WO2019167375A1 - 診断装置および診断方法 - Google Patents
診断装置および診断方法 Download PDFInfo
- Publication number
- WO2019167375A1 WO2019167375A1 PCT/JP2018/044793 JP2018044793W WO2019167375A1 WO 2019167375 A1 WO2019167375 A1 WO 2019167375A1 JP 2018044793 W JP2018044793 W JP 2018044793W WO 2019167375 A1 WO2019167375 A1 WO 2019167375A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- cluster
- attribute
- operation data
- unit
- abnormality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
- G05B23/0281—Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
Definitions
- the present invention relates to a diagnostic apparatus and a diagnostic method, and is suitable for application to a diagnostic apparatus and a diagnostic method for diagnosing abnormalities in equipment using operation data of the equipment, for example.
- ICT Information and Communication Technology
- IoT Internet of Things
- thermal power plants are required to operate in consideration of key performance evaluation indices (KPI: Key Performance Indicators) such as operating rate, environmental performance, and efficiency.
- KPI Key Performance Indicators
- ART adaptive resonance theory
- Patent Document 1 discloses a technique for diagnosing an abnormality and estimating the cause by using a cluster number and the number of appearances of each cluster.
- the present invention has been made in consideration of the above points, and intends to propose a diagnostic device and a diagnostic method capable of suppressing the occurrence of misreporting without processing the learning data and generating a normal cluster each time diagnosis is performed. To do.
- it is a diagnostic device for diagnosing an abnormality in the facility based on the operation data of the facility, and determines whether the operation data of the facility can be classified into an existing cluster, If it is determined that classification is not possible, a clustering unit that adds a cluster having an attribute as abnormal to the existing cluster, a diagnosis unit that diagnoses an abnormality in the facility based on a processing result by the clustering unit, and a predetermined period A cluster attribute setting unit configured to set an attribute of the existing cluster based on the operation data of the facility.
- a diagnosis method for diagnosing an abnormality in the equipment based on the operation data of the equipment wherein the clustering unit determines whether the operation data of the equipment can be classified into an existing cluster, and classification A second step of diagnosing an abnormality in the equipment based on a processing result by the clustering unit, and a first step of adding a cluster having an abnormal attribute to the existing cluster when it is determined that the cluster is not possible And a third step in which the cluster attribute setting unit sets the attribute of the existing cluster based on the operation data of the facility in a predetermined period.
- the attributes of the existing cluster are set based on the operation data of the equipment during a predetermined period, it is not necessary to process the learning data and generate a normal cluster each time diagnosis is performed. Can be suppressed.
- FIG. 1 shows the diagnostic system by 1st Embodiment as a whole.
- the diagnostic system 1 suppresses the occurrence of misreporting by appropriately changing the cluster attributes appropriately. Further, in the diagnosis system 1, by changing the setting of the attribute of the existing cluster, the process of processing the learning data and generating a normal cluster each time diagnosis is not required.
- such a configuration will be described with an example as appropriate.
- the diagnosis system 1 refers to, updates, etc. the plant 100, a diagnosis device 200 that diagnoses the operation status of the plant 100 (for example, an abnormality (an abnormality sign) in the plant 100), and various types of information stored in the diagnosis device 200. And an external device 300 for performing. Diagnostic device 200 is communicably connected to plant 100 and external device 300.
- the plant 100 includes a device 110 constituting the plant 100 and a control device 120 for controlling the device 110.
- the control device 120 receives a measurement signal transmitted from the device 110 and transmits a control signal.
- the diagnosis device 200 diagnoses an abnormality in the plant 100 based on the operation data of the plant 100.
- the diagnosis device 200 is a computer such as a server device or a tablet terminal, and a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), a HDD (Hard Disk Drive), a communication unit (not shown) And so on.
- CPU Central Processing Unit
- RAM Random Access Memory
- ROM Read Only Memory
- HDD Hard Disk Drive
- communication unit not shown
- Functions of the diagnostic apparatus 200 are read out from the program stored in the ROM by the CPU into the RAM, for example. It may be realized by executing (software), may be realized by hardware such as a dedicated circuit, or may be realized by combining software and hardware. Also, some of the functions of the diagnostic device 200 may be realized by another computer that can communicate with the diagnostic device 200.
- the communication unit is composed of, for example, a NIC (Network Interface Card), and performs protocol control during communication with the external device 300.
- the diagnostic apparatus 200 includes an operation data DB 260 and a calculation result DB 270 as DBs.
- DB computerized information is stored, and information is stored in a form called an electronic file (electronic data).
- the external input interface unit 210 diagnoses operation data collected in the plant 100 and external input signals created by operating the external input device 310 (the keyboard 311 and the pointing device 312) provided in the external device 300. Capture to device 200. For example, the operation data captured by the diagnostic device 200 is stored in the operation data DB 260.
- the external output interface unit 220 outputs (sends, prints, displays, etc.) the operation data stored in the operation data DB 260 and the calculation results stored in the calculation result DB 270.
- the clustering unit 230 determines whether or not the operation data of the plant 100 can be classified into an existing cluster. If it is determined that the operation data cannot be classified, the cluster having an abnormal attribute (abnormal cluster) is added to the existing cluster.
- the clustering unit 230 is equipped with a technique (clustering technique) using adaptive resonance theory, and classifies the operation data stored in the operation data DB 260 into clusters according to the similarity.
