the presence or absence of and abnormal types of abnormal railway vehicle, it is possible to determine the vehicle organization of the railway vehicle at the same time. Thus, string or abnormal presence or absence of and the abnormality type of the determined rail vehicle that such things for any vehicle knitting, the determination result of the abnormality such as the presence or absence of a railway vehicle by receiving the information about the vehicle knitting externally it is not necessary to attach, it can be easily identified.
[0016]
　Here, classifier outputs an abnormal type whether or abnormality of the rail vehicle as the vehicle condition to generate a (first classifier) is the time of machine learning, a normal rail vehicle (no abnormalities railway vehicle) and teacher data for, there is a need and teacher data about the unusual railway vehicle.
　Teacher data of normal railway vehicle includes a primary spring abnormality index of normal primary spring of the front bogie of the vehicle information (railway vehicle of a railway vehicle abnormality indicators and the rear of the truck, the secondary spring abnormality reference railcar since) is actually detectable, and it is easy to prepare. In contrast, the teacher data abnormal railcar, since itself providing a abnormal railcar difficult (to prepare Moreover many very difficult) is, it is difficult to actually detect the vehicle information , it is not easy to prepare.
　Therefore, the teacher data abnormal rail vehicle as the vehicle information, the vehicle information detected for normal railway vehicle, for example, it is preferable to use those calculated by numerical simulation using a general-purpose mechanism analysis software.
[0017]
　That is, 1 is preferable that the classifier, as the vehicle information of the training data, the primary spring abnormality index of the front bogie which is actually detected by the detection device for a normal railway vehicle, of the rear side of the carriage and next spring abnormality reference and secondary spring abnormality index of the railway vehicle, the primary spring abnormality index of the front bogie of the normal rail vehicle, 2 primary spring abnormality index and the railway vehicle of the rear of the truck using a secondary spring abnormality index follows the spring primary spring abnormality index of the front bogie abnormal railcar calculated by numerical simulation of the abnormal indication, the primary spring abnormality index and the railway vehicle of the rear of the truck together with, as the vehicle state of the training data, using the type of a known presence or absence of abnormality and the abnormality of the normal rail vehicle and said abnormal railway vehicle, the normal rail vehicle and When the vehicle information of the serial abnormal railway vehicle is input, to output a vehicle state of the normal rail vehicle and said abnormal railcar, it is generated using the machine learning.
[0018]
　According to such a preferred configuration, as teacher data for the classifier machine learning, abnormal since there is no need to prepare a actually detected vehicle information about the railway vehicle, it is possible to easily machine learning.
[0019]
　The vehicle information, but can also be detected using a sensor provided on each rail vehicle, given the time and cost of maintenance and inspection, etc., can be detected using a sensor provided on the track preferable.
　That is, the detection device preferably comprises a sensor provided in the track. However, the detection device may also be having a sensor provided on the railway vehicle. The sensor provided in the track, the wheel load sensor described in Patent Document 3 described above, as a sensor provided on the railway vehicle can be exemplified by the sensor described in Patent Documents 1 and 2 mentioned above.
[0020]
　As the classifier, support vector machines and neural networks, although it is possible to employ various configurations as long as can be generated using a machine learning, especially naive Bayes classifier having the advantage that the mechanism is simple and calculation speed is faster vessel is preferably used.
Effect of the invention
[0021]
　According to the present invention, conventional without requiring a great deal of time and effort for determining the threshold value as, for good only generate a classifier by machine learning using the known data, the vehicle state easily it is possible to determine.
BRIEF DESCRIPTION OF THE DRAWINGS
[0022]
[1] Figure 1 is a block diagram illustrating a classifier determining device comprises railway vehicle state monitoring apparatus according to a first embodiment of the present invention is provided.
FIG. 2 is an explanatory diagram for explaining a determination method by generating method and classifier classifier shown in FIG.
FIG. 3 shows the determination results obtained when you enter vehicle information abnormal railcar of judgment object data classifier shown in FIG.
FIG. 4 shows the determination results obtained when you enter vehicle information for a normal rail vehicle of the determination target data classifier shown in FIG.
FIG. 5 is a block diagram illustrating a classifier determining device comprises railway vehicle state monitoring device according to a second embodiment of the present invention is provided.
