METHOD AND ELEVATOR CONTROLLER FOR DETECTING A MALFUNCTION IN AN ELEVATOR

Abstract

An elevator controller detects a malfunction such as elevator blockage in an observed elevator utilizing a method including: acquiring first data during an application phase, the first data correlating with at least one condition in the observed elevator; acquiring further data during the application phase, the further data correlating with the at least one condition in other elevators; determining a current relative behavior of the observed elevator during the application phase based on a comparison of the first data with the further data; and detecting the malfunction in the observed elevator based on an analysis of the current relative behavior. The normal relative behavior information of the observed elevator, learned in a machine learning procedure during a preceding learning phase, is taken into account upon analyzing the current relative behavior of the observed elevator. The method enables automatically detecting malfunctions in an elevator while reducing a probability of false alarms.

Claims

1-12. (canceled)

13. A method for detecting a malfunction in an observed elevator, the method comprising the steps of: acquiring first data during an application phase, the first data correlating with at least one condition in the observed elevator; acquiring further data during the application phase, the further data correlating with the at least one condition in other elevators; determining a current relative behavior of the observed elevator during the application phase based on a comparison of the first data with the further data; and detecting a malfunction in the observed elevator based on an analysis of the current relative behavior.

14. The method according to claim 13 wherein the malfunction is a temporary blockage of the observed elevator and wherein the first data and the further data correlate with conditions of an elevator being affected by a temporary blockage.

15. The method according to claim 13 including determining a normal relative behavior of the observed elevator in a learning phase preceding the application phase by the steps of: acquiring other first data during the learning phase, the other first data correlating with the at least one condition in the observed elevator; acquiring other further data during the learning phase, the other further data correlating with the at least one condition in the other elevators; determining the normal relative behavior of the observed elevator during the learning phase based on a comparison of the other first data with the other further data acquired during the learning phase; and wherein the step of detecting the malfunction in the observed elevator includes an analysis of the current relative behavior in comparison with the normal relative behavior.

16. The method according to claim 13 wherein the other elevators have previously been determined to have a certain similarity to the observed elevator.

17. The method according to claim 16 wherein the certain similarity between the observed elevator and one of the other elevators is determined in a learning phase preceding the application phase, which application phase includes the step of detecting the malfunction in the observed elevator.

18. The method according to claim 17 wherein the certain similarity between the observed elevator and the one of the other elevators is determined based upon at least one of: information relating to a physical distance between the observed elevator and the one of the other elevators; information relating to an application of the observed elevator and the one of the other elevators; and information relating to a temporal segment in which the first data and the further data are acquired.

19. The method according to claim 13 wherein the first data is generated based on signals provided by sensors supervising conditions in the observed elevator and the further data is generated based on signals provided by sensors supervising conditions in the other elevators.

20. The method according to claim 13 wherein the first data and the further data correlate to at least one of: a number of door motions occurring during a time interval; a number of elevator trips occurring during a time interval; a change in car occupancy occurring during a time interval; distances travelled during a time interval; an amount of time passed since a last trip; and an amount of time passed since a last door motion.

21. The method according to claim 13 wherein the method is executed in an elevator controller receiving at least one of: the first data acquired by a multiplicity of sensors distributed throughout the observed elevator and the further data acquired by a multiplicity of sensors distributed throughout the other elevators; and control data generated in an elevator control unit of the observed elevator and control data generated in an elevator control unit of at least one of the other elevators.

22. The method according to claim 13 including alerting a service staff to the detected malfunction.

23. The method according to claim 13 including initiating an action to overcome the detected malfunction.

24. An elevator controller comprising: a data acquisition interface receiving the first data and the further data; and a data processor connected to the data acquisition interface and being adapted to at least one of execute, perform and control the method according to claim 13 to detect the malfunction in the observed elevator.

25. A computer program product comprising computer readable instructions which, when performed by a processor of an elevator controller, instruct the elevator controller to at least one of execute, perform and control the method according to claim 13 to detect the malfunction in the observed elevator.

26. A non-transitory computer readable medium comprising the computer program product according to claim 25 stored thereon.

Description

DESCRIPTION OF THE DRAWINGS

[0071] FIG. 1 shows an arrangement of multiple elevators in which malfunctions may be detected with a method in accordance with the present invention.

[0072] FIG. 2 shows a flowchart of a possible implementation of a method in accordance with the present invention.

[0073] The figures are only schematic and not to scale. Same reference signs refer to same or similar features.

DETAILED DESCRIPTION

[0074] FIG. 1 shows an arrangement of several elevators 1 including an observed elevator 3 and several other elevators 5 in a first building 7 and a second building 9. Each elevator 1 comprises an elevator car 11 which may be moved throughout an elevator shaft 13 such as to be accessible from various floors 15. Each of the elevator cars 11 is displaced using a drive unit 17 controlled by a control unit 19.

