JP2021116160A - Learning model generation device, estimation device, learning model generation method, and estimation method - Google Patents

Abstract

To drastically reduce workload on solving a malfunction event generated in an elevator.SOLUTION: A learning model generation device (1) includes a learning model generation unit (103) for generating a learning model (54) outputting treatment information with an event code group showing malfunction events generated in a target elevator, operation information of the target elevator, and specification information of the target elevator as input by performing machine learning using an event code group showing malfunction events generated in elevators, treatment information showing treatment for solving the malfunction events, operation information showing operation states of elevators when the malfunction events occur, and specification information showing specifications of the elevators as teacher data.SELECTED DRAWING: Figure 4

Description

The present invention relates to an apparatus or the like that supports work for resolving a malfunction event that has occurred in an elevator.

In the past, workers working to resolve elevator failures have identified appropriate response actions based on empirical rules. Patent Document 1 below discloses a technique for presenting a cause candidate having a certain degree of certainty or more and a countermeasure for the cause by inputting information on a failure that has occurred in an elevator.

JP-A-2018-167957

Conventional techniques for identifying appropriate response measures based on empirical rules impose a heavy burden on workers. Further, in the technique described in Patent Document 1, a worker needs to go to an elevator in which a failure has occurred in order to input information regarding the failure. That is, the conventional technique is insufficient for significantly reducing the work load of the worker.

One aspect of the present invention is to realize a learning model generation device, an estimation device, and the like that significantly reduce the work load related to solving a malfunction event that has occurred in an elevator.

In order to solve the above problems, the learning model generator according to the first aspect of the present invention has a code group consisting of one or a plurality of event codes indicating a failure event that has occurred in the elevator, and the problem event. By performing machine learning using the treatment information indicating the treatment of the above, the operation information indicating the operation status of the elevator at the time of the occurrence of the malfunction event, and the specification information indicating the specifications of the elevator as teacher data, the target elevator can be used. It is provided with a learning unit that generates a learning model that outputs the treatment information by inputting the code group indicating the occurrence of a malfunction event, the operation information of the target elevator, and the specification information of the target elevator.

Conventionally, a worker who works to solve a malfunction event of an elevator refers to a code group indicating a malfunction event and takes an appropriate countermeasure based on an empirical rule that takes into consideration the operation information and the specification information. It was time-consuming because it was specified.

On the other hand, according to the above configuration, a learning model in which the code group, the treatment information, the operation information, and the specification information are machine-learned as teacher data is generated. Therefore, by using the learning model, it is possible to estimate a measure for resolving the malfunction event indicated by the code group based on the code group, the operation information, and the specification information. Therefore, the work load of the worker can be significantly reduced.

In the learning model generator according to the second aspect of the present invention, in the first aspect, the event code is associated with date and time information indicating the date and time when the event code was output, and the learning unit is associated with the code group. May be identified by data in which the event codes constituting the code group are arranged in chronological order based on the date and time information.

Even if the combination of event codes included in the code group is the same, if the output order of the event codes is different, the cause of the problem may be different. Different causes of failure usually take different actions. Therefore, it is desirable to treat a code group composed of event codes having the same combination but different output orders as separate code groups, not as the same code group.

According to the above configuration, the code group is identified by the data in which the event codes constituting the code group are arranged in chronological order based on the date and time information. Then, it is possible to generate a learning model in which a code group composed of event codes having the same combination but different output orders is treated as a separate code group. Therefore, the estimation accuracy by the learning model can be improved.

In the learning model generator according to the third aspect of the present invention, in the first or second aspect, the event code, the operation information, and the specification information are associated with identification information for identifying the elevator. An acquisition unit for acquiring the operation information and the specification information used as the teacher data may be provided based on the identification information associated with the event code constituting the code group used as the teacher data.

According to the above configuration, since the operation information and the specification information are automatically acquired based on the identification information associated with the event code, it is not necessary to obtain the operation information and the specification information to be used as the teacher data in advance. As a result, the workload in generating the learning model can be reduced.

In any one of the above aspects 1 to 3, the learning model generator according to the fourth aspect of the present invention may include the specification information indicating the parts constituting the elevator.

Depending on the type of parts that make up the elevator, the work to resolve the malfunction event may differ.

According to the above configuration, a learning model for estimating the treatment considering the parts constituting the elevator in which the failure event has occurred is generated. Therefore, by using the learning model, the worker can use the learning model even if it is not based on an empirical rule. It is possible to know the appropriate treatment according to the part.

In order to solve the above problems, the estimation device according to the fifth aspect of the present invention includes a code group consisting of one or a plurality of event codes indicating a malfunction event that has occurred in the elevator, and a measure for eliminating the malfunction event. A learning model generated by performing machine learning using the treatment information indicating the above, the operation information indicating the operating status of the elevator at the time of occurrence of the malfunction event, and the specification information indicating the specifications of the elevator as teacher data. It is provided with an estimation unit that estimates the treatment information by inputting the code group indicating a malfunction event that has occurred in the target elevator, the operation information of the target elevator, and the specification information of the target elevator.

Conventionally, a worker who works to solve a malfunction event of an elevator refers to a code group indicating a malfunction event and takes an appropriate countermeasure based on an empirical rule that takes into consideration the operation information and the specification information. It was time-consuming because it was specified.

