Identification of status groups out of several single states of a mobile unit

The invention relates to a method for identification of status groups out of several single states of a mobile unit.

Commonly, mobile units comprise a sensor system to determine the state of the mobile unit. Sensor systems with a large number of sensors deliver a lot of sensor data. One might try to analyse the sensor data with the aim to find a correlation between the sensor data and fails of the mobile unit. But no correlations can be found, because there are so many sensor data and just a few fails.

An objective of the invention is to provide an effective method to enable the analysing of sensor data. This objective is accomplished by means of a method according to claim 1. In the method for identification of status groups out of several single states of a mobile unit, according to the invention, each single state of the mobile unit is de ^¬ scribed by a single state description. Such a single state description comprises an identifier and state information.

The single state descriptions are analysed in that way, that for each single state of the mobile unit a probability of the state information is determined, and that for each single state of the mobile unit a state profile is created, which state profile comprises the probable state information. Fur ^¬ ther, the created state profiles are grouped by means of a clustering algorithm on the basis of their similarity, wherein several groups are formed and each of the formed groups describes a status group of the mobile unit.

The invention is based on the idea, that it will be helpful to form status groups, in which the state information is similar. In this case, one of the status group may correlate with at least one failure of the mobile unit. There may be another status group, which does not correlate with any of the fails of the mobile unit. Further, the invention is based on the finding that the single state descriptions describing one single state of the mo ^¬ bile unit, particularly the state information of the single state descriptions describing one single state of the mobile unit, may vary partially. Thus, the idea of the invention is to create for each single state of the mobile unit a state profile, which comprises the probable state information.

Thus, each state profile advantageously comprises the proba ^¬ ble state information for the respective single state of the mobile unit.

Moreover, the state profiles may comprise an identifier for the respective single state of the mobile unit. Expediently, the created state profiles are grouped on the basis of their similarity of their probable state profiles.

The mobile system may be a vehicle, particularly a railway vehicle, or any other mobile system. For instance, the iden ^¬ tifier may be an event code and/or a diagnostic code or any other identifier. The identifier may code a single state of the mobile unit and/or an event (type) concerning the mobile unit .

Further, the single state description can be, for example, a dataset, a vector, a text (in written form) or similar.

Advantageously, the mobile unit comprises several sensors. At least one single state description may be provided, if sensor data of at least one of the sensors indicates a single state and/or an event, particularly as defined by a given require- ment . In the meaning of this invention sensor data may be the measured values of the respective sensors and/or thereof cal ^¬ culated values. For example, at least one single state de ^¬ scription may be provided, if the sensor data exceeds a given range. Moreover, the given requirement may be any logic re ^¬ quirement considering the sensor data and/or depending on the sensor data. For instance, at least one single state descrip ^¬ tion may be provided, if the sensor data oscillates. Particu- larly, at least one single state description may be provided, if the sensor data shows a failure and/or a defect and/or if the sensor data is unusual and/or abnormal. Several single state descriptions may be provided intermittently and/or con ^¬ tinuously.

Preferably, at least one sensor has been read out for the re ^¬ spective single state description. Each sensor can be read out intermittently and/or continuously. Advantageously, the state information comprises at least one sensor address of the at least one sensor, which expediently has been read out for the respective single state descrip ^¬ tion. Moreover, the state information may comprise the sensor addresses of several sensors, which expediently have been read out for the respective single state description. The sensor address may be e . g. an alphanumerical code for the respective sensor. In the meaning of this invention the term sensor address comprises any information concerning the identity of the at least one sensor suitable for identification of the at least one sensor, e.g. a sensor certificate. Hence, the state information may comprise at least one sensor cer ^¬ tificate of the at least one sensor.

Preferentially, the state information comprises a measured value of the at least one sensor, which expediently has been read out for the respective single state description. Fur ^¬ ther, the state information may comprise a thereof calculated value. This means, the state information may comprise a value, which is calculated of a measured value of the at least one sensor, which expediently has been read out for the respective single state description. Hence, the state infor ^¬ mation may comprise sensor data of the at least one sensor, which expediently has been read out for the respective single state description.

Further, it is preferred that the state information comprises a creation time, at which the respective single state de ^¬ scription expediently has been created.

Moreover, it is advantageous that the state information com ^¬ prises a description text of the respective single state.

Also, if a description text is allocated to an identifier of at least one of the single state descriptions, it may be un ^¬ derstood that the state information of the respective single state description comprises the description text.

