Patent Application
20170261403
The present invention relates to an abnormality detection procedure development apparatus and an abnormality detection procedure development method that support development of an abnormality detection procedure for a mechanical apparatus.
A mechanical apparatus for social infrastructure, such as a turbine for power generation, is required to run for 24 hours per day. In order to maintain a high operating rate of the mechanical apparatus, an unplanned interruption of the mechanical apparatus has to be prevented. In order to do this, there is a need for transition from periodic maintenance that is based on the operating time for a mechanical apparatus in the related art to state monitoring maintenance that suitably performs preventive maintenance based on a state of the mechanical apparatus. In order to realize the state monitoring maintenance, it is important for the state monitoring apparatus to play the role of analyzing operating data which is collected through various sensors that are installed in the mechanical apparatus, according to a predetermined abnormality detection procedure, and of diagnosing an indication of abnormality or a failure of the mechanical apparatus. At this point, the abnormality detection procedure refers to a flow of processing by a computer that processes data which is acquired from one or more sensors and diagnoses the indication and the like of the abnormality of the mechanical apparatus based on a result of the processing.
Incidentally, in order to improve the precision with which the state monitoring apparatus diagnoses the indication of the abnormality of the mechanical apparatus, it is important that the abnormality detection procedure which is used for the diagnosis of the indication is continuously periodically updated in such a manner as to decrease the number of false positives or false negatives on the diagnosis of the indication. It is noted that the false positive refers to a case where a normal state of the mechanical apparatus is diagnosed as being abnormal and that the false negative refers to a case where an abnormal state of the mechanical apparatus is diagnosed as being normal.
In PTL 1, as an example of a technology that develops the abnormality detection procedure for a mechanical system, the following method for creating a uniform quality evaluation of a turbine mechanical system and a system similar to this and for providing an automatic failure diagnosis tool is disclosed: “A process that operates in a computer creates a mechanical unit signature, a mechanical side signature, and a mechanical fleet signature and keeps track of these signatures (refer to blocks 110 and 120 in
PTL 1: JP-A-2005-339558
The technology that is disclosed in PTL 1 leads to a so-called optimization problem that adjustment is automatically made until a compensation parameter for compensating the data variance that is caused by the ambient condition and the fuel quality for the mechanical system, a threshold for a quality evaluation category, and a detection algorithm are to be automatically adjusted until a required performance is satisfied (refer to FIG. 3 and FIG. 4 in PTL 1).
However, when it comes to the optimization problem in the detection of the abnormality of the mechanical system, because a function indicating an index for optimization depends on various variables or parameters, there is a limitation on the automatic adjustment of the various variables or parameters using only an optimization algorithm. Accordingly, it is indispensable to utilize expert's domain knowledge (expert knowledge) of the mechanical apparatus and to make an adjustment manually, but in a case where the number of parameters to be adjusted is great, it takes working hours even for an expert on the mechanical apparatus to develop the abnormality detection procedure.
An object of the present invention, which was made to solve the technical problems in the related art, is to provide an abnormality detection procedure development apparatus and an abnormality detection procedure development method that can reduce working hours that it takes for development and is capable of developing an abnormality detection procedure that has a high detection performance.
According to an aspect of the present invention, there is provided an abnormality detection procedure development apparatus, including: a parameter setting unit that sets a parameter verification range that is a range where a value of a parameter is changed, based on data that is input by a user, with respect to the parameter relating to abnormality determination which is included in an abnormality detection procedure for a mechanical apparatus; an evaluation unit that causes the value of the parameter to be changed in the parameter verification range, and, with respect to each of the values of the parameter that are changed, evaluates abnormality detection performance of the abnormality detection procedure; and a display unit on which a performance evaluation table that shows abnormality detection performances which are evaluated by the evaluation unit is displayed with respect to each of the values of the parameter.
According to the present invention, there are provided an abnormality detection procedure development apparatus and an abnormality detection procedure development method that can reduce working hours that it takes for development and is capable of developing an abnormality detection procedure that has a high detection performance.
Embodiments of the present invention will be described in detail below referring to the drawings.
