JP7171880B1 - Anomaly predictive diagnosis device and program - Google Patents

Abstract

Kind Code: A1 An abnormality predictive diagnostic device capable of appropriately determining whether or not a diagnostic model needs to be updated. Kind Code: A1 Based on operation data D2 obtained by acquiring the state of the monitored equipment 200 and a diagnostic model D3, the presence or absence of an abnormality sign of the monitored equipment 200 is determined, and the determination result is affirmative (D5=“1”). ”), the sign detecting unit 116 for generating the anomaly sign alert data D6 indicating the occurrence of an anomaly sign alert, and the anomaly sign alert based on the operation data D2 corresponding to the anomaly sign alert data D6. The abnormality predictor diagnostic device 100 is provided with a model update control unit 120 that determines whether or not the data D6 has credibility, and determines whether or not the diagnostic model D3 needs to be updated based on the determination result. [Selection drawing] Fig. 1

Description

The present invention relates to an abnormality sign diagnosis device and program.

As a background art of this technical field, the abstract of Patent Document 1 below states, "[Problem] To provide an abnormality sign diagnosis system or the like that appropriately diagnoses the presence or absence of an abnormality sign even when mechanical equipment is restarted. [Solution] The abnormality predictive diagnostic system 1 selects one of other mechanical equipment of the same model as the mechanical equipment as a normal model to be used for diagnosing the mechanical equipment when restarting the predetermined mechanical equipment after the stop period. A normal model selection means 162 for selecting the latest normal model, and a diagnosis means 163 for diagnosing whether or not there is an anomaly sign of the machinery based on the normal model selected by the normal model selection means 162." Have been described.

Japanese Patent Application Laid-Open No. 2020-035281

By the way, in the above-described technology, there are cases where it is preferable to correctly update the diagnostic model for determining the sign of abnormality using the data of the device itself as much as possible. .
SUMMARY OF THE INVENTION The present invention has been made in view of the circumstances described above, and it is an object of the present invention to provide an abnormality predictive diagnosis apparatus and program capable of appropriately determining whether or not a diagnostic model needs to be updated.

In order to solve the above problems, the abnormality predictor diagnosis device of the present invention determines whether or not there is an abnormality predictor in the monitoring target equipment based on operation data obtained by acquiring the state of the monitoring target equipment and a diagnostic model, and determines the determination result is affirmative, a sign detection unit for generating anomaly sign alert data indicating occurrence of an anomaly sign; and the anomaly sign alert data is confirmed based on the operation data corresponding to the anomaly sign alert data. and a model update control unit that determines whether the diagnostic model is to be updated if the determination result is negative.

According to the present invention, it is possible to appropriately determine whether or not to update a diagnostic model.

1 is a block diagram of an abnormality sign diagnostic device according to a first embodiment; FIG. 1 is a block diagram of a computer; FIG. It is a figure which shows the specific example of a credibility judgment database. 2 is a flowchart (1/2) schematically showing contents of a control program and the like; 2 is a flowchart (2/2) schematically showing contents of a control program and the like; It is a figure which shows an example of the relationship between an abnormality degree and sensor data. FIG. 10 is a diagram showing another example of the relationship between the degree of anomaly and sensor data;

[Overview of embodiment]
In a general anomaly predictive diagnostic system, an anomaly predictive alert is issued when the predictive diagnosis result is "abnormality predictive present", and the anomaly predictive alert can be classified into two types: definite and false. For this reason, maintenance planners refer to the status of past alert issuance, sensor data related to alerts, etc., and obtain from past operation records, etc., whether anomaly sign alerts are classified as confirmed or false alarms. I was making decisions based on my experience. Therefore, it would be desirable if this type of determination could be automated. Further, if the abnormality predictor alert is a false alarm, it is preferable to update the diagnosis model of the abnormality predictor diagnosis so that the abnormality predictor diagnosis system will not issue a false alarm thereafter. Therefore, the embodiment described later attempts to appropriately determine whether or not to update the diagnostic model based on the result of the predictive diagnosis that indicates that there is an abnormality predictor.

Here, the significance of judging the credibility of an abnormality sign will be described in more detail, taking the diagnosis of an abnormality sign of a gas engine as an example.
(1) After the anomaly portent diagnostic device issues an anomaly portent alert, it may continue to generate an alert with similar sensor data for several days. Such alerts are likely to be false alarms. It is also assumed that although a certain anomaly portent alert was issued in the past, no corresponding failure occurred. Then, alerts similar to this alert are also likely to be false alarms. Such knowledge is information accumulated in the experience of skilled workers and the knowledge of maintenance planners.

