DE102019200031A1 - Method and device for determining the cause of a fault in switch failures in rail-bound traffic - Google Patents

Abstract

The invention relates to a method for determining the cause of a fault (11) in switches (50) in rail-bound traffic, comprising the following steps: detection of a switch state (10) of the switch (50) by means of a sensor system (3-x), detection of a switch fault an evaluation device (4) on the basis of the detected switch state (10), and if a switch fault is determined: estimating a cause of the fault (11) on the basis of a Bayesian network (5) modeling the switch (50) by means of the evaluation device (4), the Bayesian network (5) is calibrated on the basis of at least collected data of at least one existing switch monitoring system and / or on the basis of structured expert knowledge, output of the estimated cause of the error (11) as an error cause signal (12). The invention further relates to an associated device (1).

Description

The invention relates to a method and a device for determining the cause of a fault in switch failures in rail-bound traffic.

Switches are one of the central elements of the rail infrastructure because they allow trains to change tracks without stopping. In the course of a turnout, movable tongues of the turnout are moved from one locking position to another locking position, whereby in particular a target track is set.

Turnouts, including a turnout drive, a turnout lock, actuating linkage etc., are continuously exposed to the weather and, due to their complexity and high mechanical stress when crossing trains, are one of the most fault-prone and therefore maintenance-intensive components in the track network. In addition to regular on-site inspections, with the aim of optimizing the maintenance processes, special switch monitoring systems have been used for some time, which automatically switch based on the measurement of so-called control current curves of electric motors of switch drives, i.e. in particular those that affect the moving parts of the switch and the switch drive.

Such a point monitoring system is for example from the DE 10 2016 221 479 A1 known. Versions of such systems are known, for example, from Siemens (SIDIS W compact), Strukton Rail (POSS) or voestalpin (VAE Roadmaster). The WO 2016/074971 A1 describes in this context a specific sensor with which actuating current curves of electrical switches can be measured without feedback.

In the associated evaluation methods, certain features of the measured actuating current curves are described using mathematical parameters and then compared with predetermined threshold or reference values. In the event of a discrepancy, i.e. exceeding or falling below the threshold or reference values, an alarm is generated, whereupon a more detailed analysis of the measurement data is carried out by a trained switch engineer or an on-site inspection by a maintenance team.

For individual non-movable switch parts in the overflow area of the switch (eg the heart) is in the DE 10 2004 014 282 C5 describes a method for state detection based on acceleration sensors attached to the switch. A similar method for monitoring switch conditions, but with vehicle-side (instead of infrastructure-side) acceleration sensors is in the DE 10 2004 045 457 B4 described.

It is known that the above Alternating current-based switch monitoring systems often generate false alarms. This can be partially prevented by improved evaluation methods, which take into account in particular the temperature dependence of the control current curves. However, the known switch monitoring systems can ultimately only detect the presence of a switch fault. In the best case, at least a determination of the phase and thus the point in time within the turnout at which the disturbance (usually mechanical resistance in the turnout system) occurs. An experienced turnout engineer can then at least make an approximate estimate of which turnout parts could be affected, even without on-site inspection. However, a detailed and systematic diagnosis of the cause is not possible with the available means.

In addition, existing turnout monitoring systems can only ever take into account certain parts or sub-functions of the turnout (i.e. selected components).

As a result, this means that, in practice, many causes of errors for switch failures can only be identified by specialist staff on site.

The invention is based on the object of providing a method and a device for ascertaining the cause of an error in the case of switch failures in rail-bound traffic, in which the ascertainment of the cause of the error can be improved, in particular in a less complex manner.

The object is achieved according to the invention by a method with the features of claim 1 and a device with the features of claim 10. Advantageous embodiments of the invention result from the subclaims.

In particular, a method is provided for determining a cause of error in turnout disturbances in rail-bound traffic, comprising the following steps: detection of a turnout condition of a turnout using sensors, detection of a turnout failure using an evaluation device based on the detected turnout condition, and if a turnout failure is determined: Estimating an error cause on the basis of a Bayesian network modeling the turnout by means of the evaluation device, the Bayesian network being calibrated on the basis of at least collected data from at least one existing turnout monitoring system and / or on the basis of structured expert knowledge, outputting the estimated error cause as an error cause signal.

