JP2019200557A - Abnormality diagnosis apparatus, abnormality diagnosis method, and abnormality diagnosis program - Google Patents

Abstract

To improve diagnosis accuracy after disassembling facility and replacing components.SOLUTION: An abnormality diagnosis apparatus according to the present invention includes: a variable process quantity extraction unit for acquiring a relationship between a plurality of measured values acquired from a facility in a first operation cycle before a predetermined event, for acquiring a relationship between a plurality of measured values acquired from the facility at an early stage of a second operation cycle after the predetermined event, and for identifying of the measured value as a first group measured value whose relationship has changed before and after the event; and a reference data processing unit for update-processing the first group measured value among the plurality of measured values acquired in the first operation cycle based on the first group measured value among the plurality of measured values acquired at the early stage of the second operation cycle and for reference creation processing in which the updated first group measured value and the measured value other than the updated first group measured value among the plurality of measured values acquired in the first operation cycle are used as reference data for diagnosing the facility in the second operation cycle.SELECTED DRAWING: Figure 2

Description

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

  When a large-scale plant such as a power plant or a chemical plant suddenly stops, the restoration cost and the opportunity loss amount become enormous and the social impact is great. Therefore, recently, a technique in which a computer detects a sign of plant abnormality has become widespread. Many technologies diagnose the presence or absence of plant abnormalities based on the difference between the measured values measured from the plant (often referred to in the field as “process quantity”) and some reference value. Yes. The key to improving diagnostic accuracy is how to determine appropriate reference values for comparison.

  The plant abnormality diagnosis device of Patent Document 1 lowers (makes sweet) the reference value to be compared with the current process amount as the plant equipment deteriorates over time. The monitoring / diagnosis system of Patent Document 2 accumulates the inspection results of the equipment with respect to the past process amount of the equipment and sets it as a “plant chart”. The system creates a plant state prediction model from the plant chart and uses the prediction model to predict a reference value to be compared with the current process amount. The diagnostic device of Patent Literature 3 uses adaptive resonance theory and issues a warning when a new category that does not belong to an existing category occurs in a multi-dimensional space of process amount. When a new input variable is added due to the renovation of the plant, the diagnostic device builds a diagnostic model with only the new input variable, and when an existing input variable is deleted, Substitute a constant dummy value.

JP 2003-58233 A JP-A-6-331507 JP 2013-25461 A

However, in Patent Documents 1 to 3, since a past process amount within a certain range is used as a reference value retroactively from the present time, the data amount of the reference value does not change, and improvement in diagnosis accuracy due to accumulation of the reference value cannot be expected. In addition, equipment is often disassembled and part of the parts are updated during periodic inspections. And after such a periodic inspection, the relationship between a plurality of process quantities may be large and discontinuously change. However, Patent Documents 1 to 3 are premised on uniform and continuous deterioration of equipment, and a separate measure is required to cope with a large and discontinuous change between a plurality of process quantities.
Therefore, an object of the present invention is to improve diagnostic accuracy after disassembling equipment and replacing parts.

The abnormality diagnosis apparatus according to the present invention acquires a relationship between a plurality of measured values acquired from a facility during a first operation cycle before a predetermined event, and is acquired from the facility at an early stage of the second operation cycle after the predetermined event. And a variable process amount extraction unit that acquires a measurement value whose relationship has changed before and after the event as a first group measurement value, and a plurality of values acquired in the first operation cycle. Update processing for updating the first group measurement value among the measurement values based on the first group measurement value among the plurality of measurement values acquired in the initial stage of the second operation cycle, and the updated first group measurement value And a reference creation process that uses, as a reference data for diagnosing equipment in the second operation cycle, measurement values other than the updated first group measurement value among the plurality of measurement values acquired in the first operation cycle; A reference data processing unit And, characterized by.
Other means will be described in the embodiment for carrying out the invention.

  According to the present invention, it is possible to improve diagnostic accuracy after disassembling equipment and replacing parts.

It is a figure explaining the environment where the abnormality diagnosis apparatus of this embodiment operates. It is a figure explaining the structure of an abnormality diagnosis apparatus. It is a figure explaining the specific example of reference | standard data. It is a figure explaining the specific example of diagnostic object data. It is a figure explaining the correlation between two types of measured values. It is a figure explaining the change of the correlation between two types of measured values. It is a figure explaining the density ratio between two types of measured values. It is a figure explaining the change of the density ratio between two types of measured values. It is drawing explaining reexamination of a measured value. It is a figure explaining physical model information. It is a figure explaining maintenance history information. It is a flowchart of the 1st processing procedure. It is a flowchart of the 2nd processing procedure. It is a display example of a structure graph. It is a figure explaining several reference data.

  Hereinafter, a mode for carrying out the present invention (referred to as “the present embodiment”) will be described in detail with reference to the drawings. This embodiment is an example of performing abnormality diagnosis of a power plant. However, the present invention is generally applicable to equipment diagnosed by comparing a plurality of measured values with their reference values.

(environment)
An environment in which the abnormality diagnosis apparatus of this embodiment operates will be described with reference to FIG. FIG. 1 is a schematic diagram of the power plant 9. The power generation plant 9 includes a boiler device 4, a power generation device 5, a system 3a, and a system 3b. The system 3a includes a pump 41, a valve 42a, a valve 42b, and a pipe 43. Similar to the system 3a, the system 3b also includes a pump, a valve, and piping (not shown). The boiler device 4 generates steam. The pump 41 supplies the steam to the power generation device 5. The power generation device 5 rotates the steam turbine with the pressure of the steam, and rotates the generator with the rotational force.

