CN108700872B

CN108700872B - Machine sorting device

Info

Publication number: CN108700872B
Application number: CN201680082718.7A
Authority: CN
Inventors: 远山泰弘
Original assignee: Mitsubishi Electric Corp
Current assignee: Mitsubishi Electric Corp
Priority date: 2016-02-29
Filing date: 2016-02-29
Publication date: 2021-08-06
Anticipated expiration: 2036-02-29
Also published as: TWI621951B; WO2017149598A1; JPWO2017149598A1; JP6366852B2; TW201732638A; CN108700872A

Abstract

A data acquisition unit (101) acquires machine classification index data from a machine classification index database (201). A classification index quantification unit (102) converts qualitative data included in the machine classification index data into quantitative data indicating the degree of similarity between the qualitative data. A device classification unit (103) classifies devices on a device-by-device basis using the quantitative data.

Description

Machine sorting device

Technical Field

The present invention relates to a device sorting apparatus for sorting devices in units of devices constituting the devices.

Background

In the case of an elevator, an air conditioner, or the like, there are a plurality of such devices of the same kind in a plurality of environments, and it is useful to classify the devices into devices having the same characteristics. For example, in a conventional system described in patent document 1, in control of lighting and air conditioning for the purpose of energy saving of building equipment, elevators are classified into equipment having the same characteristics. In the system described in patent document 1, the elevator operation information is used to model the traffic of public areas, the number of persons in rooms, and the occupancy rate for each day of the week, each time period, and the like, and plan a control schedule. Here, the building classification is performed so that the analysis result of similar buildings having the same characteristics is used for buildings for which elevator operation information cannot be acquired.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2005-104635

Disclosure of Invention

Problems to be solved by the invention

When analyzing a failure, an abnormality, or the like in a machine constituting an elevator, an air conditioner, or the like, it is possible to expect an improvement in the accuracy of detecting the failure or the abnormality, as compared with analysis using only a single machine, by classifying and analyzing a plurality of machines and apparatuses by the same kind and the same characteristics.

However, in the conventional method, since the classification is performed on a facility unit such as an elevator, there is a problem as follows: even if the devices have different features, the devices cannot be classified as long as the device units are of the same kind and the same features. For example, in the case where the door machine constituting the elevator a and the door machine constituting the elevator B are of the same type but have different characteristics, conventionally, if the elevator a and the elevator B are judged to have the same characteristics as each other as the elevator, the door machines of both the elevators are classified into the same characteristics.

Further, in the conventional method, the indexes used for classification are elevator operation information, the use of the elevator, and the scale, but when analyzing a failure, an abnormality, or the like, classification is performed in consideration of a large amount of information such as a history of the failure, an installation environment of the equipment/equipment, and equipment replacement information, and thus improvement of classification accuracy can be expected. These pieces of information are not necessarily quantitative data consisting of numerical values, and sometimes qualitative data including character information. In the conventional method, when classification is performed based on qualitative data, evaluation of how similar different qualitative data are to each other is not considered. As a result, the cause of the abnormality cannot be sufficiently analyzed, and there is a problem that the abnormality detection accuracy is lowered.

The present invention has been made to solve the above-described problems, and an object thereof is to provide an apparatus classification device capable of accurately analyzing a failure, an abnormality, and the like of an apparatus.

Means for solving the problems

The machine classification device of the invention comprises: a data acquisition unit that acquires device classification index data, which is information unique to each device among a plurality of devices each including one or a plurality of devices, the device classification index data being acquired from monitoring data of each device; a classification index quantification unit that converts qualitative data included in the machine classification index data into quantitative data indicating a similarity between the qualitative data; and a machine classification unit that classifies the devices in units of machines using the quantitative data.

Effects of the invention

The machine classification device of the present invention converts qualitative data included in machine classification index data into quantitative data indicating the degree of similarity between the qualitative data, and classifies equipment on a machine-by-machine basis using the quantitative data. This enables accurate analysis of a failure, abnormality, or the like of the device.

Drawings

Fig. 1 is a configuration diagram of a device sorting apparatus according to embodiment 1 of the present invention.

Fig. 2 is an explanatory diagram showing an example of maintenance actual data used by the machine sorting apparatus according to embodiment 1 of the present invention.