FIG. 6 shows the determination results obtained when you enter judgment object data classification apparatus shown in FIG.
[7] FIG. 7 is a schematic diagram showing a schematic configuration of a railway vehicle state monitoring apparatus according to the first and second embodiments of the present invention.
DESCRIPTION OF THE INVENTION
[0023]
　Hereinafter, with reference to the accompanying drawings as appropriate, railway vehicle state monitoring apparatus according to an embodiment of the present invention (hereinafter, appropriately abbreviated as "monitor") will be described. First, a description will be given of the overall structure first.
[0024]
　
　FIG. 7 is a schematic diagram showing a schematic configuration of a monitoring apparatus according to an embodiment of the present invention. The first according to an embodiment the monitoring device and monitoring apparatus according to the second embodiment described later (hereinafter collectively referred to as appropriate as "monitoring device according to the present embodiment"), the determination device comprises and determination method according generating method and classifier classifier (determination content) of a different only has an overall configuration that is common as shown in FIG. 7 to both.
　As shown in FIG. 7, the monitoring device 100 according to this embodiment includes a detector 1 for detecting a vehicle information represented by the wheel loads of the wheels 31 which includes the railway vehicle 3 traveling on the track, the detected vehicle information There are input, and a determination unit 2 which comprises a classifier 21 for outputting a vehicle condition such as the presence or absence of a rail vehicle 3 abnormal.
[0025]
　Detector 1 monitoring apparatus 100 according to this embodiment is mounted on rails R of the right and left which constitutes the trajectory, the wheel load sensor 11 for measuring the wheel loads of the wheels 31 which railway vehicle 3 comprises, wheels and a calculating unit 12 connected to the heavy sensor 11. The wheel load sensor 11, the wheel load and sensor using strain gauges as described in Patent Document 3 described above, it is possible to use a load cell. Arithmetic unit 12 is measured by the wheel load sensor 11, based on the wheel loads transmitted from the wheel load sensor 11, calculates the respective abnormality reference which will be described later. Specifically, for example, computing unit 12, the calculation formula of the abnormality reference (equations (1) to (3)) are stored, and substituting the wheel load transmitted from the wheel load sensor 11 in the formula program for calculating each abnormality reference Te is the installed PC (personal computer). Then, the arithmetic unit 12 outputs the abnormality reference calculated for judgment apparatus 2 (classifier 21) as vehicle information.
[0026]
　Determining device 2 monitoring apparatus 100 according to this embodiment, for example, be generated using a machine learning program serve as a classifier 21 which outputs a vehicle state corresponding to the input vehicle information is installed It has been with the PC.
　In the monitoring apparatus 100 illustrated in FIG. 7, an operation unit 12 of the detection device 1, although the determination apparatus 2 is provided separately from, the present invention is not limited thereto, an arithmetic unit 12 determining apparatus 2 it is also possible to construct using a single PC with program to fulfill both functions.
[0027]
　Hereinafter, the monitoring apparatus 100 according to the first and second embodiments of the present invention will be described in order.
[0028]
　
　FIG 1 is a block diagram illustrating a classifier that determination device monitoring apparatus according to the first embodiment of the present invention is provided. 1 (a) is a block diagram illustrating how to generate a classifier using a machine learning is the block diagram showing a state determining vehicle state by referring to FIG. 1 (b) was generated classifier .
　Monitoring according to the first embodiment device 100 includes a first step of detecting a vehicle information represented by wheels of a wheel weight and the like comprising the railway vehicle traveling on the track detection apparatus 1, the detected vehicle information determination device input to the classifier 21 2 is provided, railway vehicles from the classifier 21 (vehicle state railway vehicle is unknown) performing a second step of outputting a vehicle condition such as the presence or absence of an abnormality (FIG. 1 (b )reference).
[0029]
　Monitoring apparatus 100 according to the first embodiment is directed to a railway vehicle to a pair comprising a carriage having wheels of left and right two pairs of longitudinal back and forth.
　Then, (vehicle information input to the classifier 21) vehicle information detected by the first step, as shown in FIG. 1 (b), railway respectively represented by the following formula (1) and (2) a primary spring abnormality index of the front of the primary spring abnormality indicators and the rear of the bogie truck of the vehicle, there is a secondary spring abnormality reference railcar represented by the following formula (3).