Furthermore, a device for determining a cause of error in switch malfunctions in rail-bound traffic is created, comprising an input device for receiving a switch state of a switch detected by means of a sensor system, and an evaluation device for determining a switch fault on the switch based on the detected switch state, the evaluation device being designed in this way is to estimate a cause of error on the basis of a Bayesian network modeling the turnout after detection of a turnout fault, the Bayesian network being calibrated on the basis of at least collected data from at least one existing turnout monitoring system and / or on the basis of structured expert knowledge, and an output device, wherein the output device is designed to output the estimated cause of the error as an error cause signal.

A Bayesian network is a directed acyclic graph, in which the nodes describe random variables and the edge-dependent dependencies between the random variables. Here, nodes can be connected to higher-level parent nodes. Each node of the network is given a conditional probability distribution for the represented random variable depending on the state of the parent node.

The conditional probability distributions are described, for example, by probability tables. For calibration, the parameters of the Bayesian network are learned, for example, using adaptive heuristic learning methods. The Bayesian network is in particular formed in a memory device of the evaluation device.

According to the invention, data collected from at least one existing switch monitoring system and / or structured expert knowledge are used for teaching. The data is obtained, for example, by compiling a history of switch faults and the associated causes of the fault, as well as subsequent statistical evaluation. The structured expert knowledge is determined, for example, by structured questioning of switch engineers. In this case, the cause of the error is queried, in particular for certain turnout failures, taking into account other circumstances, such as weather, a turnout age or a turnout type, and the results are then statistically evaluated.

If one speaks in this context of estimating a cause of error or an estimated cause of error, this means in particular a set of causes of error with respectively estimated probabilities. As a result, the Bayesian network provides estimates for the probabilities of the possible causes of the error.

The advantage of the invention is that both collected data from existing switch monitoring systems and structured expert knowledge can be used. This means that large training data sets, such as those required for training artificial neural networks, can be dispensed with and the causes of errors can be reliably determined even with significantly less data.

The estimation on the basis of the Bayesian network takes place in particular taking into account the current state of the points detected by the sensors. The condition of the turnout includes, in particular, information about whether a turnout can be circulated, ie whether the turnout can be set or not. In addition, the current turnout status can also include information as to whether turnout circulation is delayed, that is, whether turnout circulation takes longer than in the normal case. Furthermore, in particular a locked state of a movable tongue of the switch is also part of the switch state.

In one embodiment, it is provided that at least one operating characteristic of the switch derived from the condition of the switch is used as additional evidence when estimating. An operating feature can be generated, for example, by summarizing state variables of the switch state into characteristic quantities. For example, it can be provided that state variables with a threshold value be compared. Depending on whether the threshold value is fallen short of or exceeded, the operating condition is assigned the status “normal” or “abnormal”.

In a further embodiment, it is provided that at least one parameter of a control current curve of an electric motor of the switch is used as an operating feature. The actuating current is recorded in particular by means of a sensor designed for this purpose on the electric motor and recorded in high-resolution with a frequency of 50 Hz, for example. Typically, a control current curve generated over time from the control current has a steep increase in the control current at the beginning, since the electric motor has to release a switch tongue from a lock. After removal, the tongue moves and the actuating current drops again and remains almost constant while the tongue is moving into the other locking position. After locking the tongue in the other locking position, the actuating current drops again. The control current curve typically extends over 1 to 2 seconds. Characteristic values can be derived from such a control current curve. For example, the gradients of the actuating current curve and / or an area under the actuating current curve or parts thereof can be evaluated. Furthermore, parameters can also be derived on the basis of a statistical evaluation of the actuating current curve. The links between the parameters as operational characteristics and the nodes of the Bayesian network modeling the turnout state are derived from existing turnout monitoring systems and / or from the structured expert knowledge when calibrating the Bayesian network.