Fig. 3 is a hardware configuration diagram of the device sorting apparatus according to embodiment 1 of the present invention.

Fig. 4 is a flowchart showing a machine classification process of the machine classification device according to embodiment 1 of the present invention.

Fig. 5 is an explanatory diagram showing an example of quantitative data used in the machine classification device according to embodiment 1 of the present invention.

Fig. 6 is a flowchart showing a device classification process in the case of prior information having qualitative data similarity of the device classification apparatus according to embodiment 1 of the present invention.

Fig. 7A, 7B, and 7C are explanatory diagrams showing examples of prior information of the device sorting apparatus according to embodiment 1 of the present invention.

Fig. 8 is an explanatory diagram showing an example of classification of quantitative data by the machine classification device according to embodiment 1 of the present invention.

Fig. 9 is an explanatory diagram showing feature values of each device of the device sorting apparatus according to embodiment 1 of the present invention.

Fig. 10 is a configuration diagram of a device sorting apparatus according to embodiment 2 of the present invention.

Detailed Description

Hereinafter, in order to explain the present invention in more detail, a mode for carrying out the present invention will be explained based on the attached drawings.

Embodiment 1.

Fig. 1 is a configuration diagram of a monitoring system including a device sorting apparatus 100 according to the present embodiment.

In the illustrated monitoring system, the device sorting apparatus 100 is connected to the data collection management apparatus 200, and the data collection management apparatus 200 is connected to the monitoring target 400 via the network 300.

The device classification apparatus 100 includes a data acquisition unit 101, a classification index quantification unit 102, and a device classification unit 103. The data acquisition unit 101 is a processing unit that acquires machine classification index data from the machine classification index database 201 managed by the data collection management device 200. The classification index quantification unit 102 is a processing unit that converts qualitative data included in the machine classification index data into quantitative data. The device classification unit 103 is a processing unit that classifies the devices on a device-by-device basis using the quantitative data generated by the classification index quantification unit 102.

The data collection management device 200 collects monitoring data from the monitoring object 400, and stores and manages the data as the device classification index database 201. The monitoring data stored in the equipment classification index database 201 is data directly or indirectly acquired from the monitoring target 400, such as data generated from equipment information and inspection of the monitoring target 400 by a maintenance worker (for example, maintenance actual data). Fig. 2 shows an example of maintenance data, such as an elevator, as the machine classification index data stored in the machine classification index database 201.

Fig. 2 shows an example of maintenance actual data obtained from equipment information and inspection of one machine of one equipment by a maintenance person. In the example of the actual maintenance data, as an example of the data item, an equipment ID, a model ID, an equipment ID, an installation area, a worker name, a maintenance work content, presence or absence of an abnormality, and the like are described. The values of these data items are an example. The data item can be changed to store an item for maintaining actual data collected from an actual device or machine. Further, as long as the devices and the apparatuses can be distinguished, data of a plurality of devices and apparatuses may be collected into one table. Further, as long as the devices and the machines can be associated with each other, data of one machine of one device may be divided into a plurality of tables. Further, the maintenance data of maintenance jobs having different forms, such as a normal maintenance job and a maintenance job in the case where a failure or an abnormality occurs, may be divided and managed. That is, the machine classification index data stored in the machine classification index database 201 may be any information as long as it is information specific to the machine.

The monitoring object 400 is a device including one or more devices, such as an elevator or an air conditioner. Regarding the monitoring object 400, it is assumed that two or more devices composed of the same device exist. The monitoring target 400 may be directly connected to the data collection management apparatus 200 without being connected to the network 300. Regardless of the method of connecting the monitoring target 400 and the data collection management apparatus 200, the data collection management apparatus 200 and the device sorting apparatus 100 may be connected to each other via a network.