　The primary spring abnormality index of the front bogie = (P1 + P4) - (P2 + P3) · · · (1)
　primary spring abnormality index of the rear bogie = (P5 + P8) - ( P6 + P7) ··· (2)
　of the rail vehicle secondary spring abnormality index = (P1 + P3 + P6 + P8) - (P2 + P4 + P5 + P7) ··· (3)
　However, P1 is a wheel load of the front right wheel of the front bogie, P2 is a wheel load of the front left wheel of the front bogie , P3 is a wheel load of the rear right wheel of the front bogie, P4 is a wheel load of the rear left wheel of the front bogie, P5 is a wheel load of the front right wheel of the rear bogie, P6 after wheel wheel load of the front left side of the bogie, P7 is a wheel load of the rear right wheel of the rear bogie, P8 denotes a wheel load of a wheel in the rear left of the rear bogie.
　Each abnormality reference represented by the above formula (1) to (3), as described above, are calculated by the arithmetic unit 12 to the detection device 1 is provided.
　Further, the vehicle state output from the classifier 21 in the second step, as shown in FIG. 1 (b), an abnormality of the existence of and abnormal types of rail vehicles. Specifically, the classifier 21 of the first embodiment, since there is a naive Bayesian classifier, as described below, the probability railway vehicle is normal, each type of abnormality in railway vehicle (FIG. 1 (b in), for convenience, abnormal a, describes only abnormal B) calculates the probability that occurs. Then, the determination unit 2, of the probability that the classifiers of each type output from 21 abnormality has occurred, and outputs a high vehicle state the most probable as a final determination result.
[0030]
　As shown in FIG. 1 (a), the second classifier 21 used in the step, the vehicle information railcar vehicle state is known as the teacher data (combination of vehicle state of outputting the vehicle information to be input) and with vehicle state, the to output a known vehicle condition, are generated using the machine learning when the vehicle information is input.
　Specifically, when the machine learning and teacher data for normal railway vehicle (no abnormalities railway vehicle), it is necessary and teacher data for unusual railcar.
　Teacher data of normal railway vehicle includes a primary spring abnormality index of normal primary spring of the front bogie of the vehicle information (railway vehicle of a railway vehicle abnormality indicators and the rear of the truck, the secondary spring abnormality reference railcar since) is actually detectable, and it is easy to prepare. In contrast, the teacher data abnormal railcar, since itself providing a abnormal railcar difficult (to prepare Moreover many very difficult) is, it is difficult to actually detect the vehicle information , it is not easy to prepare.
　Therefore, the teacher data abnormal rail vehicle as the vehicle information, it is preferable to use those calculated by numerical simulation from the detected vehicle information about the normal rail vehicle.
[0031]
　That is, the classifier 21 used in the second step, as the vehicle information teacher data, the primary spring abnormality index of normal actually detected primary spring abnormality index of the front bogie for railway vehicles, the rear bogie and and secondary spring abnormality index of railway vehicles, were calculated by numerical simulation of the secondary spring abnormality index of the primary spring abnormality indicators and rail vehicles of normal primary spring abnormality index of the front bogie, the rear bogie the primary spring abnormality index of the front bogie of the abnormal vehicle, the use of a secondary spring abnormality index of the primary spring abnormality indicators and rail vehicle of the rear bogie, as the vehicle state of the training data, and normal railcar using a type known presence or absence of abnormality and the abnormality of abnormal railway vehicle, when the vehicle information of normal railcar and abnormal railway vehicle is input, the normal railcar and abnormal railway cars To output a vehicle state, you are preferably generated using a machine learning.
[0032]
　The classifier 21 of the first embodiment, naive Bayes classifier is used. Hereinafter, the determination method according to the generation method (machine learning methods) and classifier 21 of classifier 21 of the first embodiment, with reference to FIG. 2 will be described more specifically.
　Figure 2 is an explanatory diagram for explaining a determination method by generating method and classifier 21 of classifier 21.
