Method for determining a parameter of a wheel of an observed railway vehicle and evaluation unit

Abstract

An evaluation unit is fed with driving-related diagnostic data of a reference railway vehicle. The diagnostic data contain driving-related properties, on which a parameter of the wheel of the reference railway vehicle depends. The evaluation unit is also fed with wheel measurement data with measured values of the parameter of the wheel of the reference railway vehicle. The evaluation unit determines a function between the driving-related diagnostic data and the wheel measurement data via a supervised learning algorithm. The evaluation unit is fed with driving-related diagnostic data containing driving-related properties of an observed railway vehicle, on which a parameter of the wheel of the observed railway vehicle depends. The parameter of the wheel of the observed railway vehicle is determined by the evaluation unit using the determined function and the diagnostic data of the observed railway vehicle.

Claims

1. A method for determining a parameter of a wheel of an observed railway vehicle, the method comprising: supplying an evaluation unit with driving-related diagnostic data of at least one reference railway vehicle, the diagnostic data containing driving-related properties on which the parameter of a wheel of the reference railway vehicle depends; supplying the evaluation unit with wheel measurement data of the at least one reference railway vehicle, the wheel measurement data containing measured values of the parameter of the wheel of the reference railway vehicle; determining with the evaluation unit a function between the driving-related diagnostic data and the wheel measurement data based on a supervised learning algorithm; supplying the evaluation unit with driving-related diagnostic data of the observed railway vehicle, the diagnostic data containing driving-related properties on which the parameter of the wheel of the observed railway vehicle depends; determining with the evaluation unit the parameter of the wheel of the observed railway vehicle using the determined function and the diagnostic data of the observed railway vehicle; and determining a progression of the parameter and determining a point at which the parameter of the wheel of the observed railway vehicle reaches a given critical value.

2. The method according to claim 1, wherein the parameter of the wheel is a parameter selected from the group consisting of a diameter, a wheel flange height, and a wheel flange width.

3. The method according to claim 1, which comprises feeding to the evaluation unit driving-related diagnostic data of a plurality of reference railway vehicles, and sorting the diagnostic data of the plurality of reference railway vehicles into categories selected from the group consisting of: a railway vehicle type; a wagon type of the respective railway vehicle, a material type of the wheel of the respective railway vehicle; and a breaking system type of the respective railway vehicle; and determining the function in dependence on the diagnostic data of those reference railway vehicles which are in the same category as the observed railway vehicle.

4. The method according to claim 1, wherein the driving-related properties are properties selected from the group consisting of mileage signals, speed signals, braking signals, and sliding signals.

5. The method according to claim 1, wherein the driving-related properties comprise location signals.

6. The method according to claim 5, wherein the location signals are GPS signals.

7. The method according to claim 1, wherein: the driving-related properties comprise anti-sliding signals; and/or the driving-related properties comprise torque signals of at least one motor of the respective railway vehicle.

8. The method according to claim 1, wherein the driving-related properties comprise a weight of the train cars of the respective railway vehicle and/or a configuration of the respective railway vehicle.

9. The method according to claim 1, wherein the driving-related properties comprise axle-temperature signals, ambient-temperature signals, and/or ambient-humidity signals extending over a given time range.

10. The method according to claim 1, wherein, if the evaluation unit is fed with driving-related diagnostic data of several reference railway vehicles, checking for completeness and plausibility of the diagnostic data of the several reference railway vehicles and determining the function in dependency of the diagnostic data of those reference railway vehicles, whose diagnostic data are complete and plausible.

11. The method according to claim 1, wherein the supervised learning algorithm is configured to utilize a recurrent neural network.

12. The method according to claim 1, wherein the diagnostic data comprise data points and each of the data points comprises a time stamp.

13. The method according to claim 12, which comprises: grouping the data points from one wheel measurement to a next wheel measurement into one group by way of the time stamps thereof; and using the measured value of the parameter of the wheel, which has been measured in the next wheel measurement, as a target value for the supervised learning algorithm.

14. The method according to claim 13, which comprises using the measured value of the parameter of the wheel, which has been measured in the next wheel measurement, as a target value for the group of data points.

15. The method according to claim 1, which comprises splitting the diagnostic data into segments with a given number of data points and using interpolated wheel measurement data as target values.