- the clustering unit 230 first classifies the normal operation data to generate a plurality of clusters, and normally sets the attributes of the generated plurality of clusters (defined as normal clusters). When the clustering unit 230 determines that the result of classifying the current operation data into clusters does not belong to a normal cluster, the clustering unit 230 generates a new cluster (new cluster) using the clustering technique. The clustering unit 230 stores the clustering result in the calculation result DB 270.
- the clustering unit 230 will be described later with reference to FIG. However, as the adaptive resonance theory, those disclosed in known documents such as Patent Document 1 may be used. In other words, the clustering unit 230 is not limited to the configuration described in the present embodiment, and may be configured by mounting other clustering techniques.
- the diagnosis unit 240 diagnoses an abnormality in the plant 100 based on the processing result (clustering result) by the clustering unit 230.
- the occurrence of a new cluster by the clustering unit 230 means that the state of the diagnosis target has changed to a new state (new state). Therefore, for example, the diagnosis unit 240 can determine the occurrence of an abnormality based on the occurrence of a new cluster.
- the diagnosis unit 240 calculates the distance from the target data to the nearest normal cluster as the degree of abnormality.
- the diagnosis result by the diagnosis unit 240 is transmitted to the display device 320 via the external output interface unit 220, for example, and the operator is notified of the diagnosis result.
- the calculation result correction unit 250 includes a cluster attribute setting unit 251 and a cluster attribute change post-processing unit 252, and based on the operation data of the plant 100 in a predetermined period, the attribute of an existing cluster stored in the calculation result DB 270 is obtained. Set or correct the calculation result of the degree of abnormality.
- the cluster attribute setting unit 251 normally sets the attributes of the cluster into which the operation data of the plant 100 in a predetermined period is classified, and sets the attributes of the clusters other than the normally set cluster abnormally To do. The processing of the cluster attribute setting unit 251 will be described later with reference to FIG.
- the cluster attribute change post-processing unit 252 determines the degree of abnormality that is the distance between the driving data and the normal cluster with respect to past driving data. Recalculate. The processing of the cluster attribute change post-processing unit 252 will be described later with reference to FIG.
- the function and DB of the diagnosis apparatus 200 are provided inside the diagnosis apparatus 200. However, a part of these functions are arranged outside the diagnosis apparatus 200, and only data is transferred between the apparatuses. You may make it communicate.
- DB information which is a signal (data) stored in each DB, can be displayed on the display device 320 via the external output interface unit 220, and these information can be displayed by operating the external input device 310. This can be corrected by an external input signal generated in this manner.
- the external input device 310 includes the keyboard 311 and the pointing device 312, but a device for inputting data, such as a microphone or a touch panel for voice input, may be appropriately employed. Can do.
- the external device 300 can be realized by a computer such as a notebook computer or a tablet terminal, but is not limited to a computer.
- the external input device 310 and the display device 320 may be provided as separate devices.
- the diagnosis system 1 may be provided with other devices such as a printing device and a projector device in addition to or instead of the display device 320.
- the diagnostic apparatus 200 can be implemented as an apparatus for analyzing the operation data offline.
- the application object of the diagnostic apparatus 200 is made into the plant 100, it cannot be overemphasized that an application object is not limited and can be applied to various facilities.
- FIG. 2 is a diagram illustrating an example of data stored in a DB provided in the diagnostic apparatus 200.
- FIG. 2A shows an example of data stored in the operation data DB 260.
- FIG. 2A In the table shown in FIG. 2A, each measurement value (operation data) measured by each sensor (sensor A, sensor B, sensor C,...) Provided in the plant 100 is set for each sampling period. (Item A, Item B, Item C,).
- (B) and (C) in FIG. 2 are data obtained by operating the clustering unit 230 and the like, and show an example of data stored in the calculation result DB 270.
- attribute data for each cluster and weight coefficient data for each cluster are stored.
- operation data classification results (data indicating the relationship between operation data and time, cluster number, etc.) are stored.
- the attribute is information for defining (identifying) whether each cluster is a normal cluster or an abnormal cluster.
- the weight coefficient is coefficient information for defining the center coordinates of each cluster. Further, for abnormal clusters, events that are automatically or manually defined for each cluster are stored.
- the table shown in FIG. 2B and the table shown in FIG. 2C are provided for each group. More specifically, when operation data including a plurality of data (for example, first data to sixth data) in a plurality of sensors is divided into a plurality of groups for each device (for example, the first data in the first device). And second data (first group data) are input, and output data, third data, and fourth data (second group data) from the first device are input to a second device provided downstream of the first device. When the fifth data and the sixth data (third group data) are input to a third device that is input and provided in parallel with the first device), the clustering unit 230 creates a cluster for each group. You may comprise so that it may produce
- FIG. 3 is a diagram illustrating an example of a processing procedure related to a diagnostic process performed by the diagnostic apparatus 200.
- step S10 the external input interface unit 210 captures (collects) operation data from the control device 120 and stores it in the operation data DB 260.
- step S11 the diagnosis unit 240 determines whether the operation mode is the learning mode or the diagnosis mode. If the diagnosis unit 240 determines that the learning mode is selected, the process proceeds to step S12. If the diagnosis unit 240 determines that the diagnosis mode is selected, the process proceeds to step S13.
- the learning mode is operated when the flowchart is repeated for the first time (first time), and the processing is performed so that the diagnosis mode is operated after the second time.