  The pump 41 of the system 3a has a room 44a and a room 44b. The room 44a and the room 44b are partitioned by a seal 45 so that steam can pass therethrough. The pump 41 first receives the steam flowing from the boiler device 4 in the room 44a, and moves the received steam to the room 44b. The pump 41 distributes steam from the room 44b in three directions. The three directions are, firstly, the power generation device 5, secondly, any equipment outside the system 3a, and thirdly, any equipment inside the system 3a. The configuration and connection relationship of the system 3b are the same as the configuration and connection relationship of the system 3a. Sensors (pressure gauges) 8 a to 8 f are arranged at important points such as the pipe 43. Sensors 8 h and 8 g are also arranged in the boiler device 4 and the power generation device 5. The boiler control device 6 controls the boiler device 4. The power generation control device 7 controls the power generation device 5.

(Abnormality diagnosis device)
A configuration of the abnormality diagnosis apparatus 1 will be described with reference to FIG. The abnormality diagnosis device 1 is a general computer, and includes a central control device 11, an input device 12 such as a keyboard and a touch panel, an output device 13 such as a display and a speaker, a main storage device 14, an auxiliary storage device 15, and a communication device 16. Prepare. These are connected to each other by a bus. The auxiliary storage device 15 stores reference data 31, diagnosis target data 32, physical model information 33, and maintenance history information 34 (all of which will be described later in detail).

  The fluctuation process amount extraction unit 21 and the reference data processing unit 22 in the main storage device 14 are programs. In the following, when the operation subject is described as “XX section”, it means that the central controller 11 reads “XX section” from the auxiliary storage device 15 and loads it into the main storage device 14, and the processing described later. Means to execute. The abnormality diagnosis device 1 is connected via a network 2 to a boiler control device 6, a sensor 8 h arranged in the boiler device 4, a power generation control device 7, and a sensor 8 g arranged in the power generation device 5. The abnormality diagnosis apparatus 1 is connected via a network 2 to sensors 8a to 8f arranged at important points of piping (not shown in FIG. 2) of the system 3.

(Control value and measurement value)
The power generation control device 7 sends a control value 46a (see FIG. 1) to the power generation device 5. The control value 46a is, for example, a value indicating the amount of power (power generation amount) that the power generation device 5 should generate. The boiler control device 6 sends control values 46 b and 46 c to the boiler device 4. The control value 46b is, for example, a value indicating the temperature of steam that should be generated by the boiler device 4. The control value 46c is, for example, a value indicating the steam pressure that the boiler device 4 should generate. The sensors 8a to 8h measure the vapor pressure at each location. Generally, the control value may be a value that is a control target for the equipment belonging to the power plant 9, and is not limited to the power generation amount, the steam temperature, and the steam pressure. The measured value is also generally only a physical quantity that can be measured from the equipment belonging to the power plant 9, and may be a flow rate, a temperature, or the like. That is, the sensors 8a to 8h may be pressure gauges, flow meters, thermometers, or the like.

(Data to be diagnosed)
Data to be diagnosed by the abnormality diagnosis apparatus 1 of the present embodiment is referred to as “diagnosis target data”. The abnormality diagnosis apparatus 1 determines a diagnosis result of “normal” or “abnormal” for the diagnosis target data. The diagnosis target data is a time-series n-dimensional vector. More specifically, the diagnosis target data includes, for example, current measurement values of the sensors 8a to 8h (a specific example will be described later).

(Reference data)
The data that the abnormality diagnosis device 1 compares with the diagnosis target data when acquiring the diagnosis result is referred to as “reference data”. The reference data is a vector having the same number of dimensions as the diagnosis target data. More specifically, the reference data includes past measurement values of the sensors 8a to 8h (a specific example will be described later). It is known that the power plant 9 was in a “normal” state at the time when the reference data was measured. On the other hand, as a matter of course, it is unclear whether or not the power plant 9 is normal at the time (current) when the diagnosis target data is acquired.

(Regular inspection and operation cycle)
The power plant 9 is regularly inspected, for example, once a month. In this periodic inspection, the entire power plant 9 is stopped, and main facilities such as the boiler device 4, the power generation device 5, and the pump 41 are disassembled. After disassembling, the person inspecting visually checks the parts for damage, wear, defects, etc., and performs nondestructive inspection of the parts as necessary. If a part is damaged, the inspector replaces the part with a new part. If the part is not damaged, the inspector cleans the part and assembles the equipment to the original state.

  After such a periodic inspection, the power plant is in a “semi-new state”. The measured values of a power plant that is in a semi-new state may behave differently than before the regular inspection. This is due to the fact that the parts have been replaced with new ones. Therefore, different behaviors do not indicate “abnormality” of the power plant. However, there is a considerable difference between the diagnosis target data after the periodic inspection and the previous reference data. If the previous reference data is used as it is, the diagnosis result becomes “abnormal”.

  Therefore, it is necessary to review a part of the previous reference data after the periodic inspection. The period between one periodic inspection and the next periodic inspection is called an “operation cycle”. That is, it is necessary to prepare the latest reference data for each operation cycle. Periodic inspection is an event that divides the operating cycle. However, in addition to periodic inspections, various events that divide the operation cycle can be assumed, such as replacement of the entire facility, significant changes in the operation method of the facility, and significant changes in the natural environment.

(Time relationship between diagnosis target data and reference data)
Assume that the power plant was completed on January 1 of the year and started operation immediately. After that, it is assumed that a periodic inspection is performed on the last day of every month. On January 31, the first periodic inspection will be conducted. For simplification of explanation, it is assumed that each facility of the power plant is operating normally without receiving an abnormality diagnosis from January 1 to 30. Then, the diagnosis target data acquired from January 1 to 30 automatically becomes reference data for February 1 to 27. The diagnosis target data from February 1 to 27 is compared with the reference data (the diagnosis target data of the previous month) from February 1 to 27. Unless there is an abnormality in the facility from February 1 to 27, the union of the diagnosis target data from January 1 to 30 and the diagnosis target data from February 1 to 27 is from March 1 to This is the standard data for 30 days. The same applies to the following.