On the other hand, according to the above configuration, the code group and the operation are performed by using a learning model in which the code group, the treatment information, the operation information, and the specification information are machine-learned as teacher data. Based on the information and the specification information, it is possible to estimate the measures for resolving the malfunction event indicated by the code group. Therefore, the work load of the worker can be significantly reduced.

In the estimation device according to the sixth aspect of the present invention, in the fifth aspect, the event code is associated with date and time information indicating the date and time when the event code was output, and in the generation of the learning model, and the above. The estimation unit may identify the code group by data in which the event codes constituting the code group are arranged in chronological order based on the date and time information.

According to the above configuration, the code group is identified by the data in which the event codes constituting the code group are arranged in chronological order based on the date and time information. Then, the treatment information can be estimated by the learning model that treats the code group composed of the event codes having the same combination but different output orders as separate code groups. Therefore, the estimation accuracy by the learning model can be improved.

In the estimation device according to the seventh aspect of the present invention, in the fifth or sixth aspect, the event code, the operation information, and the specification information are associated with identification information for identifying the elevator, and the object is described. The estimation unit includes an acquisition unit that acquires the operation information and the specification information based on the identification information associated with the event code that constitutes the code group that indicates a malfunction event that has occurred in the elevator. The acquired operation information and the specification information may be input.

According to the above configuration, the operation information and the specification information are automatically acquired based on the identification information associated with the event code, so that the worker needs to obtain the operation information and the specification information of the target elevator in advance. There is no. As a result, the work load of the worker can be reduced.

In any one of the above aspects 5 to 7, the estimation device according to the eighth aspect of the present invention may include the specification information indicating the parts constituting the elevator.

Depending on the type of parts that make up the elevator, the work to resolve the malfunction event may differ.

According to the above configuration, since a learning model for estimating the treatment considering the parts constituting the elevator in which the failure event has occurred is used, the worker knows the appropriate treatment according to the parts without being based on the empirical rule. be able to.

In order to solve the above problems, the learning model generation method executed by the learning model generation device according to the ninth aspect of the present invention includes a code group consisting of one or a plurality of event codes indicating a malfunction event that has occurred in the elevator. , Machine learning using the treatment information indicating the measures for resolving the malfunction event, the operation information indicating the operating status of the elevator at the time of the occurrence of the malfunction event, and the specification information indicating the specifications of the elevator as teacher data. By performing Includes learning steps to do.

According to the above configuration, the same operation and effect as that of the learning model generator according to the first aspect is obtained.

In order to solve the above problems, the estimation method executed by the estimation device according to the tenth aspect of the present invention includes a code group consisting of one or a plurality of event codes indicating a failure event that has occurred in the elevator, and the failure event. By performing machine learning using the treatment information indicating the measures for solving the problem, the operation information indicating the operating status of the elevator at the time of the occurrence of the malfunction event, and the specification information indicating the specifications of the elevator as teacher data. Using the generated learning model, the treatment information is estimated by inputting the code group indicating the failure event that occurred in the target elevator, the operation information of the target elevator, and the specification information of the target elevator. Includes an estimation step to do.

According to the above configuration, the same operation and effect as that of the estimation device according to the fifth aspect is obtained.

According to one aspect of the present invention, it is possible to significantly reduce the work load related to the elimination of the malfunction event that has occurred in the elevator.

It is a figure which shows the outline about an example of the learning model generation which concerns on one Embodiment of this invention. It is a figure which shows the outline about an example of the treatment information estimation which concerns on one Embodiment of this invention. It is a figure which shows a specific example of the set of the original data used for the learning model generation shown in FIG. It is a block diagram which shows an example of the main part structure of the treatment information estimation system which concerns on one Embodiment of this invention. It is a figure which shows one specific example of the specification information group and the operation information group stored in the storage device shown in FIG. The estimation result may be different when the input data contains event code groups that consist of the same event code combination but are arranged in chronological order of the event code occurrence date and time in the combination. It is a figure which shows that there is. It is a flowchart which shows an example of the flow of the learning model generation processing executed by the control part of the learning model generation apparatus shown in FIG. It is a flowchart which shows an example of the flow of the estimation process executed by the control part of the estimation apparatus shown in FIG. It is a figure which shows the screen example which concerns on the modification of one Embodiment of this invention.

In the present embodiment, a learning method and an estimation method of a procedure for resolving a malfunction event that has occurred in an elevator using machine learning will be described. In addition, in this specification, "treatment" shall mean the content of repair work (repair, replacement, etc. of parts) by a repair company or the like.

(Generation of learning model)
FIG. 1 is a diagram showing an outline of an example of learning model generation according to the present embodiment. The learning model generator 1 shown in FIG. 1 performs machine learning using a set of data sets including event code group D1 (code group), treatment information D2, specification information D3, and operation information D4 as teacher data D5. A training model 54 is generated.

The event code group D1 is composed of one or a plurality of event codes indicating a malfunction event that has occurred in the elevator. The action information D2 indicates a action for resolving the malfunction event that has occurred in the elevator. Specification information D3 indicates the specifications of the elevator. The operation information D4 indicates the operation status of the elevator at the time when the failure event occurs.

As shown in FIG. 1, the learning model generation device 1 first acquires a set of previously generated data (hereinafter, referred to as “original data”) in which the event code group D1 and the treatment information D2 are associated with each other. (A1 in FIG. 1). The original data is one of the original data of the teacher data D5. Details of the original data will be described later.