The mechanical apparatus 1 is an apparatus that is a target which is monitored by the state monitoring apparatus 2. The mechanical apparatus 1 is a target for a maintenance job that is performed by the maintenance engineer 3, periodically or when an abnormality or an indication of an abnormality (which is referred to simply as an abnormality) is detected by the state monitoring apparatus 2. Various sensors (not illustrated) are mounted in the mechanical apparatus 1, and various pieces of state data of the mechanical apparatus 1, which are measured by the various sensors, are output, as pieces of operating data, toward the state monitoring apparatus 2. It is noted that any apparatus that realizes an expected function by an accompanying mechanical operation may serve as the mechanical apparatus 1.
The state monitoring apparatus 2 is configured with a display device that is not illustrated, a console, a control computer, and the like, and is connected to the mechanical apparatus 1 and the abnormality detection procedure development apparatus 10 through a wired or wireless communication system. The state monitoring apparatus 2 collects the operating data from the mechanical apparatus 1, accumulates the data, in addition, diagnoses periodically the presence and absence of the abnormality in the mechanical apparatus 1 according to a prescribed abnormality detection procedure, and reports a result of the diagnosis to the administrator 4. Based on the reporting of the result of the diagnosis from the state monitoring apparatus 2, when noticing an abnormality of the mechanical apparatus 1, the administrator 4 instructs the maintenance engineer 3 in a work site to perform an operation of maintaining the mechanical apparatus 1.
The abnormality detection procedure development apparatus 10 is configured with a personal computer or a workstation, and supports development of the abnormality detection procedure for the mechanical apparatus 1 by the expert 5. That is, the expert 5 develops the abnormality detection procedure for the mechanical apparatus 1, using the operating data for evaluation which are acquired from the state monitoring apparatus 2, or utilizing his/her own domain knowledge (an expert knowledge). At this point, the abnormality detection procedure development apparatus 10 has an evaluation means of evaluating the abnormality detection procedure that is developed by the expert 5, and reports a result of the evaluation by the evaluation means to the expert 5.
It is noted that the expert 5, for example, refers to a developer or a designer of the mechanical apparatus 1, and a person who has ample experience in development or design, a management, and maintenance of a similar mechanical apparatus.
As described above, the abnormality detection procedure for the mechanical apparatus 1, which is developed by the expert 5 and is evaluated, is sent from the abnormality detection procedure development apparatus 10 to the state monitoring apparatus 2, and is used, in the state monitoring apparatus 2, for monitoring (abnormality detection) of the mechanical apparatus 1.
At this point, a function of each of the abnormality detection procedure editing unit 13, the parameter setting unit 14, the evaluation unit 15, the reflection unit 16, the performance target value setting unit 17, the search unit 18, the display unit 19, and the communication unit 21 is realized by the arithmetical-operation processing device executing a prescribed program that is stored in the storage device. Furthermore, the abnormality detection procedure storage unit 11 and the operating data storage unit 12 are each configured by a storage device storing prescribed data. Furthermore, it is assumed that the user interface 20 includes a keyboard, a mouse, a liquid crystal display device, and the like, and further includes a control program therefor. A user can exchange data among the abnormality detection procedure editing unit 13, the parameter setting unit 14, the reflection unit 16, the performance target value setting unit 17, the search unit 18, and the display unit 19, through the user interface 20.
As illustrated in
Furthermore, the “abnormality detection procedure ID” is an identification number for identifying an abnormality detection procedure that is indicated by the abnormality detection procedure information 110, (which is hereinafter referred to simply as the abnormality detection procedure), the “failure mode” is a name of a failure that is a target which is to be detected with the current abnormality detection procedure, and the “version” is the number of times that the abnormality detection procedure is updated. Furthermore, the “old procedure ID” is information for specifying the abnormality detection procedure information 110 that is the immediately-preceding version of a current abnormality detection procedure information 110. Therefore, the “abnormality detection procedure ID” of the abnormality detection procedure information 110, which is stored in the abnormality detection procedure storage unit 11, is searched for with the “old procedure ID” as a key, and thus the immediately-preceding version of the abnormality detection procedure information 110 can be acquired.