Expediently, the probable state information comprises a per ^¬ centage, at which the at least one sensor expediently has been read out for the respective single state of the mobile unit. The percentage may be calculated on the basis of the single state descriptions, particularly for each single state of the mobile unit. Moreover, the probable state information may comprise the sensor addresses of at least a part of the sensors, particularly of all sensors, which sensors expedi ^¬ ently have been read out for the respective single state of the mobile unit.

Furthermore, the probable state information may comprise the sensors, particularly the sensor addresses of the sensors, which sensors expediently have been read out at a percentage higher than a given percentage, e.g. higher than 50%.

Moreover, the creation times of two profiles may differ less than a given time difference for most of their occurrences. Hence, the average difference between the creation time of a first profile to the creation time of a second profile and/or the average difference between the creation times of two pro ^¬ files may be an average time difference. Thus, the average time difference (between the creation times of two profiles) may be less than a given time difference.

The probable state information of the profile may comprise the average time difference, if the average time difference is less than a given time difference. The average time dif ^¬ ference may be the average difference between the creation time of the profile to the creation time of another profile. The identifier of the other profile may be included into the average time difference of the profile.

Further, at least most of the description texts of two pro ^¬ files may be similar. Moreover, the description texts of two profiles may comprise the same key word(s) . The key word(s) may be predefined and/or may be determined by means of a ma ^¬ chine learning process. The probable state information of the two profiles may comprise the part of the description text being the same and/or the key word(s) being the same. Expediently, at least a first state profile belongs to a first group.

In a preferred embodiment of the invention, it is checked, particularly by means of a similarity measure, if most of the probable state information of a second state profile is the same as the probable state information of each state profile already belonging to the first group. The similarity (meas ^¬ ure) may be determined by means of provided knowledge and/or by means of a machine leaning process.

Particularly, if at least most of the probable state informa ^¬ tion of a second state profile is the same as the probable state information of each state profile already belonging to the first group, the second state profile expediently is a part of the same first group.

Particularly, "at least most of their probable state informa ^¬ tion is the same" may mean "more than 50% of their probable state information are the same", or any specific percentage higher than 50%. For example, two state profiles may belong to the same group, if more than 50% of their probable state information are the same. For instance, two state profiles may belong to the same group, if more than 50% of their sen ^¬ sor addresses are the same.

Further, "at least most of their probable state information is the same" may mean that at least one part of the probable state information (e.g. at least one sensor address or at least any other given number of sensor addresses) is different and/or is missing. For instance, two state profiles may belong to the same group, if at least one part of the prob ^¬ able state information is different and/or is missing, par- ticularly for one of the two state profiles. Moreover, "at least most of their probable state information is the same" may mean that there is a maximum given distance between their probable state information. Further, if at least most of the probable sensor addresses of a second state profile are the same as the probable sensor addresses of each state profile already belonging to the first group, the respective sensor data (expediently that means measured value determined by the respective sensor and/or thereof calculated value) may be compared as well. The probable sensor address of a sensor comprises the sensor ad ^¬ dress and the percentage of the respective sensor. Particu ^¬ larly, if at least most of the respective sensor data of the second state profile is the same as the respective sensor data of each state profile already belonging to the first group, the second state profile may be a part of the same first group.

In an advantageous embodiment of the invention, it may be checked, particularly by means of a similarity measure, if the probable state information of a second state profile is similar to the probable state information of each state pro ^¬ file already belonging to the first group. The similarity (measure) may be determined by provided knowledge and/or by means of a machine leaning process.

Particularly, if the probable state information of a second state profile is similar to the probable state information of each state profile already belonging to the first group, the second state profile advantageously is a part of the same first group. For instance, two state profiles may belong to the same group, if their creation times differ less than a given time difference. If both of the two state profiles occur more than once, the two state profiles may belong to the same group, if their creation times differ less than a given time difference for most of their occurrences. "Most of their occurrences" may mean "more than 50% of their occurrences", or any other specific percentage bigger than 50%. Moreover, the two state profiles may belong to the same group, if their average time difference is less than a given time difference.

Further, for example, the state profiles may be grouped by their description text. Two state profiles may belong to the same group, if at least most of their description text is similar. "At least most of the description text is similar" may mean "more than 50% of the description text is similar/is the same", or any other specific percentage bigger than 50%. Moreover, the two state profiles may belong to the same group, if the description texts comprise the same key

word(s) . The key word(s) may be predefined and/or may be de- termined by means of a machine learning process.

Preferably, the clustering algorithm comprises hierarchical clustering. The hierarchical clustering may be a diversive and/or an agglomerative clustering.

Expediently, the grouping is a clustering. Hence, "crated state profiles are grouped" may mean "crated state profiles are clustered". The grouping may be an unidimensional clus- tering. It is preferred that the grouping is a multidimensional clustering.

