JP2005063385A - Monitoring method, monitoring apparatus and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a monitoring method, a monitoring apparatus and a program that can diagnose a monitored object that acquires data in real time. <P>SOLUTION: Operation data on a monitored object are acquired and recorded as time series data in a time series database (Step S1). The operation data are next vector-quantized into histograms in an observation window to create time series indexes, and attribute information is input (Step S2, S3). Time series data (reference data) to be retrieved from the time series database are decided, and a search window having the same size as the reference data is used to search the time series indexes (input data) for a similar one (Step S4, S5). If a similar time series index is present, the attribute information about the time series index is displayed (Step S6: YES, Step S7). If not, attribute information is input (Step S6: NO, Step S8). <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a monitoring method, a monitoring apparatus, and a program for quickly diagnosing a defect or the like by using past operation data accumulated for a monitoring target such as a plant.

  In general, in many facilities that require continuous operation, such as various plants, production facilities, or air conditioning facilities installed in buildings, it is difficult to avoid problems such as failures in the installed devices. However, in such equipment, since continuous operation is required, it is required to take a quick response in order to minimize the apparatus stop period. Therefore, conventionally, a technique for estimating the cause of a malfunction by extracting an appropriate similar case that occurred in the past when a malfunction has occurred with such a facility as a monitoring target has been developed.

  For example, in Patent Document 1, for example, when a failure occurs in a monitoring target, process data from the monitoring target at the time of the failure occurrence is qualitatively processed, and the obtained qualitative data and qualification data of all cases are A technique for calculating similarity and extracting case data in descending order of similarity is disclosed. Here, the process data qualification process is performed by calculating the crest factor and the fluctuation rate of the process data. The crest factor is an index indicating the presence or absence of a protruding portion of the waveform in the upward and downward directions, and is a value obtained by dividing the maximum value or the minimum value by the average value. The fluctuation rate is an index indicating the tendency of change, and is a value obtained by dividing the deviation value by the average value. The similarity is calculated by the number of coincidence between the crest rate and the fluctuation rate.

Japanese Patent Laid-Open No. 2003-15877

  However, in the technique of Patent Document 1, the maximum value, the minimum value, the average value, and the deviation value are required when calculating the crest factor and the fluctuation rate in the qualitative processing, and in order to calculate these numerical values, All process data to be determined must be confirmed. Therefore, when a failure or the like occurs in the monitoring target, it takes a long time to specify the process data at the time of the failure and perform qualitative processing. As a result, since it takes a long time to obtain a failure diagnosis result, it is difficult to deal with a monitoring target that needs to acquire data in real time and immediately resolve the failure.

  The present invention has been made in view of the above problems, and provides a monitoring method, a monitoring apparatus, and a program capable of performing diagnosis on a monitoring target for acquiring data in real time.

Means and effects for solving the problems

  In order to solve the above-described problems, a monitoring method according to the present invention includes a step of recording operation data to be monitored in a database as time-series data, and at the same time each time related to an operation data pattern of the operation data is recorded in the database. Generating and recording an index for each, determining a time-series data to be diagnosed with reference to the database, reading an index of the time-series data to be diagnosed, and using the index to perform the diagnosis And a step of searching for an index having an operation data pattern similar to the target.

  The monitoring apparatus according to the present invention records the operation data to be monitored in the database as time-series data, and records and records in the database, and simultaneously generates and records an index for each operation data pattern of the operation data. Means, referring to the database, determining time-series data to be diagnosed, reading an index of the time-series data to be diagnosed, and using the index, an operation data pattern similar to the diagnosis object is read out And a means for searching for an index.

  The program according to the present invention is a step of recording monitoring operation data as time series data in a database, and simultaneously recording in the database, generating and recording an index for each time related to an operation data pattern of the operation data, Referring to the database, determining time-series data to be diagnosed, reading an index of the time-series data to be diagnosed, and searching for an index having an operation data pattern similar to the diagnosis object by using the index And causing the computer to execute.