- step S12 the clustering unit 230 processes the learning data, generates a cluster in a normal state, and moves the process to step S10.
- step S13 the diagnosis unit 240 determines whether or not calculation result correction is necessary (whether it is a predetermined timing). If the diagnosis unit 240 determines that correction of the calculation result is necessary, the process proceeds to step S14. If the diagnosis unit 240 determines that correction of the calculation result is unnecessary, the process proceeds to step S16.
- the predetermined timing for correcting the calculation result the timing at which the characteristics of the plant 100 have changed, such as after periodic inspection, the timing at which a predetermined period has elapsed since the previous calculation result was corrected, and the calculation result from the operator It is possible to set an arbitrary timing such as a timing at which a signal for executing the correction is input.
- step S14 calculation result correction processing is performed.
- the details of the calculation result correction processing will be described later with reference to FIGS. 6 and 8, etc., but in the calculation result correction processing, automatic attribute change processing and post-attribute change processing are executed, and cluster attribute correction, abnormality Correction of the calculation result of the degree, correction of the weighting coefficient, and the like are performed.
- step S15 the external output interface unit 220 outputs the result of the calculation result correction process.
- the external output interface unit 220 displays screen information (for example, part or all of the table shown in FIG. 5B) for confirming the result of changing the cluster attribute by the cluster attribute setting unit 251. Screen information including the above information) is transmitted to the external device 300. Further, for example, the external output interface unit 220 displays screen information (for example, part or all of the information shown in the table shown in FIG. 5B) related to the screen for confirming the processing result by the cluster attribute change post-processing unit 252. And the screen information indicating the degree of abnormality as shown in FIG. 10) are transmitted to the external apparatus 300.
- the trigger for outputting the result of the calculation result correction process may be a request from the external device 300, a periodic process such as a monthly process, or another trigger.
- the clustering unit 230 processes the operation data. More specifically, the clustering unit 230 classifies the operation data into clusters. Further, when the clustering unit 230 determines that the result of classifying the operation data into clusters does not belong to the existing cluster, the clustering unit 230 generates a new cluster (new cluster) by the clustering technique. In addition, the clustering unit 230 obtains the new cluster information (cluster number, attribute, event, weighting coefficient, etc.), the result of recalculating the weighting coefficient, and the operation data classification result (time, cluster number, degree of abnormality, etc.) as the clustering result. Is stored in the calculation result DB 270.
- step S17 the diagnosis unit 240 determines whether an abnormality has occurred in the plant 100 (whether the attribute of the classified cluster is abnormal). If the diagnosis unit 240 determines that an abnormality has occurred, it moves the process to step S17, and if it determines that no abnormality has occurred, it moves the process to step S18.
- step S18 the external output interface unit 220 generates screen information including information indicating that an abnormality has been detected, such as “An abnormality has been detected”, and transmits the screen information to the display device 320 (screen display).
- step S19 the diagnosis unit 240 determines whether or not there has been a stop instruction from the operator. If the diagnosis unit 240 determines that there is an instruction to stop, the diagnosis unit 240 ends the diagnosis process. If it is determined that there is no instruction, the diagnosis unit 240 proceeds to step S10.
- FIG. 4 is a diagram illustrating an example of a configuration in a case where adaptive resonance theory (ART) is used as an example of the clustering unit 230.
- ART adaptive resonance theory
- the ART module 410 includes an F0 layer 411, an F1 layer 412, an F2 layer 413, a memory 414, and a selection subsystem 415, which are coupled to each other.
- the F1 layer 412 and the F2 layer 413 are coupled via a weighting factor.
- the weighting factor represents the prototype (prototype) of the cluster into which the input data is classified.
- the prototype represents a representative value of the cluster.
- the F0 layer 411 normalizes the input vector and removes noise.
- a suitable cluster candidate is selected by comparing the input data input to the F1 layer 412 with a weighting coefficient.
- Kw is a learning rate parameter (0 ⁇ Kw ⁇ 1), and is a value that determines the degree to which the input vector is reflected in the new weighting factor. Note that (Equation 1) and (Equation 2) to (Equation 12) described below are incorporated in the ART module 410.
- the characteristic of the data classification algorithm of the ART module 410 is the above (Process 4).
- (Process 4) when input data different from the learned pattern is input, a new pattern can be recorded without changing the recorded pattern. Therefore, it is possible to record a new pattern while recording a pattern learned in the past.
- the ART module 410 learns the given pattern. Therefore, when new input data is input to the learned ART module 410, it is possible to determine which pattern in the past is close by the above algorithm. A pattern that has never been experienced before is classified as a new cluster.
- FIG. 4B is a diagram illustrating an example of the configuration of the F0 layer 411.
- the input data I i is normalized again at each time to create a normalized input vector u i 0 that is input to the F1 layer 412 and the selection subsystem 415.
- W i 0 is calculated from the input data I i according to (Equation 2).
- a is a constant.
- FIG. 4C is a diagram illustrating an example of the configuration of the F1 layer 412.
- u i 0 obtained in (Expression 5) is held as a short-term memory, and P i to be input to the F2 layer 413 is calculated.
- the calculation formulas for the F1 layer 412 are collectively shown in (Formula 6) to (Formula 12).