Expediently, the profile is created automatically, particu- larly by an evaluation unit.

For the grouping, a reasonable number of groups may be found, which number of groups may lay in a given range. Further, for the grouping, a reasonable similarity of the profiles within one group may be achieved, which reasonable similarity may lay over a given grad of similarity.

Further, the invention relates to a usage of the method de ^¬ scribed above for statistical analysis of the status groups of the mobile unit.

Within the statistical analysis of the status groups of the mobile unit, it may be determined how often a status group occurs in respect to other status groups. Further, it may be determined how often a single state occurs in respect to the status group it belongs. Moreover, it may be determined which status groups occur together, particularly in a given time range. Furthermore, it may be determined which groups corre ^¬ late with fails of the mobile unit. If a group correlates with fails/a failure of the mobile unit, it may be determined which fails/failure the group correlates with.

The advantage of the statistical analysis of the status groups of the mobile unit is to learn more about

fails/failures, its correlations and/or its sources. On this basis, it might be possible to decrease and/or to eliminate sources of fails/failures. Further, it might be possible to identify redundant and/or needless data. Therefore, it might be possible to decrease the amount of data, which has to be analysed.

Moreover, the invention relates to a usage of the method de ^¬ scribed above to allocate a provided single state description and/or an identifier of a provided single state description to one of the formed groups.

Further the invention relates to an evaluation unit for iden- tification of status groups out of several single states of a mobile unit. According to the invention, each single state of the mobile unit is described by a single state description, which single state description comprises an identifier and state information. According to the invention, the evaluation unit is embodied to analyse the single state descriptions in that way, that for each single state of the mobile unit a probability of the state information is determined, and that for each single state of the mobile unit a state profile is created, which state profile comprises the probable state in- formation. Moreover, the evaluation unit is embodied to group the created state profiles by means of a clustering algorithm on the basis of their similarity, wherein several groups are formed and each of the formed groups describes a status group of the mobile unit.

The evaluation unit may be used for executing the method mentioned above. Thus, features, which are mentioned above (in connection with the method) may also refer to the evaluation unit .

Further the invention relates to an analysis system, which comprises the evaluation unit mentioned above, a processing device and several sensors. According to the invention, the sensors are connected with the processing device and the processing device is connected with the evaluation unit.

The sensors may be the sensors mentioned in connection with the method. The sensors may be positioned in und/or at the mobile system.

The processing device may be embodied to read out at least one of the several sensor. Moreover, the processing device may be embodied to create and/or provide single state de- scriptions. Particularly, the processing device may be embod ^¬ ied to send the single state descriptions to the evaluation unit . Even if terms are used in the singular or in a specific nu ^¬ meral form, the scope of the invention should not be re ^¬ stricted to the singular or the specific numeral form.

The previously given description of advantageous embodiments of the invention contains numerous features which are par ^¬ tially combined with one another in the dependent claims. Ex ^¬ pediently, these features can also be considered individually and be combined with one another into further suitable combi ^¬ nations. More particularly, these features may be combined with the evaluation unit, the analysis system and the method according to the respective independent claim individually as well as in any suitable combination. Furthermore, features of the method, formulated as apparatus features, may be consid ^¬ ered as features of the evaluation unit and/or of the analy- sis system. Accordingly, features of the evaluation unit and/or of the analysis system, formulated as process fea ^¬ tures, may be considered as features of the method.

The above-described characteristics, features and advantages of the invention and the manner in which they are achieved can be understood more clearly in connection with the following description of an exemplary embodiment, which will be explained with reference to the drawings. The exemplary embodi ^¬ ment is intended to illustrate the invention, but is not sup- posed to restrict the scope of the invention to combinations of features given therein, neither with regard to functional features. Furthermore, suitable features of the exemplary em ^¬ bodiment can also be explicitly considered in isolation, be removed from one of the exemplary embodiment, and/or be com- bined with any of the appended claims.

In the drawings display: FIG 1 a schematic overview of an analysis system compris ^¬ ing an evaluation unit;

FIG 2 table containing single state descriptions

FIG 3 a table showing profiles.

FIG 1 shows an analysis system 2 comprising an evaluation unit 4 for identification of status groups out of several single states of a mobile unit. Further, the analysis system 2 comprises several sensors 6 and a processing device 8. Ex- emplarily, the analysis system 2 comprises four sensors 6. Principally, the analysis system 2 may comprise any number of sensors 6. Each sensor 6 of the analysis system 2 is con- nected with the compressing device 8 via a data connection 10. Moreover, the processing device 8 is connected to the evaluation unit 4 via a data connection 12. Each of the data connections 10, 12 may be tethered or wireless.