16. An evaluation unit for determining a parameter of a wheel of an observed railway vehicle, the evaluation unit being configured to: receive driving-related diagnostic data of at least one reference railway vehicle, the diagnostic data comprising driving-related properties, on which the parameter of a wheel of the reference railway vehicle depends; receive wheel measurement data of the at least one reference railway vehicle, the wheel measurement data comprising measured values of the parameter of the wheel of the reference railway vehicle; determine a function between the driving-related diagnostic data and the wheel measurement data on a basis of a supervised learning algorithm; receive driving-related diagnostic data of the observed railway vehicle, the diagnostic data comprising driving-related properties, on which the parameter of the wheel of the observed railway vehicle depends; determine the parameter of the wheel of the observed railway vehicle by using the determined function and the diagnostic data of the observed railway vehicle; and determine a progression of the parameter and determine a point at which the parameter of the wheel of the observed railway vehicle reaches a given critical value.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING

(1) FIG. 1 a schematic overview of a supervised learning algorithm to determine a function, which is used to determine a parameter of a wheel of an observed railway vehicle; and

(2) FIG. 2 a schematic overview of a further supervised learning algorithm to determine the function.

DESCRIPTION OF THE INVENTION

(3) FIG. 1 schematically shows a supervised learning algorithm 2. The supervised learning algorithm utilises a recurrent neural network (RNN), here a long short-term memory (LSTM) network. An evaluation unit (not shown) determines a function by means of the supervised learning algorithm 2, which function is used to determine a parameter of a wheel of an observed railway vehicle.

(4) The evaluation unit is fed with driving-related diagnostic data 4 of at least one reference railway vehicle as input data. The diagnostic data 4 comprises driving-related properties, on which a parameter of the wheel of the reference railway vehicle depends. The diagnostic data 4 comprises several data points 6, which are fed to the evaluation unit.

(5) Further, the evaluation unit is fed with wheel measurement data 8 of the at least one reference railway vehicle as target data. The wheel measurement data 8 comprise measured values 9 of the parameter of the wheel of the at least one reference railway vehicle.

(6) The data points 6 of the diagnostic data 4 are grouped into groups and/or into time intervals by means of time stamps of the data points. An interval is the time from one wheel measurement to the next wheel measurement. Further, a group comprises the diagnostic data 4 with a time stamp from one wheel measurement to the next wheel measurement. In FIG. 1, a dashed line 10 shows the border of two intervals/groups schematically.

(7) The architecture of the RNN is given. Moreover, the architecture of the RNN can be adjusted manually before and/or after each cycle of approximation done by the supervised learning algorithm 2.

(8) The first data point 6 is a data point 6 with the earliest time stamp. The first data point 6 is fed to the RNN in its first state 12 (illustrated as first/top shaded box). The result fed together with the next data point 6 to the RNN in its state 12 (at that time), and so on. The wheel measurement data 8 is used as target values at the end of each interval. Hence, the wheel measurement data 8 is used as target values for the groups. The procedure using the RNN is known from elsewhere.

(9) With this procedure, the RNN learns the function between the driving-related diagnostic data 4 and the wheel measurement data 8.

(10) The function between the driving-related diagnostic data 4 and the wheel measurement data 8 is determined on the basis of the supervised learning algorithm 2 by means of the evaluation unit.

(11) The first gained function may be a first approximation. However, executing the supervised learning algorithm 2 for several cycles will lead to a better approximation, and—hence—to a better function.

(12) The parameter of the wheel of the observed railway vehicle is determined by means of the determined function on the basis of the diagnostic data of the observed railway vehicle. Therefore, the RNN (in its resulted state) is fed with the diagnostic data of the observed railway vehicle. Moreover, wheel measurement data of the observed railway vehicle may be used as target values.

(13) The determined parameter is a parameter at a time and/or mileage after the latest wheel measurement of the observed railway vehicle.

(14) The determined parameter may be a present parameter and/or a future parameter.

(15) FIG. 2 shows schematically an analogous supervised learning algorithm 13.

(16) The following description is restricted essentially to the differences from the embodiment of FIG. 1, to which is referred regarding unchanged features and functions. Essentially identical elements are generally denoted by the same reference numbers, and not mentioned features are included in the following embodiment without being described again.

(17) In this case, the diagnostic data 4 are split into segments with a given number of data points 6. In FIG. 2, a dashed line 14 schematically shows the border of two segments.

(18) Interpolated wheel measurement data 16 are used as target values. The interpolated wheel measurement data 16 comprise interpolated values 18 of the measured parameter of the wheels.

(19) Any function may be used for the interpolation in the first cycle of the supervised learning algorithm 13. However, a function gained by the method described in FIG. 1 may be used for the interpolation in the first cycle of the supervised learning algorithm 13.

(20) The resulted function of the supervised learning algorithm 13 may be used for the interpolation within the next cycle of approximation/within the next cycle of the supervised learning algorithm.

(21) While specific embodiments have been described in detail, those with ordinary skill in the art will appreciate that various modifications and alternative to those details could be developed in light of the overall teachings of the disclosure. For example, elements described in association with different embodiments may be combined. Accordingly, the particular arrangements disclosed are meant to be illustrative only and should not be construed as limiting the scope of the claims or disclosure, which are to be given the full breadth of the appended claims, and any equivalents thereof.