- a and b are constants
- f is a function shown in (Equation 4)
- T j is a fitness calculated by the F2 layer 413
- z j i is a weighting factor for the input data I i of the cluster number j. It is.
- g is as follows.
- FIG. 5 is a diagram illustrating an example of a data classification result in the clustering unit 230 and an abnormality degree calculation method in the diagnosis unit 240.
- FIG. 5 shows an example of a classification result obtained by classifying operation data into clusters.
- FIG. 5A as an example, two items of the operation data are shown, and are represented by a two-dimensional graph.
- the vertical axis and the horizontal axis indicate standardized operation data of each item.
- the operation data is divided into a plurality of clusters 500 (circles shown in FIG. 5A) by the ART module 410.
- One circle corresponds to one cluster.
- the operation data is classified into four clusters.
- the cluster 500A with the cluster number “1” is a cluster having a relatively large item A value and a small item B value.
- the cluster 500B with the cluster number “2” is a cluster in which the values of the items A and B are relatively small.
- the cluster 500C with the cluster number “3” is a cluster in which the value of the item A is relatively small and the value of the item B is relatively large.
- the cluster 500D with the cluster number “4” is a cluster in which both the values of the item A and the item B are relatively large.
- FIG. 5B is a diagram illustrating an example of the result of classifying the operation data into clusters and the operation data.
- the horizontal axis is time, and the vertical axis is operation data (measurement signal) or cluster number.
- the operation data is classified into clusters having cluster numbers “1” to “4”.
- clusters with cluster numbers “1” to “3” are generated during the learning period. Therefore, if the operation data is not classified into the clusters having the cluster numbers “1” to “3” during the diagnosis period, a cluster having the cluster number “4” (a new state that is not experienced during normal operation) is generated. Thereby, in step S17 of FIG. 3, it can be determined that an abnormality has occurred in the plant 100 (occurrence of abnormality) based on the cluster number.
- diagnosis unit 240 classifies the operation data 501 as a new cluster (in this example, the cluster having the cluster number “4”), the distance 503 between the operation data 501 and the centroid 502 of the nearest normal cluster is used as the degree of abnormality. calculate. By calculating the degree of abnormality, it is possible to evaluate how far the operation data 501 is from the normal cluster.
- FIG. 6 is a diagram illustrating an example of a processing procedure related to the attribute automatic change processing executed by the cluster attribute setting unit 251.
- the attribute automatic change process operates in step S14 in FIG.
- the cluster attribute setting unit 251 has a plurality of attribute change methods, and the user can select an attribute change method to be used as an option.
- step S20 the cluster attribute setting unit 251 determines whether to use the first attribute change option. If the cluster attribute setting unit 251 determines to use the cluster attribute setting unit 251, the process proceeds to step S ⁇ b> 21.
- the cluster attribute setting unit 251 appears as the attribute change method of the first attribute change option when processing operation data from the present to a predetermined period before (an example of a predetermined period).
- the generated cluster is set as a normal cluster, and a cluster other than the normal cluster is set as an abnormal cluster. More specifically, the cluster attribute setting unit 251 changes the attribute value (attribute value) in the table shown in FIG. For example, in the table shown in FIG. 2B, when the operation data of the cluster having the cluster number “1” does not appear in a predetermined period, the attribute value is changed from “normal” to “abnormal”.
- the predetermined period has been described as an example of a period from the present to a predetermined period before, but is not limited to this, for example, other periodical inspections to a certain period before A period may be adopted.
- the events in the table shown in FIG. 2B may be configured to be automatically updated, manually updated, or automatically and manually updated. Also good.
- step S22 the cluster attribute setting unit 251 determines whether to use the second attribute change option. If the cluster attribute setting unit 251 determines to use the cluster attribute setting unit 251, it moves the process to step S ⁇ b> 23.
- the cluster attribute setting unit 251 extracts device information adjusted in the periodic inspection.
- the cluster attribute setting unit 251 extracts device information with reference to data such as a periodic inspection plan and a result report as devices adjusted in the periodic inspection.
- step S24 the cluster attribute setting unit 251 identifies a group related to the data input to the device extracted in step S23, and processing data for a certain period (an example of a predetermined period) from the periodic inspection is processed for the group.
- the cluster that appears occasionally is set as a normal cluster, and the cluster other than the normal cluster is set as an abnormal cluster.
- the processing amount can be reduced by specifying and processing a group (for example, a group located downstream of the group) that is affected by the device adjusted in the periodic inspection. .
- the normal cluster and the abnormal cluster are appropriately reset, so that the occurrence of misreporting and misreporting can be suppressed.
- FIG. 7 is a diagram illustrating an example of processing contents performed by the cluster attribute setting unit 251.
- the operation data (learning data) of the learning period is processed by the clustering unit 230, and the clusters with the cluster numbers “1” to “3” are generated as normal clusters.
- the characteristics of the plant change, and only the cluster with the cluster number “3” is generated.
- only the cluster with the cluster number “3” is considered to be a normal cluster.
- the conventional method learns that the clusters with the cluster numbers “1” to “3” are normal. Unable to detect the occurrence of anomalies and a false alarm occurs.
- the cluster attribute setting unit 251 is operated to perform the diagnosis, so that only the cluster with the cluster number “3” is automatically changed to the normal cluster (normal state), thereby avoiding misreporting. Abnormalities (abnormal signs) can be detected.