  According to the monitoring method, the monitoring apparatus, and the program of the present invention, a search index is created and recorded in real time while collecting operation data to be monitored as time series data. At the time of diagnosis, an index of time series data corresponding to the time of diagnosis is read, and an index of past time series data similar to the diagnosis target is searched using this index. As a result, since the indexes recorded in advance are compared with each other, the similarity between the two indexes can be obtained in a short time, so that the diagnosis can be performed in a short time. Therefore, even when diagnosing a monitoring target that needs to acquire driving data in real time and immediately solve the problem, it is possible to cope with it sufficiently.

  Moreover, it is preferable that the monitoring method, the monitoring apparatus, and the program according to the present invention further cause the index to hold attribute information and display the attribute information held by the searched index.

  According to this configuration, the attribute information is held in the index, and the attribute information held in the searched index is displayed. Therefore, by providing necessary information for each application such as a defect as attribute information, an accurate diagnosis can be performed.

  Further, in the monitoring method, the monitoring apparatus, and the program according to the present invention, the index is a histogram obtained by vector quantization of operation data within a predetermined section from each time, and an index having an operation data pattern similar to the diagnosis target In the search, the similarity between the histogram of the time specified with a predetermined skip width from all the times of the database and the histogram of the time of the diagnosis target is calculated, and the time of the histogram having the highest similarity is output. preferable.

  According to this configuration, an index is created using the vector quantization technique and the histogram creation technique for the operation data within a predetermined section from each time, and the similarity is calculated by comparing the histograms. Therefore, an index can be created in real time, and a quick diagnosis can be performed based on the histogram.

  In the monitoring method, the monitoring apparatus, and the program according to the present invention, it is preferable that the predetermined skip width is set to be narrower as the similarity is higher.

  According to this configuration, the higher the similarity is, the narrower the skip width is set. Accordingly, the time of the histogram similar to the diagnosis target can be searched at high speed by performing a search in detail for a section having a low similarity and a rough search for a section having a high similarity.

  Moreover, it is preferable that the monitoring method, the monitoring apparatus, and the program according to the present invention further generate a feature vector from the operation data, and the index is generated using the feature vector.

  According to this configuration, the index is generated using the feature vector generated from the operation data. Therefore, the present invention can also be applied to a monitoring target having a plurality of operation data at the same time such as image data.

  The program according to the present invention is a removable recording medium such as a CD-ROM (Compact Disc Read Only Memory), FD (Floppy (registered trademark) Disk), or MO (Magneto-Optic), or a fixed recording medium such as a hard disk. In addition to being able to be recorded and distributed, it can be distributed via a communication network such as the Internet by wired or wireless telecommunication means.

  The best mode for carrying out the present invention will be described below with reference to the drawings. In the present embodiment, the operation state of the heat pump is monitored, and when a failure occurs, a past similar failure is detected, and the cause of the failure is analyzed and diagnosed immediately, and the countermeasure is determined. A monitoring device that supports this is assumed.

First, the monitoring apparatus according to the present embodiment will be described with reference to FIGS.
FIG. 1 shows a block diagram of a monitoring apparatus according to the present embodiment. FIG. 2 is a diagram illustrating time series data search by a time series data search unit.

The monitoring apparatus 100 includes an operation data collector 110, an index generator 130, a time series data searcher 140, a data attribute operation unit 150, and a data display unit 160. The monitoring apparatus 100 according to the present embodiment is roughly divided into a data generation unit and a data utilization unit. In the data generation unit, an operation data collector 110 and an index generator 130 are used. The data utilization unit uses a time series data search unit 140, a data attribute operation unit 150, and a data display unit 160.
In the present embodiment, the operating device 200 to be monitored is a heat pump.

The operation data collector 110 is for recording the operation data 210 of the operating device 200 in the time series database 111 as time series data. The operation data 210 is data indicating the state of each part of the operating device 200 to be monitored. For example, in a heat pump, the motor current, the supply temperature, the discharge temperature, and the like.
In the present embodiment, the operation data collector 110 is operated data stored in a controller of the monitoring target 200 in a remote place or a communication unit (not shown) provided via the network 400. Collect 210. The collected operation data 210 is stored in the time series database 111 as needed.

  In the time series database 111, the operation data 210 representing the state of the operating device 200 is stored as numerical data, and becomes time series data as shown in Table 1 over a long period of time.

