DIAGNOSIS OF TECHNICAL SYSTEMS

Abstract

A method for diagnosing a technical system, an apparatus, and a computer program product are provided for carrying out a main component analysis of predefined values for n variables for describing a normal state of the technical system, where n≥2, and at least one main component of the n variables is ascertained. The at least one main component is predefined for describing the normal state of the technical system. Deviations of values for the n variables for describing a current state of the technical system are ascertained from the at least one predefined main component for describing the normal state, in order to infer a fault. A data carrier is also provided.

Claims

1-13. (canceled)

14. A method for diagnosing a technical system by ascertaining deviations of values relating to at least n, with n≥2, variables for describing a current state of the technical system at a specific time from predefined values for describing a normal state to infer a fault, the method comprising: a. performing a main component analysis of predefined values relating to the at least n, with n≥2, variables for describing the normal state of the technical system and ascertaining at least one main component of the n variables; b. predefining the at least one main component for describing the normal state of the technical system; and c. ascertaining the deviations of the values relating to the n variables for describing the current state of the technical system from the at least one predefined main component for describing the normal state to infer the fault.

15. The method according to claim 14, which further comprises carrying out steps a. and b. during a learning phase and carrying out step c. during an operating phase.

16. The method according to claim 14, which further comprises carrying out step a. for each of m, with m≥2, different operating modes, and carrying out step b. by predefining the at least one main component for describing the normal state based on a current operating mode.

17. The method according to claim 14, which further comprises at least one of using the n variables to describe the same physical quantity or using the values of the n variables to describe the state of the same component of the technical system.

18. The method according to claim 14, which further comprises carrying out step b. by predefining no more than x, with 1≤x<n, main components of the n variables for describing the normal state.

19. The method according to claim 18, which further comprises predefining only the main components together covering at least 70% of a total variance of the values of the n variables for describing the normal state.

20. The method according to claim 14, which further comprises carrying out step b. by predefining only a dominant main component.

21. The method according to claim 14, which further comprises carrying out a step d. following step c. by taking an extent of the deviation as a basis for inferring a predefined fault.

22. The method according to claim 14, which further comprises carrying out step c. by ascertaining the deviation by ascertaining a distance of the values of the n variables for describing the current state from the at least one predefined main component for describing the normal state.

23. A non-transitory computer program product with instructions stored thereon that, when executed by a mobile terminal, cause the mobile terminal to perform the steps of claim 14.

24. A data carrier on which the computer program product according to claim 23 is stored.

25. An apparatus, comprising: a technical system and an evaluation unit; said evaluation unit configured to: perform a main component analysis of predefined values relating to n, with n≥2, variables for describing a normal state of the technical system and ascertaining at least one main component of the n variables; predefine the at least one main component for describing the normal state of the technical system; and ascertain deviations of values relating to the n variables for describing a current state of the technical system from the at least one predefined main component for describing the normal state to infer a fault.

26. The apparatus according to claim 25, wherein the technical system is a rail vehicle.

Description

[0044] The invention permits numerous embodiments. It is explained in more detail on the basis of the figures below, which each depict an exemplary configuration. Identical elements in the figures are provided with identical reference signs.

[0045] FIG. 1 shows a graph for a main component analysis relating to two variables,

[0046] FIG. 2 shows a drive component of a rail vehicle with multiple sensors.

[0047] FIG. 1 graphically depicts a main component analysis in a graph. In this exemplary embodiment, only values relating to two variables of the same physical quantity are taken into consideration, in order to ensure a simple representation and to avoid a multidimensional representation of multiple variables.

[0048] The graph in FIG. 1 has two coordinate axes T.sub.1 and T.sub.2, in order to illustrate the functional relationship between the two variables T.sub.1 and T.sub.2. The data points Y.sub.j, j=1, 2, 3 . . . , with the values (T.sub.1Yj, T.sub.2Yj), which represent the same times t.sub.j, are predefined for a known operating mode of the technical system and characterize the normal state of the technical system in the predefined operating mode. The predefined values may also be measured values relating to the two variables T.sub.1 and T.sub.2—that is to say here temperature measured values of different temperature sensors. The measured values relating to each data point have then been recorded at the same time. They may also have been recorded for the same component of the technical system.

[0049] During a learning phase, an operating-mode-dependent normal curve is now learned by computing for each operating mode, for example “accelerating”, “braking”, “traveling at speed”, “traveling through a tunnel”, “workshop”, etc., independently of one another, the main component of the multivariate distribution (T.sub.1, T.sub.2, of the variables T.sub.1 and T.sub.2, . . . predefined for monitoring and diagnosis. In other words, the dominant eigenvector of the covariance matrix of these sensor signals is computed.

[0050] The curve H is now a one-dimensional representation of the most plausible sensor value combinations for the respective underlying driving mode. The curve H is the main component of the predefined values relating to the variables T.sub.1 and T.sub.2 for describing the normal state of the technical system in the predefined operating mode.

