JPWO2017090098A1 - Equipment management apparatus and method - Google Patents

Abstract

Provided is an apparatus operating state identification apparatus and method capable of automatically and accurately determining the operating state of an apparatus with low network load and low cost. A data acquisition unit that acquires data related to the operating status of the device, a feature amount extraction unit that extracts feature amounts based on the data acquired by the data acquisition unit, and classifies the feature amounts extracted by the feature amount extraction unit A clustering unit to be created, a labeled data creation unit for creating data in which feature quantities classified by the clustering unit are labeled with the operating state of the device attached to the cluster to which the classified feature quantity belongs, and labeled A storage unit that stores data created by the data creation unit, a feature amount extracted by the feature amount extraction unit, and data stored in the storage unit, and determines the operating state of the device and outputs a determination result A state determination unit.

Description

  The present invention relates to an equipment management apparatus and method.

  As a background art in this technical field, there is JP 2010-257010 A (Patent Document 1). This publication describes that “a machine tool control device capable of preventing the occurrence of machining defects by performing machining while preventing abnormal wear and chipping of a rotary tool” (see summary). ).

  Moreover, there exists Unexamined-Japanese-Patent No. 2004-070424 (patent document 2). This gazette states that “the machine tool operating status is automatically acquired in all operations and work processes, the performance data is collected easily and accurately, the maintenance timing is obtained, and it is effective for energy conservation measures in the process. In addition, it provides an operational information collection system that performs cost calculation in production management and accumulates basic data for improving production efficiency. "(See summary).

  Moreover, there exists Unexamined-Japanese-Patent No. 2004-151843 (patent document 3). This publication provides “a machine tool operating status management system that can automatically determine the operating status even when using a machine tool that is not designed in consideration of the operating status”. (See summary).

JP 2010-257010 A Japanese Patent Laid-Open No. 2004-070424 JP 2004-151843 A

  As a method for visualizing the operating state of equipment (apparatus), there is a method in which a plurality of sensors such as current sensors and vibration sensors are installed and acquired data is transferred to a host system. However, with this method, acquired data is transferred to the host system as it is, so when applied to all devices in the factory, the number of sensors and the amount of data increase, and the network load, system construction cost, and maintenance cost increase. To do.

  On the other hand, a method of reducing the network load by performing a process of extracting only meaningful information from data and then transferring it to a higher system is conceivable. Patent Document 1 describes a method of extracting a feature quantity such as a peak value or a peak value of a frequency component from sensor data and performing state identification using a manually set threshold value. However, since the method of Patent Document 1 uses a fixed threshold method for discrimination of state identification, the feature amount changes depending on the type of equipment, tool / work change, or change of machining conditions, and the state identification of the apparatus is accurately performed. Is difficult to do. For this reason, it is necessary to reset the threshold each time the machining conditions are changed, and it is difficult to automate the identification of the state of the apparatus.

  On the other hand, Patent Document 2 describes a method of creating a feature amount database in advance for each tool and machining condition, and a method of predicting feature amounts by cutting simulation from an NC program. However, the tool may be made by a field engineer, and the processing conditions may be changed by the operator's judgment. Therefore, it is difficult to grasp the feature amount in advance for all facilities.

  Patent Document 3 describes a method for determining an operating state by monitoring a control signal of an NC machine. However, this method cannot be applied to non-NC machines. In addition, the tool idle state and the actual cutting state cannot be distinguished, and the effective cutting time and the like are not accurately known.

  Therefore, an object of the present invention is to provide an equipment management apparatus and method capable of automatically determining the operating state of an apparatus while suppressing a network load.

  In order to solve the above problems, for example, the configuration described in the claims is adopted.