Next, the learning model generator 1 acquires the specification information D3 and the operation information D4 corresponding to the original data for each of the acquired original data (a2 in FIG. 1). Then, the learning model generation device 1 generates the teacher data D5 from the event code group D1, the treatment information D2, the specification information D3, and the operation information D4. Then, the learning model generation device 1 generates the learning model 54 by performing machine learning using the generated teacher data D5. The learning model 54 is a learning model that outputs treatment information by inputting event code group, specification information, and operation information.

(Estimation of treatment information)
FIG. 2 is a diagram showing an outline of an example of treatment information estimation according to the present embodiment. The estimation device 2 shown in FIG. 2 uses the learning model 54 to estimate the treatment information D12 by inputting the event code group D11, the specification information D13, and the operation information D14.

As shown in FIG. 2, the estimation device 2 first acquires the event code group D11 (a11 in FIG. 2). The event code group D11 is composed of one or a plurality of event codes indicating the malfunction event in the elevator in which the malfunction event has occurred (hereinafter, may be referred to as "target elevator"). Next, the estimation device 2 acquires the specification information D13 and the operation information D14 of the target elevator (a12 in FIG. 2). The specification information D13 indicates the specifications of the target elevator. The operation information D14 indicates the operation status of the target elevator at the time when the failure event occurs. Then, the estimation device 2 generates input data D40 from the event code group D11, the specification information D13, and the operation information D14, and inputs the input data D40 to the learning model 54 (a13 in FIG. 2). The learning model 54 to which the input data D40 is input outputs the treatment information D12 (a14 in FIG. 2). The treatment information D12 is an estimation result of the treatment for resolving the malfunction event that has occurred in the target elevator.

<Generation of original data>

FIG. 3 is a diagram showing a specific example of the original data and the defect occurrence data and the defect countermeasure data which are materials for generating the original data. The defect occurrence data is data related to a defect event that has occurred in the elevator. The defect handling data is data related to countermeasures for resolving the defect event. Hereinafter, the set of defect occurrence data will be referred to as "fault occurrence data group 61", the set of defect treatment data will be referred to as "fault treatment data group 62", and the set of original data will be referred to as "original data group 51". Describe. In FIG. 3, the defect occurrence data group 61, the defect treatment data group 62, and the original data group 51 are shown in a table format. Each of the records of the defect occurrence data group 61 is the defect occurrence data, each of the records of the defect countermeasure data group 62 is the defect countermeasure data, and each of the records of the original data group 51 is the original data.

(Details of defect occurrence data)
As an example, the defect occurrence data includes defect identification information, elevator identification information, event code, and event code occurrence date / time (date / time information), as shown in FIG. The defect occurrence data group 61 is stored in a predetermined storage device as an example, and new defect occurrence data is stored every time a defect event occurs. The predetermined storage device may be a storage device 5 described later.

The defect identification information is information for identifying a defect event that has occurred in the elevator. The elevator identification information is information for identifying an elevator in which a malfunction event has occurred. As described above, the event code indicates a malfunction event that has occurred in the elevator. The event code group is one or more event codes associated with one defect identification information. In the example of FIG. 3, the event codes E001, E002, E003, and E000 associated with the defect identification information A001 are one event code group. Further, E011, E012, and E000 associated with the defect identification information A002 are one event code group. Since the event code E000 is an event code indicating an operation in which the control panel of the elevator outputs the event code group to the outside, one event code E000 is included for each event code group. In short, the operation is an operation of notifying the occurrence of a malfunction event to the outside. Hereinafter, the event code E000 will be referred to as a "report event code". The "external" may be, for example, a system for remotely monitoring an elevator, or an information processing device of a trader who repairs, maintains, or inspects the elevator. The event occurrence date and time is information indicating the date and time when the event code was generated.

(Details of defect handling data)
As an example, the defect handling data includes defect identification information, elevator identification information, cause detailed information, and treatment detailed information, as shown in FIG. The defect handling data group 62 may be stored in the same storage device as the defect occurrence data group 61, or may be stored in a different storage device. Every time a measure for resolving a trouble event is taken, new trouble-shooting data is added to the trouble-shooting data group 62. Since the defect identification information and the elevator identification information have already been described, the description will not be repeated here.
The detailed cause information is information indicating the cause of the malfunction event. The detailed treatment information is information indicating a treatment in which the malfunction event is resolved. As an example, this information is input by the worker who solved the malfunction event.

(Details of original data)
As an example, the original data included in the original data group 51 includes defect identification information, elevator identification information, event code, event occurrence date / time, cause code, and action code, as shown in FIG. As shown in FIG. 3, the set of event codes associated with one defect identification information corresponds to the event code group D1, and the set of the cause code and the action code associated with one defect identification information is the action. Corresponds to information D2.
Since the defect identification information, the elevator identification information, the event code, and the event occurrence date and time have already been described, the description will not be repeated here.

The cause code is a code for classifying the detailed cause information included in the defect handling data. Specific cause codes include, for example, "C0: No entry / unknown", "C1: Object detection door sensor abnormality near door pocket", "C2: String object detection door sensor abnormality", "C3: External factor near door", There are "C4: Brake" and "C5: Other".

The action code is a code for classifying the detailed action information included in the defect action data. Specific action codes include, for example, "E0: No entry / unknown", "E1: Replacement of abnormal parts", "E2: Repair of abnormal parts", "E3: Removal of cause of abnormality", "E4: Others". There is. Here, the "abnormal cause" refers to an external factor that causes an abnormality in a part. That is, "removal of the cause of abnormality" is a measure for removing the external factor.