In a further embodiment it is provided that at least one external influencing variable is used as additional evidence when estimating. Such an external influencing variable can be, for example, a temperature of an environment of the switch and / or a temperature of the switch, which can be detected, for example, by means of a temperature sensor. External influencing variables can also be a date of a last maintenance of the turnout and / or an age of the turnout etc. Other weather properties, such as moisture, precipitation, etc., can also be taken into account as external factors. Furthermore, bedding of the switch can also be taken into account as an external influencing variable, it being possible to differentiate here, for example, between a ballast bed or a solid carriageway.

In one embodiment it is provided that the Bayesian network has a modular structure, with individual modules being selected depending on the switch being examined and being combined to form the Bayesian network. The switch model is broken down and encapsulated in particular into individual sub-models, the sub-models being connected to the other sub-models via defined interfaces or interface nodes. As a result, individual modules of the switch model or the Bayesian network can be exchanged more easily if the method is to be used for a different switch type. The selection and compilation can be done both manually and automatically, for example on the basis of tables in which the respective sub-models are stored for each turnout type. Corresponding sub-models relate, for example, to switch tongues, a switch drive, switch locks or switch heating.

In one embodiment it is provided that at least one recommended action for troubleshooting is derived on the basis of the estimated cause of the error and is output as a recommended action signal. This enables a search for a cause of the error to be carried out in a targeted manner. The recommended action is derived on the basis of the results provided by the Bayesian network, i.e. the estimated probabilities for individual causes of errors. For example, the cause of the error that was assigned the greatest probability by the Bayesian network after performing the method can first be checked.

It can be provided here that corresponding recommendations for action are transmitted to a smartphone or tablet computer and displayed there. Furthermore, it can be provided that a result of an action, for example a test or a check on site by a switch engineer, can then be entered into the switch model or the Bayesian network immediately via smartphone or tablet computer, so that an improved estimate of the cause of the error can be made can be provided.

In a further embodiment, it is provided that a sensitivity analysis is carried out to derive the at least one recommended action, the sensitivity analysis being used to determine which informative added value a certain action delivers after an implementation, and an action based on the respective informative added value is recommended . Here, the strengths of influence, that is the statistical influences, between the individual nodes of the calibrated Bayesian network. Nodes that influence each other particularly strongly provide less information than nodes that influence each other less strongly. If there is evidence for one of these nodes, the evidence obtained for the other node provides only a small informative added value, since the influence is great. In the opposite case, i.e. with little influence, the informative added value of a determined evidence is greater for the other node. Based on this dependency, recommendations for action are selected that provide great informative added value. Standard methods can be used for sensitivity analysis. For example, test evidence can be used in the individual nodes of the Bayesian network to determine the degree of influence. Carrying out a sensitivity analysis is particularly advantageous if several recommendations for action can be considered. A turnout engineer then only needs to carry out such actions or tests or checks that quickly lead to a result when searching for the cause of the error, that is, they provide great informative added value.

In a further embodiment it is provided that the Bayesian network is at least partially updated or recalibrated, taking into account data that was collected after a previous calibration. In particular, it is provided that the Bayesian network learns continuously, that is to say that continuously newly acquired information on switch faults and the causes of errors are taken into account in the conditional probabilities of the individual nodes of the Bayesian network. In this way, a knowledge or experience base of the Bayesian network can be continuously expanded and updated.