In the case where a skilled worker or the like determines that it is a "false alarm", it may be preferable to ignore the alert by prioritizing the experience of the skilled worker or the like over the alert issued by the anomaly predictive diagnostic device. many. Then, it is desirable that the event is identified as "false alarm" and the alert is not issued in the first place.

(2) In sensor data, if a trend similar to past failures occurs, the diagnostic model may generate an anomaly sign alert. However, there are cases where this anomaly portent alert should be judged as "a false alarm occurring too early". Too large a weighting factor when weighting the sensor data in the diagnostic model is often the cause of this type of false alarm. In this case, false alarms can be suppressed by changing the weighting coefficients.

(3) Workers, etc., when hesitant to judge whether an anomaly sign alert is correct or not, need to check the trend graph of the reacting sensor data and related sensor data, maintenance history and parts replacement information around the sensor. to determine whether it is false or confirmed.
In addition, if the sensor data shows a trend different from normal, or if the interlocking state of the related sensor data is different from normal, the reliability of the outside temperature, outside pressure, humidity, fuel information, etc. should be suspected. In addition, there are cases where an artificial factor (for example, a foreign object blocking the intake port) that causes a problem occurs. Therefore, it is preferable to estimate and identify such factors and take countermeasures.
In addition, if an abnormality sign alert occurs in a highly reliable part even though the elapsed time since the last maintenance was short, it should be judged as a false alarm in many cases.
In addition, an abnormality sign alert may be generated when a worker or the like performs an abnormal operation. However, this is often found to be false after the fact as a result of collating human information. In addition, in many cases of this kind, it turns out to be a false alarm with a delay of several years.

In these cases, an operator or the like may change the model coefficients of the diagnostic model. In addition, the operator may classify the anomaly sign alert as a false alarm in association with the state of the trend of the sensor data indicating the abnormal transition as described above.
Therefore, in the embodiment described later, the above (1) to (3) are systematized by the credibility determination database, and based on the credibility determination database, it is determined whether the abnormal sign is confirmed or false.

[First embodiment]
<Configuration of the first embodiment>
FIG. 1 is a block diagram of an abnormality predictor diagnosis device 100 according to the first embodiment. The abnormality predictor diagnostic device 100 is a system for diagnosing an abnormality predictor of the equipment 200 to be monitored. The monitored equipment 200 is, for example, a gas engine generator. However, the monitoring target facility 200 is not limited to this, and may be a chemical plant, a nuclear power plant, medical equipment, communication equipment, or the like. In addition, the equipment to be monitored 200 may have all components installed in the vicinity, or each component may be installed at a plurality of discrete locations.

The abnormality predictor diagnosis device 100 includes a predictor analysis section 110 , a model update control section 120 , a sensor section 150 and a DB section 160 . "DB" means "database". The DB unit 160 includes an operation database 162 , a model management database 164 , a diagnosis result database 165 , a cause diagnosis database 166 and a credibility determination database 168 . A magnetic disk device, an optical disk device, a semiconductor storage device, or the like can be applied to each database described above.

The sign analysis unit 110 also includes an operation monitoring unit 112 , a diagnosis model selection unit 114 , a sign detection unit 116 (sign detection means), and a cause diagnosis unit 118 . The model update control unit 120 (model update control means) includes a predictor experience determination unit 122 , a predictor credibility determination unit 124 , and a model update determination unit 126 .

The sensor unit 150 includes a plurality of sensors (not shown) that measure the state of the equipment 200 to be monitored. The operation database 162 accumulates operation data D2 based on these sensor detection values (sensor data). The operation data D2 includes information such as the identification information of each sensor, the identification information of the equipment in which each sensor is installed among the equipment constituting the monitored facility 200, each sensor data, and the date and time of detection. This operation data D2 is supplied to the predictor analysis unit 110 .

The operation monitoring unit 112 monitors the operating state of the monitored equipment 200 based on the operation data D2. The model management database 164 stores a plurality of diagnostic models D3 respectively corresponding to a plurality of predictive diagnostic items. Here, the predictive diagnostic items are, for example, "engine air-fuel mixture temperature abnormality", "cooling water pressure drop", "oil leakage", and the like. Here, each diagnostic model D3 is data describing operation details of the monitoring target equipment 200 in a normal state according to the status of the monitoring target equipment 200 and predictive diagnostic items to be monitored. The diagnostic model selection unit 114 selects one diagnostic model D3 according to the state of the monitoring target equipment 200 and the predictive diagnostic items from among these multiple diagnostic models D3.