　First, as shown in FIG. 2 (a), by entering the teacher data to the classifier 21, the frequency distribution of the vehicle information for each vehicle state is formed. In the example shown in FIG. 2 (a), the vehicle state (normal, abnormal A, abnormal B) are shown the frequency distribution of the primary spring abnormality index of the formed front side of the carriage for each actually (primary spring abnormality index of the rear bogie, the secondary spring abnormality reference railcar) other vehicle information similar frequency distribution also is formed. Further, in the example shown in FIG. 2 (a), for convenience, the vehicle condition is normal, abnormal A, shows a case where three abnormal B but in practice the number corresponding to the number of abnormality type frequency distribution is formed.
[0033]
　Next, as shown in FIG. 2 (b), the normal distribution based on the frequency distribution that is formed as described above (the probability density distribution) are formed. In the example shown in FIG. 2 (b), the vehicle state (normal, abnormal A, abnormal B) are shown superimposed normal distribution of the primary spring abnormality index of the formed front side of the carriage for each actually (primary spring abnormality index of the rear bogie, the secondary spring abnormality reference railcar) other vehicle information (see FIG. 2 (c)) like the normal distribution is formed also stored.
　As described above, the classifier 21 is generated.
[0034]
　In the case where actually detect the vehicle information about the abnormal railway vehicle, as described above with reference to FIG. 2 (a), the abnormal railcar (abnormal A, the abnormal B) also actually detected and forming a frequency distribution of the vehicle information, it is possible to form a normal distribution (probability density distribution) shown in Figure based on the frequency distribution 2 (b).
　Use, however, as mentioned above, for it to actually detect the vehicle information about the abnormal railcar is difficult, for the unusual railcar, the vehicle information, the vehicle information detected for normal railcar it is considered to be calculated by the stomach numerical simulation. Specifically, for example, in the example shown in the following (1) to the procedure of (4), the normal distribution of the vehicle information about the abnormal railcar (probability density distribution) (FIG. 2 (b), the anomaly A, it is conceivable to form a normal distribution) of abnormal B.
　(1) assuming a normal railway vehicle, a general-purpose mechanism analysis software (for example, multi-body dynamics analysis tool steel shim pack Japan Co., Ltd. "SIMPACK") to perform the numerical simulation using, vehicle information (the front side of the truck to the calculated secondary spring abnormality index) numerical results of the primary spring abnormality index, the rear of the primary spring abnormality indicators and a railway vehicle bogie.
　(2) assuming an abnormal railcar running numerical simulation using the general-purpose mechanism analysis software, vehicle information (primary spring abnormality index of the front of the primary spring abnormality index of the truck, the rear bogie and calculating a secondary spring abnormality index) numerical results of the railway vehicle. In this case, (in the example shown in FIG. 2 (b), abnormal A, abnormal B) types of assumed abnormal calculating the numerical results for each.
　(3) Based on the above (1) and (2), in a normal rail vehicle and abnormal railcar, how changing whether the vehicle information to determine the amount of change. That is, by subtracting the numerical results of normal railway vehicle calculated from the numerical results of abnormal railway vehicle calculated in (1) above (2), obtaining the amount of change.
　(4) Average As described above, a normal distribution formed by using the actually detected vehicle information about the normal rail vehicle (probability density distribution) (see FIG. 2 (b)), without changing its standard deviation σ only the value μ and Shifts only the change amount obtained in the above (3), the Shifts the normal distribution is calculated as a normal distribution of the vehicle information about the abnormal railcar. Calculating This is a normal distribution of the vehicle information on the normal distribution and abnormal railcar vehicle information on the successful railway vehicles, although the average value μ of each other different, the standard deviation σ is based on the assumption that it will be equal to it is a method.
[0035]
　Next, as shown in FIG. 2 (c), the classifier 21 generated in the manner described above, the primary spring abnormality index of each vehicle information (the front side of the carriage is detected in the first step, the rear bogie the primary spring abnormality indicators, the secondary spring abnormality reference railcar) is input indicating the respective vehicle information point is input labeled "detection value" (FIG. 2 (c)). Classifier 21, in accordance with the value of the vehicle information input, calculates the probability of each vehicle state. In the example shown in FIG. 2 (c), according to the value of the primary spring abnormality index of the front bogie entered, the probability railway vehicle is normal, the probability of abnormal A railway vehicle has occurred, the railway vehicle probability of abnormal B occurs with n respectively 1 , a 1 , b 1 is calculated as. Further, according to the value of the primary spring abnormality reference side of the truck after the input, the probability railway vehicle is normal, the probability of abnormal A railway vehicle has occurred, the abnormality B railcars occurring probability the n respectively 2 , a 2 , b 2 is calculated as. Furthermore, depending on the value of the secondary spring abnormality index of railway vehicle entered, the probability railway vehicle is normal, the probability of abnormal A railway vehicle has occurred, the probability of abnormal B railcars occurs respectively n 3 , a 3 , b 3 is calculated as.