As an example, the original data is generated by the original data generation device 6 described later based on the defect occurrence data and the defect treatment data, and is stored in the storage device. In the present embodiment, the storage device is the storage device 5 described later.
The original data generation device 6 reads the defect occurrence data from the storage device based on the user's operation. Further, the original data generation device 6 reads out from the storage device the defect handling data in which the read defect occurrence data and the defect identification information match, based on the user's operation.

Then, the user performs an operation of designating the cause code corresponding to the cause detailed information included in the read trouble-shooting data. As a result, the original data generation device 6 identifies the cause code corresponding to the detailed cause information. In addition, the user performs an operation of designating the corresponding action code for the action detailed information included in the read defect action data. As a result, the original data generation device 6 identifies the treatment code corresponding to the detailed treatment information.

Then, the original data generation device 6 generates the original data by associating the defect identification information, the elevator identification information, the event code, the event occurrence date and time, and the specified cause code and action code included in the read defect occurrence data. do. The original data generation device 6 adds the generated original data to the original data group 51 stored in the storage device 5.

<Structure of the main part of the treatment information estimation system>
FIG. 4 is a block diagram showing an example of a main part configuration of the treatment information estimation system 100 according to the present embodiment. As shown in FIG. 4, the treatment information estimation system 100 includes a learning model generation device 1, an estimation device 2, an input device 3, an output device 4, a storage device 5, and an original data generation device 6. The learning model generation device 1, the estimation device 2, or the input device 3 may also have a function as the original data generation device 6 to generate the original data. In this case, the original data generation device 6 is unnecessary.

The input device 3 receives an input from the user and outputs an input signal based on the input to the learning model generation device 1 or the estimation device 2.

The output device 4 outputs the information generated by the estimation device 2. The output mode by the output device 4 is not particularly limited. The output device 4 may be, for example, a display device that displays the information as an image, or a printing device that prints the information.

The storage device 5 stores programs and data used in the treatment information estimation system 100. As an example, the storage device 5 stores the original data group 51, the specification information group 52, the operation information group 53, and the learning model 54. Note that these data may be stored in separate storage devices. For example, the treatment information estimation system 100 may include four storage devices that individually store each of the original data group 51, the specification information group 52, the operation information group 53, and the learning model 54. Further, the storage device 5 may store the defect occurrence data group 61 and the defect treatment data group 62.

(Main part configuration of learning model generator 1)
As shown in FIG. 4, the learning model generation device 1 includes a control unit 10. The control unit 10 controls each unit of the learning model generation device 1 in an integrated manner, and is realized by a processor and a memory as an example. In this example, the processor accesses the storage (not shown), loads the program (not shown) stored in the storage into the memory, and executes a series of instructions included in the program. As a result, each unit of the control unit 10 is configured.

As each of the units, the control unit 10 includes a source data acquisition unit 101, a teacher data generation unit 102 (acquisition unit), and a learning model generation unit 103 (learning unit).

The original data acquisition unit 101 acquires the original data. Specifically, the original data acquisition unit 101 reads out the original data group 51 from the storage device 5 based on the input signal (hereinafter, referred to as the learning start signal) indicating the learning start instruction from the input device 3. The original data acquisition unit 101 outputs the read original data group 51 to the teacher data generation unit 102.

The teacher data generation unit 102 generates the teacher data D5. Specifically, first, when the teacher data generation unit 102 acquires the original data group 51 from the original data acquisition unit 101, for each of the original data, the specification information group 52 starts from the specification information group 52 based on the elevator identification information included in the original data. , Read the specification information. Further, the teacher data generation unit 102 reads out the operation information from the operation information group 53 based on the elevator identification information included in the original data for each of the original data. The specification information group 52 is a set of elevator specification information. The operation information group 53 is a set of elevator operation information.

Then, the teacher data generation unit 102 generates the teacher data D5 by associating the event code group D1 and the treatment information D2, the specification information D3, and the operation information D4 included in the original data with the elevator identification information as a key. The processing is performed for each of the original data included in the original data group 51 to generate a set of teacher data D5. Then, the teacher data generation unit 102 outputs the set of the generated teacher data D5 to the learning model generation unit 103.

(Details of specification information group 52 and operation information group 53)
FIG. 5 is a diagram showing a specific example of the specification information group 52 and the operation information group 53. In FIG. 5, the specification information group 52 and the operation information group 53 are shown in a table format, but the formats of the specification information group 52 and the operation information group 53 are not limited to this example.

As an example, the specification information included in the specification information group 52 includes elevator identification information, specifications, and a part code, as shown in FIG. Since the elevator identification information has already been described, the description will not be repeated here.

The specifications are information indicating specifications such as building use, capacity, speed, etc. of the elevator indicated by the elevator identification information. The information included in the specifications is not limited to this information. The building use is information indicating the use of the building in which the elevator is installed. Building uses include, but are not limited to, apartment buildings (condominiums, apartments, etc.) and hotels, as well as hospitals, offices, etc., as shown in FIG. The capacity is information indicating the capacity of the elevator. The speed is information indicating the moving speed of the elevator, and in the example of FIG. 5, the unit is "m / min".

The part code is the identification information of the parts used in the elevator indicated by the elevator identification information.