In one embodiment it is provided that the estimation is carried out independently of the detection of a turnout fault, wherein a turnout condition is estimated for the estimation and / or an estimated operating characteristic of the turnout and / or an estimated external influencing variable is used in the estimation as at least one additional evidence. In contrast to the procedure described so far, in which a cause of an error is only determined after a turnout fault has been determined, future turnout failures together with an associated error cause can be predicted even before the turnout failure occurs. Since switch failures can be avoided, this enables predictive or predictive switch maintenance and maintenance. For example, it can be provided that operating characteristics of the switch, for example parameters of the actuating current curve, are tracked over time and a trend is determined from this. A future value of an operating characteristic can then be extrapolated on the basis of the determined trend. Using the estimated future value, a future point in time for falling below or exceeding threshold values can be estimated. An operational feature estimated for this future point in time will then be used as evidence in the Bayesian network. For the future, it is therefore possible to estimate both the occurrence of a turnout fault and an associated cause of the fault. If, for example, a housing of a gear unit arranged between an electric motor and a tongue of the switch leaks, as a result of which moisture penetrates into the housing and the gear, the gear will corrode as time progresses. As a result, turnout circulation from one locking position to another locking position is becoming increasingly difficult. This manifests itself in the parameters of the actuating current curve, in particular through an overall increasing actuating current. A trend can be derived from the increase, which can be extrapolated into the future. If the values for the actuating current extrapolated from the trend exceed a threshold value at a future point in time, the corresponding operating characteristics, in particular the extrapolated actuating current and possibly further parameters of the actuating current curve, are used as evidence in the calibrated Bayesian network and an error cause is estimated. In this way, the occurrence of a switch failure can be estimated for a future point in time by specifying the cause of the fault (e.g. a stiff gear unit).

The advantages of the individual embodiments of the device are the same as those of the embodiments of the method. In particular, it is provided that the evaluation device is designed to carry out the individual embodiments of the method.

Parts of the device, in particular the evaluation device, can be designed individually or in combination as a combination of hardware and software, for example as program code, which is executed on a microcontroller or microprocessor.

• 1 eine schematische Darstellung einer Ausführungsform der Vorrichtung zum Ermitteln einer Fehlerursache bei Weichenstörungen im schienengebundenen Verkehr;
• 2 eine schematische Darstellung eines mittels eines Bayesschen Netzes erstellten Weichenmodells zur Verdeutlichung der Erfindung;
• 3 ein schematisches Ablaufdiagramm einer Ausführungsform des Verfahrens zum Ermitteln einer Fehlerursache bei Weichenstörungen im schienengebundenen Verkehr.

The invention is explained in more detail below on the basis of preferred exemplary embodiments with reference to the figures. Here show:

• 1 a schematic representation of an embodiment of the device for determining a cause of error in switch failures in rail traffic;
• 2nd a schematic representation of a switch model created by means of a Bayesian network to illustrate the invention;
• 3rd a schematic flow diagram of an embodiment of the method for determining a cause of error in switch failures in rail-bound traffic.

In 1 is a schematic representation of an embodiment of the device 1 to determine the cause of a fault in switch failures of switches 50 shown in rail traffic. The device 1 comprises an input device 2nd for receiving a turnout condition detected by means of a sensor system 3-x 10th a switch 50 . The sensor system 3-x comprises, for example, an actuating current sensor 3-1 on an electric motor of a point machine 51 , an infrastructure-side acceleration sensor 3-2 on the switch 50 and an on-vehicle acceleration sensor 3-3 on the vehicle 52 . The device further comprises 1 an evaluation device 4th to detect a turnout fault on the turnout 50 based on the recorded state of the turnout 10th . For this, the evaluation device evaluates 4th for example one using the control current sensor 3-1 detected control current curve and detects a switch failure when a threshold value specified for the control current is exceeded.

After determining a turnout fault, the evaluation device estimates 4th a cause of error 11 based on a the switch 50 modeling Bayesian network 5 . The Bayesian network 5 is, for example, in a memory device (not shown) of the evaluation device 4th educated. It is envisaged that the Bayesian network 5 is calibrated on the basis of at least collected data of at least one existing switch monitoring system and / or on the basis of structured expert knowledge.

The device 1 further comprises an output device 6 which is the estimated cause of the error 11 as an error cause signal 12th issues. The error cause signal 12th can be designed, for example, as a digital data packet.

The error cause signal 12th can, for example, be transmitted to a mobile tablet computer or a smartphone (both not shown) to assist a turnout engineer in troubleshooting.

It can be provided that at least one from the soft state during the estimation 10th derived operating characteristic 13 the soft 50 as additional evidence in the Bayesian network 5 is used. For this purpose, the evaluation device guides 4th such an operational characteristic 13 from the turnout state 10th ab, in particular this can be a parameter derived from the actuating current curve.