  The present application includes a plurality of means for solving the above-described problems. To give an example, a data acquisition unit that acquires data related to the operating state of the apparatus, and a feature amount is extracted based on the data acquired by the data acquisition unit. A feature amount extraction unit, a clustering unit that classifies the feature amount extracted by the feature amount extraction unit and creates a cluster, and the feature amount classified by the clustering unit is attached to the cluster to which the classified feature amount belongs A labeled data creation unit that creates data with a label indicating the operating state of the apparatus, a storage unit that stores data created by the labeled data creation unit, a feature amount extracted by a feature amount extraction unit, and the storage unit And a state determination unit that determines the operating state of the device based on the data stored in and outputs a determination result.

  ADVANTAGE OF THE INVENTION According to this invention, the equipment management apparatus and method which can perform the determination of the operating state of an apparatus automatically, suppressing a network load can be provided.

  Problems, configurations, and effects other than those described above will become apparent from the following description of embodiments.

Configuration diagram of an example of a facility management apparatus in Embodiment 1 6 is a flowchart of an example of a principal component vector acquisition method according to the first embodiment. Example of distribution of feature amount in embodiment 1 Flowchart of an example of facility management method in embodiment 1 Configuration diagram of an example of a facility management apparatus in the second embodiment Flowchart of an example of facility management method in Embodiment 2 Configuration diagram of an example of a facility management apparatus in the third embodiment Flowchart of an example of facility management method in embodiment 3 Configuration diagram of an example of a facility management apparatus in Embodiment 4 Flowchart of an example of facility management method in embodiment 4 Configuration diagram of an example of a facility management apparatus in the fifth embodiment Flowchart of an example of facility management method in embodiment 5

  Hereinafter, examples will be described with reference to the drawings.

  In this embodiment, using FIGS. 1 to 3, the feature pattern is learned from the data acquired by the current sensor installed on the motor power line of the equipment or device such as a machine tool, and the type of machine tool, tool, work, An example of an equipment management method and apparatus that can estimate the operating state of equipment (also referred to as an apparatus) in general even when the processing conditions are different will be described.

  FIG. 1 is an example of an equipment management apparatus according to the present embodiment. The equipment management device 99 includes a machine tool 1 (sometimes referred to as a device), a sensor unit 2, and a equipment state identification unit 3.

  The sensor unit 2 (data acquisition unit) has a current sensor installed on the motor power line of the machine tool 1, acquires time-series current data, and transfers the current data to the equipment state identification unit 3. The sensor is not limited to a current sensor, and any sensor that can acquire data related to the operating state of the apparatus, such as a vibration sensor and a rotation sensor, and output the data can be used. Moreover, the place where the sensor is installed is not limited to the motor power line, and may be any place where data relating to the operating state of the apparatus can be acquired.

  The equipment state identification unit 3 includes a feature amount extraction unit 5, a statistical learning unit 4, a labeled feature amount memory 9 (storage unit), and a state determination unit 10, and learns feature amounts from data acquired by the sensor unit 2. The operating state is determined, and the determination result is transferred to the production management system 11 via the network 12.

  The feature quantity extraction unit 5 extracts the feature quantity from the data acquired by the sensor unit 2 and transfers the feature quantity to the statistical learning unit 4 and the state determination unit 10. For example, in the present embodiment, as the feature amount, a value obtained by performing fast Fourier transform on current data for each predetermined section and reducing the dimension to about three dimensions by calculating the inner product of the amplitude value of each frequency component and the principal component vector is used.

  Note that other feature quantity extraction methods include a method using wavelet transform and a method using factor analysis.

  FIG. 2 shows a flowchart of an example of acquiring a principal component vector.

  When the machine tool 1 starts operation (200), the sensor unit 2 transfers the acquired data to the feature amount extraction unit 5 (201). The feature quantity extraction unit 5 performs fast Fourier transform on the transferred data, and transmits the result to the production management system 11 via the network 12 (202). The production management system 11 performs principal component analysis on the received data, and transfers the generated principal component vector to the equipment state identification unit 3 (203).

  The equipment state identification unit 3 receives the principal component vector and ends the processing (204). In the present embodiment, for example, the production management system 11 transfers the generated principal component vector to the feature amount extraction unit 5, and the feature amount extraction unit 5 receives the principal component vector. The principal component vector may be received by the statistical learning unit.