As an example, the operation information included in the operation information group 53 includes elevator identification information, operation value classification, acquisition date and time, and acquisition value, as shown in FIG. Since the elevator identification information has already been described, the description will not be repeated here.

The operating value classification is information for classifying the measured values related to the operation of the elevator indicated by the elevator identification information. The operating value classification includes, as an example, the number of activations, the mileage, and the number of times the car door is opened and closed, but is not limited to this example. The number of activations is the total number of movements of the elevator car from the time the elevator is installed to the date and time of acquisition. Here, the movement of the elevator car means "from the start of the movement to the stop of the movement". For example, if the elevator departs from the 1st floor, stops once on the 3rd floor, resumes movement, and stops on the 6th floor, the number of movements in this series of flows is 2 times (from the 1st floor to the 3rd floor and the 3rd floor). 6th floor). The mileage is the total distance traveled by the elevator car from the time the elevator is installed to the date and time of acquisition. The number of times the car door opens and closes is the total number of times the elevator car door opens and closes between the time the elevator is installed and the date and time of acquisition.

The acquired value is a measured value related to the operation of the elevator. The acquisition date and time is information indicating the date and time when the acquisition value was acquired. The acquired value included in the operation information group 53 is acquired from the control panel at the time of periodic inspection, for example.

With reference to FIG. 4 again, the description of the main part configuration of the learning model generator 1 will be returned.

The learning model generation unit 103 generates a learning model 54 by performing machine learning using a set of teacher data D5 generated by the teacher data generation unit 102. As an example, the learning model generation unit 103 generates a learning model 54 by using a known machine learning method such as a neural network (NN). Further, the learning model generation unit 103 may generate the learning model 54 by using the existing machine learning service. The learning model generation unit 103 stores the generated learning model 54 in the storage device 5.

The learning model generation unit 103 arranges the event codes constituting the event code group D1 in chronological order based on the event occurrence date and time for each of the event code groups D1 of the teacher data D5 (hereinafter, "time series event code group"). ”). That is, the learning model generation unit 103 performs machine learning on the event code group D1 of the teacher data D5 in consideration of the event occurrence date and time, and generates the learning model 54. Specifically, the learning model generation unit 103 is an event code group composed of the same combination of event codes, and the event code groups in which the order of the event codes in the combination is arranged in chronological order are different. It is identified as a group and a learning model 54 is generated. As a result, a learning model 54 is generated that identifies the event code group consisting of the same event code but having a different order when the event codes are arranged in chronological order as different event code groups.

(Main part configuration of estimation device 2)
The estimation device 2 includes a control unit 20 as shown in FIG. The control unit 20 controls each unit of the estimation device 2 in an integrated manner, and is realized by a processor and a memory as an example. In this example, the processor accesses the storage (not shown), loads the program (not shown) stored in the storage into the memory, and executes a series of instructions included in the program. As a result, each unit of the control unit 20 is configured.

As the respective units, the control unit 20 includes an input data generation unit 201 (acquisition unit) and an estimation unit 202.

The input data generation unit 201 first acquires the event code group input from the input device 3. Specifically, the event code group D11 indicating the failure event that has occurred in the target elevator, which is input by the user of the treatment information estimation system 100 using the input device 3, is acquired. In the present embodiment, it is assumed that each event code included in the input event code group D11 is associated with the event occurrence date and time.

When the input data generation unit 201 acquires the event code group D11, the input data generation unit 201 reads the specification information D13 and the operation information D14 of the target elevator from the specification information group 52 and the operation information group 53, respectively. Specifically, the input data generation unit 201 reads out the specification information D13 and the operation information D14 including the same elevator identification information as the elevator identification information included in the acquired event code group D11. Then, the input data generation unit 201 generates the input data D40 by associating the acquired event code group D11 and the read specification information D13 and operation information D14 with the elevator identification information as a key. The input data generation unit 201 outputs the generated input data D40 to the estimation unit 202.

The estimation unit 202 acquires the treatment information D12 using the learning model 54. Specifically, the estimation unit 202 inputs the event code group D11, the specification information D13, and the operation information D14 of the target elevator into the learning model 54, and acquires the treatment information D12 output from the learning model 54. Then, the estimation unit 202 causes the output device 4 to output the acquired treatment information D12 as an estimation result.

The estimation unit 202 identifies the event code group D11 included in the input data D40 as a time-series event code group. That is, the estimation unit 202 identifies the event codes constituting the event code group D11 by the data arranged in chronological order based on the event occurrence date and time.

FIG. 6 shows an estimation result when the input data includes event code groups consisting of the same combination of event codes and having different orders when arranged in chronological order of the occurrence date and time of the event codes in the combination. It is a figure which shows that may be different.

As shown in FIG. 6, the input data D40a includes a time-series event code group D11a in which the time order of the occurrence date and time is the order of event codes E001, E002, and E000. Further, the input data D40b includes a time-series event code group D11b in which the time order of the occurrence date and time is the order of event codes E002, E001, and E000. It is assumed that the specification information D13 and the operation information D14 are common to the input data D40a and D40b. In other words, the input data D40a and D40b differ only in the order in which the event codes included in the event code group are generated.

As described above, the learning model 54 shown in FIG. 6 is a model trained in consideration of the time-series event code group. When the input data D40a is input to the learning model 54, the estimation unit 202 shall acquire the treatment information D12a. On the other hand, when the input data D40b is input to the learning model 54, the estimation unit 202 may acquire the treatment information D12b different from the treatment information D12a. In this way, in order to consider the time-series event code group, even if the event code group consists of the same combination of event codes, if the order when the event codes in the combination are arranged in chronological order is different, the input data D40 Even if the other data in the data are the same, different treatment information D12 may be obtained as an estimation result.