Fig. 3 is a block diagram showing a hardware configuration of a machine classification device for implementing the present embodiment. Fig. 3 shows an example in which the machine sorting apparatus 100 and the data collection and management apparatus 200 shown in fig. 1 are configured by hardware. The device sorting apparatus 100 and the data collection management apparatus 200 include a processor 11, a memory (memory)12, a communication I/F (interface) device 13, a storage (storage)14, and an output device 15. The processor 11 is a processor for realizing the functions of the machine sorting apparatus 100 and the data collection management apparatus 200. The memory 12 is a storage unit such as a ROM or a RAM used as a program memory for storing various programs corresponding to functions of the device sorting apparatus 100 and the data collection management apparatus 200, a work memory used when the processor 11 performs data processing, a memory for developing signal data, and the like. The communication I/F device 13 is a communication interface with the outside such as the network 300. The memory 14 is a memory for storing various data and programs. The output device 15 is a device for outputting the processing result to the outside.

The processing performed by the data acquisition unit 101, the classification index quantification unit 102, and the machine classification unit 103 in fig. 1 is executed by the processor 11 reading a program stored in the memory 12. The data stored in the device classification index database 201 is stored in the memory 14 from the monitoring target 400 via the network 300 by the communication I/F device 13. The processing result of the device classification unit 103 is stored in the memory 14 as necessary, and is output to the outside by the output device 15. The device sorting apparatus 100 and the data collection management apparatus 200 may be configured by different hardware.

Next, the operation of the device sorting apparatus 100 according to the present embodiment will be described.

The data collection management device 200 continuously or intermittently inputs the device classification index data acquired from the monitoring target 400 to the device classification index database 201. The machine classification device 100 acquires and processes machine classification index data from the machine classification index database 201. Fig. 4 is a flowchart showing the processing of the machine sorting apparatus 100.

First, the data obtaining unit 101 obtains the machine classification index data from the machine classification index database 201 (step ST 1). In addition, when a plurality of data items are included in the machine classification index data, the flow of fig. 4 is executed for each data item. For example, when a machine ID is input as an index of the machine classification index data, a list of the classified machine IDs is output. The form of the list is not limited, and for example, a sort ID is assigned to each sort, and each machine ID and the corresponding sort ID are stored in a table form in a line and output. As another example of the list, there is a method of: one file is generated for each classification, and the machine ID belonging to the classification is stored in the file.

The classification index quantification unit 102 converts qualitative data included in the device classification index data acquired from each device into quantitative data composed of numerical values in a form that allows similarity determination. In step ST2, whether or not the input machine classification index data is quantitative data is determined by whether or not the data is a numerical value, and the subsequent processing branches. In step ST2, if the data is quantitative data (yes in step ST 2), the classification index quantification unit 102 ends the process. That is, the device classification index data input to the classification index quantifying unit 102 is directly output to the device classifying unit 103. On the other hand, in step ST2, in the case where it is not quantitative data (step ST 2: NO), the process of step ST3 is performed. In step ST3, the distance between the qualitative data is calculated as the similarity between the qualitative data, and a value corresponding to the distance is assigned to each data to be used as the quantitative data. The distance between qualitative data is calculated by a character string analysis method such as hierarchical clustering analysis of n-gram, and a numerical value corresponding to the distance is set as quantitative data. Here, regarding the qualitative data, when the relationship between the position of the character and the influence on the distance (the character in the front represents a classification of a larger class and thus has a large influence on the distance, the character in the rear represents a classification of a smaller class and thus has a small influence on the distance, and the like) is known, processing such as weighting the character having a large influence on the distance may be performed when calculating the distance. For example, if the first half of the character string of the device ID indicates the major update version number and the second half indicates the minor update version number, the distance may be affected more greatly as the character string in front is used.

Fig. 5 shows an example of quantitative data obtained by converting qualitative data. Fig. 5 shows, as a simple example, a quantitative data example in which the name of the machine ID is represented by connecting the major update version number and the minor update version number of the machine by a hyphen "-". In this quantitative data example, devices with device IDs AAA-01, AAA-02, and AAA-03 have the same major update version number and only the minor update version number is different, and therefore, similar values are assigned. The machine ID BBB-01, BBB-02 machine has a different major update version number than AAA-01, AAA-02, AAA-03, and therefore is assigned a more distant value.

When the machine classification index data is quantitative data in step ST2 or after the processing in step ST3 is performed, the machine classification unit 103 classifies machines whose input values (i.e., feature values in multivariate analysis or the like) are close to each other by a multivariate analysis method, a machine learning method, or the like (step ST 4). Specific classification examples are described later.