The index generator 130 is for generating an index related to the operation data pattern at each time based on the operation data 210 stored in the time series database 111.
In the present embodiment, the index generator 130 receives the operation data 210 collected by the operation data collector 110, generates an index, and stores it in the time series index 131. Here, a corresponding index always exists for the operation data 210 at a certain time in the time series database 111 and is stored in the time series index 131. The time-series index 131 can also have an ID of attribute information (hereinafter, abbreviated as “attribute ID”) storing the device information of the operating device 200 at that time. Note that the attribute ID is appropriately input by the operator 300 via the data attribute operation unit 150 described later.

Here, index generation by the index generator 130 will be described in more detail.
In the index generator 130, the acquired operation data 210 is divided into numerical values so that the state can be identified by an ID (hereinafter, abbreviated as “state ID”). Here, each prepared state ID is called a bin, and the number of prepared state IDs is called a bin number. For example, if the state is identified by the boundary values as shown in Table 2, six bins are prepared. Specifically, a data table of a state ID column and a numerical value column of operation data 210 is prepared. Here, the state ID column is divided into six stages of bin numbers “h1” to “h6” which are state IDs. In the numerical value column, the numerical value range of the operation data 210 is set for each section. That is, “0.0 or less”, “0.0 to 1.0”, “1.0 to 2.0”, “2.0 to 3.0”, “3.0 to 4.0” and “ Six types of ranges of “4.0 or more” are set corresponding to each of the six sections. Then, for example, if the numerical value of the acquired operation data 210 is “2.5”, it is determined that the state is “h4” based on the data table of Table 2, and “2.5” of the operation data 210 is set. Vector quantization is performed with the state ID “h4”. The boundary value (state ID) for identification can be determined for each application.

  Then, an attention window having a predetermined time width is set, vector quantization is performed on all operation data 210 in the section, and a histogram is created by counting how many state IDs exist. . For example, when an attention window having a time width of 50 is set, the breakdown of the number of states of the 50 operation data 210 (which state ID in Table 2 is classified) is shown below. Count. That is, in the histogram creation at the time Ti, the state of h1 to h6 that is the state ID of Table 2 is counted for 50 data from the past time Ti-50 to the current time Ti. Similarly, the histogram creation at time Ti + 1 is performed for 50 data from time Ti-49 to time Ti + 1. Then, for example, a histogram as shown in Table 3 is created.

  In the present embodiment, the histogram for each time series data thus created is stored as the time series index 131. That is, the acquired time series data always has a histogram, and this histogram has a structure in which the time series data is stored in the time series database 111 and simultaneously recorded as the time series index 131.

The time series data search unit 140 searches for an index (histogram) having an operation data pattern similar to the time determined as time series data to be diagnosed by the operator 300 in accordance with a command from the data attribute operation unit 150 described later. Is for.
In the present embodiment, a histogram of a predetermined time determined as time series data to be diagnosed by the operator 300 is acquired from the time series index 131 as reference data, and the past histogram recorded in the time series index 131 is acquired. Is searched for whether there is a histogram similar to the input data in the past. The reference data is not limited to the time series data in the attention window having the predetermined time width set by the index generator 130 described above, and may be time series data in a time width that is an integral multiple of the attention window.

Here, the search of the time series data by the time series data search unit 140 will be described in more detail based on FIG.
As shown in FIG. 2, when the interval of the reference data is longer than the attention window set by the index generator 130, a search window for input data having the same time width as that interval is obtained by connecting a plurality of attention windows. Set. Then, the similarity is calculated for each histogram of the reference data and the input data in order by moving the search position of the search window with respect to the input data.

  Here, the calculation of the similarity is performed for each attention window. This similarity means the overlapping ratio of histograms and takes a numerical value of 0.0 to 1.0.

  Furthermore, the similarity in the entire search window is defined as follows.

  Then, the lowest numerical value among the similarities of the respective attention windows is set as the similarity S of the search windows. When this numerical value is high, the similarity between the input data in the search window and the reference data is high. On the other hand, when this numerical value is low, it means that there are few similar patterns of input data and reference data in the search window. Therefore, when the next search time is determined, the search window is moved with the following skip width w. That is, in the input data, the search window is largely moved in the vicinity of a low similarity, and the search window is moved in a small vicinity in the vicinity of a high similarity to perform a detailed investigation.