<Flow of learning model generation process>
FIG. 7 is a flowchart showing an example of the flow of the learning model generation process executed by the control unit 10 of the learning model generation device 1.

When the original data acquisition unit 101 acquires the learning start signal from the input device 3 (step S1, “hereinafter, the description of the step” is omitted), the original data group 51 is read from the storage device 5 (S2). The original data acquisition unit 101 outputs the read original data group 51 to the teacher data generation unit 102.

The teacher data generation unit 102 reads the specification information from the specification information group 52 and the operation information from the operation information group 53 based on the elevator identification information included in the original data for each of the acquired original data (S3). Specifically, the teacher data generation unit 102 reads out the specification information D3 and the operation information D4 including the same elevator identification information as the elevator identification information included in the original data for each of the original data. Subsequently, the teacher data generation unit 102 generates a set of teacher data D5 by associating the specification information D3 and the operation information D4 with which the elevator identification information matches with each of the original data (S4). The teacher data generation unit 102 outputs the set of the generated teacher data D5 to the learning model generation unit 103.

The learning model generation unit 103 generates a learning model 54 from the set of acquired teacher data D5 (S5, learning step). Specifically, the learning model generation unit 103 generates the learning model 54 by performing machine learning using the set of teacher data D5. The learning model generation unit 103 stores the generated learning model 54 in the storage device 5 (S6). This completes the learning model generation process.

(Update of learning model 54)
The defect occurrence data and the defect countermeasure data are stored in the storage device each time the worker takes action on the defect event of the target elevator. The user of the original data generation device 6 generates new original data using the original data generation device 6 based on the new defect occurrence data and the defect countermeasure data, and stores the new original data in the original data group 51. That is, the user updates the original data group 51. At this time, the user may regenerate the learning model 54 by using the learning model generation device 1 based on the updated original data group 51. Since the flow of the process of regenerating the learning model 54 is the same as the flow of the process of generating the learning model described above, the description is not repeated here.

<Flow of estimation processing>
FIG. 8 is a flowchart showing an example of the flow of the estimation process executed by the control unit 20 of the estimation device 2.

The input data generation unit 201 acquires an event code group D11 indicating a malfunction event that has occurred in the target elevator from the input device 3 (S11). Subsequently, the input data generation unit 201 reads the specification information D13 from the specification information group 52 and the operation information D14 from the operation information group 53 based on the elevator identification information of the acquired event code group D11 (S12). Specifically, the input data generation unit 201 reads out the specification information D13 and the operation information D14 including the same elevator identification information as the elevator identification information included in the event code group D11. Then, the input data generation unit 201 associates the event code group D11, the specification information D13, and the operation information D14, and generates the input data D40 (S13). The input data generation unit 201 outputs the generated input data D40 to the estimation unit 202.

The estimation unit 202 inputs the acquired input data D40 to the learning model 54, and acquires the output treatment information D12, that is, a set of the cause code and the treatment code (S14, estimation step). Then, the estimation unit 202 outputs the acquired treatment information D12 as an estimation result (S15). The output information preferably includes the cause indicated by the cause code and the action indicated by the action code. For example, when "C1" is acquired as the cause code and "E1" is acquired as the treatment code, the estimation unit 202 displays "abnormality of the object detection door sensor near the door pocket" and "replacement of the object detection door sensor near the door pocket" on the screen. It is preferable to output. As a result, the user or worker of the treatment information estimation system 100 (estimation device 2) can clearly grasp the cause of the failure of the target elevator and the estimation result of the treatment.

<Effect>
As described above, the learning model generation device 1 according to the present embodiment performs machine learning using the event code group D1, the treatment information D2, the specification information D3, and the operation information D4 as the teacher data D5. As a result, the learning model generator 1 inputs the event code group indicating the malfunction event that has occurred in the target elevator, the operation information of the target elevator, and the specification information of the target elevator, and outputs the treatment information for the malfunction event. Generate a learning model 54 to be used.

Further, the estimation device 2 according to the present embodiment uses the learning model 54 to obtain an event code group D11 indicating a malfunction event that has occurred in the target elevator, specification information D13 of the target elevator, and operation information D14 of the target elevator. Is used as an input to estimate the treatment information D12 for the malfunction event.

In the past, workers who work to resolve elevator malfunctions have referred to the event code group and specified appropriate measures based on empirical rules that take into account operation information and specification information, which is troublesome. Was hanging.

On the other hand, the learning model generation device 1 generates a learning model 54 in which the event code group D1, the treatment information D2, the specification information D3, and the operation information D4 are machine-learned as the teacher data D5. Therefore, the estimation device 2 uses the learning model 54 to estimate measures for resolving the malfunction event indicated by the event code group based on the event code group D11, the specification information D13, and the operation information D14 of the target elevator. be able to. Therefore, the work load of the worker can be significantly reduced.

Further, each event code input to the learning model generation device 1 and the estimation device 2 is associated with an event occurrence date and time. The learning model generation device 1 and the estimation device 2 identify the input event code group as a time-series event code group.