The predictor detection unit 116 outputs the degree of abnormality D4 corresponding to the predictive diagnosis item based on the current and past operation data D2 stored in the operation database 162 and the selected diagnostic model D3. Here, the degree of abnormality D4 is an index representing the degree of deviation from the normal state. In addition, the portent detection unit 116 performs anomaly portent diagnosis of the monitoring target equipment 200 based on the degree of anomaly D4. Here, the "abnormality sign diagnosis" is not limited to diagnosing whether or not the monitoring target equipment 200 reaches an abnormal state in which it cannot be operated. It also includes diagnosing the degree of performance degradation when 200 performance degradation is occurring. Then, the sign detection unit 116 diagnoses the presence or absence of an abnormality sign in the monitoring target equipment 200 based on the degree of abnormality D4, and outputs the result as determination data D5. The determination data D5 is binary data having a value of "0" (no sign of abnormality) or "1" (with a sign of abnormality).

For example, when the degree of abnormality D4 is equal to or greater than a predetermined threshold value, the sign detection unit 116 sets the determination data D5 to "1" (there is a sign of abnormality), and otherwise sets the determination data D5 to "0" (there is no sign of abnormality). It is conceivable that Further, the sign detection unit 116 determines whether or not the degree of abnormality D4 is credible based on the relationship between the sign diagnosis item, the operation data D2, and the degree of abnormality D4. If there is, and the degree of abnormality D4 is equal to or greater than a predetermined threshold value, the determination data D5 is set to "1" (with a sign of abnormality); otherwise, the determination data D5 is set to "0" (no sign of abnormality). is also conceivable.

The predictor detection unit 116 generates abnormality predictor alert data D6 and writes it to the diagnostic result database 165 when the judgment data D5 is “1” (there is an abnormality predictor). The abnormality portent alert data D6 is obtained by linking the portent diagnostic item, the operation data D2, the degree of abnormality D4, and the credibility data D10. Here, the credibility data D10 is data indicating whether the anomaly predictive alert data D6 is "confirmed", "false alarm" or "undetermined". “Undecided”. However, the credibility data D10 is changed to "correct report" or "false report" by the model update control unit 120. FIG.

Here, "confirmed report" means that an abnormal state appears in the monitoring target facility 200 after a certain period of time has elapsed after the abnormality prediction alert data D6 is generated if no countermeasures are taken against the abnormality prediction alert data D6. means that it is likely to Further, "false alarm" means that when no countermeasures are taken against the anomaly predictive alert data D6, an abnormal state corresponding to the monitoring target facility 200 is kept even after a certain period of time has elapsed after the anomaly predictive alert data D6 is generated. It means that there is a high possibility that it will not appear.

Then, the portent detection unit 116 supplies the abnormality portent alert data D<b>6 to the model update control unit 120 . By the way, the credibility data D10 included in the anomaly portent alert data D6 is "undecided" when the anomaly portent alert data D6 is newly generated. Therefore, the credibility data D10 of "definite report" or "false report" is added later to the abnormality portent alert data D6 by the processing described later. In this way, the anomaly predictive alert data D6 that is a "definite report" or a "false alarm" may be referred to as "evaluated anomaly predictor alert data D6A".

Furthermore, the symptom detection unit 116 learns a predetermined diagnostic model indicating the normal range of the operation information of the monitored facility 200 based on a statistical method (data mining) using the operation data D2 as learning data. The symptom detection unit 116 registers the learning result in the model management database 164 as a diagnostic model D3.

The cause diagnosis unit 118 may identify an "effect sensor" for the abnormality portent alert data D6. The influence sensor is the sensor that outputs the sensor data that has the greatest influence as the cause of the judgment that the sign detection unit 116 has determined that there is a sign of abnormality among the sensor data of the sensors belonging to the sensor unit 150. is. The cause diagnosis unit 118 also generates cause diagnosis data D8 and writes it to the cause diagnosis database 166. FIG. Here, the cause diagnosis data D8 is data in which the predictive diagnostic item, the pattern of the influence sensor, and the abnormality predictive alert data D6 are linked.

As described above, the model update control unit 120 receives the abnormality portent alert data D6 from the portent analysis unit 110 .
The predictor experience determination unit 122 of the model update control unit 120 stores existing evaluated abnormality predictor alert data D6A similar to the abnormality predictor alert data D6 received this time in the cause diagnosis database 166 (hereinafter referred to as "existing similar data"). ) is recorded.

Here, the meaning of "similar" will be explained. When the operational data D2 includes n sensor data, these sensor data correspond to one point in the n-dimensional space. Points in the n-dimensional space related to the operation data D2 included in the existing evaluated anomaly portent alert data D6A and points in the n-dimensional space related to the operation data D2 included in the anomaly portent alert data D6 received this time is equal to or less than a predetermined value, the abnormality portent alert data D6 of both are assumed to be "similar".