[0051] H can be represented in vector notation as H=(m.sub.1, m.sub.2, m.sub.3, . . . )+a* (v.sub.1, v.sub.2, v.sub.3, . . . ), where m.sub.1 indicates the median of the first sensor signal, m.sub.2 indicates the median of the second sensor signal, m.sub.3 indicates the median of the third sensor signal, etc., and v.sub.1, v.sub.2, v.sub.3, . . . indicates the main component vector.

[0052] Data points from values relating to the two variables for describing a current operating state are now also entered into the defined state space with the two variables as dimensions for describing the normal state. Each data point represents the values recorded for the two variables during operation at a time t.sub.0, t.sub.1, t.sub.2 . . . . The values T.sub.1S0 and T.sub.2S0 of the data point S.sub.0 have been recorded by the temperature sensors at a time t.sub.0 during the operation of the technical system.

[0053] For these data points, the deviation from the learnt normal curve is ascertained. To this end, the vector D=(D.sub.1, D.sub.2, . . . ) with the shortest distance from the applicable operating-mode-dependent normal curve is computed for each data point—illustrated as D.sub.S0=(D.sub.1S0, D.sub.2S0). Each entry D.sub.i therein indicates how far the value of the sensor i deviates from the most plausible value in the overall picture.

[0054] On the basis of the norm of the vector D, it is then first decided whether said vector describes a normal state. If not, a fault picture is attributed in a second step on the basis of the pattern of deviations (of the entries in the distance vector D).

[0055] Specifically, the value PR=(D.sub.1{circumflex over ( )}4+D.sub.2{circumflex over ( )}4+ . . . ) can be computed. If this value is less than a system-dependent limit value E.sub.1, that is to say PR<E.sub.1<2, then a sensor fault in the sensor with the greatest entry in D is involved. If the value is greater than a system-dependent limit value E.sub.2, that is to say N>PR>E.sub.2>N/2, then a deviation produced by an operating mode is involved. If, additionally, K components each having L similar sensors are assessed in the same vehicle, then it holds that 3*L/2>PR>L/2 for a fault in the component to which the sensor with the greatest entry D belongs, while a value of PR close to K can likewise indicate a driving mode that has not yet been recorded.

[0056] A pattern that occurs repeatedly in succession and indicates a component failure can be used to generate maintenance requirements.

[0057] The method is used to produce automated standard curves that detect irregularities in the operation of components and to derive the fault that exists from the pattern of each irregularity.

[0058] FIG. 2 outlines a drive component of a rail vehicle. Said rail vehicle comprises two wheels on an axle, a driving engine, also called a traction engine, and a gear for transmitting force from a drive shaft to the axle.

[0059] To distinguish between sensor, fan and bearing damage in the drive of high-speed trains, besides the exterior temperature and the speed of travel, which are identical for all drive components of a rail vehicle, the time series of the sensors are recorded for each drive component of identical design or of similar design. In the present example, these are:

[0060] L.sub.1 measures the temperature of a first traction engine bearing,

[0061] L.sub.2 measures the temperature of a laminated stator core,

[0062] L.sub.3 measures the temperature of a second traction engine bearing,

[0063] L.sub.4 measures the temperature of a first gear bearing of a small pinion,

[0064] L.sub.5 measures the temperature of a second gear bearing of the small pinion,

[0065] L.sub.6 measures the temperature of a first gear bearing of a large gearwheel,

[0066] L.sub.7 measures the temperature of a second gear bearing of the large gearwheel.

[0067] In addition, wheelset inner bearing temperatures and wheelset outer bearing temperatures can also be recorded.

[0068] From a technical point of view, the coupling of the temperatures T.sub.L1 to T.sub.L7 exists as a result of the ventilation. The cooling air is taken in at L.sub.1, heats up in the traction engine (L.sub.2) and is then taken away by the remaining components L.sub.3 to L.sub.7.

[0069] The temperatures of the drive component, which are recorded at different locations, but at the same time in each case, are intended to be used to monitor the function of the drive component. Since the component temperatures influence one another, but an engineering model is not available, the normal operating characteristic curve of the engine (typical distribution of the sensor values) is first determined statistically. Large deviations of a measurement from this characteristic curve are then used to detect a critical state.

[0070] The input quantities used are the cited temperatures of each of the drive components of the same design in the rail vehicle, and their normal distribution is produced therefrom. If a main component analysis of a test dataset over a lengthy period reveals for example that over 80% of the variation in the dataset is explained by the first component, this dominant main component is first used as the characteristic curve for the normal drive operation.

[0071] The distance from this normal distribution is then computed for new measured values. From the fault patterns, it is recognized whether a sensor fault, problems with a fan, operating-mode-dependent faults or communication problems are involved. Fault reports are produced from the irregularities.