The sensors 6 as well as the processing device 8 are arranged in and/or at a mobile unit 14. In this example the mobile unit 14 is a railway vehicle, particularly a train. Further, the evaluation unit 4 may be situated at the landside, e. g. in a (primary) control centre (not shown) .

The processing device 8 reads out the sensors 6. Moreover, the processing device 8 creates and/or provides single state descriptions 18 (see FIG 2) . At least one sensor 6 is read out for the respective single state description 18. Particu ^¬ larly, if the sensor data of at least one sensor 6 fulfils a given requirement, at least one single state description 18 is created. Moreover, the single state description ( s ) 18 is/are sent to the evaluation unit 4, particularly by the processing device 8.

The evaluation unit 4 identifies status group out of several single states of the mobile unit 14. Each single state of the mobile unit 14 is described by at least one single state de- scription 18 (see FIG 2) . Each single state description 18 comprises an identifier 20 and state information 22 (see FIG 2) . The single state descriptions 18 are analysed in the way that for each single state of the mobile unit 14 the probability of the state information 22 is determined. Fur ^¬ ther, the single state descriptions 18 are analysed in the way that for each single state of the mobile unit 14 a state profile 30 is created, which comprises the probable state in ^¬ formation 32 (see FIG 3) . The analysis takes place in the evaluation unit 4. Moreover, the analysis is done automati ^¬ cally. Hence, the state profiles 30 are created automati ^¬ cally.

Further, the created state profiles 30 are grouped by means of a clustering algorithm on the basis of their similarity, particularly by means of the evaluation unit 4. Therein several groups are formed and each of the formed groups de ^¬ scribes a status group of the mobile unit 14. Also, the grouping is done automatically.

The method for identification of status groups out of several single states of the mobile unit 14 may be used for statisti ^¬ cal analysis of the status groups of the mobile unit 14.

Within the statistical analysis of the status groups of the mobile unit 14, it is determined, for example, how often a status group occurs in respect to other status groups. More ^¬ over, it can be determined which status groups occur to ^¬ gether, particularly in a given time range. Furthermore, it can be determined which groups correlate which fails of the mobile unit 14.

Further, the method for identification of status groups out on several single states of a mobile unit 14 may be used to allocate a provided singles state description 18 and/or to allocate an identifier 20 of a provided single state descrip ^¬ tion 18 to one of the formed groups. For example, after the grouping, each (into the evaluation unit 4) incoming singles state description 18 is allocated to one of the formed groups .

Further, for each of the formed groups/status groups an in- struction about how to act may be provided.

FIG 2 schematically shows a table 16 containing single state descriptions 18. Each line of the table 16 schematically shows one single state description 18 at least partially in form of a data set.

Each single state description 18 comprises an identifier 20 and state information 22. The state information 22 comprises a sensor address 24 of at least one sensor 6 which has been read out for the respective single state description 18. The sensor address 24 may be an alphanumerical code for the re ^¬ spective sensor 6. The state information 22 may also comprise a measured value of the at least one sensor 6, which has been read out for the respective single state description 18, and/or a thereof calculated value (not shown) . Moreover, the state information 22 comprises a creation time 26 at which the respective single state description 18 has been created. Further, the state information 22 comprises a description text of the respective single state (not shown) . The descrip- tion text describes the meaning of the identifier 20. More ^¬ over, the description text describes the single state of the mobile unit 14 shown in FIG 1.

The single state descriptions 18 are analysed in the follow- ing way, particularly by the evaluation unit 4 shown in

FIG 1:

For each single state of the mobile unit 14 a probability of the state information 22 is determined. The probable state information 32 is gained, wherein the probable state informa ^¬ tion 32 comprises a percentage 34 at which the at least one sensor 6 has been read out for the respective single state of the mobile unit 14. For each single state of the mobile unit 14 the percentage 34 is calculated on the basis of the single state descriptions 18.

For example, for the single state descriptions 18 comprising the identifier 20 "CC" the Sensor IV has been read out at 100%, and the Sensor III has been read out at 33%. Further, for the identifier "AA" Sensor I and Sensor II are read out, both at 100%. Moreover, for each single state of the mobile unit 14 a state profile 30 (see FIG 3) is created, wherein the state profile 30 comprises the probable state information 32. The state profiles are created in the evaluation unit 4 shown in FIG 1. FIG 3 shows a table 28 comprising the created state profiles 30 for each single state of the mobile unit 14. Each state profile 30 comprises an identifier 20 and the probable state information 32 for the respective single state of the mobile unit 14. In this example, the probable state information 32 comprises the sensor addresses 24 of the sensors 6, which have been reed out for the respective single state of the mo ^¬ bile unit 14, and the percentage 34, at which the respective sensors 6 have been read out. Principally, the state profiles 30 could also comprise other probable state information based on the respective single state descriptions 18.