In addition, when there is no existing data similar to the newly generated abnormality predictor alert data D6, the predictor credibility determination unit 124 performs follow-up observation of the monitoring target equipment 200 to determine whether the abnormality predictor alert data D6 determine credibility. Furthermore, if there is existing similar data for the abnormality predictor alert data D6, the predictor credibility determination unit 124 determines whether this existing similar data is a "definite report" or a "false report".

Then, the portent credibility determination unit 124 adds the same credibility data D10 as the credibility data D10 of the “confirmed report” or “false report” included in the existing similar data to the newly generated abnormality portent alert data D6. do. As a result, the anomaly portent alert data D6 becomes evaluated anomaly portent alert data D6A. Further, when the credibility data D10 of the evaluated anomaly predictive alert data D6A is "confirmed", the predictive credibility determination unit 124 stores the contents of the evaluated anomaly predictive alert data D6A in the cause diagnosis database 166. Also store in

The model update determination unit 126 determines whether or not the corresponding diagnostic model D3 needs to be updated when the newly generated abnormality sign alert data D6 is "false alarm". Furthermore, when updating the diagnostic model D3, the model update determination unit 126 determines which of "new creation" and "partial update" is adopted as the update mode.

Here, the significance of "new creation" and "partial update" will be explained. The diagnostic model D3 includes an algorithm and parameters that are constants applied to the algorithm. Here, changing only the constants without changing the algorithm is called "partial update", and changing the algorithm itself is called "new creation".

FIG. 2 is a block diagram of computer 900 . The abnormality predictor diagnosis device 100 shown in FIG. 1 includes this computer 900 .
2, computer 900 includes CPU 901, RAM 902, ROM 903, HDD 904, communication I/F 905, input/output I/F 906, and media I/F 907. Communication I/F 905 is connected to communication circuit 915 . Input/output I/F 906 is connected to input/output device 916 . A media I/F 907 reads and writes data from a recording medium 917 .

The ROM 903 stores control programs executed by the CPU, various data, and the like. The CPU 901 implements various functions by executing application programs loaded into the RAM 902 . The inside of the abnormality predictor diagnosis apparatus 100 shown in FIG. 1 is a block diagram of functions realized by an application program or the like. However, the abnormality predictor diagnosis apparatus 100 may be configured using one computer 900 or may be configured using a plurality of computers 900 connected via a communication circuit 915 .

<Specific example of credibility determination database>
FIG. 3 is a diagram showing a specific example of the credibility judgment data D30 stored in the credibility judgment database 168. As shown in FIG.
In FIG. 3, the credibility determination data D30 has a predictive diagnosis item portion 30, a determination criteria portion 34, and an evaluation information portion 36 (evaluation information). The predictive diagnosis item section 30 stores a plurality of predictive diagnosis items 30a, 30b, 30c, and the like. The judgment criteria unit 34 stores a plurality of judgment criteria #1 to #4 for each predictive diagnosis item.

Also, the evaluation information section 36 stores the evaluation information of "O" or "X" according to the combination of these judgment criteria #1 to #4. Here, the evaluation information of "O" means that when an abnormality predictor occurs for the predictive diagnosis item, the abnormality predictor is reliable (considered to be a definite report). In addition, the evaluation information of "x" means that when an abnormality predictor occurs for the predictive diagnosis item, the abnormality predictor is unreliable (considered to be a false alarm). The content of the credibility determination data D30 is systematized based on the experience of skilled workers and the knowledge of maintenance planners.

In the illustrated example, the predictive diagnosis item 30a is "engine air-fuel mixture temperature abnormality", and as judgment criterion #1, "measured value difference ΔT between inlet temperature (intake air temperature) and outlet temperature (exhaust temperature)" is "predetermined temperature k°C or higher” or “less than k°C”. In addition, for each of the results of the judgment criterion #1, the judgment criterion #2 is whether the “measured value H of the cooling water level height” is “a predetermined height h1 or more” or “less than h1”. is mentioned. Also, for each combination of judgment results of judgment criteria #1 and judgment criteria #2, judgment criteria #3 is set as "Yes" or "No" for "whether the cooling tank was filled with water at the time of the previous maintenance". , are mentioned.

Further, the predictive diagnosis item 30b is "cooling water pressure drop", and whether or not "the pressure drop speed V exceeds the predetermined speed V1" is listed as the criterion #1. Then, when "the pressure drop speed V exceeds the predetermined speed V1" for the judgment criterion #1, "Yes" or "No" for the "presence or absence of periodicity of the pressure drop" is given as the judgment criterion #2. there is As a judgment criterion #3, "T1<T≤T2" and "T>T2" (where T1 and T2 are predetermined temperatures) are listed for the "cooling water heater temperature T".