　The classifier 21 is based on the probabilities calculated as described above, the probability P railway vehicle is normal N (equation in Figure 2 (c) (4)) , abnormal A railway vehicle has occurred probability P a (formula (5 in FIG. 2 (c))), the probability P is abnormal B railcars has occurred B is calculated (equation (6) in FIG. 2 (c)) the.
　Finally, decision device 2, the probability P classifier 21 calculates N , P A , P B is output as a final determination result high vehicle state the most probable of.
[0036]
　Hereinafter, an example of a result of determining an abnormality type whether or abnormality of the railway vehicle by monitoring apparatus 100 according to the first embodiment.
　Five vehicles organization of X series (a knitted, b knitting, c knitting, d knitting, e knitting) for, under the following conditions, the primary spring of the vehicle information (the front side of the carriage as it passes through the path curve section the primary spring abnormality indication of abnormality reference, the rear bogie, and detecting secondary spring abnormality index) of a railway vehicle to generate a classifier 21 by a machine learning using the detected vehicle information, the generated classifier 21 It was determined vehicle condition by inputting the detected vehicle information.
　(A) Target curved section
　　, the entrance relaxation curve: length 47m
　　-yen Curve: Length 60.1M, radius 251m, Kant 0.065 m, slack 0.009 m
　　-exit transition curve: length 47m
　　-wheel load sensor installation position: position of 15m from the starting point of the circle curve
　(B) using the data
　　, teacher data: the curved section detects the normal rail vehicle passing through the vehicle information (10 days), and normal railway the detected abnormal railcar calculated by numerical simulation of the vehicle information about the vehicle vehicle information and the abnormality of type
　　-determination target data: the curved section detects the normal rail vehicle passing through the vehicle information (after the teacher data acquisition 13 days), and calculated by numerical simulation using a general-purpose mechanism analysis software from the vehicle information about the normal railway vehicle the detected Abnormal vehicle information railcar
　　Incidentally, for any type of abnormality is determined by abnormality of type and determination target data included in the training data,
　(1) air spring provided on the inner rail side of the front bogie connected leveling valve has failed, the abnormality remains performing the discharging operation (abbreviated as "front inside軌排air"),
　(2) connected to an air spring provided on the outer rail side of the front bogie by leveling valve has failed, will remain subjected to the discharging operation (abbreviated as "front outer軌排gas"),
　(3) leveling connected to an air spring provided on the inner rail side of the front bogie valve has failed, the air supply operation to leave abnormally been (abbreviated as "front inside軌給air"),
　(4) a leveling which is connected to an air spring provided on the outer rail side of the front bogie valve has failed, to remain subjected to the air supply operation That abnormal (abbreviated as "front outer軌給gas"),
　(5) breaking the axial spring forward in rail side of the front bogie (abbreviated as "1-shaft軌折loss"),
　(6) the front side of the carriage breakage of the shaft spring outer front rail side of (abbreviated as "1 off-axis軌折loss")
six were assumed as the type of abnormality.
[0037]
　Figure 3 shows the determination results obtained when you enter vehicle information abnormal railcar of the determination target data to the classifier 21.
　As shown in FIG. 3, the determination result abnormality output as the type were fully consistent with the assumed (simulating) abnormal type.
　Thus, according to the monitoring apparatus 100 according to the first embodiment, it can be seen that the abnormality of the type for abnormal railcar is accurately determinable.
[0038]
　Figure 4 shows the determination results obtained when you enter vehicle information for a normal rail vehicle of the determination target data to the classifier 21.
　As shown in FIG. 4, the majority determination result is determined to be a normal.
　Thus, according to the monitoring apparatus 100 according to the first embodiment, it can be seen that even a relatively accurately determinable for a normal rail vehicle.