On the other hand, when the similarity between the qualitative data is known as the prior information, the similarity of the prior information may be used. With regard to the prior information, only a part of the qualitative data may be specified. For example, only the similarity of the main update number in the machine ID is specified, and the like. Further, the prior information may be assigned a weighting rule for each character position of the qualitative data. For example, the ratio of the weight of the primary update number to the secondary update number in the machine ID is specified, and the like. Furthermore, qualitative data that is not converted into quantitative data may be given as advance information.

Fig. 6 shows a classification index quantification flow in the case of prior information. The same processes as those in fig. 4 are denoted by the same step numbers. When the step of determining whether or not the data is quantitative data at step ST2 is not quantitative data (no at step ST 2), the classification index quantification unit 102 determines whether or not there is prior information on the degree of similarity with the input device classification index data, and branches the subsequent processing (step ST 5). In step ST5, if there is no prior information of similarity (no in step ST5), the process of step ST3 is performed in the same manner as the flow of fig. 4. In the case of the prior information having the similarity (YES in step ST5), in the process of step ST6, a numerical value corresponding to the similarity of the given prior information is assigned. Here, when the degree of similarity of the prior information is qualitative data instead of quantitative data, the distance between the qualitative data is calculated as the degree of similarity between the qualitative data, and a value corresponding to the distance is assigned to each data, as in step ST3, thereby setting the data as quantitative data. The distance between qualitative data is calculated by a method of calculating the distance between words such as hierarchical clustering analysis of n-grams, and a numerical value corresponding to the distance is set as quantitative data. The method of calculating the distance between the qualitative data may also use a method different from step ST 3.

Fig. 7 shows an example of the prior information. An example of specifying the similarity of qualitative data is shown in fig. 7A. Fig. 7A is an example of specifying the similarity of the front 3 characters of the machine ID, showing the following case: the similarity of machines with machine IDs AAA and BBB is high, and the similarity of machines with machine ID CCC is low compared to machines with machine IDs AAA and BBB. Fig. 7B shows an example of a weighting rule for specifying each character position of qualitative data. Fig. 7B is an example of a weighting rule for each character position of the machine ID, and is an example of a case where: the weight of the 1 st to 3 rd characters of the machine ID is increased, so that the weight of the 1 st to 3 rd characters is set to 10, the weight of the 5 th to 6 th characters is set to be smaller than that of the 1 st to 3 rd characters, and the weight of the 5 th to 6 th characters is set to be 1. In addition, since the 4 th character is a hyphen, it is excluded from the weighting rule. Fig. 7C shows an example of specifying qualitative data that is not converted into quantitative data. Fig. 7C shows a case where the device ID is not quantified. Fig. 7A, 7B, and 7C are examples of the designated information, and the manner of giving the information may be changed.

As another example, a free text described as a result of a maintenance work of the machine may be used as the machine classification index data. For example, words such as "abnormal", "treatment completed", and "event a cause" included in a free text in which a result of a maintenance operation is described are extracted by morphological analysis, and similar numerical values are assigned to texts having a large number of similar morphemes.

Since the machine classification unit 103 classifies the machines according to the machines having similar characteristics, the quantitative data converted by the classification index quantifying unit 102, which is a plurality of machines, is input, and the machines having similar values of the quantitative data are classified. The quantitative data may be input as one data item, or may be output by aggregating a plurality of data items. In step ST4 in fig. 4 or 6, a general multivariate analysis method such as hierarchical clustering analysis such as dendrograms and non-hierarchical clustering analysis such as k-means method, and a general machine learning method such as support vector machine may be used. An example of classification is shown in fig. 8.

In fig. 8, as an example of classification of quantitative data, quantitative data of a plurality of data items of three machines is input, and feature 1 and feature 2 are schematically displayed on a two-dimensional scattergram as a feature space when a multivariate analysis method such as principal component analysis is performed. The following is shown in fig. 8: the eigenvalue 801 and the eigenvalue 802 are closer in distance on the scatter plot and are therefore summarized as a classification 804. The following is shown: the feature quantity value 803 is far from the feature quantity values 801 and 802 on the scatter diagram, and is therefore set to a classification 805 different from the classification 804. As a method of performing the classification in this way, a general clustering analysis method such as a nearest neighbor method in which distances between the

characteristic quantities

801, 802, and 803 are calculated and classification is performed based on a threshold value of the distances, and a k-means method in which the number of classifications is determined in advance can be used.