Further, the predictive diagnosis item 30c is "oil leakage", and "Yes" or "No" for "Is oil pressure low?" If the judgment criterion #1 is "Yes", the judgment criterion #2 is whether the cooling water amount Q is "a predetermined value q1 or more" or "less than q1". Then, as a judgment criterion #3, "Yes" or "No" is given for "whether the valve was normal at the time of the previous maintenance".

<Operation of the first embodiment>
4 and 5 are flow charts schematically showing the contents of the control program and the like applied to this embodiment. 4 and 5, the solid lines mainly indicate the flow of processing, and the dashed lines indicate the flow of data.
When the process proceeds to step S2 in FIG. 4, the sign detection unit 116 creates a new diagnostic model D3. Next, when the process proceeds to step S4, the sign detection unit 116 determines whether or not the created diagnostic model D3 is appropriate. If "No" is determined here, the process returns to step S2, and the symptom detection unit 116 recreates the diagnostic model D3. Thereafter, steps S2 and S4 are repeated until the diagnostic model D3 is determined to be appropriate. On the other hand, if determined as "Yes" in step S4, the process proceeds to step S6. Here, the symptom detection unit 116 registers the created new diagnostic model D3 in the model management database 164 .

Next, when the process proceeds to step S<b>7 , the operation monitoring unit 112 acquires operation data D<b>2 from the operation database 162 . Further, the diagnostic model selection unit 114 selects a diagnostic model D3 corresponding to the predictive diagnostic item to be diagnosed, and reads it from the model management database 164 . Next, when the process proceeds to step S8, the symptom detection unit 116 calculates the degree of abnormality D4 based on the operation data D2 and the selected diagnostic model D3, and determines the determination data D5 based on the degree of abnormality D4. Calculate Furthermore, in step S<b>8 , the sign detection unit 116 generates abnormality sign alert data D<b>6 and registers it in the diagnostic result database 165 when the judgment data D<b>5 is “1” (there is an abnormality sign). As described above, the abnormality portent alert data D6 is data in which the portent diagnostic item, the operation data D2, and the degree of abnormality D4 are linked.

Next, when the process proceeds to step S10, the predictor experience determination unit 122 of the model update control unit 120 performs to determine whether or not existing similar data (similar existing evaluated anomaly sign alert data D6A) exists. If "No" is determined in step S10, the process proceeds to step S50 in FIG. 5, and a follow-up observation process is executed. Here, the follow-up observation process is a process of observing whether or not a failure has occurred in the monitored equipment 200 during a predetermined follow-up observation period.

Next, when the process proceeds to step S52, the predictor detection unit 116 determines whether or not a failure has occurred according to the predictor diagnostic items related to the abnormality predictor alert data D6 during the follow-up observation period of step S50. If it is determined as "No" here, the process proceeds to step S53, and the predictive experience determination unit 122 checks the diagnosis result database 165 so that the credibility data D10 of the newly registered abnormality predictive alert data D6 is " Record that it was a false alarm.

On the other hand, if determined as "Yes" in step S52, the process proceeds to step S54. Here, a worker or the like performs maintenance work to deal with the failure. Next, when the process proceeds to step S56, the cause diagnosis unit 118 evaluates the response of the abnormality predictor diagnosis device 100 in steps S52 and S53 according to the result of the maintenance work in step S54. First, as described above, in step S52, the sign detection unit 116 determined whether or not a failure occurred during the follow-up observation period. The cause diagnosing unit 118 evaluates the content of this judgment and stores the correctness or incorrectness in the diagnosis result database 165 . The cause diagnosis unit 118 also causes the cause diagnosis database 166 to store the correctness/incorrectness of the failure cause related to the abnormality portent alert data D6. Note that the process of step S56 described above may be performed by an operator or the like.

If it is determined "Yes" (there is existing similar data) in step S10 of FIG. 4, the process proceeds to step S12. In step S12, the portent credibility determination unit 124 determines whether the newly generated abnormality portent alert data D6 is a confirmed report or a false report according to the following procedure based on the credibility determination data D30.

That is, the predictor credibility determination unit 124 selects a predictor diagnosis item included in the abnormality predictor alert data D6 from among the predictor diagnosis items 30a, 30b, 30c, etc. (see FIG. 3). Next, the predictor credibility determination unit 124 collates the judgment criteria #1 to #4, etc. in the selected predictive diagnosis item with the operation data D2 included in the abnormality predictor alert data D6, and sends the evaluation information unit 36 Based on the stored evaluation information, it is determined whether the abnormality sign alert data D6 is credible (determined) or not credible (false alarm).

If the newly generated abnormality portent alert data D6 is "false alarm", it is determined as "No" in step S12, and the process proceeds to step S13. Since the anomaly predictive alert data D6 is a false alarm, the diagnostic model D3 used to generate the anomaly predictive alert data D6 is updated in subsequent processing. However, as described above, there are two types of update mode of the diagnostic model D3: "new creation" and "partial update".