It can further be provided that when estimating at least one external influencing variable 7-1 as additional evidence in the Bayesian network 5 is used. The at least one external influencing variable 7-1 can, for example, by means of a further sensor system 14 (eg a temperature sensor) can be detected and also via the input device 2nd be received.

It can also be provided that the Bayesian network 5 has a modular structure, with individual modules depending on the switch being examined 50 selected and to the Bayesian network 5 be put together.

Furthermore, it can be provided that based on the estimated cause of the error 11 at least one recommended action 15 derived for troubleshooting and as an action recommendation signal 16 is issued.

Provision can be made for deriving the at least one recommended action 15 a sensitivity analysis is carried out, the sensitivity analysis being used to determine the informative added value that a specific action delivers after an implementation, and an action based on the respective informative added value being recommended.

It can also be provided that the Bayesian network 5 is at least partially updated or recalibrated taking into account data collected after a previous calibration. This allows the statistical dependencies in the Bayesian network 5 be updated based on new knowledge.

Provision can be made for the estimation to be carried out independently of the detection of a switch fault, with a switch condition for the estimation 10th is estimated and / or when estimating as at least one additional piece of evidence an estimated operating characteristic of the switch 50 and / or an estimated external influencing variable 7-1 is used. This enables predictive turnout maintenance and turnout maintenance.

In 2nd is a schematic representation of a Bayesian network 5 created switch model shown to illustrate the invention.

• - „Externe Einflussgrößen“ 7-1: Dies sind Umfeldbedingungen, die ggf. das Weichenverhalten und hierdurch Fehlerwahrscheinlichkeiten von außen beeinträchtigen, beispielsweise eine Temperatur oder eine Witterung. Im gezeigten Ausführungsbeispiel umfassen die externen Einflussgrößen 7-1 die folgenden Netzknoten 5-1, 5-2:
  • - Netzknoten 5-1 für eine „Bettung“ der Weiche mit den Zuständen „Schotterbett“ und „Feste Fahrbahn“,
  • - Netzknoten 5-2 für eine Temperatur des Umfeldes der Weiche mit den Zuständen „niedrig“, „mittel“ und „hoch“, wobei die Zustände jeweils vorgegebenen Temperaturbereichen zugeordnet sind;
• - „Potentielle Fehlerursachen“ 7-2: hiervon sind die folgenden Netzknoten 5-3, 5-4, 5-5 umfasst:
  • - Netzknoten 5-3 für „Fremdkörper im Bereich der Weichenzungen“ mit den Zuständen „ja“ und „nein“,
  • - Netzknoten 5-4 für einen „Klemmenden Weichenverschluss“ (= klemmende Verriegelung in mindestens einer der beiden Verriegelungspositionen) mit den Zuständen „ja“ und „nein“,
  • - Netzknoten 5-5 für eine „Unzureichende Schmierung der Gleitstühle“ mit den Zuständen „ja“ und „nein“;
• - „Aggregierte Fehlerzustände“ 7-3: hiervon ist der folgende Netzknoten 5-6 umfasst:
  • - Netzknoten 5-6 allgemein für eine „Mechanische Blockade der Weichenzungen“ mit den Zuständen „ja“ und „nein“,
• - „Beobachtete Wirkung“ 7-4: hiervon sind die Netzknoten 5-8, 5-9 umfasst:
  • - Netzknoten 5-8 für „Kein Weichenumlauf möglich“ mit den Zuständen „ja“ und „nein“,
  • - Netzknoten 5-9 für „Weichenumlauf verzögert“ mit den Zuständen „ja“ und „nein“.