  More generally, the diagnosis target data of the nth operation cycle is compared with the reference data of the nth operation cycle. The reference data of the nth operation cycle is the union of the diagnosis target data of the operation cycles before the n−1th operation cycle after all. However, each time a new nth operation cycle starts, the content of the reference data of the nth operation cycle is reviewed. The initial data of the diagnosis target data in the nth operation cycle is used for this review (detailed later). The “first operation cycle” and the “second operation cycle” in the claims correspond to the “n-1th operation cycle” and the “nth operation cycle”, respectively.

(Specific examples of standard data)
A specific example of the reference data 31 will be described with reference to FIG. In the reference data 31, a measurement value is stored in the measurement value column 102 and a control value is stored in the control value column 103 in association with the date and time stored in the date / time column 101.
The date / time in the date / time column 101 is the year / month / day / hour / minute / second when the sensor 8 including the pressure gauge acquires the measured value of an arbitrary location of the power plant 9.

  The measurement value in the measurement value column 102 is the above-described measurement value. The sub-columns 102a to 102f store the pressure measured by the sensor 8a, the pressure measured by the sensor 8b, and so on. “#” Expresses different values in an abbreviated manner. Here, the measurement value is six dimensions for convenience of space, but may be eight dimensions, which is the same as the number of pressure gauges in FIG.

The control value in the control value column 103 is the control value described above. The sub-columns 103a to 103c store the boiler temperature, boiler pressure, and power generation amount as control values. “♭” expresses different values in an abbreviated manner.
Note that “X ₁ ” or the like is applied to the control value, and “Y ₁ ” or the like is applied to the measured value. This is to intuitively understand the relationship between the input “X ₁ ” and the like and the output “Y ₁ ” and the like for the power plant 9.

  As can be seen from FIG. 3, the reference data 31 is a time-series set of measured values acquired from the sensor 8 of the power plant 9 and control values given to the power plant 9. As described above, it is known that the power plant is normal at each time point.

(Specific example of diagnosis target data)
A specific example of the diagnosis target data 32 will be described with reference to FIG. The configuration of the diagnosis target data 32 is almost the same as the configuration of the reference data 31 (FIG. 3). The only difference is that the diagnosis target data 32 has a summary column 114. Note that lowercase letters “x ₁ ” and the like are applied to the control values, and lowercase letters “y ₁ ” and the like are applied to the measurement values. This is for distinguishing from the uppercase letters of the reference data. As long as the subscripts “i” in the lower right match, “X _i ” and “x _i ” are the same control values, and “Y _i ” and “y _i ” are the same measured values. is there. The difference between uppercase and lowercase is only the difference between whether the value is a reference or a diagnosis target. For example, the following can be seen from FIG.

・ The power plant received a periodic inspection on March 31, 2018.
-After that, on April 1, 2018, 10:00:00, the power plant 9 restarted.
“Initial” is stored in the summary column 114 of records from April 1, 2018, 10:00 to 00:10 to 10:50:00. Of the records of the diagnosis target data 32, the record in which “initial” is stored is hereinafter also referred to as “initial record”. The initial record is a record related to the power plant 9 which is in a semi-new state. As described above, the measurement value of the initial record may indicate a behavior that is not seen in the previous reference data 31.

Now, the reference data 31, focusing on the measurement value Y ₁ and the measurement values Y _2, as the measurement value Y ₁ is increased, the measurement value Y ₂ is also increased to that recognized trending. In the data to be diagnosed before the periodic inspection, when the measured value y ₁ and the measured value y ₂ are focused, it is assumed that the measured value y ₂ tends to increase as the measured value y ₁ increases. However, in the initial record of the diagnostic object data after periodic inspection, as the measured value y ₁ is greater, it may be measured values y ₂ is smaller. The abnormality diagnosis apparatus 1 of the present embodiment extracts a change in the relationship between such two types of measurement values.

(Relationship between measured values)
The relationship between the measured values will be described with reference to FIGS. This embodiment employs a correlation or a density ratio as the relationship between the measured values.

(correlation)
A correlation between two types of measurement values will be described with reference to FIG. Cases 51 a, 52 a, and 53 a are obtained by drawing a point ● representing a combination of “Y ₁ ” and “Y ₂ ” of the simultaneous points in the reference data 31 on the coordinate plane. In case 51a, “Y ₁ ” and “Y ₂ ” have a strong negative correlation. In case 53a, “Y ₁ ” and “Y ₂ ” have a strong positive correlation. In case 52a, “Y ₁ ” and “Y ₂ ” appear to have no correlation.

Cases 51b, 52b, and 53b are drawn on the coordinate plane with points ● representing the combination of “y ₁ ” and “y ₂ ” of the simultaneous points in the initial record. In case 51b, “y ₁ ” and “y ₂ ” have a strong negative correlation. In case 53b, “y ₁ ” and “y ₂ ” have a strong positive correlation. In case 52b, “y ₁ ” and “y ₂ ” appear to have no correlation.

  Now, the range of “−1 or more and 1 or less” that can be correlated is divided into three, a range close to “−1”, a range including “0”, close to “0”, and a range close to “1”. Representative correlations “−1”, “0”, and “1” are assigned to the range (see reference numeral 54).

  A change in correlation between two types of measurement values will be described with reference to FIG. Six types of measurement values are indicated by circular nodes. A solid line drawn between two nodes indicates that the measured values of the two nodes have a representative correlation “1”. A broken line drawn between two nodes indicates that the measured value of the two nodes has a representative correlation “−1”. The absence of a line segment between two nodes indicates that the measured values of the two nodes have a representative correlation “0”. The relationship between the measurement values of the reference data and the relationship between the measurement values of the initial record can be expressed by a “structure graph” that is a set of such nodes and line segments.