It is noted that the abnormality detection procedure information 110 is individually created for every “type” or “ID” of the mechanical apparatus 1, and further for every “failure mode.” That is, the header information 111 can be said to be information for identifying the abnormality detection procedure that is indicated by the abnormality detection procedure information 110 in which the header information 111 itself is included.
Furthermore, as illustrated in
At this point, as the “sensor” information in the procedure information 112, names of one or more sensors that are used in the abnormality detection procedure, or names of pieces of data that are acquired by the one or more sensors are set. It is noted that a load rate, a temperature, and a pressure are set as the “sensor” information in an example in
Furthermore, as the “preprocessing” information in the procedure information 112, conversion processing of sensor data and a state separation condition are set. The conversion processing of the sensor data refers to processing that is performed on the sensor data which is acquired through the sensor before applying a diagnosis algorithm. For example, the conversion processing of the sensor data refers to filtering processing for noise removal or movement averaging processing. Furthermore, the state separation condition refers to a condition that defines a steady state of the mechanical apparatus 1.
It is noted that, in the example in
Generally, a state of the mechanical apparatus 1 is divided into a steady state in which the mechanical apparatus 1 operates in a stabilized manner, and a transient state before the steady state is reached. For example, an engine, which is not sufficiently warmed up immediately after starting, is in the transient state of not operating in a stabilized manner during that, but when a fixed period of time has elapsed, is in the steady state of operating in a stabilized manner. Therefore, when the state of the mechanical apparatus 1 is not separated into the steady state and the transient state, that is, when the abnormality of the mechanical apparatus 1 is diagnosed using all pieces of data on the mechanical apparatus 1, many false positives on the results of the diagnosis occur due to instability of the operation in the transient state. In contrast, when the abnormality of the mechanical apparatus 1 is diagnosed by separating the operating data in the steady state from among pieces of operating data on the mechanical apparatus 1 and using the separated operating data, the number of false positive is reduced, and thus precision of the result of the abnormality diagnosis can be improved in the mechanical apparatus 1.
This extraction of the steady state of the mechanical apparatus 1 is referred to as state separation. Accordingly, according to the present embodiment, the state separation condition is set using a sensor or data that is obtained from the sensor as information of the state separation condition. For example, as a state separation condition for extracting a steady state of the engine, a condition that a temperature of engine oil be equal to or higher than 60 degrees, for example, is set.
Furthermore, as the “algorithm” information in the procedure information 112, a name of an algorithm for determining the abnormality of the mechanical apparatus 1 and parameter information that is used for the algorithm are set. In the example in
It is noted that, an algorithm for abnormality determination is not limited to a cluster analysis that uses the “k-means,” and may be a “main component analysis,” or the like.
Furthermore, as the “postprocessing” information in the procedure information 112, condition data for the abnormality determination that is used in processing which determines the abnormality of the mechanical apparatus 1, which results after applying the diagnosis algorithm, is set. In the example in
In a general cluster analysis, if the operating data, which is constructed from n pieces of sensor data, is defined as being acquired at every prescribed timestamp, an n-dimensional vector space that has the n pieces of sensor data as components thereof can be assumed. Therefore, the operating data that has n components at every timestamp is divided into clusters in the n-dimensional vector space. Then, in a case where there is operating data that does not belong to any cluster, with the operating data, it is determined that an abnormality, that is, an abnormality or an indication of an abnormality appears of the mechanical apparatus 1.
At this point, in order to determine whether or not the n-dimensional operating data belongs to any one of the clusters, a concept of an abnormality level is introduced. For example, the abnormality level can be defined based on a position that is indicated by the operating data and on a Euclidean distance from the center of the closest cluster to the position. Then, in a case where an abnormality level is at a prescribed threshold or above, it is determined that the operating data that has such an abnormality level is abnormality data that does not belong to any of the clusters.