  As described above, the search window is moved with respect to the input data until the similarity S becomes equal to or higher than the set value, and a search is performed for data similar to the reference data.

The data attribute manipulator 150 is for causing the time series index 131 to hold attribute information. Here, the attribute information is information necessary for each application, and can be freely set by the operator 300. In addition, the data attribute operation unit 150 receives the time determined as time series data to be diagnosed from the operator 300 and also transmits a command to the time series data operation unit 140.
In the present embodiment, an attribute ID is added to the time series index 131 and stored. If the monitoring device 100 is a failure diagnosis system, as shown in the following Table 4, the failure cause and countermeasures are described in advance in the attribute database 151 as attribute information corresponding to the attribute ID. This information of past similar time-series data searched by the data searcher 140 can be used for failure diagnosis.

The data display unit 160 is for displaying information of the time series database 111 and information of the attribute database 151 to the operator 300.
In the present embodiment, the data display unit 160 acquires information related to the operation data 210 from the time series database 111 and displays the information on a display or the like of a personal computer. In addition, the data display unit 160 uses a time-series data search device via the data attribute operation device 150 for the time inputted by the operator 300 as the time when the investigation is determined to be necessary from the information regarding the operation data 210 displayed in the graph. 140 retrieves similar time-series data in the past, acquires the attribute ID, and displays attribute information corresponding to the attribute ID. If similar time-series data has not been found in the past, the operator 300 can newly input an attribute ID to the data attribute operation unit 150 as a new pattern.

Next, a processing procedure of the monitoring method according to the present embodiment using the above-described monitoring device will be described with reference to FIGS.
FIG. 3 is a flowchart illustrating the processing procedure of the monitoring method according to the present embodiment. FIG. 4 is a flowchart illustrating in more detail a procedure of processing for searching for a time series index similar to reference data.
Here, a case where an active search method is used as the time series data search process will be described. The active search method can be divided into two stages, a time series index creation stage and a data search stage.

First, in the time series index creation stage, the following processing is performed.
First, as shown in FIG. 3, the operation data to be monitored is collected and recorded in the database as time series data (step S1).
In the present embodiment, the operation data collector 110 collects the operation data 210 of the operating device 200 to be monitored via the network 400 and records it in the time series database 111 as time series data.

Then, after the operation data is vector-quantized, a histogram based on the window of interest is created and a time series index is generated (step S2).
In the present embodiment, in the index generator 130, the operation data 210 collected by the operation data collector 110 is divided into numerical values for every predetermined time width (attention window) (vector quantization), and the above-mentioned A histogram is created by identifying with the state ID and recorded as a time series index 131.

Also, attribute information is added to the time series index as appropriate (step S3).
In the present embodiment, if the operator 300, for example, corresponds to the operation data 210 at the time of failure, the attribute ID of the time series index 131 at the time is the attribute ID described above corresponding to the cause of the failure and the countermeasure. Enter.

Subsequently, in the data search stage, the following processing is performed.
First, time series data to be searched from the time series database is determined as reference data (step S4). The reference data is an integer multiple of the attention window.
In the present embodiment, the operator 300 selects time-series data (reference data) to be searched from the time-series database 111 displayed on the data display unit 160 and notifies the data attribute operator 150 of the time. To do.

Then, a time series index similar to the reference data is searched (step S5).
In the present embodiment, the time-series data search unit 140 searches for the time of the reference data input by the operator 300 to the data attribute operation unit 150.

Here, the process of searching for a time-series index similar to the reference data will be described in more detail based on FIG.
First, the size of the search window is acquired from the reference data (step S501). Note that, as described above, the reference data has a size that is an integral multiple of the window of interest. Then, the search start position of the search window for the input data as the search source is designated (step S502). Next, each of the reference data and the input data in the search window is divided into attention windows (step S503), and a histogram which is a time series index is acquired for each attention window (step S504). Then, the similarity is calculated for each attention window for the histograms of the reference data and the input data (step S505), and the minimum value S of the similarity is calculated (step S506). Next, it is confirmed whether or not the minimum value S exceeds a predetermined value (step S507). When the minimum value S of the similarity does not exceed the predetermined value (step S507: NO), the skip width is calculated (step S508), and the search window search start position for the input data is calculated based on the calculated skip width. (Step S509), and the process returns to Step S503. On the other hand, when the minimum value S of similarity exceeds the predetermined value (step S507: YES), the process is terminated.