[0039]
　
　FIG. 5 is a block diagram illustrating a classifier that determination device monitoring apparatus according to the second embodiment of the present invention is provided. 1 (a) is a block diagram illustrating how to generate a classifier using a machine learning is the block diagram showing a state determining vehicle state by referring to FIG. 1 (b) was generated classifier .
　Also monitoring apparatus 100 according to the second embodiment, like the first embodiment, a first step of detecting vehicle information represented by wheels of a wheel weight and the like comprising the railway vehicle traveling on the track detection apparatus 1 are input to the classifier 21 to the determination unit 2 the detected vehicle information includes, railway vehicles from the classifier 21 (vehicle state railway vehicle is unknown) performing a second step of outputting the vehicle state (FIG. 5 (b) reference).
　In the second embodiment, the vehicle state output from the classifier 21 in the second step is not a presence or absence of abnormality of and abnormal type, only a point vehicle organization of the railway vehicle is different from the first embodiment. More specifically, classifier 21, a railway vehicle to be determined is calculated probability of each vehicle formation, judgment apparatus 2 of the probability of each vehicle formation output from the classifier 21, most probable and outputs as a final determination result of high vehicle knitting. It is not necessary to determine the presence or absence of and abnormal types of abnormality, in the second embodiment may be prepared teacher data for normal railway vehicle as teacher data, the teacher data for unusual railcar always necessary Absent.
[0040]
　Hereinafter, an example of a result of determining the vehicle organization of the railway vehicle by monitoring apparatus 100 according to the second embodiment.
　10 kinds of vehicle organization of X series (a knitted, b knitting, c knitting, d knitting, e knitting, f knitting, g knitting, h knitting, i knitting, j knitting) and 11 kinds of vehicle organization of the Y-series (k knitting, l knitting, m knitting, n knitting, o knitting, p knitting, q knitting, r knitting, s knitting, t knitting, the u knitting), under the following conditions, vehicle information when passing through the path curve section (primary spring abnormality index of the front bogie, the primary spring abnormality index of the rear bogie, the secondary spring abnormality reference railcar) detects, a classifier 21 by a machine learning using the detected vehicle information produced was determined vehicle condition by inputting the vehicle information detected in the generated classifier 21.
　(A) Target curved section
　　, the entrance relaxation curve: length 47m
　　-yen Curve: Length 60.1M, radius 251m, Kant 0.065 m, slack 0.009 m
　　-exit transition curve: length 47m
　　-wheel load sensor installation position: the start position of 15m from the circle curve
　(B) using the data
　　, teacher data: vehicle information (10 days) was detected for normal railway vehicle has passed through the curved section
　　, the determination target data: the curved section It passed through was detected for normal railcar vehicle information (20 days after the teacher data acquisition)
[0041]
　Figure 6 shows the determination results obtained when inputting the determination target data to the classifier 21.
　As shown in FIG. 6, the majority of the determination results are determination result is obtained that matches the actual vehicle knitting.
　Thus, according to the monitoring apparatus 100 according to the second embodiment, it can be seen that even a relatively accurately determinable for a vehicle formation.
DESCRIPTION OF SYMBOLS
[0042]
1 ... detector
2 ... determination unit
3 ... railway vehicle
11 ... wheel load sensor
12 ... computing unit
21 ... classifier
31 ... wheel
100 ... railway vehicle status monitoring apparatus

claims
[Claim 1]
　A detection device for detecting vehicle information railway vehicle is represented by the wheel loads, etc. of the wheels comprising running on a track,
　the detected vehicle information, and outputs a vehicle state such as the presence or absence of abnormality of the rail vehicle and a determination device having a classifier,
　the classifier, the when using the vehicle information and the vehicle state of the railway vehicle the vehicle state is known as the teacher data, the vehicle information is input so as to output a known vehicle condition, railway vehicle condition monitoring apparatus characterized by being produced using a machine learning.
[Claim 2]
　The vehicle information input to the classifier is detected by the detection device, the railway vehicle is represented by the wheel loads of the wheels comprising,
　the classifier, as the vehicle state, the abnormal presence or absence of the rail vehicle and rail vehicle state monitoring device according to claim 1, characterized in that for outputting a type of abnormality.