- the abnormality can be detected by using the diagnostic device 200 in the case where the range that has been learned as normal in the past when the plant characteristics change due to secular change, periodic inspection, etc. becomes abnormal now. become.
- FIG. 8 is a diagram illustrating an example of a processing procedure related to post-attribute-change processing executed by the post-cluster-attribute-change processing unit 252.
- the post-attribute change process operates as a post-process after the cluster attribute setting unit 251 is operated in step S14 of FIG.
- the cluster attribute change post-processing unit 252 is provided with a plurality of post-processing methods, and the user can select the post-processing method to be used as an option.
- step S30 the cluster attribute change post-processing unit 252 determines whether to use the first post-processing option. If the cluster attribute change post-processing unit 252 determines that the cluster attribute is to be used, the process proceeds to step S31. If it is determined that the cluster attribute is not used, the process proceeds to step S32.
- step S31 the post-cluster attribute change processing unit 252 recalculates the degree of abnormality, which is the distance between the operation data and the normal cluster, for the past operation data. Since the calculation method of the degree of abnormality is the same as the calculation method in the diagnosis unit 240, the description thereof is omitted. The recalculated result will be described later with reference to FIG.
- step S32 the cluster attribute change post-processing unit 252 determines whether to use the second post-processing option. If the cluster attribute post-change processing unit 252 determines that the cluster attribute is to be used, the process proceeds to step S33, and if it is determined not to be used, the process proceeds to step S34.
- step S33 the cluster attribute change post-processing unit 252 adjusts the weighting factor of the cluster associated with the event.
- the cluster attribute change post-processing unit 252 adjusts the change width with the normal cluster so that it matches the change width of the operation data when an abnormality has occurred in the past. For a specific adjustment method, see FIG. A) will be described later.
- step S34 the cluster attribute change post-processing unit 252 determines whether or not to use the third post-processing option. If it is determined that the third post-processing option is used, the process proceeds to step S35. End post-change processing.
- step S35 the cluster attribute change post-processing unit 252 adjusts the weighting factor in consideration of the boundary with the abnormality for the cluster whose attribute has been changed normally.
- the cluster attribute change post-processing unit 252 calculates the distance between the operation data belonging to the cluster changed to the normal cluster and the nearest existing normal cluster, and the operation data having the maximum distance is the boundary between normal and abnormal
- the cluster weight coefficient is adjusted so that a specific adjustment method will be described later with reference to FIG.
- the post-attribute-change process recalculates the degree of abnormality, so that past data can be properly grasped. Further, according to the attribute post-processing, the weight coefficient is adjusted and the cluster is more appropriately defined, so that an abnormality in the plant 100 can be detected with higher accuracy.
- FIG. 9 is a diagram illustrating an example of a result of adjusting the cluster weighting coefficient by the operation of the cluster attribute change post-processing unit 252.
- the adjustment content of the weighting factor in step S33 is shown in FIG. 9A, and the adjustment content of the weighting factor in step S35 is shown in FIG. 9B.
- clusters with cluster numbers “1” and “2” are normal clusters
- clusters with cluster numbers “3” and “4” are abnormal clusters
- clusters with cluster number “4” are sensors.
- a drift abnormality (abnormality in which a sensor malfunctions and a certain value is raised) is associated as an event (see the left figure in FIG. 9A).
- the change width 902 with the normal cluster nearest to the cluster is the operation data when the sensor drift occurs.
- the weight coefficient of the cluster number “4” is adjusted so that the change width 901 becomes (see the right diagram in FIG. 9A).
- the cluster attribute change post-processing unit 252 first adds to the existing clusters (clusters with cluster numbers “1”, “2”, and “3”) due to the drift abnormality that increases the value of the operation data.
- Cluster cluster number “4”
- the attribute of the second cluster cluster number “3”
- the cluster attribute setting unit 251 When changed, the change width (change width 901) in the first cluster and the third cluster (cluster of cluster number “2”) in which the attribute before drift abnormality corresponding to the second cluster is normal; Weight coefficient of the first cluster so that the change width (change width 902) between the cluster after adjustment of the weight coefficient indicating the center of the first cluster and the second cluster are equal. Adjust to.
- clusters with cluster numbers “1”, “2”, and “3” are normal clusters, and clusters with cluster number “4” are abnormal clusters (the left side of FIG. 9B). (See figure). From this state, when the cluster with the cluster number “4” is changed to a normal cluster by the cluster attribute setting unit 251, the post-cluster attribute change processing unit 252 and the existing normal cluster closest to each data belonging to the normal cluster And the cluster weighting coefficient is adjusted so that the data with the maximum distance (data 903 in this example) is the boundary between normal and abnormal (right diagram in FIG. 9B). reference).
- FIG. 10 is a diagram illustrating an example of a result of recalculating the degree of abnormality (a screen for confirming the processing result by the cluster attribute change post-processing unit 252) by the operation of the cluster attribute change post-processing unit 252.
- the abnormality degree is recalculated by the cluster attribute change post-processing unit 252 using the first post-processing option. Results are shown.
- the calculation result of the abnormality level is corrected, and the abnormality level corresponding to the cluster whose attribute has been changed normally decreases. That is, when referring to past data, the abnormal degree of the normal cluster is not high and the abnormal degree of the abnormal cluster is not low.
- the display device 320 may be configured to display both the diagnosis result before the attribute change and the diagnosis result after the attribute change, or may display either one.