  Returning to FIG. 1, the configuration of the statistical learning unit 4 will be described. The statistical learning unit 4 includes a clustering unit 6 and a labeled data creation unit 13.

  The clustering unit 6 performs clustering processing for classifying feature quantities and creating clusters, and creates clusters that are aggregates of the classified feature quantities. In addition, the clustering unit 6 transfers the position vector of the center of each cluster to the labeling unit 8. As the clustering algorithm, for example, a mixed Gaussian model or a k-means method may be used.

  The labeled data creation unit 13 creates data in which the feature quantities classified by the clustering unit 6 are labeled with the operating state labels of the devices attached to the cluster to which the feature quantities belong. Specifically, the labeled data creation unit 13 includes a labeling unit 8 and a prior knowledge unit 7.

  The labeling unit 8 calculates the norm of the position vector at the center of each cluster generated by the clustering unit 6 and assigns a unique label number to each cluster. For example, it is a unique number such as cluster 0, cluster 1, cluster 2, and the like. Note that the label is not limited to numbers, and may be any label such as a symbol such as an alphabet that can identify each cluster individually.

  The labeling unit 8 sends the norm of the position vector at each cluster center and the unique label number associated with each cluster to the prior knowledge unit 7.

  The prior knowledge unit 7 converts the label number received from the labeling unit 8 into a label indicating the operating state of the machine tool, and transfers the converted label to the labeling unit 8. For example, the label is converted by adding the meaning of the operating state of the apparatus to a unique number, with the machine tool stopped state being 0, the idling state being 1, the cutting state being 2, and the like. As a method for converting the label, for example, in a machine tool, the label is converted into an operating state label of the apparatus based on the relative positional relationship of the positions at the centers of the clusters. For example, since the norm of the position vector at the cluster center increases in the order of stop <idling ≦ cutting, the norms of the cluster centers received from the labeling unit 8 may be sorted in ascending order, and the array number may be used as a label.

  In addition, the label of the operating state of the apparatus is not limited to the stopped state, the idling state, and the cutting state, and an appropriate label may be created as appropriate according to the operation unique to each apparatus. The same applies to the following description.

  FIG. 3 shows an example of the distribution of feature amounts.

  Each point in FIG. 3 represents a feature amount extracted by the feature amount extraction unit 5 from time-series data of a certain period acquired by the sensor unit 2. Since feature values extracted from the operating state of the same equipment tend to have similar values, points belonging to the same cluster can be considered to have been generated from the same equipment state. The cluster center generated by the clustering unit 6 is the coordinates of a point representing each cluster, and the value varies depending on how to obtain the principal component vector, the type of equipment, the tool, the machining conditions, and the like.

  However, since the feature amount of the machine tool reflects the energy consumption, the norm at the center of each cluster increases in the order of stop <idling ≦ cutting, and the relative positional relationship does not depend on the type of equipment. Therefore, by sorting the norms at the center of each cluster in ascending order, each cluster can automatically correspond to the equipment state. For example, in FIG. 3, stop, idling, and cutting clusters are arranged in order from the lower left to the upper right.

  Next, returning to FIG. 1, the labeling unit 8 attaches the label to the feature quantity belonging to the cluster corresponding to the label of the operating state of the machine tool 1 transferred from the prior knowledge unit 7, and the labeled feature quantity memory 9 is stored. The cluster feature amount may be, for example, a parameter (average / variance) representing the distribution of the cluster, or may be a part or all of the feature amount belonging to the cluster.

  The labeled feature quantity memory 9 stores the feature quantity and the operation status label of the cluster device to which the feature quantity belongs as data linked to create a database.

  The state determination unit 10 compares the unlabeled feature amount transferred from the feature amount extraction unit 5 with the labeled feature amount data of the operating state of the apparatus in the database stored in the labeled feature amount memory 9, The label to be attached to the unlabeled feature amount is determined, and the determination result and the time stamp are transferred to the production management system 11 via the network 12.