One application of the present invention is to analyze a failure or abnormality of a machine. For example, when predicting the time of future failure in order to plan maintenance of the machine, the following method is available: the time required for maintenance is estimated by predicting the probability of future failure (failure risk) from data acquired from the machine and from the statistical failure occurrence frequency and degradation tendency.

Here, since the probability of occurrence of a failure and the probability of deterioration of machines having similar characteristics are also similar, it is also useful to classify machines for each similar characteristic in order to predict the risk of failure. Classifying by more similar machines can improve the accuracy of the prediction of the risk of failure. The data used for calculating the risk of failure may be the same as the data used in the machine classification device 100 of the present embodiment, or may be other data. In estimating the failure risk, the failure risk can be predicted on a machine-by-machine basis, but the failure risk of a plurality of machines may be predicted on a device-by-device basis by comprehensively determining the failure risk of a plurality of machines such as the correlation between a plurality of machines constituting the same facility.

Next, the effect of embodiment 1 will be explained. In fig. 9, as an example of the feature amount of each machine, two machines, i.e., the machine 1 and the machine 2, are represented on a two-dimensional scattergram generated by collecting data from three devices, i.e., the device a, the device b, and the device c, calculating the feature amount, and based on the feature amounts of the two machines. The device characteristic amount of the device 1 is represented as 901, and the device characteristic amount of the device 2 is represented as 902.

In the conventional method, when the devices a, b, and c are classified into devices having the same characteristics due to the classification of the device units, the devices are classified into the same classification regardless of the characteristics of the

devices

1 and 2. On the other hand, in embodiment 1, for example, the following classifications can be made in units of machines: even when the device a and the device b are classified into one category and the device c is classified into the other category in the device feature 901 of the device 1, the device a and the device c are classified into one category and the device b is classified into the other category in the device feature 902 of the device 2.

By classifying the devices having the same characteristics, it is possible to expect improvement in the accuracy of predicting the risk of failure of the devices, the accuracy of detecting a failure or an abnormality, and the like. Further, by classifying the devices in accordance with the similarity, when a failure, an abnormality, or the like is found in a certain device, by extracting the devices having the same characteristics and performing maintenance, it is possible to prevent the failure or abnormality of the other device from occurring in advance, and it is possible to expect the efficiency of maintenance of scheduling of the maintenance job for each device. For example, when the torque of the door opening/closing motor of the elevator a is reduced and the elevator is trapped, it is possible to reduce the number of failures and accidents by checking whether the door opening/closing motor of another elevator having the same characteristics has a sign of the reduced torque and performing maintenance. As another example, when a decrease in torque of the door opening/closing motor of the elevator a is detected, although there is a possibility that a decrease in torque may occur in the door opening/closing motor of another elevator having the same characteristics, if it is not necessary to immediately respond to the decrease, it is possible to expect an increase in efficiency of the work by appropriately scheduling the maintenance work.

As described above, according to the machine sorter of embodiment 1, the machine sorter includes: a data acquisition unit that acquires device classification index data, which is information unique to each device, obtained from monitoring data of each device among a plurality of devices each including a single device or a plurality of devices; a classification index quantification unit that converts qualitative data included in the machine classification index data into quantitative data indicating a similarity between the qualitative data; and a device classification unit that classifies the devices in units of devices using the quantitative data, and therefore can analyze failures, abnormalities, and the like of the devices with high accuracy.

Embodiment 2.

In embodiment 1, the device classification unit 103 classifies each device based on the device classification index data quantified by the classification index quantification unit 102. In contrast, before the machine classification index data quantified by the classification index quantification unit 102 is input to the machine classification unit 103, the machine classification index data may be converted into feature quantities so as to emphasize the difference in the features of the respective machines, and the machine classification unit 103 may classify the respective machines based on the feature quantities, which will be described as embodiment 2.