Therefore, in step S13, the model update determination unit 126 selects either "new creation" or "partial update" as the update mode for the diagnostic model D3 related to the existing similar data that is "false alarm". Details of the selection method will be described later. Next, when the process proceeds to step S14, it is determined whether or not the selected update mode is new creation. Here, if the determination is “Yes”, the process returns to step S<b>2 , and the sign detection unit 116 newly creates a diagnostic model D<b>3 and registers its contents in the model management database 164 . After that, the processes after step S4 described above are repeated.

On the other hand, if partial update is selected as the update mode of diagnostic model D3, the process proceeds to step S22, and symptom detection unit 116 selects diagnostic model D3 to be updated from among diagnostic models D3 stored in model management database 164. Select diagnostic model D3. Next, when the process proceeds to step S<b>24 , the sign detection unit 116 temporarily registers the diagnostic model D<b>3 to be updated in the model management database 164 .

Next, when the process proceeds to step S<b>26 , the symptom detection unit 116 re-learns the provisionally registered diagnosis model D<b>3 based on the operation data D<b>2 stored in the operation database 162 . Next, when the process proceeds to step S28, the symptom detection unit 116 determines whether or not the provisionally registered diagnostic model D3 is appropriate.

If it is determined as "Yes" (appropriate) in step S28, the process proceeds to step S29, and the symptom detection unit 116 officially registers the provisionally registered diagnostic model D3. On the other hand, if the determination in step S28 is "No" (inappropriate), the process returns to step S22, and the processes of steps S22 to S28 are repeated. When the process of step S29 ends, the process returns to step S8, and the operations after step S8 described above are repeated.

If it is determined "Yes" (the newly generated abnormality sign alert data D6 is "confirmed") in step S12, the process proceeds to step S30 (see FIG. 5). Here, the predictor credibility determination unit 124 registers in the cause diagnosis database 166 that the existing similar data was "confirmed". Next, when the process proceeds to step S32, the portent credibility determination unit 124 displays a predetermined checklist on the display included in the input/output device 916 (see FIG. 2). Workers, etc., enter various data via this checklist.

By the way, when the monitored equipment 200 is introduced, a predetermined initial investigation has been performed for the monitored equipment 200 (S34). Next, when the process proceeds from step S32 to step S36, the predictive credibility determination unit 124 generates an abnormality predictive alert/ Diagnose the cause of data D6.

Next, when the process proceeds to step S38, the portent credibility determination unit 124 proposes countermeasures to prevent failures and the like to the worker or the like in response to the anomaly portent alert data D6. Next, when the process proceeds to step S<b>40 , the operator or the like creates a maintenance plan for the monitored equipment 200 . Next, in step S<b>54 , a worker or the like performs maintenance work on the monitored equipment 200 . Next, when the process proceeds to step S56, the cause diagnosis unit 118 evaluates the response of the abnormality predictor diagnosis device 100 in steps S30 to S38 according to the result of the maintenance work in step S54. Then, the cause diagnosis unit 118 stores the evaluation results in the diagnosis result database 165 and the cause diagnosis database 166 as described above.

<Details of step S13>
Details of the method of selecting new creation or partial update as the update mode for the diagnostic model D3 in step S13 described above will be described.
As described above, the operational data D2 includes multiple sensor data. Therefore, the model update determination unit 126 performs a correlation analysis between the time-series data of the degree of abnormality D4 and the sensor data contributing to the degree of abnormality D4, and calculates the respective degrees of correlation. The cause of the increase in the degree of anomaly D4 is often deterioration due to wear and tear of one or more devices associated with a specific sensor. Therefore, in such a case, the sensor data tend to gradually increase or decrease as the degree of abnormality D4 increases.

However, if the algorithm of diagnostic model D3 is inadequate, the sensor data may deviate significantly from escalating or decrementing. For example, there is a case where sensor data having a strong correlation with the degree of abnormality D4 is not included in the diagnostic model D3 in the first place. In addition, the monitoring target equipment 200 may be in various different operating states such as start, steady state, stop, and idling. If the diagnosis model D3 does not incorporate an accident based on the driving state, the relationship between the degree of abnormality D4 and the sensor data becomes unnatural. In this way, when the algorithm of the diagnostic model D3 itself is inappropriate, the model update determination unit 126 selects "new creation" as the update mode in step S13, and otherwise selects "partial update". select.