  Here, the time stamp is information on the time determined by the state determination unit 10 or information on the time when the determination result is output. For example, when a mixed Gaussian model is used in the clustering unit, the label determination method directly calculates the probability of belonging to a cluster with an unlabeled feature amount from the model parameters average, variance, and mixture ratio. Is determined to belong to the cluster. When the k-means method is used, it is determined that the Euclidean distance belongs to the same cluster as the labeled feature quantity with the closest distance.

  Since the number of data is small at the beginning of learning, the state determination may be erroneous. Therefore, by holding a determination result with a time stamp on a server (storage device) (not shown) and making it possible to correct the determination result later from an input unit (not shown) or the like, production simulation accuracy can be improved.

  The production management system 11 acquires a label indicating the operating state of the machine tool 1 from the state identification unit 3 via the network 12 and executes a production simulation. In the production simulation, for example, a time transition of a production process is simulated from the operating states of a plurality of facilities, and calculation for determining processing priority between products is performed. In addition, a production plan is generated from the execution result of the production simulation, and an instruction is transmitted to workers and equipment via the network 12.

  FIG. 4 is an example of a flowchart for determining the operating state of the apparatus in this embodiment. When the machine tool 1 starts operation (400), the sensor unit 2 transfers the acquired data to the feature amount extraction unit 5 (401).

  The feature amount extraction unit 5 performs fast Fourier transform on the data received from the sensor unit 2, calculates the inner product of the absolute value of the amplitude and the principal component vector, and transfers the result to the statistical learning unit 4 and the state determination unit 10. (402). Note that the transfer result is a feature amount as described above.

  In the state determination unit 10, first, the clustering unit 6 performs clustering processing on the feature amount, and transfers the position vector of each cluster center to the labeling unit (403).

  The labeling unit 8 calculates the norm of the position vector at the center of the cluster, and refers to the prior knowledge unit 8 to label stop / idling / cutting and operating state of the apparatus in ascending order of the norm, and labeled feature quantity memory The cluster center and label are stored in 9 (404).

  When the state determination unit 10 receives the unlabeled feature amount from the feature amount extraction unit 5, the state determination unit 10 refers to the labeled feature amount memory 9, determines the label of the operating state of the device of the cluster to which the unlabeled feature amount belongs, and the time stamp Is transmitted to the production management system 11 (405).

  The production management system 11 executes a production simulation to generate a production plan (406), and transmits an instruction to the worker (407).

  The worker receives an instruction from the production management system and ends the process (408).

  In the method of judging the equipment state using the threshold value once set and fixed, if the feature value changes due to changes in the type of machine tool, tool, workpiece or machining condition, it is necessary to reset the threshold value each time Therefore, it is difficult to automatically identify the operating state of the apparatus, and the accuracy of the determination result is not high. On the other hand, in this embodiment, the feature amount distribution (cluster) is automatically created in the operation state identification of the apparatus, and labeling is performed based on the relative positional relationship between the clusters, so that the feature amount change is automatically performed. Can be handled. Therefore, in this embodiment, the operating state of the apparatus can be automatically determined with high accuracy.

  In addition, since the data acquired by the sensor unit 2 is transferred to the production management system 11 by focusing on meaningful information such as data processed at the edge or the result determined by the state determination unit 10, the network load can be reduced while reducing the network load. Visualization of the operating status of all facilities can be realized, and production simulation can be made highly accurate.

  In this embodiment, referring to FIGS. 5 and 6, in the equipment management apparatus, when there is no prior knowledge about the cluster that is the feature amount distribution, or there is an unknown cluster that cannot be labeled with prior knowledge due to cutting abnormality or the like. An example will be described in which a mechanism for performing semi-supervised learning is provided by sending an alert to the user and entrusting the labeling to the worker when it is discovered, so that it can cope with an unknown state.

  FIG. 5 is a configuration diagram of an example of the facility management apparatus according to the present embodiment.

  The facility management device 499 includes a posterior knowledge unit 401 and a user input unit 402 in addition to the prior knowledge unit 7 of the facility management device 99 of FIG.

