KR102702121B1 - Elevator Failure Diagnosis System Using Recurrent Neural Network Model - Google Patents

Abstract

An elevator fault diagnosis system utilizing a recurrent neural network model is disclosed, which includes: an elevator control panel that controls the operation of an elevator and records fault data related to a fault in real time; a cloud that receives and accumulates fault data recorded in the elevator control panel; a preprocessing unit that extracts key data to be used for analyzing the cause of a fault occurrence of an elevator from the fault data stored in the cloud and performs preprocessing to list them by time zone; a labeling unit that combines the data preprocessed by the preprocessing unit with a fault cause label indicating the actual cause of a fault occurrence of the elevator to form a learning data set; a learning modeling unit that performs recurrent neural network learning with the learning data set to generate a fault diagnosis model that infers the actual cause of a fault occurrence of the elevator; and a storage management unit that stores and distinguishes and manages the data preprocessed by the preprocessing unit, the learning data set labeled by the labeling unit, and the fault diagnosis model generated by the learning modeling unit.

Description

{Elevator Failure Diagnosis System Using Recurrent Neural Network Model}

The present invention relates to an elevator failure diagnosis system utilizing a recurrent neural network model, and more specifically, to an elevator failure diagnosis system that performs machine learning using a recurrent neural network model to identify an actual cause of a failure from a large number of failure data that occur continuously and provides the data to a maintenance worker, thereby supporting maintenance work.

Various types of high-rise buildings constructed for residential, commercial, and other purposes are equipped with elevator systems to ensure smooth movement between floors for passengers entering and exiting the building.

Typically, an elevator system is composed of an elevator car that moves along a vertical shaft formed inside a building while carrying passengers, a mechanical part including a motor unit and a traction machine that generate power to raise and lower the elevator car, and a control unit that performs control related to the operation of the elevator car.

Meanwhile, elevator maintenance is carried out when a maintenance worker goes to the site where the elevator is installed when an inspection request or a report of a malfunction is received for the elevator installed in the building, diagnoses the condition of the elevator, and then takes necessary measures.

In general, the elevator system is equipped with a monitoring device that monitors the operating status of the elevator and transmits a fault code indicating the type of fault to the maintenance center when a fault occurs. When a fault occurs in the elevator, the monitoring device transmits a fault signal containing a fault code to the maintenance center, and the maintenance center that receives the fault information dispatches a maintenance worker to the site of the elevator where the fault occurred to handle the fault. Here, the fault code refers to a number or a code combining numbers and English letters output from the elevator control panel when a fault occurs in the elevator.

If we look at the process of handling an elevator failure in more detail, when a report of an elevator failure is received, a maintenance worker dispatched to the site of the elevator failure connects the maintenance terminal to the elevator control panel to check the fault code that has occurred, and takes action according to the corresponding repair instructions by referring to the manual according to the fault code.

However, since the difficulty of fault analysis varies for each fault that occurs due to the nature of the elevator system in which various devices are organically connected, the time required to analyze the fault and resolve the problem may vary depending on the skill level of the maintenance personnel.

In addition, since in most cases when an elevator malfunctions, multiple fault codes occur simultaneously rather than a single fault code, it is difficult for unskilled maintenance personnel to accurately identify the actual cause of the malfunction, which hinders rapid fault resolution and increases elevator downtime.

In other words, conventional elevator maintenance has a large difference in the time for analyzing and processing the cause of a failure depending on the skill level of the worker, and if a failure is misanalyzed due to the lack of skill of the maintenance worker, there is a risk that other problems will arise as a result, and if the dispatched worker is unable to resolve the issue, other personnel from the research institute must be mobilized, which is very likely to result in additional personnel and time costs.

The operation data generated from various components (traction machine motor, brake, door, inverter, control panel, etc.) that make up the elevator system have the characteristic of being generated continuously. In order to accurately analyze a failure occurring in an elevator due to the characteristics of the elevator system in which operation data is generated continuously, it is necessary to check the data history generated before and after the time of occurrence of the failure data, classify the scope, and comprehensively analyze it. However, there is currently no clear standard for classifying the scope.