Even if the combination of event codes included in the event code group is the same, if the output order of the event codes is different, the cause of the malfunction event may be different. Different causes of failure events usually take different actions. Therefore, it is desirable that event code groups composed of event codes having the same combination but different output orders are treated as separate event code groups rather than being treated as the same event code group.

On the other hand, the learning model generation device 1 and the estimation device 2 identify the input event code group as a time-series event code group. Then, the learning model generation device 1 can generate a learning model 54 that handles an event code group composed of event codes having the same combination but different output orders as separate event code groups. Further, the estimation device 2 inputs an event code group composed of event codes having the same combination but different output orders to the learning model 54 as separate event code groups. As described above, the estimation accuracy by the learning model 54 can be improved.

Further, in the original data input to the learning model generation device 1, the event code group D1 is associated with the elevator identification information. Further, elevator identification information is associated with each of the specification information D3 and the operation information D4. The learning model generation device 1 reads the specification information D3 and the operation information D4 used as the teacher data D5 from the storage device 5 based on the elevator identification information associated with the event code group D1 used as the teacher data D5.

Further, the elevator identification information is also associated with the event code group D11 input to the estimation device 2. The estimation device 2 reads the specification information D13 and the operation information D14 constituting the input data D40 together with the event code group D11 from the storage device 5 based on the elevator identification information associated with the event code group D11.

That is, the learning model generation device 1 and the estimation device 2 automatically acquire the specification information and the operation information based on the elevator identification information associated with the input event code group. As a result, the user of the learning model generator 1 does not need to obtain the specification information D3 and the operation information D4 to be used as the teacher data D5 in advance. Further, the user of the estimation device 2 does not need to obtain the specification information D13 and the operation information D14 used as the input data D40 in advance. As a result, the workload in generating the learning model can be reduced.

Further, the specification information includes information indicating the parts constituting the elevator indicated by the specification information.

Depending on the type of parts that make up the elevator, the work to resolve the malfunction event may differ. On the other hand, the learning model generation device 1 generates a learning model 54 that estimates a treatment in consideration of the components constituting the target elevator. By inputting the input data D40 including the specification information D13 of the target elevator into the learning model 54, the estimation device 2 outputs the treatment information considering the parts constituting the target elevator. As a result, the worker can know the appropriate treatment according to the part without being based on the rule of thumb.

For example, in the case of an elevator that does not include the door pocket object detection door sensor, the treatment information D12 that is output as the estimation result does not output the treatment information related to the door pocket object detection door sensor. As a result, even in the same event code group D11 (more specifically, the same time-series event code group), the output treatment information D12 differs depending on the information indicating the parts constituting the elevator. .. Therefore, the estimation device 2 can output more appropriate treatment information D12 in the elevator to be treated.

[Modification example]
The estimation device 2 may identify the event code group by the data arranged in the input order of the event codes input by the user of the estimation device 2.

FIG. 9 is a diagram showing a screen example according to this modified example. Specifically, FIG. 9 is a diagram showing a specific example of an input screen of the event code group D11 and an output screen of the treatment information D12. Note that FIG. 9 shows an example in which the input screen and the output screen are configured as one screen. In other words, FIG. 9 shows an example of an input / output screen G1 including an input area G2 corresponding to the input screen and an output area G3 corresponding to the output screen. 3 and the output device 4 are configured as one information processing device.

The input area G2 includes an event code group input field G4, an elevator identification information input field G5, and an estimation execution button G6. As an example, the event code group input field G4 is composed of a plurality of cells (records) arranged in the vertical direction of the screen, and the user of the treatment information estimation system 100 inputs one event code in one cell. At this time, as an example, the user inputs the event code from the top cell of the event code group input field G4 in the order of earliest event occurrence date and time. As a result, the estimation device 2 identifies the event code group D11 as a time-series event code group. The elevator identification information input field G5 accepts input of elevator identification information. The estimation execution button G6 accepts an operation of inputting the event code group D11 to the estimation device 2. In other words, the input device 3 estimates the elevator identification information input in the event code group D11 and the elevator identification information input field G5 input in the event code group input field G4 based on the operation on the estimation execution button G6. Output (transmit) to device 2.

The output area G3 includes a treatment information output column G7. In the "cause" column of the action information output column G7, the estimation result of the cause of the malfunction event indicated by the event code group D11 input in the event code group input field G4 is displayed. In the column of "action content" of the action information output column G7, the estimation result of the action for the malfunction event is displayed. Further, in the "accuracy" column of the action information output column G7, the accuracy of the combination of the cause and the action, that is, the numerical value indicating the cause of the failure event indicated by the input data D40 and the certainty as the action is displayed.

For example, the worker may verify the combination of the cause and the treatment shown in the treatment information output column G7 for the malfunction event that occurred in the target elevator in descending order of accuracy. In other words, it is possible to avoid verifying the combination with low accuracy first, so that the worker can significantly reduce the work load.

In the event code group input field G4, the event code may be input from the top cell in the order of the latest event occurrence date and time. In this example, when the target elevator is the target of the remote monitoring service, the event code entered in the top cell is the notification event code. When the trouble report data stores the event code group in the order of the event occurrence date and time, the user of the estimation device 2 may input the event code in the reverse order from the report event code of the event code group D11 to be input. Therefore, the burden of inputting the event code group D11 by the user of the estimation device 2 is reduced.

Further, the estimation device 2 displays three combinations of causes and treatments in the treatment information output column G7 shown in FIG. 9 in descending order of accuracy, but the number of display of the combinations is not limited to this example. .. Further, the estimation device 2 may display only the combination of the cause and the treatment whose accuracy is equal to or higher than the predetermined value in the treatment information output column G7.