  In the present embodiment, the description of the configuration denoted by the same reference numeral shown in FIG. 1 and the same function of the configuration will be omitted.

  In this embodiment, when the clustered unit 6 classifies the feature quantity and creates a new unlabeled cluster, the labeled data creation unit 13 attaches an alert label to the feature quantity belonging to the new cluster. Create the data. At this time, the labeling unit 8 gives an identification number not used for a known cluster to the new cluster. The prior knowledge unit 7 converts the identification number assigned by the labeling unit 8 into an identification number obtained by adding the meaning of the alert, and outputs the identification number to the labeling unit 8. The labeling unit 8 creates and outputs data in which an alert label is added to a feature quantity belonging to a cluster to which an alert identification number is assigned.

  The user input unit 402 re-labels the operation status of the apparatus to the data labeled with the alert via the network 12 according to the input from the user. The posterior knowledge unit 401 stores data that has been re-labeled by the user input unit 402 with respect to the operating state of the apparatus. The labeling unit 8 is relabeled by the user, and learns that the unknown cluster corresponds to the label of the operating state of the apparatus based on the data stored in the posterior knowledge unit 401. Thereby, thereafter, the labeling unit can create data in which the feature amount belonging to the unknown cluster is labeled with the operating state of the apparatus.

  FIG. 6 is an example of a flowchart showing a processing method when an unknown cluster is found in this embodiment.

  When the clustering unit 6 creates an unknown cluster (600), the labeling unit 8 assigns an alert label to the unknown cluster and stores it in the labeled feature quantity memory 9 (601).

  The state identification unit 10 transfers the alert label to the production management system 11 via the network 12 (602).

  Upon receiving the alert label, the production management system 11 transmits an alert to the user via the network 12 (603).

  The user who received the alert labels the unknown label to which the alert label is assigned (604).

  The process ends (605), but the labeling performed by the user is stored in the posterior knowledge unit 401, and the labeling unit 8 thereafter refers to the posterior knowledge unit 401 and performs labeling by semi-supervised learning. Do.

  In the present embodiment, when the clustering unit 6 creates an unknown cluster, it is possible to determine the operation state of the apparatus with high accuracy by prompting the user to give a label indicating the correct operation state of the apparatus. In particular, since an appropriate label can be assigned when the management device starts to monitor the operation state of the device, the determination accuracy of the operation state of the device after learning is increased.

  In this embodiment, referring to FIGS. 7 and 8, when information on tools and machining conditions is available in advance in equipment management equipment from equipment / equipment such as machine tools, the feature database is adapted. An example of installing a feature quantity with a label to improve the accuracy of determining the operating state of the apparatus will be described.

  FIG. 7 is a configuration diagram of an example of the facility management apparatus according to the present embodiment. In the present embodiment, the description of the configuration denoted by the same reference numeral as shown in FIG. 1 and the same function of the configuration will be omitted.

  The facility management apparatus 699 has a configuration in which the management apparatus 99 in FIG. 1 is connected to the feature amount database 601 via the network 12.

  The feature quantity database 601 receives, for example, process information such as the type of machine tool, tool, workpiece, and machining conditions, and feature quantities with labels such as a stop state, an idling state, and a cutting state obtained in the process in the past. , Saved. The feature quantity database 601 searches the labeled feature quantity to which the label indicating the operating state of the apparatus is attached based on the process information such as the type of machine tool, tool, workpiece, machining condition, etc., and the label of the equipment state identification unit 3 It is transferred to the attached feature amount memory 9. Thereby, even if there is not enough data in the labeled feature quantity memory 9 at the initial stage of operating the facility management apparatus 699, the operating state of the apparatus can be accurately determined based on the input labeled feature quantity data. Can do.

  FIG. 8 is a flowchart for explaining the processing of this embodiment.

  When the machine tool 1 starts operation (800), the machine tool 1 transfers process information such as tools and machining conditions to the feature amount database 601 via the network 12 (801).