As described above, it is confirmed whether or not there is a similar time series index (step S6). If it is confirmed that a similar time series index exists (step S6: YES), attribute information is displayed (step S7), and the process is terminated. On the other hand, when it is not confirmed that a similar time series index exists (step S6: NO), the attribute information is input as a new operation data pattern (step S8), and the process is terminated.
In this embodiment, when it is confirmed that a similar time series index exists, the attribute ID of the time series index 131 corresponding to the time of the input data confirmed to be similar is checked and acquired from the attribute database 151. The attribute information corresponding to the attribute ID is displayed on the data display unit 160. On the other hand, if it is not confirmed that a similar time series index exists, the operator 300 inputs the attribute ID via the data attribute operation unit 150 and stores the attribute ID in the time series index 131 corresponding to the time of the reference data. To remember.

  The execution of the process of the monitoring method according to the present embodiment described above can be read and executed from a recording medium by a CPU (central processing unit) of a computer as a program according to the present embodiment. .

As described above, in the monitoring device 100, the monitoring method, and the program according to the present embodiment, the operation data 210 of the operation device 200 to be monitored by the operation data collector 110 is collected as time series data in the time series database 111. The index generator 130 creates a time series index 131 for search in real time, and the time series data searcher 140 searches past time series data similar to the diagnosis target using the similarity of the time series index 131 as an evaluation criterion. (Steps S1 and S2 and Steps S4 and S5 shown in FIG. 3).
As a result, it is possible to make a diagnosis on a monitoring target for acquiring driving data in real time based on past time series data similar to the diagnosis target.

Further, in the monitoring device 100, the monitoring method, and the program according to the embodiment, the attribute held by the time series index 131 obtained by making the time series index 131 hold the attribute ID and searching the data display unit 160 with reference to the attribute database 151. The attribute information corresponding to the ID is displayed (steps S3 and S7, S8 shown in FIG. 3).
As a result, accurate diagnosis can be performed by providing necessary information for each application such as a defect as attribute information in association with the attribute ID.

In the monitoring device 100, the monitoring method, and the program according to the embodiment, the index generator 130 applies the vector quantization technique and the histogram creation technique to the operation data 210 in a predetermined section (attention window) from each time. The time series index 131 is used to create the similarity, and the time series data searcher 140 calculates the similarity by comparing the histograms (step S5 shown in FIG. 3 and FIG. 4).
As a result, the time-series index 131 can be created in real time, and a quick diagnosis can be performed based on the histogram.

In the monitoring apparatus 100, the monitoring method, and the program according to the embodiment, the time series data searcher 140 sets the skip width of the search window narrower as the similarity is higher (step S508 shown in FIG. 4).
As a result, it is possible to search for the time of the histogram similar to the diagnosis target at high speed by performing a detailed search for a section with low similarity and a detailed search for a section with high similarity.

  The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above-described embodiments, and various modifications can be made as long as they are described in the claims.

  In the above-described embodiment, the heat pump is the operating device 200 to be monitored, but is not limited thereto. For example, waste incinerators, gasification melting furnaces, sludge / ash melting furnaces and other waste treatment plants, blast furnaces / converter and other steel manufacturing plants, water treatment plants, factories with multiple production machines, or A general-purpose machine such as a general-purpose compressor or a construction machine may be used.

In the above-described embodiment, the search is performed using one piece of operation data, but the present invention is not limited to this. For example, a search may be performed using a plurality of operation data. In addition, when the operating device 200 is a camera and the operation data 210 is image data, a process for digitization is necessary. Specifically, as shown in FIG. 5, a feature vector generator 120 is provided in the monitoring apparatus 100, and the operation data 210 collected by the operation data collector 110 by the feature vector generator 120 is, for example, a specific part. The RGB luminance values (red, green, blue) are converted into data and output to the index generator 130.
In this case, the present invention can also be applied to a monitoring target that includes a plurality of operation data at the same time such as image data.