Incidentally, in the example in
Additionally, as illustrated in
In a case where an algorithm that is designated with the “algorithm” information in the procedure information 112 is machine learning, as the “learning-period-of-time” information, a period for which the learning is performed is set. In an example of the evaluation information 113 in
As the “diagnosis-period-of-time” information, a period of time for which the number of false positives and the number of false negatives are evaluated are set. In the example of the evaluation information 113 in
As the “abnormality-period-of-time” information, a period of time for which the mechanical apparatus 1 within the “diagnosis period of time” is abnormal is set. In the example of the evaluation information 113 in
Furthermore, as a result of evaluating an abnormality detection result which is determined on the operating data during the “diagnosis period of time” by the “postprocessing” in the procedure information 112, the number of cases where a normal state is erroneously determined as abnormality detection and the number of cases where an abnormal state occurs but is overlooked, are stored under the headings of “number of false positives and “number of false negatives, respectively. In the example of the evaluation information 113 in
It is noted that, at this point, the false positive and the false negative are expressed in number of cases, but, for example, may be expressed with a total of duration times of the false positive and a total of duration times of the false negative, respectively.
Additionally, as illustrated in
In an example of the parameter verification result information 114 in
It is noted that generally, when a value of each of these parameters is increased, because the abnormality is difficult to catch, the number of false positives decreases and the number of false negatives increases. On the other hand, when the value of the parameter is decreased, because the abnormality is easy to catch, the number of false positives increases and the number of false negatives decreased.
The machine table 121 is configured with pieces of data, such as “category,” “type,” “ID,” and “operating data ID,” and designates the mechanical apparatus 1 that is a diagnosis target and the operating data 122 that is acquired from the mechanical apparatus 1. At this point, the “category,” the “type,” and the “ID” have the same meanings as the “category,” the “type,” and the “ID” that are referred to in the header information 111 in the abnormality detection procedure information 110 which is illustrated in
The operating data 122 is created for every mechanical apparatus 1 that is designated with the “category,” the “type,” and the “ID” of the machine table 121. That is, at least one piece of operating data 122 or normally a plurality of pieces of operating data 122 are stored in one operating data storage unit 12.
Furthermore, as illustrated in
It is noted that in the operating data 122 in an example in
A description is provided referring back to
The communication unit 21 is connected to the state monitoring apparatus 2 through a wired or wireless communication system (not illustrated), and is connected to the mechanical apparatus 1 through the state monitoring apparatus 2 (refer to
The abnormality detection procedure editing unit 13 receives input of the abnormality detection procedure ID that is input through the user interface 20, and reads the abnormality detection procedure information 110 that is designated with the abnormality detection procedure ID, from the abnormality detection procedure storage unit 11. Then, based on the abnormality detection procedure information 110 that is read, the abnormality detection procedure editing unit 13 displays an abnormality detection procedure development screen 50 on the display device through the display unit 19, as is next illustrated in
At this point, in a case where the abnormality detection procedure information 110 that corresponds to the abnormality detection procedure ID is not stored in the abnormality detection procedure storage unit 11, new development of the abnormality detection procedure is assumed. In a case where the abnormality detection procedure information 110 is stored in the abnormality detection procedure storage unit 11, update of the abnormality detection procedure is assumed.
The header information 51 is the same as the header information 111 that is illustrated in
Among pieces of editing-in-progress data 524, the user (the expert 5) can freely edit editing-target data 521 (data, such as “sensor,” “preprocessing,” “algorithm,” “postprocessing,” “learning period of time,” “diagnosis period of time,” or “abnormality period of time),” through the abnormality detection procedure development screen 50. Then, in an editing task, the pre-editing data 523, as is, may be copied to the editing-in-progress data 524, and a change may be made to a necessary part.
However, at least evaluation result data 522 (the number of false positives and the number of false negatives) in the editing-in-progress data 524 can be neither copied, nor arbitrary edited by the user. The evaluation result data 522 (the number of false positives and the number of false negatives) in the editing-in-progress data 524 is displayed after the evaluation of the abnormality detection procedure is ended in the evaluation unit 15.