[Claim 3]
　The railway vehicle includes a pair provided before and after the carriage having wheels of the right and left two pairs of front and rear
　vehicle information detected by the detection device, the train represented respectively by the following formulas (1) and (2) a primary spring abnormality index of the front of the primary spring abnormality indicators and the rear of the bogie truck of the vehicle, a secondary spring abnormal indication of the railway vehicle is expressed by the following equation (3),
　the output from the classifier to the vehicle state, the railway vehicle state monitoring device according to claim 1, wherein the abnormal presence or absence of and abnormal types of rail vehicles.
　The primary spring abnormality index of the front bogie = (P1 + P4) - (P2 + P3) · · · (1)
　primary spring abnormality index of the rear bogie = (P5 + P8) - ( P6 + P7) ··· (2)
　of the rail vehicle secondary spring abnormality index = (P1 + P3 + P6 + P8) - (P2 + P4 + P5 + P7) ··· (3)
　However, P1 is a wheel load of the front right wheel of the front bogie, P2 is a wheel load of the front left wheel of the front bogie , P3 is a wheel load of the rear right wheel of the front bogie, P4 is a wheel load of the rear left wheel of the front bogie, P5 is a wheel load of the front right wheel of the rear bogie, P6 after wheel wheel load of the front left side of the bogie, P7 is a wheel load of the rear right wheel of the rear bogie, P8 denotes a wheel load of a wheel in the rear left of the rear bogie.
[Claim 4]
　The railway vehicle includes a pair provided before and after the carriage having wheels of left and right two pairs of front and rear
　vehicle information detected by the detection device, the train represented respectively by the following formulas (1) and (2) a primary spring abnormality index of the front of the primary spring abnormality indicators and the rear of the bogie truck of the vehicle, a secondary spring abnormality index of the rail vehicle represented by the following formula (3),
　the output from the classifier to the vehicle state, the railway vehicle state monitoring device according to claim 1, characterized in that the vehicle organization of the railway vehicle.
　The primary spring abnormality index of the front bogie = (P1 + P4) - (P2 + P3) · · · (1)
　primary spring abnormality index of the rear bogie = (P5 + P8) - ( P6 + P7) ··· (2)
　of the rail vehicle secondary spring abnormality index = (P1 + P3 + P6 + P8) - (P2 + P4 + P5 + P7) ··· (3)
　However, P1 is a wheel load of the front right wheel of the front bogie, P2 is a wheel load of the front left wheel of the front bogie , P3 is a wheel load of the rear right wheel of the front bogie, P4 is a wheel load of the rear left wheel of the front bogie, P5 is a wheel load of the front right wheel of the rear bogie, P6 after wheel wheel load of the front left side of the bogie, P7 is a wheel load of the rear right wheel of the rear bogie, P8 denotes a wheel load of a wheel in the rear left of the rear bogie.
[Claim 5]
　The classifier, as the vehicle information of the training data, the primary spring abnormality index of the front bogie which is actually detected by the detection device for a normal railway vehicle, the primary spring abnormality index of the rear of the carriage and and secondary spring abnormality index of the railway vehicle, the primary spring abnormality index of the front bogie of the normal rail vehicle, from the secondary spring abnormality index of the primary spring abnormality index and the railway vehicle of the rear of the truck the primary spring abnormality index of the front bogie abnormal railcar calculated by numerical simulation, with use of a secondary spring abnormality index of the primary spring abnormality index and the railway vehicle of the rear of the truck, the teacher data Examples vehicle condition, known with the presence or absence of and abnormal types of abnormality, the normal rail vehicle and vehicle of said abnormal railway vehicle for the normal rail vehicle and said abnormal railcar When the information is entered, the railway vehicle according to claim 3 so as to output a vehicle state of the normal rail vehicle and said abnormal railway vehicle, characterized in that it is produced using a machine learning status monitoring device.
[Claim 6]
　The detection device, railway vehicle state monitoring device according to any one of claims 1 to 5, characterized in that it comprises a sensor provided in the track.
[Claim 7]
　The detection device, railway vehicle state monitoring device according to any one of claims 1 to 5, characterized in that it comprises a sensor provided on the railway vehicle.
[8.]
　The classifier railway vehicle state monitoring device according to any one of claims 1 to 7, characterized in that the naive Bayes classifier.