Accordingly, the present invention provides a fault diagnosis system capable of accurately and quickly identifying the actual cause of a fault from a large number of fault data occurring continuously using a recurrent neural network model capable of learning time series data, thereby supporting maintenance work, thereby reducing fault interpretation errors and shortening the time required for problem solving regardless of the skill level of maintenance personnel.

In addition, the present invention aims to reduce the skill gap of maintenance personnel, reduce the possibility of errors in fault interpretation, and improve the stability of maintenance by providing maintenance personnel with a new cause class indicating the actual cause of a fault based on learning about the continuous sequence of fault codes observed when an elevator breaks down.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

According to one aspect of the present invention for achieving the above object, an elevator fault diagnosis system utilizing a recurrent neural network model may be provided, which includes: an elevator control panel that controls the operation of an elevator and records fault data related to a fault in real time; a cloud that receives and accumulates the fault data recorded in the elevator control panel; a preprocessing unit that extracts key data to be used for fault cause analysis of the elevator from the fault data stored in the cloud and performs preprocessing to list them by time zone; a labeling unit that combines data preprocessed by the preprocessing unit with a fault cause label indicating an actual fault cause of the elevator to form a learning data set; a learning modeling unit that performs recurrent neural network learning with the learning data set to generate a fault diagnosis model that infers the actual fault cause of the elevator; and a storage management unit that stores and distinguishes data preprocessed by the preprocessing unit, the learning data set labeled by the labeling unit, and the fault diagnosis model generated by the learning modeling unit.

The above main data is a fault code included in the above fault data, and the above preprocessing unit can perform grouping to list the fault codes by time zone and group those within a set similar time range into one group, thereby separating the fault codes into a continuous sequence.

The above fault cause label is defined as a cause class that indicates a fault cause of the elevator that actually occurred when a continuous sequence of the fault codes is observed in a specific pattern, and the labeling unit can perform a labeling task of combining the continuous sequence of the fault codes and the corresponding cause class by referring to past fault history data.

The above fault diagnosis model can receive preprocessed fault data of a specific elevator as input data and output the corresponding cause class as a fault diagnosis result.

The above fault diagnosis model can be updated by repeatedly performing learning using the data input during fault diagnosis of the specific elevator and the output results thereof as feed data.

A fault diagnosis system according to one aspect of the present invention may further include a web server that is accessible via a user terminal and provides the user with a fault diagnosis result of the specific elevator selected by the user.

The above web server can provide a web application including a service screen that allows the user to select the specific elevator and visually display the fault diagnosis results.

Bidirectional LSTM (bidirectional Long Short-Term Memory) can be used as the above recurrent neural network model.

According to the present invention as described above, by applying a recurrent neural network model suitable for the characteristics of an elevator in which operation data is continuously generated, the advantages of deep learning can be utilized to derive accurate diagnosis results for faults occurring in an elevator and provide them to maintenance personnel, thereby reducing the deviation in time required for fault analysis and problem solving regardless of the skill level of the maintenance personnel.

In addition, the present invention can find the actual cause of an elevator failure by comprehensively analyzing a plurality of failure data that occur in time series rather than diagnosing a failure using only one failure data, and thus has the effect of significantly improving the accuracy of elevator failure analysis compared to conventional methods.

In particular, according to the present invention as described above, not only is it possible to reduce misanalysis of failures occurring in an elevator possible, but it also has the effect of expanding the scope of failure analysis by providing failures of unknown causes that cannot be explained by existing failure codes or failures related to elevator parts as new cause classes.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

Figure 1 is a schematic diagram showing the configuration of an elevator failure diagnosis system according to the present invention.
FIG. 2 is a drawing showing an example of labeling work between a fault code sequence and a cause class performed by an elevator fault diagnosis system according to the present invention.

The purpose and technical configuration of the present invention and the detailed operation and effect thereof will be more clearly understood by the detailed description based on the drawings attached to the specification of the present invention.

The terminology used herein is only used to describe specific embodiments and is not intended to limit the present invention. For example, terms such as "consisting of" or "comprising" used herein should not be construed as necessarily including all of the various components or various steps described in the invention, but should be construed as not including some of the components or some of the steps, or may further include additional components or steps. In addition, the singular expression used herein includes the plural expression unless the context clearly indicates otherwise.