The operation information and the specification information may be input by the user of the treatment information estimation system 100 together with the original data or the event code group D11 via the input device 3. Further, the information indicating the parts used in each elevator does not have to be integrated with the specification information and may be separate data.

The teacher data D5 and the input data D40 may include information on the temperature, humidity, season, date, etc. at the time of the failure event, and information on the installation location (region, country, etc.) of the elevator in which the failure event occurred. ..

Further, the learning model generation device 1 and the estimation device 2 may be configured as an integrated device. Further, the estimation device 2 may receive the learning model 54 generated by the learning model generation device 1 from the learning model generation device 1 and store it in the storage of its own device.

[Example of realization by software]
The control blocks (particularly the control unit 10 and the control unit 20) of the learning model generation device 1 and the estimation device 2 may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or software. It may be realized by.

In the latter case, the learning model generation device 1 and the estimation device 2 include a computer that executes a program instruction, which is software that realizes each function. This computer includes, for example, at least one processor (control device) and at least one computer-readable recording medium that stores the program. Then, in the computer, the processor reads the program from the recording medium and executes it, thereby achieving the object of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. As the recording medium, a "non-temporary tangible medium", for example, a ROM (Read Only Memory) or the like, a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. Further, a RAM (Random Access Memory) for expanding the above program may be further provided. Further, the program may be supplied to the computer via an arbitrary transmission medium (communication network, broadcast wave, etc.) capable of transmitting the program. It should be noted that one aspect of the present invention can also be realized in the form of a data signal embedded in a carrier wave, in which the above program is embodied by electronic transmission.

1 Learning model generator 2 Estimator 54 Learning model 102 Teacher data generation unit (acquisition unit)
103 Learning model generation unit (learning unit)
201 Input data generation unit (acquisition unit)
202 Estimator D1, D11 Event code group (code group)
D2, D12 Treatment information D3, D13 Operation information D4, D14 Specification information

Claims (10)

A code group consisting of one or more event codes indicating a malfunction event that has occurred in an elevator, treatment information indicating a measure for resolving the malfunction event, and an operating status of the elevator at the time of the occurrence of the malfunction event. By performing machine learning using the indicated operation information and the specification information indicating the specifications of the elevator as teacher data, the code group indicating a malfunction event that has occurred in the target elevator, the operation information of the target elevator, and the target A learning model generation device including a learning unit that generates a learning model that outputs the treatment information by inputting the specification information of the elevator. The event code is associated with date and time information indicating the date and time when the event code was output.
The learning model generation device according to claim 1, wherein the learning unit identifies the code group by data in which the event codes constituting the code group are arranged in chronological order based on the date and time information. The event code, the operation information, and the specification information are associated with identification information for identifying the elevator.
1. A claim 1 comprising an acquisition unit that acquires the operation information and the specification information used as the teacher data based on the identification information associated with the event code constituting the code group used as the teacher data. Alternatively, the learning model generator according to 2. The learning model generator according to any one of claims 1 to 3, wherein the specification information includes information indicating a component constituting the elevator. A code group consisting of one or more event codes indicating a malfunction event that has occurred in an elevator, treatment information indicating a measure for resolving the malfunction event, and an operating status of the elevator at the time of the occurrence of the malfunction event. Using a learning model generated by performing machine learning using the operation information shown and the specification information indicating the specifications of the elevator as teacher data, the code group indicating a malfunction event that has occurred in the target elevator and the target elevator An estimation device including an estimation unit that estimates the treatment information by inputting the operation information of the above and the specification information of the target elevator. The event code is associated with date and time information indicating the date and time when the event code was output.
The estimation according to claim 5, wherein in the generation of the learning model, and the estimation unit identifies the code group by data in which the event codes constituting the code group are arranged in chronological order based on the date and time information. Device. The event code, the operation information, and the specification information are associated with identification information for identifying the elevator.
It is provided with an acquisition unit that acquires the operation information and the specification information based on the identification information associated with the event code that constitutes the code group that indicates a malfunction event that has occurred in the target elevator.
The estimation device according to claim 5 or 6, wherein the estimation unit inputs the acquired operation information and the specification information. The estimation device according to any one of claims 5 to 7, wherein the specification information includes information indicating a component constituting the elevator. A code group consisting of one or more event codes indicating a malfunction event that has occurred in an elevator, treatment information indicating a measure for resolving the malfunction event, and an operating status of the elevator at the time of the occurrence of the malfunction event. By performing machine learning using the indicated operation information and the specification information indicating the specifications of the elevator as teacher data, the code group indicating a malfunction event that has occurred in the target elevator, the operation information of the target elevator, and the target A learning model generation method executed by a learning model generator, which includes a learning step of generating a learning model that outputs the treatment information by inputting the specification information of the elevator. A code group consisting of one or more event codes indicating a malfunction event that has occurred in an elevator, treatment information indicating a measure for resolving the malfunction event, and an operating status of the elevator at the time of the occurrence of the malfunction event. Using a learning model generated by performing machine learning using the operation information shown and the specification information indicating the specifications of the elevator as teacher data, the code group indicating a malfunction event that occurred in the target elevator and the target An estimation method executed by an estimation device including an estimation step for estimating the treatment information by inputting the operation information of the elevator and the specification information of the target elevator.

Images