  When the feature quantity database 601 receives the process information, the feature quantity database 601 searches the feature quantity database for the feature quantity with the matching process information and transfers the labeled feature quantity representing the operation state of the apparatus to the labeled feature quantity memory 9. (802), the process is terminated (803).

  Since other processes are the same as those in the first embodiment, description thereof is omitted.

  In this embodiment, since the clustering unit 6 automatically creates a feature amount distribution (cluster) and the labeled data creation unit 13 learns by attaching a label corresponding to the cluster to the feature amount, no prior database is required. However, there are cases where the number of data is small at the initial stage of learning and labeling is wrong. Therefore, when a labeled feature amount indicating the operating state of the apparatus is available in advance, it can be used to determine the operating state of the apparatus at the initial learning stage with high accuracy.

  In the present embodiment, an example in which the operation of the machine tool is stopped when a cutting abnormality such as tool breakage is detected in the equipment management apparatus will be described with reference to FIGS. In the present embodiment, the description of the configuration denoted by the same reference numeral as shown in FIG. 1 and the same function of the configuration will be omitted.

  FIG. 9 is a configuration diagram of an example of the facility management apparatus according to the present embodiment.

  The facility management apparatus 899 has a configuration in which an abnormality detection unit 801 is added to the management apparatus 99 of FIG.

  The anomaly detection unit 801 receives a feature quantity without a label from the feature quantity extraction unit 5 and refers to a feature quantity memory 9 with a label that stores data of a feature quantity with a label representing an operating state of the apparatus. Calculate the degree of abnormality. The abnormality detection unit 801 determines that the abnormality is abnormal when the abnormality degree is equal to or greater than the threshold, detects the abnormality, and transmits (outputs) an alert to the production management system 11 via the network 12.

  As the definition of the degree of abnormality, for example, when the clustering unit 6 uses a mixed Gaussian model algorithm, it is conceivable to use a negative log likelihood. As another definition of the degree of abnormality, the difference between norms of position vectors representing the positions of unlabeled feature quantities and labeled feature quantities may be taken.

  When the production management system 11 receives the alert, it sends a stop command to the machine tool 1 via the network 12 to stop the machine tool 1. The operator may be notified of the stop command by displaying it on a display (not shown) or the like via the network 12, or may be notified by voice or an alarm.

  FIG. 10 is an example of a flowchart for explaining the processing of this embodiment.

  When the machine tool 1 starts operation (1000), the same processing as in the first embodiment is performed, and the feature amount extraction unit 5 transfers the extracted unlabeled feature amount to the state determination unit 10 and the abnormality detection unit 801.

  Upon receipt of the unlabeled feature value from the feature value extraction unit 5, the abnormality detection unit 801 refers to the labeled feature value memory 9 that stores the feature value data with the label indicating the operation state of the apparatus, and has no label. The degree of abnormality of the feature quantity is calculated (1001).

  If the degree of abnormality is equal to or greater than the threshold, it is determined that the abnormality is present, and the abnormality detection unit 801 transmits (outputs) an alert to the production management system 11 via the network 12 (1002).

  Upon receiving the alert from the abnormality detection unit 801, the production management system 11 transmits a stop command to the worker and the machine tool 1 via the network 12 (1003).

  The machine tool 1 stops its operation and ends the processing (1004).

  Since other processes are the same as those in the first embodiment, description thereof is omitted.

  According to the present embodiment, since the apparatus can be stopped when the apparatus becomes in an abnormal operation, a safe facility management apparatus can be provided. In addition, clustering processing is performed on feature quantities determined to be abnormal in the clustering unit, and feature quantity data with abnormal operation labels is created in the labeled data creation unit, thereby creating cluster data of feature quantities that cause abnormal operation. Can be acquired, and the management of the apparatus can be made more accurate.

  In the present embodiment, an example in which an error in the determination result of the operating state of the apparatus is corrected later in the equipment management apparatus will be described with reference to FIGS. In the present embodiment, the description of the configuration denoted by the same reference numeral as shown in FIG. 1 and the same function of the configuration will be omitted.