Hereinafter, the present invention will be described in detail by describing preferred embodiments of the present invention with reference to the attached drawings. The embodiments described below are provided so that those skilled in the art can easily understand the technical idea of the present invention, and should not be construed as limiting the present invention thereby, and it is obvious that the embodiments of the present invention can have various applications to those skilled in the art.

Fig. 1 is a schematic diagram showing the configuration of an elevator fault diagnosis system according to the present invention. Fig. 2 is a drawing showing an example of a labeling task between a fault code sequence and a cause class performed by an elevator fault diagnosis system according to the present invention.

Referring to FIG. 1, the elevator failure diagnosis system according to the present invention comprises: an elevator control panel (10) that controls operations related to the operation of an elevator and records various operation data of the elevator in real time, including failure data related to a failure; a cloud (20) that receives and accumulates failure data recorded in the elevator control panel (10); a preprocessing unit (30) that reads data from the cloud (20) and preprocesses it into the form of input data of a learning model; a storage management unit (40) that stores data preprocessed by the preprocessing unit (30); a labeling unit (50) that combines the preprocessed data stored in the storage management unit (40) with a failure cause label to form a learning data set and stores it again in the storage management unit (40); a learning modeling unit (60) that fetches a labeled learning data set from the storage management unit (40) and performs recurrent neural network learning to generate a learning model and stores the model on which learning is completed again in the storage management unit (40). And it can be configured to include a web server (70) that can be accessed through a user terminal (80) and provides the analysis results of a learning model that inputs failure data that occurred in a specific elevator selected by the user as a failure diagnosis result.

The elevator control panel (10) performs driving control related to the operation of the elevator, and can record parameters related to the operation of various devices constituting the elevator and the control panel itself in real time as operation data.

To this end, the elevator control panel (10) may include a detection device that periodically monitors the operating status of the elevator and collects signals related to the elevator operation as an operation history for each system, and the monitoring device may receive data measured from a number of sensors that collect parameters related to the operation of various devices constituting the elevator.

The elevator control panel (10) can generate failure data including a failure code that identifies a failed device when a failure of a device included in the elevator is detected, and various operation data of the elevator including the failure data are transmitted from the elevator control panel (10) to the cloud (20) and accumulated and stored.

In the cloud (20), various operation data recorded in the elevator control panel (10) can be collected on a continuous basis, and in particular, failure data related to elevator failures can be collected and accumulated, including information on the time of occurrence.

The preprocessing unit (30) defines and extracts key data that affect the interpretation of the cause of failure from the original data collected in the cloud (20), and preprocesses them into an input data format that fits the learning model so that they can be applied to recurrent neural network learning. Here, "preprocessing into an input data format that fits the learning model" may mean performing a function of converting data into a format suitable for input into the learning model generated by the learning modeling unit (60) described below.

In addition, the 'original data' above may mean only the failure data including the failure code, or may mean all of the elevator's operation data including the failure data. The failure data is input/output data generated when the elevator fails, and may include data exchanged between the main board, inverter, door inverter, and COP (Car Operating Panel) board used for button operation.

The preprocessing performed by the preprocessing unit (30) may include extracting key data that affect the interpretation of the cause of an elevator failure from the original data, sorting the extracted key data based on the time of failure, and separating the sorted key data into a sequence suitable for input to the learning model. The preprocessing method used at this time may be a general method such as error filtering, data interpolation, over/under sampling, or dimension reduction.

As will be described later, the present invention is characterized by inferring an actual cause of a failure from a continuous sequence of a plurality of failure codes that occur simultaneously and defining a new cause class. In the above, the "key data that influences the interpretation of the cause of a failure" may mean a failure code. That is, the preprocessing unit (30) may extract a failure code as key data from the failure data collected in the cloud (20).

In addition, the preprocessing unit (30) can sort the extracted fault codes according to the occurrence time and separate them into sequences suitable for inputting the learning model. At this time, the preprocessing unit (30) can separate the sequences by performing grouping, which groups fault codes that occurred at the same time into one group.