Furthermore, in the abnormality detection procedure development screen 50, the “reflection” button 53, the “evaluation” button 54, and the “parameter evaluation” button 55 are buttons that activate the reflection unit 16, the evaluation unit 15, and the parameter setting unit 14, respectively. Accordingly, when the user clicks in the “parameter evaluation” button 55, the parameter setting unit 14 is activated, and the parameter setting unit 14 displays a parameter verification screen 60 (refer to
When the parameter setting unit 14 is first activated, only the sub-screen 61 is displayed. Then, a parameter setting table 610, an “evaluation” button 611, an “addition” button 612, and a message 613 that alerts the user to the time to the ending of the evaluation are displayed on the sub-screen 61.
The parameter setting table 610 is a table for setting a range and a stride of parameters that are verification target, and every parameter has the headings of a name of a parameter, a minimum value and a maximum value of the parameter, and a stride. The user appropriately inputs a name or a numerical value under each of the heading in the parameter setting table 610, and thus can set the range and the stride of the parameter that is the verification target.
However, in a case where the abnormality detection procedure development screen 50 (refer to
It is noted that in an example of the parameter setting table 610 in
Furthermore, on the sub-screen 61, an “addition” button 612 is clicked on, one empty row is appended in the parameter setting table 610. The user can set a name of a new parameter, a maximum value of the parameter, a minimum value of the parameter and a stride, in the empty row.
Furthermore, when the “evaluation” button 611 is clicked on, the evaluation unit 15 is activated. At this time, the evaluation unit 15 causes a value of each parameter to be changed for the range and the stride of the parameter that are set in the parameter setting table 610, and, with respect to all combinations of the parameters, evaluates an abnormality detection performance using an algorithm for the editing-in-progress data 524 on the abnormality detection procedure development screen 50 (refer to
When the evaluation of the evaluation unit 15 by the abnormality detection performance ends, the sub-screen 62 and the sub-screen 63 are displayed on the parameter verification screen 60. Then, a parameter verification result table 620, a “reflection” button 621, and a “past-search” button 622 are displayed on the sub-screen 62. The parameter verification result table 620 shows the number of false positives and the number of false negatives with respect to all combinations of parameters that are obtained by the evaluation unit 15.
Therefore, the parameter verification result table 620 can be said to be a performance evaluation table that shows a listing of abnormality detection performances of the abnormality detection procedures with respect to each of the combinations of the parameters. Furthermore, the parameter verification result table 620 shows the parameter verification result information 114 in the abnormality detection procedure information 110 (refer to
It is noted that in an example in
Furthermore, a target value input box 630 for the abnormality detection performance is displayed on the sub-screen 63. The target value input box 630 is for narrowing the number of combinations that satisfy a performance which is desired by the user, among combinations of the number of false positives and the number of false negatives that are shown in the parameter verification result table 620 on the sub-screen 62. That is, when target values that are desired by the user are set in the target value input boxes 630, boxes that satisfy the target value in the parameter verification result table 620 on the sub-screen 62 are displayed in an emphasized manner, such as, for example, in a thick-line frame.
In the example in
Subsequently, the user (the expert 5) selects a parameter that is to be used for the abnormality detection procedure, from among the combinations of parameters that correspond to the boxes which are displayed in an emphasized manner in the parameter verification result table 620. For example, in the example in
Because the selection of the parameters is performed by the user (the expert 5), the user (the expert 5) can harness his own domain knowledge. For example, in the example in
However, even in the case of the expert 5, in a situation where which relationship parameters and a performance of the abnormality detection procedure have is not understood, suitable parameters are difficult to determine. Only after understanding the relationship between the parameters as illustrated in the parameter verification result table 620 and the abnormality detection procedure performance, the expert 5 can determine suitable parameters. That is, the parameter verification result table 620 assists the expert 5 in determining optimal parameters.