FIG. 6 is a diagram showing an example of the relationship between the degree of abnormality D4 and sensor data.
The horizontal axis of FIG. 6 is the time t, and the vertical axis is the sensor data and the degree of abnormality D4. In FIG. 6, SD1 is sensor data having the highest contribution to the degree of abnormality D4. The degree of contribution indicates the ratio of the contribution of the sensor data that has the greatest influence on the magnitude of the degree of anomaly D4 to the entire sensor data. In the illustrated example, since the degree of correlation between the sensor data SD1 and the degree of abnormality D4 is high, there is a strong tendency to select "partial update" as the "update mode" in step S13.

FIG. 7 is a diagram showing another example of the relationship between the degree of abnormality D4 and sensor data.
The horizontal axis of FIG. 7 is the time t, and the vertical axis is the sensor data and the degree of abnormality D4. In FIG. 7, SD1, SD2, and SD3 are sensor data having the first, second, and third highest contributions to the degree of abnormality D4. The sensor data SDX is actually sensor data that is not referred to in the diagnostic model D3.

For example, in the following cases, the model update determination unit 126 tends to select "new creation" as the "update mode".
(1) Sudden changes in the sensor data SD1 at times t1 and t2 are not included in the diagnostic model D3.
(2) The weighting of sensor data SD2 is extremely low compared to the weighting of sensor data SD1.
(3) Although the degree of correlation of the sensor data SDX with the degree of abnormality D4 is high, the sensor data SDX is not included in the diagnostic model D3.

[Effects of Embodiment]
As described above, according to the above-described embodiment, the abnormality predictor diagnosis device 100 detects an abnormality predictor of the monitored equipment 200 based on the operation data D2 obtained by acquiring the state of the monitored equipment 200 and the diagnostic model D3. Presence/absence is determined, and when the determination result is affirmative (D5="1"), the sign detector 116 generates anomaly sign alert data D6 indicating the occurrence of an anomaly sign, and the anomaly sign alert data D6. a model update control unit 120 that determines whether or not the abnormality sign alert data D6 has credibility based on the operating data D2, and determines whether or not the diagnostic model D3 needs to be updated based on the determination result; Prepare.

In this manner, the model update control unit 120 can determine whether or not the diagnostic model D3 needs to be updated based on the contents of the operation data D2 corresponding to the abnormality predictor alert data D6. can be determined appropriately.

The model update control unit 120 also includes a predictor experience determination unit 122 that determines whether existing similar data that satisfies a predetermined similarity criterion exists for the abnormality predictor alert data D6, and a predictor experience determination unit 122. a portent credibility determination unit 124 that determines whether or not the abnormality portent alert data D6 has credibility based on the credibility determination data D30 and the operation data D2 when the determination result is affirmative; In addition, the credibility determination data D30 includes predictive diagnosis items 30a, 30b, and 30c related to signs of abnormality, judgment criteria (#1 to #4) corresponding to the predictive diagnosis items 30a, 30b, and 30c, and predictive diagnosis items 30a, 30b, and 30c. Data including 30b, 30c and evaluation information (36) that determines credibility corresponding to the judgment criteria (#1 to #4) is more preferable.

As a result, the experience of skilled workers and the knowledge of maintenance planners can be reflected in the predictive diagnosis items 30a, 30b, 30c, judgment criteria (#1 to #4), and evaluation information (36). credibility can be determined more reliably.

Further, the model update control unit 120 determines that the determination result of the omen experience determination unit 122 is affirmative (“Yes” in S10) and the determination result of the omen credibility determination unit 124 is negative (“No” in S12). In some cases, it is more preferable to further include a model update determination unit 126 that selects either partial update or new creation as an update mode of the diagnostic model D3. As a result, it is possible to appropriately partially update the diagnostic model D3 or create a new diagnostic model D3, thereby increasing the credibility of the sign of abnormality.

Further, the model update control unit 120 determines that the determination result of the omen experience determination unit 122 is affirmative ("Yes" in S10) and the determination result of the omen credibility determination unit 124 is affirmative ("Yes" in S12). In some cases, it is more preferable to have a function of updating the cause diagnosis database 166 (registration of confirmed information in S30). Thereby, the cause diagnosis database 166 can be updated by selecting the abnormality portent alert data D6 with high credibility.

In addition, when the determination result of the predictor experience determiner 122 is negative, the model update control unit 120 generates a newly generated abnormality predictor alert based on the state of the monitoring target equipment 200 during the subsequent follow-up observation period (S50).・When the determination result of the confirmed/false alarm determination function (S52) that determines whether or not the data D6 was a confirmed report and the determination result of the confirmed/false alarm determination function (S52) is affirmative (in the case of a confirmed report), a new It is more preferable to further include a function (S56) of inputting the contents of the abnormality sign alert data D6 to the cause diagnosis database 166. FIG. As a result, when the anomaly predictive alert data D6 is a confirmed report, the anomaly predictive alert data D6 can be input to the cause diagnosis database 166, and the credibility of the anomaly predictive alert data D6 for which there is no existing similar data can be obtained. Those with high probability can be registered in the cause diagnosis database 166 .