In the above, 'simultaneous time zone' does not mean exactly matching up to the hour/minute/second, but can be interpreted to mean that data that occurred in a relatively close time zone are grouped together. For example, data that occurred within a time range of several seconds or minutes can be grouped together, and the time range that serves as the basis for grouping is a design feature of this system that can be changed.

In the storage management unit (40), data preprocessed by the preprocessing unit (30) can be stored. Here, the 'preprocessed data' can be a set of fault codes sorted and grouped by time zone as described above.

In addition, the storage management unit (40) can store and manage the model results (diagnosis results) output when the learning data set labeled by the labeling unit (50) described below, the learning model generated by the learning modeling unit (60), and the preprocessed data of the failure data occurring in a specific elevator are input into the learning model.

As will be described later, the elevator fault diagnosis system according to the present invention can proceed with processes divided into a 'model learning process' and a 'fault diagnosis process', and the storage management unit (40) can manage learning data by the model learning process and input data by the fault diagnosis process separately, and can update and manage the learning model by periodically updating the data input during fault diagnosis and the results of the learning model accordingly.

The labeling unit (50) can read data preprocessed by the preprocessing unit (30) from the storage management unit (40) and combine it with a fault cause label to form a data set for model learning.

The term 'failure cause label' as mentioned above may mean a category indicating an actual cause of an elevator failure predicted when a specific pattern of preprocessed data, i.e. a specific pattern of a sequence of failure codes grouped by time zone, is observed, and the present invention defines this as a new 'cause class'.

The 'cause class' described in the present invention is a different concept from the 'fault code' defined previously. While the existing fault code indicates the possibility of a failure of an individual device constituting an elevator, the cause class is a new fault cause regulation that has not been defined previously and indicates in which part of the elevator a failure actually occurred when a continuous sequence of multiple fault codes is observed.

Referring to FIG. 2, the labeling operation between the fault code sequence and the cause class by the elevator fault diagnosis system according to the present invention will be examined in more detail.

Referring to Fig. 2, first, fault codes Fault A, Fault B, and Fault F that occurred in the time period of 11:02:09 to 11:02:10 can be grouped into one group, and fault codes Fault D, Fault E, and Fault B that occurred in the time period of 11:10:41 to 11:10:42 can be grouped into another group. This grouping task is performed by the preprocessing unit (30).

In this state, a labeling task of correspondingly associating a sequence of fault codes grouped into one group with a new cause class can be performed by the labeling unit (50). For example, the labeling unit (50) can correspond and associate a continuous sequence of fault codes occurring in the order of Fault A → Fault B → Fault F with the Cause A cause class, and can correspond and associate a continuous sequence of fault codes occurring in the order of Fault D → Fault E → Fault B with the Cause B cause class. Here, Cause A and Cause B each define an actual cause of a fault in the elevator.

At this time, the labeling unit (50) can perform combination (labeling) between preprocessed data and failure cause labels (cause classes) by referring to past failure history data stored in the cloud (20) or a separately provided failure server, and as the learning modeling unit (60) described below performs repetitive learning, the accumulated data can be used again as feed data.

As described above, the labeling unit (50) can be understood as having a function of mapping preprocessed data of failure data observed in a failure state of an elevator with the actual cause of the failure, and recurrent neural network learning can be performed with a learning data set configured through data labeling.

The learning modeling unit (60) can create a ‘fault diagnosis model’ by performing machine learning of a specific algorithm using the learning data set configured by the labeling unit (50).

The learning modeling unit (60) can use a 'recurrent neural network algorithm' that learns time information, which is a major feature of input data, and as fault data accumulates, it can periodically repeat additional learning and perform a task of updating the model by comparing it with an existing model.

As mentioned above, elevator operation data including failure data has the property of time series data that occurs continuously, so it is difficult to accurately identify the cause with a single data alone, and the cause of the failure can be accurately analyzed by comprehensively judging and interpreting other data that occurred before and after the failure.

Accordingly, the present invention selects a recurrent neural network (RNN) model as the most suitable learning model for analyzing time-series occurrence failure data and inferring the actual cause of failure. The recurrent neural network algorithm is a class of neural networks that models sequence data such as time series, and is a model mainly applied to data sets where there is a correlation between the preceding/following data.