The Bayesian network shown 5 describes the stochastic dependencies (symbolized in the 2nd by the arrows) between the individual network nodes 5-x with their respective possible states, which may have different probabilities depending on the evidence constellation. The quantitative description of the stochastic dependencies between neighboring network nodes 5-x is based on the Bayesian network concept 5 following on conditional probability distributions (or in the discrete case on corresponding probability tables, see the following tables), which in the course of the calibration of the Bayesian network 5 be determined. The network nodes occurring in the switch model 5-x can generally (ie not only in the exemplary embodiment outlined here) roughly be classified into the following classes:

• - "External influencing factors" 7-1 : These are environmental conditions that may affect the turnout behavior and thereby the likelihood of errors from the outside, for example a temperature or weather. In the exemplary embodiment shown, the external influencing variables include 7-1 the following network nodes 5-1 , 5-2 :
  • - network node 5-1 for "bedding" the switch with the states "ballast bed" and "slab track",
  • - network node 5-2 for a temperature of the area around the switch with the states "low", "medium" and "high", the states being assigned to predetermined temperature ranges;
• - "Potential causes of errors" 7-2 : of which are the following network nodes 5-3 , 5-4 , 5-5 includes:
  • - network node 5-3 for "foreign bodies in the area of the switch tongues" with the states "yes" and "no",
  • - network node 5-4 for a "clamping turnout lock" (= locking in at least one of the two locking positions) with the states "yes" and "no",
  • - network node 5-5 for "insufficient lubrication of the sliding chairs" with the states "yes" and "no";
• - "Aggregated error states" 7-3 : of which is the following network node 5-6 includes:
  • - network node 5-6 generally for a "mechanical blockade of the switch tongues" with the states "yes" and "no",
• - "observed effect" 7-4 : of which are the network nodes 5-8 , 5-9 includes:
  • - network node 5-8 for "No turnout possible" with the states "yes" and "no",
  • - network node 5-9 for "turnout delayed" with the states "yes" and "no".

Additional network nodes 5-x can optionally be provided for derived operating features. In the example shown there is a network node 5-7 for a derived operating characteristic that classifies the turnout condition as "normal" or "abnormal", in the "aggregated fault conditions" class 7-3 intended. In principle, network nodes 5-x can also be provided for derived operating characteristics in other classes.

The individual network nodes 5-x (without the network node 5-7 ) are conditionally linked to one another using the following probability tables used as model parameters. Bedding Ballast bed 96% Fixed carriageway 4% temperature Low 12% medium 71% High 17% Foreign body in the area of the switch tongues Bedding Ballast bed Slab track Yes 5% 1 % No 95% 99% Clamping switch lock temperature Low medium High Yes 20% 8th % 1 % No 80% 92% 99% Inadequate lubrication of the sliding chairs temperature Low medium High Yes 30% 10% 5% No 70% 90% 95% Mechanical blockage of the switch tongues Foreign body in the area of the switch tongues Yes No Clamping switch lock Clamping switch lock Yes No Yes No Yes 100% 100% 100% 0% No 0% 0% 0% 100% No turnout possible Mechanical blockage of the switch tongues Yes No Yes 80% 2% No 20% 98% Turnout delayed Inadequate lubrication of the sliding chairs Yes No Mechanical blockage of the switch tongues Mechanical blockage of the switch tongues Yes No Yes No No turnout possible No turnout possible No turnout possible No turnout possible Yes No Yes No Yes No Yes No Yes 0% 90% 0% 5% 0% 85% 0% 2% No 100% 10% 100% 95% 100% 15% 100% 98%

• - Netzknoten 5-3: Fremdkörper im Bereich der Weichenzungen (Ja): 19,8 % (a-priori: 4,8 %)
• - Netzknoten 5-4: Klemmender Weichenverschluss (Ja): 33,9 % (a-priori: 8,3 %)
• - Netzknoten 5-5: Unzureichende Schmierung der Gleitstühle (Ja): 19,2 % (a-priori 11,6 %)

Now lies, for example, in the network node 5-9 a delayed turnout before ("turnout delayed""Yes": 100%), this results in addition to the exclusion of other malfunctions (eg network nodes 5-8 "No turnout possible""Yes": 0%) the following probabilities for the potential causes of errors (a priori probabilities in the event of ignorance of the existence or non-existence of a fault are shown in brackets for comparison):

• - network node 5-3 : Foreign body in the area of the switch tongues (yes): 19.8% (a-priori: 4.8%)
• - network node 5-4 : Clamping turnout lock (yes): 33.9% (a priori: 8.3%)
• - network node 5-5 : Inadequate lubrication of the sliding chairs (yes): 19.2% (a-priori 11.6%)