  In the above-described embodiment, the monitoring device 100 acquires the operation data 210 of the operating device 200 via the network 400, but the present invention is not limited to this. For example, the operating device 200 may be directly connected to the monitoring device 100. Further, the operation data collector 110 may receive the operation data 210 from the operating device 200 via a secondary storage medium such as a memory card. Alternatively, the operation data collector 110 may be connected to a mail server, and the operation data 210 may be acquired from an email sent from the operating device 200 to a specific email address.

1 is a block diagram of a monitoring apparatus according to the present embodiment. It is a figure explaining the search of the time series data by a time series data search device. It is the flowchart explaining the procedure of the process of the monitoring method which concerns on this Embodiment. It is the flowchart explaining in detail about the procedure of the process which searches the time series index similar to reference data. It shows a block diagram of a monitoring device according to a modification of the present embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Monitoring apparatus 110 Operation data collector 111 Time series database 120 Feature vector generator 130 Index generator 131 Time series index 140 Time series data searcher 150 Data attribute operation unit 151 Attribute database 160 Data display part 200 Operation apparatus 210 Operation data

Claims (15)

Recording monitored operation data in a database as time-series data;
Generating and recording an index for each time related to the operation data pattern of the operation data simultaneously with recording in the database;
Determining time-series data to be diagnosed with reference to the database; and
Reading an index of time series data of the diagnosis target, and using the index to search for an index having an operation data pattern similar to the diagnosis target;
A monitoring method comprising: Having attribute information in the index;
Displaying attribute information held by the searched index;
The monitoring method according to claim 1, further comprising: The index is a histogram obtained by vector quantization of operation data within a predetermined section from each time,
The step of searching for an index having an operation data pattern similar to the diagnosis target includes:
Calculating the degree of similarity between the histogram of the time specified by a predetermined skip width from all the times of the database and the histogram of the time of the diagnosis target;
Outputting the time of the histogram having the highest degree of similarity;
The monitoring method according to claim 1, further comprising: The monitoring method according to claim 3, wherein the predetermined skip width is set to be narrower as the similarity is higher. Generating a feature vector from the driving data,
The index is
The monitoring method according to claim 1, wherein the monitoring is performed using the feature vector. Means for recording operation data to be monitored in a database as time-series data;
Means for generating and recording an index for each time related to the operation data pattern of the operation data simultaneously with recording in the database;
Means for referring to the database to determine time-series data to be diagnosed;
Means for reading an index of the time-series data of the diagnosis target, and searching for an index having an operation data pattern similar to the diagnosis target using the index;
A monitoring device comprising: Means for holding attribute information in the index;
Means for displaying attribute information held by the searched index;
The monitoring apparatus according to claim 6, further comprising: The index is a histogram obtained by vector quantization of operation data within a predetermined section from each time,
Means for searching for an index having an operation data pattern similar to the diagnosis target,
Means for calculating the similarity between the histogram of the time specified by a predetermined skip width from all the times of the database and the histogram of the time of the diagnosis target;
Means for outputting the time of the histogram having the highest degree of similarity;
The monitoring apparatus according to claim 6 or 7, further comprising: The monitoring apparatus according to claim 8, wherein the predetermined skip width is set to be narrower as the similarity is higher. Means for generating a feature vector from the driving data,
The index is
The monitoring device according to claim 6, wherein the monitoring device is generated using the feature vector. Recording the monitored operation data in the database as time series data,
Generating and recording an index for each time related to the operation data pattern of the operation data simultaneously with recording in the database;
Determining time-series data to be diagnosed with reference to the database;
Reading an index of the time series data of the diagnosis target, and using the index to search for an index having an operation data pattern similar to the diagnosis target;
A program that causes a computer to execute. Causing the index to hold attribute information;
Displaying attribute information held by the searched index;
The program according to claim 11, further causing a computer to execute. The index is a histogram obtained by vector quantization of operation data within a predetermined section from each time,
The step of searching for an index having an operation data pattern similar to the diagnosis target includes:
Calculating a degree of similarity between a histogram of times specified by a predetermined skip width from all times of the database and a histogram of times of the diagnosis target;
Outputting the time of the histogram having the highest degree of similarity;
13. The program according to claim 11, wherein the program is executed by a computer. The program according to claim 13, wherein the predetermined skip width is set to be narrower as the similarity is higher. Generating a feature vector from the driving data;
The index is
The program according to claim 11, wherein the program is generated using the feature vector.

Images