  The structure graph 61a shows the relationship between the measured values of the reference data 31, and the structure graph 62a shows the relationship between the measured values of the initial record. Data used when creating the structure graph 61a is a matrix 61b. Data used when creating the structure graph 62a is the matrix 62b. Each matrix stores a representative correlation between the measurement value of the row and the measurement value of the column in the cell at the intersection. Incidentally, when the structure graph 61a and the structure graph 62a are compared, the presence / absence of the line segment or the shape of the line segment (solid line / broken line) is different in two places. This difference is indicated by a double line segment 63a and 63b in the structure graph 63. This difference is due to the fact that parts have been replaced in the periodic inspection.

(Density ratio)
A density ratio between two types of measurement values will be described with reference to FIG. “P (y)” is a probability density when the variable “y” takes a certain range of values. When there are a plurality of “p (y)” such as p ₁ (y), p ₂ (y), p ₃ (y),..., The ratio of the two probability densities can be defined. For example, the density ratio probability density _p 1 (y) of the probability density _p 2 (y) is defined as follows.
Density ratio = p ₂ (y) / p ₁ (y)

For convenience of explanation, if “Y ₁ ” and “Y ₂ ” are regarded as the same variable, and normalized so that they can be compared more easily, the following density ratio can be defined.
Density ratio = p (Y ₂ ) / p (Y ₁ )

Now, let p (Y ₁ ) be as in case 71a in FIG. 7, and p (Y ₂ ) be as in case 71b in FIG. That is, the average and variance of p (Y ₁ ) are assumed to be substantially the same as the average and variance of p (Y ₂ ). Then, the density ratio graph is almost horizontal in the entire section of Y ₁ (Y ₂ ) (case 71c).

Similarly, if “y ₁ ” and “y ₂ ” are considered to be the same variable, and normalized so that they are easier to compare, the following density ratio can be defined.
Density ratio = p (y ₂ ) / p (y ₁ )

Now, let p (y ₁ ) be as in case 72a in FIG. 7, and p (y ₂ ) be as in case 72b in FIG. That is, the variance of p (y ₁ ) is almost the same as the variance of p (y ₂ ), but the average of p (y ₁ ) is smaller than the average of p (y ₂ ). Then, in all the sections of y ₁ (y ₂ ), the graph of the density ratio is an approximately upper right curve (case 72c).

Similarly, it is assumed that p (y ₁ ) is as in case 73a in FIG. 7, and p (y ₂ ) is as in case 73b in FIG. That is, the variance of p (y ₁ ) is almost the same as the variance of p (y ₂ ), but the average of p (y ₁ ) is greater than the average of p (y ₂ ). Then, in the entire section of y ₁ (y ₂ ), the graph of the density ratio is a curve that descends to the right (Case 73c).

Similarly, it is assumed that p (y ₁ ) is as in case 74a in FIG. 7 and p (y ₂ ) is as in case 74b in FIG. That is, the average of p (y ₁ ) is almost the same as the average of p (y ₂ ), but the variance of p (y ₁ ) is smaller than the variance of p (y ₂ ). Then, in the entire section of y ₁ (y ₂ ), the density ratio graph is a curve that is convex downward (case 74c).
In addition, the graph of the density ratio of the cases 71c to 74c is an unrealistic straight line that emphasizes the feature of the shape for the purpose of explanation.

  A change in density ratio between two types of measurement values will be described with reference to FIG. The matrix 81 stores the shape of the graph of the density ratio between the measured value of the reference data 31 in the row and the measured value of the reference data 31 in the column in the cell at the intersection. The shape of the density ratio graph is, for example, “horizontal”, “upper right”, “lower right”, “convex downward” as described above, but other examples can be used as long as they can be categorized graphically. May be included. "#" Expresses different shapes (types) of the density ratio graph in an abbreviated manner.

The matrix 82 stores the change in the shape of the graph of the density ratio between the measured value of the initial record in the row and the measured value of the initial record in the column in the cell at the intersection. Here, “Yes” indicates that the shape of the density ratio graph has changed as compared to the matrix 81 (for example, horizontal → upper right), and “No” indicates that the shape of the density ratio graph is the matrix. It shows that there is no change compared to 81. The double line segment 83a in the structure graph 83 is compared with the shape of the density ratio graph of Y ₁ and Y ₄ in the reference data 31 in the shape of the density ratio graph of y ₁ and y ₄ in the initial record. Shows that it is changing. Segment 83b of the double line is changed compared to the shape of the graph of the density ratio of Y ₂ and Y ₄ in the y ₂ and the reference data 31 is the shape of the graph of the density ratio of y ₄ in the initial record It shows that. These changes are also caused by the replacement of parts and the like in the periodic inspection.

(Review of measured values of standard data)
The review of the measurement value will be described with reference to FIG. FIG. 9 is also the same reference data 31 as FIG. The date and time of the record in FIG. 9 is the same as in FIG. However, in FIG. 9, some of the measured values Y ₁ , Y ₂ and Y ₄ are reviewed as compared to FIG. The review of measured values is to recognize the behavior of measured values after periodic inspection using the initial record, delete the measured values that deviated greatly from the behavior, and delete the measured values. It is to update with another measured value.

The structure graph 63 in FIG. 6 and the structure graph 83 in FIG. 8 indicate that the measurement values Y ₁ , Y _2, and Y ₄ in the reference data 31 should be reviewed. Therefore, the abnormality diagnosis apparatus 1 first deletes all measured values in the columns Y ₁ , Y _2, and Y ₄ from the reference data 31. Next, the abnormality diagnosis apparatus 1 determines the measurement values Y _r1 , Y _r2, and Y _r4 for which the deleted measurement values are to be updated for each record of the reference data. Y _r1 , Y _r2 and Y _r4 are called “updated measurement values”. “$” In FIG. 9 indicates different update measurement values in an abbreviated manner.