- time for processing the learning data and generating a normal cluster is not required each time diagnosis is performed, and the occurrence of misreporting can be suppressed.
- information such as programs, tables, and files for realizing each function is stored in a memory, a hard disk, a storage device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD. Can be put in.
- SSD Solid State Drive
- Diagnosis system 100 ... Plant, 200 ... Diagnosis apparatus, 210 ... External input interface part, 220 ... External output interface part, 230 ... Clustering part, 240 ... Diagnosis part, 250 ... Calculation result Correction unit 251 ... Cluster attribute setting unit, 252 ... Cluster attribute change post-processing unit, 260 ... Operation data DB, 270 ... Calculation result DB, 300 ... External device.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Automation & Control Theory (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Probability & Statistics with Applications (AREA)
- Pure & Applied Mathematics (AREA)
- Algebra (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
図1において、1は全体として第1の実施の形態による診断システムを示す。診断システム1では、クラスタの属性を適切に自動変更することにより、失報の発生を抑制する。また、診断システム1では、既存のクラスタの属性の設定を変更することにより、診断の都度、学習データを処理して正常クラスタを生成させる処理は不要となる。以下では、かかる構成について、適宜に例を挙げて説明する。
F0レイヤー411により、入力ベクトルを正規化し、ノイズを除去する。
F1レイヤー412に入力された入力データと重み係数との比較により、ふさわしいクラスタの候補を選択する。
選択サブシステム415により選択されたクラスタの妥当性がパラメータρとの比により評価される。妥当であると判断された場合、入力データは、そのクラスタに分類され、(処理4)に進む。一方、妥当であると判断されない場合、そのクラスタはリセットされ、他のクラスタからふさわしいクラスタの候補を選択する((処理2)を繰り返す)。パラメータρの値を大きくするとクラスタの分類が細かくなる。すなわち、クラスタサイズが小さくなる。逆に、ρの値を小さくすると分類が粗くなる。クラスタサイズが大きくなる。このパラメータρをビジランス(vigilance)パラメータと呼ぶ。
(処理2)において、全ての既存のクラスタがリセットされると、入力データが新規クラスタに属すると判断され、新規クラスタのプロトタイプを表す新しい重み係数が生成される。
入力データがクラスタJに分類されると、クラスタJに対応する重み係数WJ(new)は、過去の重み係数WJ(old)および入力データp(または入力データから派生したデータ)を用いて下記の(式1)により更新される。
なお上述の実施の形態においては、本発明を診断装置200に適用するようにした場合について述べたが、本発明はこれに限らず、この他種々の診断装置および診断方法に広く適用することができる。
Claims (9)
- 設備の運転データに基づいて前記設備における異常を診断する診断装置であって、
既存のクラスタに前記設備の運転データを分類できるか否かを判定し、分類できないと判定した場合、属性を異常としたクラスタを前記既存のクラスタに追加するクラスタリング部と、
前記クラスタリング部による処理結果に基づいて、前記設備における異常を診断する診断部と、
所定の期間における前記設備の運転データに基づいて、前記既存のクラスタの属性を設定するクラスタ属性設定部と、
を備えることを特徴とする診断装置。 - 前記クラスタ属性設定部は、前記既存のクラスタついて、前記所定の期間における前記設備の運転データが分類されるクラスタの属性を正常に設定し、正常に設定したクラスタ以外のクラスタの属性を異常に設定する、
ことを特徴とする請求項1に記載の診断装置。 - 前記設備を構成する各機器に入力するデータのグループごとに既存のクラスタが設けられ、
前記クラスタ属性設定部は、定期検査から一定期間または現在から一定期間前までにおける前記設備の運転データに基づいて、前記定期検査にて調整した機器に入力するデータに関わるグループにおける既存のクラスタの属性を設定する、
ことを特徴とする請求項1に記載の診断装置。 - 前記クラスタ属性設定部により前記既存のクラスタの属性が変更された際に、過去の運転データについて、運転データと属性が正常であるクラスタとの距離である異常度を再計算するクラスタ属性変更後処理部を備える、
ことを特徴とする請求項1に記載の診断装置。 - 運転データの値がかさ上げされるドリフト異常により、前記既存のクラスタに第1のクラスタが追加され、かつ、前記第1のクラスタに対応する前記ドリフト異常前の第2のクラスタの属性が前記クラスタ属性設定部により正常に変更された場合、前記第1のクラスタと前記第2のクラスタに対応する前記ドリフト異常前の属性が正常である第3のクラスタとにおける変化幅と、前記第1のクラスタの中心を示す重み係数の調整後のクラスタと前記第2のクラスタとにおける変化幅とが等しくなるように、前記重み係数を調整するクラスタ属性変更後処理部を備える、
ことを特徴とする請求項1に記載の診断装置。 - 前記クラスタ属性設定部により属性が正常に変更されたクラスタに属する複数の運転データの各々と、前記既存のクラスタのうち最近接の属性が正常であるクラスタとの距離を計算し、前記複数の運転データのうち距離が最大となる運転データが正常と異常との境界に位置するように、前記正常に変更されたクラスタの中心を示す重み係数を調整するクラスタ属性変更後処理部を備える、
ことを特徴とする請求項1に記載の診断装置。 - 前記クラスタ属性設定部によりクラスタの属性が変更された結果を確認するための画面に係る画面情報を出力する外部出力インターフェイス部を備える、
ことを特徴とする請求項1に記載の診断装置。 - 前記クラスタ属性変更後処理部による処理結果を確認するための画面に係る画面情報を出力する外部出力インターフェイス部を備える、
ことを特徴とする請求項4から6の何れか1項に記載の診断装置。 - 設備の運転データに基づいて前記設備における異常を診断する診断方法であって、