[0075] In each of the elevators 1, a multiplicity of sensors 21 is provided for measuring parameters relating to various physical conditions. For example, a door motion sensor 23 may measure parameters which may change upon a car door or a shaft door being opened and/or closed. An acceleration sensor 25 may measure accelerations acting onto the elevator car 11. A position sensor 27 may measure a current position of the elevator car 11 throughout the elevator shaft 13. A car load sensor 29 may measure a load currently acting onto the elevator car 11 thereby allowing indirectly determining changes in car occupancy.

[0076] Signals provided by the sensors 21 or data correlating therewith may be transmitted from each of the elevators 1 via transmitters 31 towards a central elevator controller 33. Such central elevator controller 33 may be located in a remote control center being far away from the buildings 7, 9. The central elevator controller 33 may comprise a data acquisition interface 37 for receiving the signals or data transmitted by the transmitters 31. Furthermore, the central elevator controller 33 comprises a data processor 35 for processing the data received via the data acquisition interface 37.

[0077] The central elevator controller 33 may execute or control a method for detecting a malfunction in the observed elevator 3 in accordance with an embodiment of the present invention.

[0078] For such purpose, first data correlating with at least one condition in the observed elevator 3 may be acquired during an application phase. For example, such first data may be derived from signals of one or more of the sensors 21. Particularly, first data may be derived from signals of one or more sensors 21 supervising conditions which are influenced upon occurrence of the malfunction to be detected, i.e. for example supervising conditions which are influenced upon occurring of a blockage of the observed elevator 3.

[0079] Furthermore, further data is acquired during the application phase, such further data also relating to the same one condition as supervised in the observed elevator 3 but prevailing in one of the other elevators 5. For example, such further data may be derived from signals of one or more sensors 21 supervising conditions which are influenced upon the same type of malfunction to be detected, i.e. for example supervising conditions which are influenced upon occurrence of a blockage in the other elevator 5.

[0080] Subsequently, based on the acquired first data and further data and processing these first and further data in the data processor 35, the elevator controller 33 determines a current relative behavior of the observed elevator 3 during the application phase based on a comparison of the first data with the further data.

[0081] Finally, based on an analysis of the current relative behavior, the malfunction in the observed elevator 3 may be detected.

[0082] In a preferred embodiment, not only the current relative behavior of the observed elevator 3 is analyzed, but also a normal relative behavior of this observed elevator 3 is taken into account upon detecting the malfunction in the observed elevator 3. For such purpose, the normal relative behavior is determined in a machine learning procedure during a learning phase. In principle, the normal relative behavior is determined using a similar procedure as used for determining the current relative behavior but based on first and further data acquired during the learning phase instead of the application phase.

[0083] Furthermore, in a preferred embodiment, only the further data acquired in other elevators 5 having previously been determined to have a certain similarity to the observed elevator 3 are taken into account upon determining the current relative behavior of the observed elevator 3.

[0084] For example, other elevators 5 being located in the same first building 7 as the observed elevator 3 may be assumed to have a sufficient similarity to the observed elevator 3 as an elevator usage profile throughout the same first building 7 may be assumed to be similar for all of the elevators 1 comprised in this first building 7.

[0085] Overall, a system is proposed which may learn regular operational behavior of an observed elevator 3 from its deviance with respect to other elevators 5 for example located nearby and/or having another type of similarity with respect to the observed elevator 3.

[0086] Subsequently, a specific implementation of an embodiment of the method for detecting a malfunction in the observed elevator 3 will be explained with reference to the flowchart shown in FIG. 2.

[0087] Roughly summarized, such embodiment may comprise five steps. In a first step S1, a time series between elevator installations may be correlated. A pairwise correlation matrix resulting therefrom may then be processed in a second step S2 in which basis elevators are selected based on for example correlation threshold. Resulting from such second step, clusters of semantically related elevator installations are used in a third step S3 for deriving a differential model between semantic neighbors. Two other possible ways to select clusters of semantically related elevator installations used in the third step are by similar physical distance (e.g. elevators in the same building) or by activation time (e.g. elevators operating during the same time periods like 9-5 on workdays). Upon learning the relationship of the observed elevator installation as related to basis elevators during a learning phase, the observed elevator may be monitored and malfunctions may be detected based on a maximum differential tolerance. For such purpose, in a fourth step S4, it is then decided whether or not a malfunction such as a blockage of the elevator is present in the observed elevator. For such blockage decision, first and further current data 39 and a current relative behavior of the observed elevator 3 derived therefrom may be compared with data representing a normal relative behavior of the observed elevator 3 as determined during the learning phase. If, in such decision step, a blockage of the observed elevator 3 is detected, service staff may be alerted in a final step S5 or other actions for overcoming the detected malfunctions may be initiated.