  FIG. 11 is a configuration diagram of an example of the facility management apparatus according to the present embodiment.

  The facility management apparatus 1099 has a configuration in which a determination result holding server 1001 (storage device) is added to the facility management apparatus 99 of FIG.

  The determination result holding server 1001 holds the time stamp transmitted by the state determination unit 10 and the determination result. The determination result holding server 1001 holds the determination result label held when the state determination unit 10 outputs a conversion rule for converting an incorrect label attached to the determination result into a correct label via the network 12. Is rewritten according to the conversion rule. In other words, the state determination unit 10 holds the determination result holding server 1001 when the data stored in the labeled memory 9 is replaced with the feature amount data labeled with the correct device operating state through learning. Correct the judgment result. Thereby, the facility can be managed with higher accuracy.

  Note that the case where the feature amount data labeled with the correct device operating state is replaced through learning will be described in the following processing.

  FIG. 12 is a flowchart for explaining processing when an error is found in the determination result of the past state determination unit 10 in this embodiment.

  When an error is found in the determination result of the past state determination unit 10, the process is started (1200), and the state determination unit 10 determines a conversion rule for converting an incorrect label into a correct label via the network 12. The data is transferred to 1001 (1201). As the conversion rule, for example, in the labeled feature quantity memory label used for the determination in the past state determination unit 10, the label indicating “cut” is “2” and the label indicating “idling” is “1”. When a result determined to be “idling” is to be converted to “cutting”, a numerical array such as [12] is transferred.

  When receiving the label conversion rule from the state determination unit 10, the determination result holding server 1001 converts the label held according to the conversion rule, and notifies the production management system 11 that the label has been converted (1202).

  Upon receiving the notification that the label has been converted from the determination result holding server 1001, the production management system 11 executes the production simulation again using the time stamp in the determination result holding server 1001 and the corrected determination result (1203). The process ends (1204).

  Since other processes are the same as those in the first embodiment, description thereof is omitted.

  As a method for finding an error in the past determination result, for example, there is the following method. Since the labeled feature quantity memory 9 stores the cluster labels corresponding to the feature quantities of each cluster, when the labeling unit 8 updates the labeled feature quantity memory 9, the labeled feature quantity before the update is used. And the updated feature quantity with the label are compared. If the label is different even though the feature quantity is the same, it can be determined that there is an error in the past determination result.

  In this case, the labeled feature quantity memory 9 transfers information indicating that there is an error in the past determination result and correct label information to the state determination unit 10. The state determination unit 10 re-labels the determination result of the determination result holding server 1001 based on the transferred information. Thereby, the management information of the apparatus can be automatically corrected.

  According to the present embodiment, the facility management apparatus 1099 can automatically correct the past determination result, so that the accuracy of the determination result can be increased.

In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.
Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a recording device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
Further, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

DESCRIPTION OF SYMBOLS 99 ... Equipment management apparatus 1 ... Machine tool, 2 ... Sensor part, 3 ... Equipment state identification part, 4 ... Statistical learning part, 5 ... Feature quantity extraction part, 6 ... Clustering part, 7 ... Prior knowledge part, 8 ... Label Attaching unit, 9 ... Labeled feature amount memory, 10 ... State determination unit, 11 ... Production management system, 12 ... Network, 13 ... Labeled data creation unit, 499, 699, 899, 1099 ... Equipment management device, 401 ... Subsequent Knowledge unit, 601 ... feature-like database, 801 ... abnormality detection unit, 1001 ... determination result holding server

Claims (15)