クラスタリング部が、既存のクラスタに前記設備の運転データを分類できるか否かを判定し、分類できないと判定した場合、属性を異常としたクラスタを前記既存のクラスタに追加する第1のステップと、
診断部が、前記クラスタリング部による処理結果に基づいて、前記設備における異常を診断する第2のステップと、
クラスタ属性設定部が、所定の期間における前記設備の運転データに基づいて、前記既存のクラスタの属性を設定する第3のステップと、
を備えることを特徴とする診断方法。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201880084942.9A CN111542792A (zh) | 2018-03-01 | 2018-12-05 | 诊断装置和诊断方法 |
| EP18907849.6A EP3761133A4 (en) | 2018-03-01 | 2018-12-05 | DIAGNOSTIC DEVICE AND DIAGNOSTIC PROCEDURE |
| PH12020551136A PH12020551136A1 (en) | 2018-03-01 | 2020-07-27 | Diagnosis apparatus and diagnosis method |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2018037007A JP7180985B2 (ja) | 2018-03-01 | 2018-03-01 | 診断装置および診断方法 |
| JP2018-037007 | 2018-03-01 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2019167375A1 true WO2019167375A1 (ja) | 2019-09-06 |
Family
ID=67805749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2018/044793 Ceased WO2019167375A1 (ja) | 2018-03-01 | 2018-12-05 | 診断装置および診断方法 |
Country Status (5)
| Country | Link |
|---|---|
| EP (1) | EP3761133A4 (ja) |
| JP (1) | JP7180985B2 (ja) |
| CN (1) | CN111542792A (ja) |
| PH (1) | PH12020551136A1 (ja) |
| WO (1) | WO2019167375A1 (ja) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114365125A (zh) * | 2019-09-11 | 2022-04-15 | 日立造船株式会社 | 信息处理装置、运转辅助系统、信息处理方法、以及信息处理程序 |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7483341B2 (ja) * | 2019-09-26 | 2024-05-15 | キヤノン株式会社 | 情報処理方法、情報処理装置、機械設備、物品の製造方法、プログラム、記録媒体 |
| WO2021199194A1 (ja) * | 2020-03-30 | 2021-10-07 | 三菱電機株式会社 | 診断システム、診断方法及びプログラム |
| WO2021229815A1 (ja) * | 2020-05-15 | 2021-11-18 | 富士通株式会社 | 情報処理装置、評価方法、および評価プログラム |
| DE112021005251T5 (de) * | 2020-10-05 | 2023-09-07 | Fanuc Corporation | Zustandsbestimmungsvorrichtung und zustandsbestimmungsverfahren |
| JP7743432B2 (ja) * | 2020-11-16 | 2025-09-24 | 住友重機械工業株式会社 | 表示装置、制御装置、制御方法及びコンピュータプログラム |
| JP7624382B2 (ja) * | 2021-12-24 | 2025-01-30 | 株式会社日立製作所 | 点検対象物情報取得システム及び点検対象物情報取得方法 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070088550A1 (en) * | 2005-10-13 | 2007-04-19 | Dimitar Filev | Method for predictive maintenance of a machine |
| JP2010237893A (ja) | 2009-03-31 | 2010-10-21 | Hitachi Ltd | プラントの異常診断装置及び異常診断方法 |
| JP2015018389A (ja) | 2013-07-10 | 2015-01-29 | 株式会社日立パワーソリューションズ | 異常予兆診断装置及び異常予兆診断方法 |
| JP2015162140A (ja) * | 2014-02-28 | 2015-09-07 | 株式会社日立製作所 | プラントと制御装置の診断装置 |
| JP2016033778A (ja) * | 2014-07-31 | 2016-03-10 | 株式会社日立パワーソリューションズ | 異常予兆診断装置及び異常予兆診断方法 |
| WO2017022784A1 (ja) * | 2015-08-05 | 2017-02-09 | 株式会社日立パワーソリューションズ | 異常予兆診断システム及び異常予兆診断方法 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5150590B2 (ja) * | 2009-09-25 | 2013-02-20 | 株式会社日立製作所 | 異常診断装置及び異常診断方法 |
| CN102576227B (zh) * | 2009-10-02 | 2014-06-25 | 株式会社日立制作所 | 设备诊断装置 |
| JP5544418B2 (ja) * | 2010-04-08 | 2014-07-09 | 株式会社日立製作所 | プラントの診断装置、診断方法、及び診断プログラム |
| JP5647626B2 (ja) * | 2012-01-05 | 2015-01-07 | 株式会社日立製作所 | プラント状態監視装置、プラント状態監視方法 |
| JP5530019B1 (ja) * | 2013-11-01 | 2014-06-25 | 株式会社日立パワーソリューションズ | 異常予兆検知システム及び異常予兆検知方法 |
| JP6116466B2 (ja) * | 2013-11-28 | 2017-04-19 | 株式会社日立製作所 | プラントの診断装置及び診断方法 |
| CN103676937B (zh) * | 2013-12-10 | 2017-05-10 | 中联重科股份有限公司 | 控制器中控制参数的处理方法及装置 |
| JP2016012158A (ja) * | 2014-06-27 | 2016-01-21 | 株式会社日立製作所 | プラント異常予知装置、異常予知装置、表示装置及び表示方法 |