The purpose of the conversion into the feature quantities is to clarify the difference between the respective machines when the machines are classified from the plurality of machine classification index data. In the case of machine classification based on only one piece of machine classification index data, similar values are assigned to similar machine classification index data at the time of conversion to quantitative data, and therefore classification can be performed only by the values of the machine classification index data. However, when a device is classified based on a plurality of pieces of device classification index data, even if a device has a similar value of one piece of device classification index data, other pieces of device classification index data may have a distant value. In such a case, if the value of the machine classification index data is kept as it is, the difference between the machines cannot be clearly understood, and thus the machines cannot be accurately classified. Therefore, by obtaining a feature amount for clarifying the difference between devices from a plurality of device classification index data, the devices can be classified more accurately. For example, there is a method as follows: the feature amount is set using the distance in the MT method based on a plurality of pieces of machine classification index data such as model IDs and installation areas. The feature amount may be a general method such as a multivariate analysis method of each principal component in the principal component analysis, a regression coefficient and error in the regression analysis, a similarity in the pattern matching method, or the like.

Fig. 10 is a configuration diagram of a monitoring system to which the device sorting apparatus 100a according to embodiment 2 is applied. The device classification apparatus 100a according to embodiment 2 includes a data acquisition unit 101, a classification index quantification unit 102, a device classification unit 103a, and a feature value conversion unit 104. Here, the data acquisition unit 101 and the classification index quantification unit 102 are the same as those in embodiment 1. The feature value conversion unit 104 is a processing unit that converts the machine classification index data quantified by the classification index quantification unit 102 into feature values. The device classification unit 103a is a processing unit that classifies devices using the feature values converted by the feature value conversion unit 104. In fig. 10, the data collection management device 200, the network 300, and the monitoring target 400 are the same as those in embodiment 1 shown in fig. 1.

In the device classification apparatus 100a configured as described above, the feature value conversion unit 104 converts the quantitative data generated by the classification index quantification unit 102 into feature values before the quantitative data is input to the device classification unit 103 a. The device classification unit 103a acquires the feature values from the feature value conversion unit 104, and classifies devices having similar feature values as devices having similar features. The feature value conversion unit 104 may be configured as follows: the quantitative data generated by the classification index quantifying unit 102 is not entirely converted into feature data, but is partially converted. When only a part of the data is converted, the device classification unit 103a classifies the data using both the quantitative data and the converted feature amount.

As described above, according to the machine sorter of embodiment 2, the machine sorter includes: a data acquisition unit that acquires device classification index data, which is information unique to each device, obtained from monitoring data of each device among a plurality of devices each including a single device or a plurality of devices; a classification index quantification unit that converts qualitative data included in the machine classification index data into quantitative data indicating a similarity between the qualitative data; a feature value conversion unit that converts the quantitative data into a feature value representing a difference in the feature of each device; and a device classification unit that classifies devices having similar characteristics as devices having similar characteristics, and can analyze failures, abnormalities, and the like of the devices with higher accuracy.

In the present invention, the embodiments may be freely combined, any component of the embodiments may be modified, or any component may be omitted in the embodiments, within the scope of the present invention.

Industrial applicability

As described above, the device sorting apparatus according to the present invention is suitable for a plurality of devices, each of which is classified for each device included in the plurality of devices, and is applied to a device such as a lifter or an air conditioner, and a plurality of the devices of the same type exist in different environments.

Description of the reference symbols

100. 100 a: a machine classification device; 101: a data acquisition unit; 102: a classification index quantification unit; 103. 103 a: a machine classification section; 104: a feature value conversion unit; 200: a data collection management device; 201: a machine classification index database; 300: a network; 400: an object is monitored.

Claims

1. A machine sorter, comprising:

a data acquisition unit that acquires device classification index data, which is information unique to each device, obtained from monitoring data of each device among a plurality of devices each including one or a plurality of the devices;

a classification index quantification unit that converts qualitative data included in the machine classification index data into quantitative data indicating a similarity between the qualitative data; and

and a device classification unit that classifies the devices in units of devices according to the devices having the same characteristics, using the quantitative data.

2. A machine sorter, comprising:

a classification index quantification unit that converts qualitative data included in the machine classification index data into quantitative data indicating a similarity between the qualitative data;

a feature value conversion unit that converts the quantitative data into a feature value representing a difference in a feature of each of the devices; and

and a device classification unit that classifies the devices having similar characteristics as devices having similar characteristics, and classifies the devices in units of the device according to the devices having similar characteristics.