[Modification]
The present invention is not limited to the embodiments described above, and various modifications are possible. The above-described embodiments are illustrated for easy understanding of the present invention, and are not necessarily limited to those having all the described configurations. Further, other configurations may be added to the configurations of the above embodiments, and part of the configurations may be replaced with other configurations. Also, the control lines and information lines shown in the drawings are those considered to be necessary for explanation, and do not necessarily show all the control lines and information lines necessary on the product. In practice, it may be considered that almost all configurations are interconnected. Possible modifications to the above embodiment are, for example, the following.

(1) Since the hardware of the abnormality predictor diagnosis apparatus 100 in the above embodiment can be realized by a general computer, the flowcharts shown in FIGS. or distributed via transmission channels.

(2) The processes shown in FIGS. 4 and 5 and the other processes described above have been described as software processes using a program in the above embodiment, but some or all of them are implemented in an ASIC (Application Specific Integrated Circuit). ; application-specific IC) or FPGA (Field Programmable Gate Array) or the like may be substituted for hardware processing.

(3) Various processes executed in the above embodiments may be executed by a server computer via a network (not shown), and various data stored in the above embodiments may also be stored in the server computer. .

30a, 30b, 30c Predictive diagnosis item 36 Evaluation information part (evaluation information)
100 Abnormality sign diagnostic device 116 Sign detection unit (sign detection means)
120 model update control unit (model update control means)
122 Omen experience judgment unit 124 Omen credibility judgment unit 126 Model update judgment unit 166 Cause diagnosis database 200 Monitoring target facility 900 Computer D2 Operation data D3 Diagnosis model D6 Abnormality omen alert data D30 Credibility judgment data #1 to #4 Judgment criteria

Claims (6)

Determining whether or not there is a sign of abnormality in the equipment to be monitored based on operation data obtained by obtaining the state of the equipment to be monitored and a diagnostic model, and if the result of the determination is affirmative, a sign of abnormality indicating the occurrence of a sign of abnormality a sign detection unit that generates alert data;
Based on the operation data corresponding to the anomaly predictive alert data , it is determined whether or not the anomaly predictive alert data is a confirmed report, and if the determination result is negative, the diagnostic model is updated. and a model update control unit for determining. The model update control unit
a predictor experience determination unit that determines whether or not existing similar data satisfying a predetermined similarity criterion exists for the abnormality predictor alert data;
A sign credibility determination unit that determines whether or not the abnormality sign alert data is a confirmed report based on the credibility determination data and the operation data when the determination result of the predictor experience determination unit is affirmative. and
The credibility determination data includes a predictive diagnosis item related to an abnormality predictive sign, a criterion corresponding to the predictive diagnosis item, and evaluation information that determines whether or not the predictive diagnosis item and the criterion correspond to the predictive diagnosis item and the determination criterion. The abnormality predictive diagnosis device according to claim 1, wherein the data includes and. The model update control unit
When the determination result of the predictive experience determination unit is positive and the determination result of the predictive credibility determination unit is negative, either partial update or new creation is selected as an update mode of the diagnostic model. 3. The abnormality predictive diagnosis device according to claim 2, further comprising a model update determination unit for determining whether to update the model. The predictor experience determination unit determines whether or not the existing similar data is stored in a cause diagnosis database that stores the existing abnormality predictor alert data,
The model update control unit
A function of registering in the cause diagnosis database that the existing similar data is a confirmed report when the determination result of the predictor experience determination unit is positive and the determination result of the predictive credibility determination unit is positive. The abnormality predictive diagnosis device according to claim 3, characterized by comprising: The model update control unit
When the determination result of the predictor experience determination unit is negative, based on the state of the monitored equipment during the subsequent follow-up observation period, a confirmed/false alarm determination is made to determine whether the abnormality predictor alert data is a confirmed report. function and
5. The method according to claim 4, further comprising a function of inputting the content of the abnormality predictor alert data into the cause diagnosis database when the judgment result of the positive/false alarm judgment function is affirmative. Abnormal sign diagnosis device. the computer,
Determining whether or not there is a sign of abnormality in the equipment to be monitored based on operation data obtained by obtaining the state of the equipment to be monitored and a diagnostic model, and if the result of the determination is affirmative, a sign of abnormality indicating the occurrence of a sign of abnormality prognostic detection means for generating alert data;
Based on the operation data corresponding to the anomaly predictive alert data , it is determined whether or not the anomaly predictive alert data is a confirmed report, and if the determination result is negative, the diagnostic model is updated. model update control means for judging;
A program to function as

Images