In this way, when the user (the expert 5) selects one box from boxes that are displayed in an emphasized manner in the parameter verification result table 620 on the sub-screen 62, the selected box is displayed in emphasized manner. Then, a combination of parameters that corresponds to the box is determined as parameters that are to be used for the abnormality detection procedure. It is noted that in the example in
Subsequently, when the user clicks on the “reflection” button 621, parameters (abnormality level, duration time) that the user (the expert 5) determines as being optimal based on the parameter verification result table 620 are reflected in the editing-in-progress data 524 in the abnormality detection procedure editing information 52 in
Furthermore, when the user clicks on the “past-search” button 622, the search unit 18 is activated, and the search unit 18 displays a search result screen 70 (refer to
As illustrated in
As a result of the determination, in a case where the processing is activated by the abnormality detection procedure editing unit 13 (the “evaluation” button 54 in
At this point, the processing for the evaluation in Step S13 is described in detail, using an example of the editing-in-progress data 524 on the abnormality detection procedure development screen 50 in
Next, the evaluation unit 15 writes an evaluation result that is obtained with the evaluation in Step S13 to a box for the evaluation result data 522 (number of false positive, number of false negative) in the editing-in-progress data 524 on the abnormality detection procedure development screen 50 in
On the other hand, as a result of the determination in Step S12, in a case where the activation is not caused by the abnormality detection procedure editing unit 13 (the “evaluation” button 54 in
In the processing operations that are repeatedly performed, the evaluation unit 15 first generates the abnormality detection procedure with respect to a combination of parameters that is designated in the processing (Step S15) that is first repeated (Step S16). Subsequently, the evaluation unit 15 evaluates the performance of the abnormality detection procedure that is generated in Step S16, using the operating data 122 that is acquired in Step S11 (Step S17). It is noted that the processing for the evaluation in Step S17 is basically the same as the processing for the evaluation in Step S13 described above.
Next, the evaluation unit 15 displays a result of evaluating the performance of the abnormality detection procedure, which is obtained with the repeatedly-performed processing operations in Step S16 and Step S17, as the parameter verification result table 620, on the display device (Step S19), and ends the processing.
As a result of the determination, in a case where the activation is caused by the abnormality detection procedure editing unit 13 (the “reflection” button 53 in
It is noted that at the time of writing to the abnormality detection procedure storage unit 11, the header information 51 is together written, and a box for “Ver” is updated.
On the other hand, in the determination in Step S31, in a case where the activation is not caused by the abnormality detection procedure editing unit 13 (the “reflection” button 53 in
Furthermore, in a case where as a result of the determination in Step S41, the target value is not input into the target value input box 630, processing operations in Step S42 and Step S43 are skipped, and the processing is ended.
Subsequently, the search unit 18 acquires the parameter information on the verification target (Step S52). Specifically, the parameter information can be acquired from the parameter setting table 610 on the parameter verification screen 60.
Next, the search unit 18 searches for information for verifying a similar abnormality detection procedure, referring to the abnormality detection procedure storage unit 11 (Step S53). That is, the search unit 18 extracts a category, a type, and an ID that are included in the header information 111, which are consistent with or similar to those which are included in information on the mechanical apparatus 1 that is the search target, which is acquired in Step S51, and parameter information that is included in the parameter verification result information 114, which is consistent with or similar to the parameter information that is acquired in Step S52, from the abnormality detection procedure information 110 (refer to
Then, the search unit 18 displays a result of the search on the display device (Step S54), and ends the processing.
Next, when the user selects one row (which, in an example in
As described above, according to the embodiment of the present invention, it is easy for the expert 5 who is a user of the abnormality detection procedure development apparatus 10 to develop the abnormality detection procedure for the mechanical apparatus 1. Particularly, because a relationship between a value of a parameter that is used for an algorithm for the abnormality detection procedure and the abnormality detection performance can be easily verified, it is possible that a parameter that achieves a maximum abnormality detection performance is selected. That is, the time that it takes to develop the abnormality detection procedure can be shortened. Furthermore, as described above, in selecting a parameter, because the domain knowledge of the expert 5 can be harnessed, it is possible that the abnormality detection procedure is developed in accordance with a situation of the abnormality of the mechanical apparatus 1.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2014-183084 | Sep 2014 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2015/074831 | 9/1/2015 | WO | 00 |