In addition, the present invention can use a bidirectional LSTM (bidirectional Long Short-Term Memory), which is a type of recurrent neural network model, in the process of performing a learning model generation algorithm. Bidirectional LSTM is a complement to the existing LSTM, which has lower accuracy as the length increases due to the variability of the length of the separated sequences. While the existing LSTM can only process forward data, the bidirectional LSTM can be said to be a recurrent neural network algorithm that can process forward and reverse bidirectional data.

More specifically, the learning modeling unit (60) can generate a learning model that automatically derives a cause class indicating an actual cause of a failure when preprocessed data of newly occurring failure data is input through repeated learning of a learning data set consisting of a combination of a sequence of failure codes occurring at the same time and a corresponding cause class.

Accordingly, as described below, when a user who accesses the web server (70) through a user terminal (80) selects a specific elevator, the learning modeling unit (60) can receive preprocessed data of a failure data that occurred in the selected elevator as input data for the learning model and derive a result. The result derived by the learning modeling unit (60) can be stored in the storage management unit (40) as a failure diagnosis result.

The web server (70) can receive a specific elevator number desired by the user and provide the learning model result, which is a fault diagnosis result of the corresponding elevator, to the user. The web server (70) can be accessed through a portable user terminal (80), and can provide a web application including a service screen that allows the user to select a specific elevator and visually display the fault diagnosis result of the selected elevator. Here, the 'user' can be a maintenance worker who wants to receive a fault diagnosis service.

For reference, a fault code occurring in an elevator only indicates a possibility of a failure, and the occurrence of a fault code itself does not mean that the elevator has actually failed. Therefore, it was difficult to find the exact cause of a failure in an elevator using only a conventional fault code, but the present invention has the advantage of being able to diagnose the actual cause of a failure in an elevator with high accuracy by generating a fault diagnosis model that infers the actual cause of a failure based on a sequence of fault codes that occur continuously through recurrent neural network learning.

Meanwhile, the elevator fault diagnosis system utilizing the recurrent neural network model according to the present invention can be largely performed by including a 'model learning process' and a 'fault diagnosis process'. Here, the model learning process is a process related to the operation of generating a fault diagnosis model, and the fault diagnosis process can be understood as a process related to the operation of diagnosing the cause of a fault that actually occurred in an elevator by utilizing the generated learning model.

Below, the process operations of the elevator failure diagnosis system according to the present invention will be examined.

Model learning process

First, the 'model learning process' operation of the elevator failure diagnosis system according to the present invention can be performed through the following process.

1. A record including failure data recorded on the elevator control panel (10) is accumulated in the cloud (20).

2. The preprocessing unit (30) reads data from the cloud (20), preprocesses it into input data for the learning model, and stores it in the storage management unit (40).

3. The labeling unit (50) combines preprocessed data and fault cause labels (cause classes) to match the model input to create a learning data set and stores it again in the storage management unit (40).

4. The learning modeling unit (60) creates a fault diagnosis model by bringing in a labeled learning data set and learning a recurrent neural network model, and stores the model for which learning is complete in the storage management unit (40).

Fault Diagnosis Process

Next, the ‘fault diagnosis process’ operation of the elevator fault diagnosis system according to the present invention is performed through the following process.

1. A user (maintenance worker) accesses a web server (70) through a user terminal (80) and selects an elevator for which a fault analysis is desired.

2. The preprocessing unit (30) reads the failure data of the selected elevator from the cloud (20), preprocesses it into input data for the learning model, and stores it in the storage management unit (40).

3. The learning modeling unit (60) inputs the preprocessed data stored in the storage management unit (40) as input data for the learning model and stores the results derived from the learning model, i.e., the fault diagnosis results, in the storage management unit (40).

4. The web server (70) retrieves the fault diagnosis results from the storage management unit (40) and displays them on the service screen on the user terminal (80), thereby providing the user with information indicating the actual cause of the elevator fault.

As described above, the elevator fault diagnosis system according to the present invention generates a model that infers the actual cause of a fault from a sequence of fault codes that occur continuously through recurrent neural network learning, and when a maintenance worker inputs a desired fault analysis target elevator, the fault diagnosis results can be provided on-site to a user terminal (80) to support maintenance work.