| CN105825242B (zh) * | 2016-05-06 | 2019-08-27 | 南京大学 | 基于混合网格分层聚类的集群通信终端轨迹实时异常检测方法与系统 |
-
2018
- 2018-03-01 JP JP2018037007A patent/JP7180985B2/ja active Active
- 2018-12-05 WO PCT/JP2018/044793 patent/WO2019167375A1/ja not_active Ceased
- 2018-12-05 CN CN201880084942.9A patent/CN111542792A/zh active Pending
- 2018-12-05 EP EP18907849.6A patent/EP3761133A4/en not_active Withdrawn
-
2020
- 2020-07-27 PH PH12020551136A patent/PH12020551136A1/en unknown
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070088550A1 (en) * | 2005-10-13 | 2007-04-19 | Dimitar Filev | Method for predictive maintenance of a machine |
| JP2010237893A (ja) | 2009-03-31 | 2010-10-21 | Hitachi Ltd | プラントの異常診断装置及び異常診断方法 |
| JP2015018389A (ja) | 2013-07-10 | 2015-01-29 | 株式会社日立パワーソリューションズ | 異常予兆診断装置及び異常予兆診断方法 |
| JP2015162140A (ja) * | 2014-02-28 | 2015-09-07 | 株式会社日立製作所 | プラントと制御装置の診断装置 |
| JP2016033778A (ja) * | 2014-07-31 | 2016-03-10 | 株式会社日立パワーソリューションズ | 異常予兆診断装置及び異常予兆診断方法 |
| WO2017022784A1 (ja) * | 2015-08-05 | 2017-02-09 | 株式会社日立パワーソリューションズ | 異常予兆診断システム及び異常予兆診断方法 |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP3761133A4 |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114365125A (zh) * | 2019-09-11 | 2022-04-15 | 日立造船株式会社 | 信息处理装置、运转辅助系统、信息处理方法、以及信息处理程序 |
| EP4030249A4 (en) * | 2019-09-11 | 2023-10-04 | Hitachi Zosen Corporation | INFORMATION PROCESSING DEVICE, OPERATION SUPPORT SYSTEM, INFORMATION PROCESSING METHOD, AND INFORMATION PROCESSING PROGRAM |
Also Published As
| Publication number | Publication date |
|---|---|
| EP3761133A1 (en) | 2021-01-06 |
| EP3761133A4 (en) | 2021-11-03 |
| JP7180985B2 (ja) | 2022-11-30 |
| JP2019153018A (ja) | 2019-09-12 |
| CN111542792A (zh) | 2020-08-14 |
| PH12020551136A1 (en) | 2021-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019167375A1 (ja) | 診断装置および診断方法 | |
| JP5484591B2 (ja) | プラントの診断装置及びプラントの診断方法 | |
| CN112655004B (zh) | 用于异常检测和/或预测性维护的计算机实现的方法 | |
| JP5292477B2 (ja) | 診断装置及び診断方法 | |
| WO2021071911A1 (en) | Sensor contribution ranking | |
| CN114450646A (zh) | 用于使用深度学习来检测风力涡轮操作异常的系统和方法 | |
| JP5150590B2 (ja) | 異常診断装置及び異常診断方法 | |
| JP2017010232A (ja) | プラント診断装置及びプラント診断方法 | |
| CN110084326A (zh) | 一种基于模糊集的工业设备异常检测方法 | |
| JP6830414B2 (ja) | 診断装置及び診断方法 | |
| CN117330906A (zh) | 设备电弧故障检测方法、装置、设备及存储介质 | |
| CN117499103A (zh) | 一种用户异常行为检测方法、系统、装置及可读存储介质 | |
| CN116611022A (zh) | 智慧校园教育大数据融合方法及平台 | |
| JP2022186541A (ja) | ソースコード解析装置およびソースコード解析方法 | |
| CN115146890A (zh) | 企业运营风险告警方法、装置、计算机设备和存储介质 | |
| JP5647626B2 (ja) | プラント状態監視装置、プラント状態監視方法 | |
| CN114358421A (zh) | 磁盘故障预测方法、装置、存储介质和电子设备 | |
| CN118410463B (zh) | 一种基于数据行为偏执的行为异常检测方法及系统 | |
| CN119939277A (zh) | 设备故障识别方法、系统及储存介质 | |
| CN116032016B (zh) | 智慧风电场运行状态在线监测预警系统及预警方法 | |
| WO2021140942A1 (ja) | 診断装置、診断方法およびプログラム | |
| JP6676508B2 (ja) | セキュリティ診断装置およびセキュリティ診断方法 | |
| CN119720051B (zh) | 基于工业自动化的设备预测性维护遥测系统 | |
| JP2020135739A (ja) | 障害予兆検知システム | |
| CN117215868A (zh) | 一种磁盘异常检测方法、装置、电子设备及存储介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 18907849 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2018907849 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2018907849 Country of ref document: EP Effective date: 20201001 |
|
| WWW | Wipo information: withdrawn in national office |
Ref document number: 2018907849 Country of ref document: EP |