• - Netzknoten 5-3: Fremdkörper im Bereich der Weichenzungen (Ja): 27,8 %
• - Netzknoten 5-4: Klemmender Weichenverschluss (Ja): 5,7 %
• - Netzknoten 5-5: Unzureichende Schmierung der Gleitstühle (Ja): 9,5 %

In addition, the likelihood of a “low” temperature increases from 12% (a priori) to 18.1%, because especially low temperatures reduce the lubricity of the lubricants used for the sliding chairs or icing in winter can lead to jamming switch closures. However, it is known, for example, that there are actually high temperatures at the time of the fault, which is an additional evidence in the network node 5-2 corresponds, the probabilities for the temperature-dependent causes of error are reduced again and instead a foreign body in the area of the switch tongues becomes more likely as the cause of the error for the observed, delayed turnout circulation:

• - network node 5-3 : Foreign body in the area of the switch tongues (yes): 27.8%
• - network node 5-4 : Clamping turnout lock (yes): 5.7%
• - network node 5-5 : Inadequate lubrication of the sliding chairs (yes): 9.5%

Even if only a small part of the error causes of a turnout malfunction of a turnout is taken into account in the example described, the example nevertheless clearly shows how, with the method according to the invention or the turnout model used therein, appropriate conclusions about error causes and their effects on the functionality of turnouts are made and plausible and traceable diagnoses of the causes of errors can be provided. In addition to suitability from a functional point of view, the method according to the invention thus also fulfills requirements with regard to practical suitability, i.e. i.a. a use of procedures that are intuitively understandable for humans.

In 3rd is a schematic flow diagram of an embodiment of the method for determining a cause of error in switch malfunctions in rail traffic is shown. The starting point is at the beginning 100 a calibrated Bayesian network, which maps a turnout as a turnout model (cf. e.g. 2nd ). The Bayesian network is calibrated on the basis of at least collected data from at least one existing switch monitoring system and / or on the basis of structured expert knowledge.

In one step 101 the a priori probabilities are determined for all network nodes. In particular, the a priori probabilities for the network nodes representing the individual causes of the failure are determined. For this purpose, the local probabilities are propagated through the Bayesian network in a manner known per se.

In the procedural step 102 a state of the turnout is detected by means of sensors on the turnout. For example, this can be a detected control current or a control current curve.

In the procedural step 103 the state probabilities of the individual network nodes of the Bayesian network are updated on the basis of the acquired sensor data, the acquired sensor data being used as evidence.

In the procedural step 104 After updating the state probabilities, the cause of the error is estimated and output as an error cause signal. The cause of the fault or the cause of the fault signal comprises in particular the respective probabilities for individual causes of the points fault, which can be read from the probabilities of the individual network nodes associated with the causes of the fault.

It can be done in one step 105 it should be provided that operational characteristics are derived from the detected switch state, for example in the form of parameters of a recorded actuating current curve. The derived operating characteristics are in process step 103 then also used to update states in associated network nodes of the Bayesian network.

It can be in process step 104 it should be provided that, based on the estimated cause of the error, at least one recommended action for troubleshooting is derived and output as a recommended action signal. For example, on the basis of the probabilities for the individual causes of the error, a recommendation for action can be given with regard to a part of the switch which has to be repaired.

Accordingly, a recommendation can be given as to which tool a turnout engineer should have available and carry with him during an on-site inspection.

Furthermore, in one process step 106 it is provided that a sensitivity analysis is carried out in order to derive the at least one recommended action, the sensitivity analysis being used to determine which informative added value a specific action delivers after an implementation, and an action based on the respective informative added value is recommended. In particular, an action strategy can be generated and issued with several action recommendations to be carried out one after the other.

In a subsequent process step 107 the at least one recommended action or strategy can be implemented and tests or checks can be carried out, in particular by a switch engineer on site at the switch. The results determined for the condition of the points are again, together with possibly changed results of the sensors, in the process step 103 used to update the states of the network nodes of the Bayesian network and / or to derive the operating features.