A data acquisition unit for acquiring data relating to the operating state of the device;
A feature amount extraction unit that extracts a feature amount based on the data acquired by the data acquisition unit;
A clustering unit for creating a cluster by classifying the feature amount extracted by the feature amount extraction unit;
A labeled data creation unit that creates data with a label indicating an operation state of the device attached to a cluster to which the classified feature amount belongs, to the feature amount classified by the clustering unit;
A storage unit for storing the data created by the labeled data creation unit;
A state determination unit that determines the operating state of the device based on the feature amount extracted by the feature amount extraction unit and the data stored in the storage unit, and outputs a determination result;
An equipment management apparatus comprising: The equipment management device according to claim 1,
The labeled data creation unit
A labeling unit for attaching the label to the cluster created by the clustering unit;
An equipment management device comprising: The facility management device according to claim 2,
The labeled data creation unit
Based on a relative position between the clusters created by the clustering unit, a prior knowledge unit that converts a label of the cluster labeled by the labeling unit into a label of an operating state of the device;
Have
The labeling unit attaches, to the feature quantity classified by the clustering unit, a label indicating the operating state of the device of the cluster to which the feature quantity belongs, converted by the prior knowledge unit. . The equipment management device according to claim 3,
The state determination unit outputs a determination result of the determined operating state of the apparatus with information on the determined time,
The facility management apparatus further includes a storage device that stores the determination result with the time information output from the state determination unit,
The determination result stored in the storage device can be changed. The equipment management device according to claim 1,
The labeled data creation unit creates data in which an alert label is attached to a feature amount belonging to the new cluster when the clustering unit classifies the feature amount and creates a new cluster. Equipment management device. The facility management device according to claim 5,
A user input unit that attaches a label of an operating state of the device to data labeled with an alert in the labeled data creation unit;
A posterior knowledge unit storing data labeled with the operating state of the device in the user input unit;
With
The facility management apparatus, wherein the labeled data creation unit creates data with a label of an operating state of the apparatus based on data stored in the posterior knowledge unit. The equipment management device according to claim 1,
Comprising a feature quantity database capable of inputting feature quantity data labeled with the operating state of the device;
The feature quantity database outputs the input data to the storage unit. The equipment management device according to claim 1,
An abnormality detection unit for detecting an abnormality based on the feature amount extracted by the feature amount extraction unit and the data stored in the storage unit;
The abnormality detection unit outputs an alert when an abnormality is detected. The facility management apparatus according to claim 8,
An equipment management apparatus comprising a production management system that stops the apparatus when the abnormality detection unit outputs an alert. The equipment management device according to claim 1,
A storage device for storing the determination result output from the state determination unit;
The facility management apparatus, wherein the state determination unit corrects the determination result stored in the storage device when information on the label of the data created by the labeled data unit is rewritten. A data acquisition process for acquiring data relating to the operating state of the device;
A feature amount extraction step for extracting a feature amount based on the data acquired in the data acquisition step;
A clustering step of creating a cluster by classifying the feature amounts extracted in the feature amount extraction step;
A labeled data creation step for creating data with a label indicating the operating state of the device attached to the cluster to which the feature amount belongs to the feature amount classified in the clustering step;
A storage step for storing the data created in the labeled data creation step;
A state determination step of determining an operating state of the device based on the feature amount extracted in the feature amount extraction step and the data stored in the storage step and outputting a determination result;
An equipment management method comprising: The facility management method according to claim 1,
The labeled data creation step includes:
A labeling step for labeling the clusters created in the clustering step;
A facility management method comprising: The facility management method according to claim 2,
The labeled data creation step includes:
A prior knowledge step of converting a label of the cluster labeled in the labeling step into a label of an operating state of the device based on a relative position between the clusters created in the clustering step;
In the labeling step, the facility management method is characterized in that the feature quantity classified in the clustering process is attached with a label indicating the operating state of the device of the cluster to which the feature quantity belongs, converted in the prior knowledge process. . The facility management method according to claim 3,
In the state determination step, the determination result of the determined operating state of the apparatus is attached with information on the determined time and output,
The facility management method further includes a determination result storage step for storing the determination result with the time information output from the state determination step,
The facility management method, wherein the determination result stored in the determination result storage step can be corrected. The facility management method according to claim 1,
In the labeled data creation step, when a new cluster is created by classifying feature quantities in the clustering unit, data in which an alert label is attached to a feature quantity belonging to the new cluster is created. Equipment management method.

Images