[0088] In a specific example of this embodiment, in the first step S1, multi-dimensional signals may be given for all elevator installations in a fleet. A pairwise correlation matrix may then be sought, where entries of such correlation matrix depict a correlation in the input signals for pairs of elevator installations in a portfolio.

[0089] A calculation of a calculation metric may be determined by a practitioner, e.g. Spearman's rho may be applied for rank based correlation.

[0090] Furthermore, multi-dimensional input signals may be first projected onto a lower-dimensional space (e.g. a single dimensional time series) before calculating the correlation matrix, applying for example Principle Component Analysis.

[0091] Then, in the second step S2, for the whole portfolio of elevator installations, elevator installations may be clustered into “semantically similar” groups. Parameters for clustering such installations may depend on the practitioner's choosing. For example, physical distance between elevator installations (e.g. within one building, one city block, etc.) may be chosen. Alternatively or additionally, existing semantic grouping of buildings (e.g. groups installations of different hotel/hospital/school buildings) may be chosen. As a further alternative or supplement, a temporal segment in which to compare the correlation threshold (e.g. only correlation of input signals from the last 4 weeks are considered) may be chosen.

[0092] Subsequently, in the third step S3, given installations within one semantic cluster, a differential baseline is constructed by subtracting each elevator installation's input signals from a weighted average of the other elevator installation's input signals. The weighting may be given by the semantic similarity of installations defined in Step S2.

[0093] As such, the differential baseline of each installation i.e. may be calculated as:

[00001]$D_i=\frac{1}{\mathrm{len}\left(S\right)-1}\underset{j}{.Math.}{w}_{i,j}*M_j\forall j\ne i\in S$

where D.sub.i is the differential baseline for installation i; S is the set of installations belonging to the same semantic cluster; w.sub.i,j is the semantic similarity between installation i and j; M.sub.j is the matrix of input signals where the columns are the different signal types and the rows are mapped semantic time periods (e.g. hour 0 of a weekday, hour 1 of a weekday, . . . hour 23 of a weekday, hour 0 of a weekend day, hour 1 of a weekend day . . . ).

[0094] Additionally, there are many other possible ways to calculate a differential baseline of each installation.

[0095] It is to be noted that semantically meaningful time periods may be used to aggregate absolute time in the training period. The specific aggregation method and interval of aggregation remains a choice of the practitioner (e.g. averages of hourly time bins for weekdays and weekend days).

[0096] For each installation, a model may be learned from the ‘normal’ behavior of the differential baseline using standard anomaly detection methods, e.g. One-Class Support Vector Machines. These models may learn from a matrix of data (D.sub.i) to define a ‘normal’ space of operations. In this context, the historical differential baseline provided by D.sub.i characterizes how a specific installation operates with respect to other semantically related installations.

[0097] Finally, having trained the model, it may be applied on current data. The blockage estimation step requires that current data is processed the same way as the training data to derive differential vectors. These differential vectors may then be fed to the anomaly detection model to determine whether the vector is anomalous or not. If anomalous, operations personnel may be notified. Otherwise, the data may be transferred back into the training pipeline as additional training data for continuous model updates.

[0098] Briefly summarized, embodiments of the present invention allow overcoming shortcomings of conventional approaches for detecting malfunctions in an elevator. Particularly, it may no more be necessary that a customer complaint has to trigger a call-back and/or a site visit upon a malfunction occurring in an elevator. Instead, the approach proposed herein allows automatically detecting malfunctions such as blockage in an elevator with a low probability of false alarms. Particularly, the proposed approach is less susceptible to false positives (false alarms) that may arise from the examination of single installations. For example, lack of operation of a specific elevator at the beginning of a school day would not raise a blockage alert if all elevators in the building are also lacking in operation. Provisioning and consideration of operational rules may be significantly reduced or minimized as blockage detection may be based on relative behavior of installations against its neighbors. An unsupervised approach may eliminate a need for manual labelling of blockage scenarios. A solution may be independent of an input signal source/automatic pruning of irrelevant input signals. The present approach may enable a proactive service call before customer realize a malfunction. Furthermore, the present approach may be applicable in modernization or NI (new installation) installations where additional sensing hardware is deployed without connection to a shaft information system or elevator control unit.

[0099] Finally, it should be noted that the term “comprising” does not exclude other elements or steps and the “a” or “an” does not exclude a plurality. Also, elements described in association with different embodiments may be combined.

[0100] In accordance with the provisions of the patent statutes, the present invention has been described in what is considered to represent its preferred embodiment. However, it should be noted that the invention can be practiced otherwise than as specifically illustrated and described without departing from its spirit or scope.