In one step 108 It can further be provided that the turnout state determined by means of the sensor system and / or the results of the additional tests and checks are used to optimize the Bayesian network. In particular, the Bayesian network is recalibrated, taking the current turnout condition and the associated estimated cause of error into account.

The procedural steps 103 , 106 , 107 such as 102 and 105 can also be repeated cyclically or iteratively in order to gradually improve the estimation of the cause of the error and to narrow down the troubleshooting using the recommended actions.

Reference list

1

contraption

2nd

Entrance facility

3-x

Sensors

3-1

Control current sensor

3-2

Accelerometer (infrastructure side)

3-3

Accelerometer (vehicle side)

4th

Evaluation device

5

Bayesian network

5-x

Network node

6

Output device

7-1

External influencing variables (class)

7-2

Potential causes of errors (class)

7-3

Aggregated error states (class)

7-4

Observed effect (class)

10th

Turnout condition

11

Cause of error

12th

Error cause signal

13

Operating characteristic

14

further sensors

15

Recommended action

16

Action recommendation signal

50

Switch

51

Point machine

52

vehicle

100-108

Procedural steps

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102016221479 A1 [0004]
• WO 2016/074971 A1 [0004]
• DE 102004014282 C5 [0006]
• DE 102004045457 B4 [0006]

Claims (10)

Method for determining a cause of error (11) in the case of switch failures in rail-bound traffic, comprising the following steps: Detection of a switch state (10) of a switch (50) by means of a sensor system (3-x), Detection of a switch failure using an evaluation device (4) on the basis of the detected switch condition (10), and if a switch failure is detected: Estimating an error cause (11) on the basis of a Bayesian network (5) modeling the turnout (50) by means of the evaluation device (4), the Bayesian network (5) on the basis of at least collected data of at least one existing turnout monitoring system and / or on the basis is calibrated by structured expert knowledge, Output of the estimated cause of error (11) as an error cause signal (12). Procedure according to Claim 1 , characterized in that when estimating at least one operating characteristic (13) of the switch derived from the switch condition (10) is used as additional evidence. Procedure according to Claim 2 , characterized in that at least one parameter of a control current curve of an electric motor of the switch (50) is used as the operating feature (13). Method according to one of the preceding claims, characterized in that at least one external influencing variable (7-1) is used as additional evidence when estimating. Method according to one of the preceding claims, characterized in that the Bayesian network (5) has a modular structure, individual modules being selected as a function of the switch (50) examined and being combined to form the Bayesian network (5). Method according to one of the preceding claims, characterized in that at least one recommended action (15) for troubleshooting is derived on the basis of the estimated cause of the error (11) and is output as a recommended action signal (16). Procedure according to Claim 6 , characterized in that a sensitivity analysis is carried out in order to derive the at least one recommended action (15), the sensitivity analysis determining what informative added value a specific action delivers after an implementation, and an action based on the respective informative added value being recommended . Method according to one of the preceding claims, characterized in that the Bayesian network (5) is at least partially updated or recalibrated taking into account data which have been collected after a previous calibration. Method according to one of the preceding claims, characterized in that the estimation is carried out independently of the detection of a turnout fault, wherein a turnout condition (10) is estimated for the estimation and / or an estimated operating characteristic (13) of the turnout (13) is estimated as at least one additional evidence. 50) and / or an estimated external influencing variable (7-1) is used. Device (1) for determining the cause of a fault (11) in the event of switch failures in rail-bound traffic, comprising: an input device (2) for receiving a turnout state (10) of a turnout (50) detected by means of a sensor system (3-x), and an evaluation device (4) for determining a turnout fault on the turnout (50) on the basis of the detected turnout condition (10), The evaluation device (4) is designed to estimate a cause of the fault (11) on the basis of a Bayesian network (5) modeling the switch (50) after a switch fault has been determined, the Bayesian network (5) at least on the basis of collected data an existing switch monitoring system and / or calibrated on the basis of structured expert knowledge, and an output device (6), the output device (6) being designed to output the estimated cause of the error (11) as an error cause signal (12).

Images