The abnormality diagnosing device 1 is configured such that “(Y ₁ , Y ₂ , Y ₃ , Y ₄ , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )” among “(Y ₁ , Y ₂ , Y ₃ , X ₃ , X ₃ , X ₃ )”. Y ₂ , Y ₄ ) ”is updated with“ (Y _r1 , Y _r2 , Y _r4 ) ”, and“ (Y _r1 , Y _r2 , Y ₃ , Y _r4 , Y ₅ , Y ₆ , X ₁ , X ₂ ) is updated. , X ₃ ) ”. At this time, the abnormality diagnosis apparatus 1 does not randomly determine the updated measurement value. The abnormality diagnosis apparatus 1 determines the update measurement value so that the vector of each record of the reference data 31 including the update measurement value is as similar as possible to the initial record. There are various methods for determining such updated measurement values. For example, the abnormality diagnosis apparatus 1 uses the following method.

<Method 1: k-nearest neighbor method>
<1-1> abnormality diagnostic apparatus 1, the records that reference data from the _{"(Y 1, Y 2,} Y 3, Y 4, Y 5, Y 6, X 1, X 2, X 3)" of , Y ₁ , Y ₂ and Y ₄ are deleted to create “(Y ₃ , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )”.
<1-2> The abnormality diagnosis apparatus 1 is a vector having the closest Euclidean distance to “(Y ₃ , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )”, the second closest vector,. , K-th vector is extracted from “(y ₃ , y ₅ , y ₆ , x ₁ , x ₂ , x ₃ )” of the initial record. For the sake of simplicity, it is assumed that k = 2.
<1-3> abnormality diagnosis apparatus 1, the average value of the _{y 1} corresponding to the vector close to _{y 1} and the second corresponding to the closest vector to _{Y r1.}
<1-4> The abnormality diagnosis apparatus 1 similarly obtains Y _r2 and Y _r4 .
<1-5> The abnormality diagnosis apparatus 1 repeats the processes <1-1> to <1-4> for all records of the reference data.

<Method 2: Collaborative filtering method>
Method 2 is the same as Method 1 except for the following <2-2>.
<2-2> The abnormality diagnosis apparatus 1 is a vector having the highest cosine similarity with “(Y ₃ , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )”, the second largest vector,. The kth largest vector is extracted from “(y ₃ , y ₅ , y ₆ , x ₁ , x ₂ , x ₃ )” of the initial record. The cosine similarity is the cosine of the angle formed by two vectors (-1 to 1), and the greater the cosine similarity (closer to “1”), the more the two vectors are considered to be similar. The For the sake of simplicity, it is assumed that k = 2.

<Method 3: Miss Forest method>
<3-1> The abnormality diagnosis apparatus 1 uses “(y ₁ , y ₂ , y ₃ , y ₄ , y ₅ , y ₆ , x ₁ , x ₂ , x ₃ )” of all initial records as learning data. Create a predictive model. The prediction model is a mathematical formula including these nine variables.
<3-2> abnormality diagnosis apparatus 1, compared prediction _{model, "(Y 3, Y 5} , Y 6, X 1, X 2, X 3)" by substituting a unknown _"(Y 1, Y ₂ , Y ₄ ) ”.
<3-3> The abnormality diagnosis device 1 sets the obtained “(Y ₁ , Y ₂ , Y ₄ )” as “(Y _r1 , Y _r2 , Y _r4 )”.
<3-4> The abnormality diagnosis apparatus 1 repeats the processes <3-1> to <3-3> for all records of the reference data.

<Method 4: Method using probability distribution>
<4-1> The abnormality diagnosis apparatus 1 obtains the probability distribution of y ₁ of the initial record.
<4-2> The abnormality diagnosis apparatus 1 generates Y _r1 so as to follow the probability distribution of y ₁ . Now, if there are 100 records of reference data, the abnormality diagnosis apparatus 1 generates 100 values of Y _r1 . These 100 values are in accordance with the probability distribution of _{y 1.}
<4-3> The abnormality diagnosis apparatus 1 similarly generates Y _r2 and Y _r4 .
<4-4> The abnormality diagnosis apparatus 1 updates “(Y ₁ , Y ₂ , Y ₄ )” of each record of the reference data with “(Y _r1 , Y _r2 , Y _r4 )”. Y _r1 used in a record with reference data cannot be used in other records. The same applies to Y _r2 and Y _r4 . Then, the number in the case where “(Y ₁ , Y ₂ , Y ₄ )” of 100 records is updated with “(Y _r1 , Y _r2 , Y _r4 )” is (100 × 99 × 98 ×... × 1) ³

<4-5> The abnormality diagnosis apparatus 1 obtains the center of gravity of the initial record.
<4-6> The abnormality diagnosis apparatus 1 includes the reference data “(Y _r1 , Y _r2 , Y ₃ , Y _r4 , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )” and each of the initial records. Find the sum of the distance to the center of gravity.
<4-7> The abnormality diagnosis apparatus 1 sets the case where the sum of distances is minimized as the updated reference data.

  In addition to the above, the abnormality diagnosis apparatus 1 may use a multiple substitution method. Eventually, the abnormality diagnosis method performs arbitrary processing such that the relationship (correlation or density ratio) between the measured values of the reference data after the update is as close as possible to the relationship between the measured values of the initial record.

  In the above description, there is a premise that if a certain operation cycle actually starts and an initial record of diagnosis target data is not acquired, a change in the relationship between measured values due to the periodic inspection is not found. However, changes in the relationship between measured values can be predicted to some degree theoretically or technically.