The state profiles 30 are grouped by means of a clustering algorithm on the basis of their similarity. In this example, the clustering algorithm is hierarchical clustering, particu- larly agglomerative clustering. Several groups are formed, and each of the formed groups describes a status group of the mobile unit 14. The groups are formed in the evaluation unit 4 shown in FIG 1. The groups are formed as following:

At least a first state profile 30 belongs to first group. For the grouping it is checked by means of a similarity measure, if most of the probable state information 32 of a second state profile 30 is the same as the probable state informa ^¬ tion 32 of each state profile 30 already belonging to the first group. In a first example, two state profiles 30 belong to the same group, if their probable state information 32 is the same. Particularly, two state profiles 30 belong to the same group 30, if their sensor addresses 24 are the same. In this exam ^¬ ple, the state profiles 30 with the identifiers 20 "AA" and "BB" comprise the same sensor addresses 24, namely "SI" and "SII", so that these profiles can be grouped into one group, namely " (AA, BB) " .

Further, the state profile 30 with the identifier 20 "AB" has more sensor addresses 24 than any other state profile 30, so that the state profile 30 with the identifier 20 "AB" stays alone in one group.

In a second example, which can be examined additionally and/or alternatively to the first example, a grouping can be done as following:

Two state profiles 30 belong to the same group 30, if at least most of their probable state information 32 is the same. Particularly, two state profiles 30 belong to the same group 30, if their sensor addresses 24 are the same and/or if most of their sensor 24 addresses are the same.

Particularly, it is checked by means of a similarity measure, if most of their probable state information 32 (hier: sensor addresses 24) are the same. The evaluation unit 4 comprises provided knowledge and/or machine leaning. The similarity measure is determined by means of the provided knowledge and/or a machine leaning process. Most of their probable state information 32 is the same, if their measured similar ^¬ ity is higher than a given value. Two state profiles 30 belong to the same group 30, if - in this example - at least one sensor address 24 is missing for one of the two state profiles 30. Hence, two state profiles 30 belong to the same group 30, if their maximum distance is One (in this example) .

In this example, the state profiles 30 with the identifiers 20 "AA" and "BB" comprise the same sensor addresses 24, namely "SI" and "SII". Their distance is Zero. Thus, the state profiles 30 with the identifiers 20 "AA" and "BB" are grouped to one group, particularly because their similarity is higher than the given value.

Further, the state profiles 30 with the identifier 20 "DD" comprise one sensor address 24 more (namely "SIV") than the state profiles 30 with the identifiers 20 "AA" and "BB". Hence, one sensor address 24 is missing for the state pro ^¬ files 30 with the identifiers 20 "AA" and "BB" in respect to the state profile 30 with the identifier 20 "DD". The dis- tance between the state profile with the identifier 20 "DD" to the state profiles 30 with the identifiers 20 "AA" and "BB" is One. Therefore, the state profiles 30 with the iden ^¬ tifiers 20 "AA", "BB" and "DD" are grouped to one group, namely " (AA, BB, DD) ", particularly because their similarity is higher than the given value.

Also the state profiles 30 with the identifier 20 "AB" are allocated to one sensor address 24 more (namely "SIII") than the state profile 30 with the identifier 20 "DD". Hence, one sensor address 24 is missing for "DD" in respect to "AB" - their distance is One. But two sensor addresses 24 are miss ^¬ ing for "AA" and "BB" in respect to "AB" - thus, their distance is Two. Therefore, "AA", "BB", "DD" and "AB" are not grouped into one group 30, particularly because their simi- larity is not higher than the given value.

In this second example, five different state profiles 30 are grouped into three different groups, namely into the groups " (AA, BB, DD)", "AB" and "CC". Hence, also the seven different single state descriptions 18 are grouped into the three different groups.

Additionally and/or alternatively, the identifiers 20 can also be grouped by other aspects than their sensor ad ^¬ dress (es) 24, e. g. by the massage time 27, by the descrip ^¬ tion text and/or by their measured values. Hence, the group ^¬ ing can be a multidimensional grouping.

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 disclo ^¬ sure. For example, elements described in association with different embodiments may be combined. Accordingly, the par ^¬ ticular 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.