According to the present invention, by applying a recurrent neural network model suitable for the characteristics of an elevator in which operation data is continuously generated, the advantages of deep learning can be utilized to derive accurate diagnosis results for faults occurring in an elevator and provide them to maintenance personnel, thereby reducing the deviation in time required for fault analysis and problem solving regardless of the skill level of the maintenance personnel.

In addition, the present invention can find the actual cause of an elevator failure by comprehensively analyzing a plurality of failure data that occur in time series rather than diagnosing a failure using only one failure data, and thus has the effect of significantly improving the accuracy of elevator failure analysis compared to conventional methods.

In particular, according to the present invention as described above, not only is it possible to reduce misanalysis of failures occurring in an elevator possible, but it also has the effect of expanding the scope of failure analysis by providing failures of unknown causes that cannot be explained by existing failure codes or failures related to elevator parts as new cause classes.

The elevator fault diagnosis system according to the present invention can be implemented by a computer device having one or more processors and a memory storing one or more programs executed by the one or more processors. To this end, the elevator fault diagnosis system according to the present invention can be implemented in the form of a program or software including one or more computer-executable commands and stored in the memory.

The present invention is not limited to the described embodiments, and it is obvious to those skilled in the art that various modifications and variations can be made without departing from the spirit and scope of the present invention. Accordingly, such modifications or variations should fall within the scope of the claims of the present invention.

10: Elevator control panel
20: Cloud
30: Preprocessing section
40: Storage Management Department
50: Labeling section
60: Learning Modeling Department
70: Web server
80: User terminal

Claims (8)

Elevator control panel that controls the operation of the elevator and records failure data related to failure in real time;
A cloud that receives and accumulates the failure data recorded on the elevator control panel;
A preprocessing unit that extracts key data to be used for analyzing the cause of a failure of the elevator from the failure data stored in the cloud and performs preprocessing to list them by time zone;
A labeling unit that combines the data preprocessed by the above preprocessing unit with a failure cause label indicating the actual cause of a failure of the elevator to form a learning data set;
A learning modeling unit that performs recurrent neural network learning with the above learning data set to generate a fault diagnosis model that infers the actual cause of the fault of the elevator; and
It includes a storage management unit that stores and manages separately the data preprocessed by the above preprocessing unit, the learning data set labeled by the above labeling unit, and the fault diagnosis model generated by the above learning modeling unit.
The above main data is the fault code included in the above fault data,
The above preprocessing unit lists the above fault codes by time zone and performs grouping to group those within a set similar time range into one group, thereby separating the above fault codes into a continuous sequence.
Elevator failure diagnosis system using recurrent neural network model.
delete In claim 1,
The above fault cause label is defined as a cause class that indicates the actual fault cause of the elevator that occurred when a continuous sequence of the above fault codes is observed in a specific pattern.
The labeling unit is characterized in that it performs a labeling task that combines the continuous sequence of the fault codes and the corresponding cause class by referring to past fault history data.
Elevator failure diagnosis system using recurrent neural network model.
In claim 3,
The above fault diagnosis model is characterized in that it receives preprocessing data of a specified fault data of the elevator as input data and outputs the corresponding cause class as a fault diagnosis result.
Elevator failure diagnosis system using recurrent neural network model.
In claim 4,
The above fault diagnosis model is characterized in that it is updated by performing repetitive learning using the data input during fault diagnosis of the specific elevator and the output result thereof as feed data.
Elevator failure diagnosis system using recurrent neural network model.
In claim 5,
Further comprising a web server that is accessible through a user terminal and provides the user with the results of fault diagnosis of the specific elevator selected by the user.
Elevator failure diagnosis system using recurrent neural network model.
In claim 6,
The above web server is characterized in that it provides a web application including a service screen that allows the user to select the specific elevator and visually display the fault diagnosis results.
Elevator failure diagnosis system using recurrent neural network model.
In claim 1,
The above recurrent neural network model is characterized by using bidirectional LSTM (bidirectional Long Short-Term Memory).
Elevator failure diagnosis system using recurrent neural network model.

Images