(Physical model information)
The physical model information 33 will be described with reference to FIG. In the physical model information 33, in association with the equipment name stored in the equipment name column 121, the maintenance content column 122 stores the maintenance content, and the theoretical influence measurement value column 123 stores the theoretical influence measurement value. ing.
The equipment name in the equipment name column 121 is the name of the equipment constituting the power plant 9. Here, for convenience of explanation, the code at the end of the equipment name (such as 42b) is made to match the code in FIG.
The maintenance content in the maintenance content column 122 is the content of maintenance work performed on the equipment.
It is predicted theoretically or technically that the relationship between the theoretically affected measured value in the theoretically affected measured value column 123 changes compared to before the maintenance work is performed on the equipment. It is a measured value.

For example, pay attention to the second line in FIG. “Y ₃ ” is the pressure in the pipe immediately upstream of the valve 42b (FIG. 1). “Y ₅ ” is the pressure in the pipe immediately downstream of the valve 42b. The relationship between “Y ₃ ” and “Y ₅ ”, such as the correlation or density ratio described above, changes before and after the overhaul of the valve 42b. The combination of the measured values that change can be predicted theoretically or technically based on the design conditions of the equipment, or natural laws such as equations of motion and laws of inertia. And the prediction is supported with considerable accuracy by computer simulation. In general, such a design condition is called a deductive “physical model” as a concept for an inductive statistical model. The reason for naming “physical model information” 33 also exists here.

(Maintenance history information)
The maintenance history information 34 will be described with reference to FIG. In the maintenance history information 34, the maintenance content is stored in the maintenance content column 132 and the maintenance time is stored in the maintenance time column 133 in association with the equipment name stored in the equipment name column 131.
The equipment name in the equipment name column 131 is the same as the equipment name in FIG.
The maintenance content in the maintenance content column 132 is the same as the maintenance content in FIG.
The maintenance time in the maintenance time column 133 is a time when maintenance work is performed on the equipment. For example, “immediately after the end of the (n-1) th operation cycle” means that after the completion of the (n-1) th operation cycle, maintenance work (periodic inspection) was performed immediately before the start of the nth operation cycle.

  Hereinafter, the processing procedure of this embodiment will be described. As processing procedures, there are a first processing procedure and a second processing procedure. The second processing procedure utilizes the physical model information 33 and the maintenance history information 34 described above. The second processing procedure includes the contents of the first processing procedure (details will be described later).

(First processing procedure)
The first processing procedure will be described with reference to FIG. It is assumed that the reference data 31 (FIG. 3) is already stored in the auxiliary storage device 15 as a premise for starting the first processing procedure. It is assumed that the periodic inspection is completed and the power plant 9 is restarted now that the nth operation cycle has started.

In step S <b> 201, the fluctuation process amount extraction unit 21 of the abnormality diagnosis apparatus 1 acquires diagnosis target data 32. Specifically, the variable process amount extraction unit 21 acquires measurement values y _{1 to} y ₆ from the sensor 8, acquires x ₂ and x ₃ from the boiler control device 6, and acquires x ₁ from the power generation control device 7. To do. Thereafter, the variable process amount extraction unit 21 accumulates the acquired data as a record of the diagnosis target data 32 (FIG. 4).

In step S202, the variable process amount extraction unit 21 specifies an initial record. Specifically, first, the variable process amount extraction unit 21 waits for a trial operation period of a predetermined length to elapse. The trial operation period is a period (for example, 60 minutes) long enough to know the behavior of each device after the periodic inspection.
Second, after the trial run period has elapsed, the variable process amount extraction unit 21 stores “initial” in the summary column 114 of the records of the diagnosis target data 32 acquired during the trial run period (initial record).

  In step S203, the variable process amount extraction unit 21 compares the initial record with the reference data. Specifically, the fluctuation process amount extraction unit 21 creates a combination of two types of measurement values by the method described with reference to FIGS. 5 and 6, and for all the combinations, the correlation in the reference data and the initial record Compare with correlation. At this time, the fluctuation process amount extraction unit 21 creates a combination of two types of measurement values by the method described in FIGS. 7 and 8, and for all the combinations, the shape of the density ratio graph in the reference data, You may compare with the shape of the graph of the density ratio in an initial record.

  In step S204, the fluctuation process amount extraction unit 21 extracts a change in the relationship between the measurement values. Specifically, first, the variable process amount extraction unit 21 extracts a measurement value whose correlation has changed by the method described with reference to FIGS. 5 and 6. At this time, the fluctuating process amount extraction unit 21 may extract a measurement value in which the shape of the density ratio graph is changed by the method described with reference to FIGS. 7 and 8. The fluctuating process amount extraction unit 21 extracts two measured values whose relationship has changed, but ignores the overlap.

  Second, the variable process amount extraction unit 21 displays the measurement value extracted in “first” in step S204 on the output device 13 in the form of a structure graph. An example of the screen display at this time is shown in FIG. The fluctuating process amount extraction unit 21 may hide nodes of measurement values that have not changed in relationship. Further, the variable process amount extraction unit 21 may display a system diagram of the system including the nodes appearing in the structure graph and highlight the nodes appearing in the structure graph.

  In step S205, the reference data processing unit 22 of the abnormality diagnosis apparatus 1 updates a part of the reference data. Specifically, the reference data processing unit 22 determines the update measurement value using any one of the update measurement value determination methods described above.

In step S206, the reference data processing unit 22 stores the updated reference data. Specifically, the reference data processing unit 22 updates (overwrites) the measurement value extracted in “first” of step S204 with the updated measurement value determined in step S205 among the measurement values of the reference data 31. And remember). As a result, according to the above example, each record of the reference data 31 has “(Y _r1 , Y _r2 , Y ₃ , Y _r4 , Y ₅ , Y ₆ , X ₁ , X ₂ , X ₃ )”. Data is stored.
Thereafter, the first processing procedure is terminated.

(Second processing procedure)
The second processing procedure will be described with reference to FIG. As a premise for starting the second processing procedure, it is assumed that the reference data 31 (FIG. 3), the physical model information 33 (FIG. 10), and the maintenance history information 34 (FIG. 11) are already stored in the auxiliary storage device 15. . It is assumed that the periodic inspection is completed and the power plant 9 is restarted now that the nth operation cycle has started.

  The processing in steps S201 to S204 in FIG. 13 is the same as the processing in steps S201 to S204 in FIG.

  In step S204b, the fluctuating process amount extraction unit 21 of the abnormality diagnosis apparatus 1 extracts a theoretical affected measurement value. Specifically, first, the variable process amount extraction unit 21 refers to the maintenance history information 34 (FIG. 11), and what kind of maintenance is performed on which equipment immediately before the current operation cycle starts. Identify For example, in the maintenance time column 133 of the maintenance history information 34, the closest to the present (the nth operation cycle) is “immediately after the end of the n-1th operation cycle” in the first row. That is, the variable process amount extraction unit 21 recognizes that the seal replacement of the pump 41 has been performed in the immediately preceding periodic inspection.

Second, the variable process amount extraction unit 21 refers to the physical model information 33 (FIG. 10) using “pump 41” and “seal replacement” identified in “first” in step S204b as search keys, and the corresponding record. Is obtained as “qL”. Here, “qL = {Y ₁ , Y ₂ , Y ₃ , Y ₄ }” is acquired.

Third, the variable process amount extraction unit 21 acquires the measurement value extracted in “first” in step S204 as “PL”. Here, it is assumed that “PL = {Y ₁ , Y ₂ , Y ₄ }” has been acquired. “QL” is a set of measurements whose relationship should change theoretically or technically. On the other hand, “PL” is a set of measurement values in which a change in the relationship is recognized as a result of statistical processing of actual measurement values regardless of the theory.
Note that “PL” and “qL” here correspond to “first group measurement value” and “second group measurement value” in the claims, respectively.

  In step S204c, the variable process amount extraction unit 21 determines whether or not PL⊆qL is established. Specifically, when the PL is a subset of qL (step S204c “Yes”), the variable process amount extraction unit 21 proceeds to step S205, and when the PL is not a subset of qL (step S204c “No”). ), The process proceeds to step S207. Incidentally, the case where PL is not a subset of qL means that there is a measurement value (referred to as “singular measurement value”) that is an element of PL and not an element of qL. The presence of a singular measurement value indicates that there is a high possibility that an abnormality has occurred in the equipment related to the singular measurement value.

  The processes in steps S205 and S206 in FIG. 13 are the same as the processes in steps S205 and S206 in FIG. However, it should be noted that in step S205 of FIG. 13, the reference data processing unit 22 of the abnormality diagnosis apparatus 1 determines an updated measurement value for all measurement values included in the PL, and ignores qL as a result. is there. After step S206, the second processing procedure ends.

  In step S207, the reference data processing unit 22 of the abnormality diagnosis apparatus 1 deletes the reference data 31 (FIG. 3). Specifically, the reference data processing unit 22 deletes all records of the reference data 31.

In step S208, the reference data processing unit 22 outputs an alarm. Specifically, the reference data processing unit 22 outputs an alarm indicating that an abnormality has occurred in the equipment related to the singular measurement value to the output device 13 by voice or text. An example of the alarm here is “please immediately check the equipment related to the measured value Y _i ”. The reference data processing unit 22 may display a system diagram including the equipment from which the singular measurement value Y _i is acquired. Thereafter, the second processing procedure is terminated.

(Modification of the second processing procedure)
The reference data processing unit 22 may execute the following processing in steps S207 and S208 described above.

  In step S207, the reference data processing unit 22 determines the update measurement value using any one of the above-described update measurement value determination methods. However, at this time, the reference data processing unit 22 determines updated measurement values for all measurement values included in qL, not the measurement values included in PL.

  In step S208, the reference data processing unit 22 updates (overwrites and stores) all the measurement values included in qL among the measurement values of the reference data 31 with the updated measurement values determined in step S207. Thereafter, the second processing procedure is terminated.

(Abnormal diagnosis)
The abnormality diagnosis apparatus 1 diagnoses the power plant 9 by comparing the diagnosis target data with the reference data after the first processing procedure or the second processing procedure. At this time, the abnormality diagnosis apparatus 1 may or may not include the initial record in the diagnosis target data to be compared. Furthermore, the abnormality diagnosis apparatus 1 can arbitrarily determine a specific method for comparing diagnosis target data with reference data. For example, the abnormality diagnosis apparatus 1 calculates the Euclidean distance between the representative value of the diagnosis target data (for example, the center point of the cluster) and the representative value of the reference data in a multidimensional space, and applies a predetermined threshold to the distance. Alternatively, “abnormal” or “normal” may be output as the diagnosis result.

(Operation cycle in which reference data is acquired)
In the above, the reference data of the nth operation cycle is the union of the diagnosis target data before the n−1th operation cycle. That is, each time a new operation cycle starts, the reference data is accumulated while being partially updated. As a result, the data amount of the reference data gradually increases, but a plurality of reference data does not exist.

  It is desirable that the power generation amount of the power plant is maintained at a certain level. Therefore, the control value is often set to approximately the same value. In many cases, the power generation amount is adjusted by changing the number of systems to be operated. Therefore, in the above description, the abnormality diagnosis apparatus 1 has been based on the assumption that there is no room for selecting reference data from a plurality of candidates.

However, the abnormality diagnosis apparatus 1 may select reference data suitable for comparison with the diagnosis target data of the current operation cycle from a plurality of candidates. Hereinafter, specific processing will be described with reference to FIG.
The reference data processing unit 22 of the abnormality diagnosis apparatus 1 cuts out past diagnosis target data for each operation cycle and prepares it as n−1 reference data candidates that can be used for diagnosis in the n-th operation cycle (see FIG. 15 code 91).
The reference data processing unit 22 reviews each of the n−1 candidate reference data using the initial record of the diagnosis target data of the nth operation cycle (reference numeral 92 in FIG. 15).

The reference data processing unit 22 stores the revised reference data candidates in the auxiliary storage device 15 (reference numeral 93 in FIG. 15). Here, “first operation cycle base” or the like indicates that the reference data of the nth operation cycle is created based on the diagnosis target data of the first operation cycle.
- reference data processing unit 22, among the diagnostic object data of the n operating cycles, to obtain a control value x _i at the time of actually performing the diagnosis.
The reference data processing unit 22 selects reference data having a control value X _i that is similar to the acquired control value x _i that satisfies a predetermined reference from among the revised reference data candidates.

(Effect of this embodiment)
The effects of the abnormality diagnosis device of this embodiment are as follows.
(1) The abnormality diagnosis apparatus can create reference data that is truly suitable for comparison with the current diagnosis target data of the facility.
(2) The abnormality diagnosis apparatus can accumulate the reference data while reviewing the measurement values only when a change between the measurement values that can be assumed theoretically or technically occurs.
(3) The abnormality diagnosis device can output a warning when a change between measured values that cannot be assumed theoretically or technically occurs.
(4) The abnormality diagnosis device can maintain diagnosis accuracy even after parts are replaced by inspection or the like.
(5) The abnormality diagnosis apparatus can statistically recognize a change in the relationship between the measurement values.
(6) The abnormality diagnosis apparatus can create reference data close to the current measurement value of the facility.
(7) The abnormality diagnosis apparatus can select an appropriate one for the current control value from among a plurality of reference data candidates.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a recording device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
Further, the control lines and information lines are those that are considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. In practice, it may be considered that almost all the components are connected to each other.

DESCRIPTION OF SYMBOLS 1 Abnormality diagnosis apparatus 2 Network 3 System | strain 4 Boiler apparatus 5 Electric power generation apparatus 6 Boiler control apparatus 7 Electric power generation control apparatus 8 Sensor (pressure gauge)
9 Power Plant 11 Central Controller 12 Input Device 13 Output Device 14 Main Storage Device 15 Auxiliary Storage Device 16 Communication Device 21 Fluctuating Process Amount Extraction Unit 22 Reference Data Processing Unit 31 Reference Data 32 Diagnosis Target Data 33 Physical Model Information 34 Maintenance History Information

Claims (9)

Acquire the relationship between multiple measured values acquired from the equipment in the first operating cycle before a predetermined event,
Acquiring a relationship between a plurality of measured values acquired from the facility at an early stage of the second operation cycle after the predetermined event;
A variable process amount extraction unit that identifies a measurement value whose relationship has changed before and after the event as a first group measurement value,
The first group measurement value of the plurality of measurement values acquired in the first operation cycle is based on the first group measurement value of the plurality of measurement values acquired in the initial stage of the second operation cycle. Update process to update,
The updated first group measurement value and a measurement value other than the updated first group measurement value among a plurality of measurement values acquired in the first operation cycle are used as the equipment in the second operation cycle. A reference data processing unit for performing reference creation processing as reference data for diagnosing
An abnormality diagnosis device characterized by the above. A storage unit that stores physical model information in which a plurality of measurement values that are theoretically or technically predicted to change in relation to the predetermined event after the predetermined event are stored as a second group measurement value With
The variable process amount extraction unit
With reference to the physical model information, the second group measurement value corresponding to the event that actually occurred,
The reference data processing unit is
The update process and the reference creation process are performed only when the first group measurement value is a subset of the second group measurement value.
The abnormality diagnosis apparatus according to claim 1. The reference data processing unit is
If the first group measurement value is not a subset of the second group measurement value, an abnormality has occurred in the equipment related to the measurement value that is not included in the second group measurement value and included in the first group measurement value. Output a warning that
The abnormality diagnosis apparatus according to claim 2. The predetermined event is:
Inspection of the equipment,
The inspection
Including disassembly of the equipment or replacement of parts of the equipment;
The abnormality diagnosis apparatus according to claim 3. The relationship between the measured values is
A correlation or density ratio between the two measured values,
The abnormality diagnosis apparatus according to claim 4. The reference data processing unit is
Among the measurement values acquired in the first operation cycle, the measurement value other than the first group measurement value, and among the measurement values acquired in the initial stage of the second operation cycle, the measurement value other than the first group measurement value Performing the update process based on the distance to
The abnormality diagnosis apparatus according to claim 5. The reference data processing unit is
In the case where the predetermined event repeatedly occurs, for each of the plurality of first operation cycles delimited by the event, create a plurality of the reference data for diagnosing the equipment in the second operation cycle,
Selecting one of the plurality of created reference data based on a control value for the facility at the time of diagnosis of the facility;
The abnormality diagnosis apparatus according to claim 6. The fluctuation process amount extraction unit of the abnormality diagnosis device
Acquire the relationship between multiple measured values acquired from the equipment in the first operating cycle before a predetermined event,
Acquiring a relationship between a plurality of measured values acquired from the facility at an early stage of the second operation cycle after the predetermined event;
A measurement value in which the relationship has changed before and after the event is identified as a first group measurement value,
The reference data processing unit of the abnormality diagnosis device is
The first group measurement value of the plurality of measurement values acquired in the first operation cycle is based on the first group measurement value of the plurality of measurement values acquired in the initial stage of the second operation cycle. Update process to update,
The updated first group measurement value and the measurement values other than the updated first group measurement value among the plurality of measurement values acquired in the first operation cycle are the facilities in the second operation cycle. A reference creation process that is used as reference data for diagnosing
An abnormality diagnosis method for an abnormality diagnosis apparatus characterized by the above. For the variable process quantity extraction part of the abnormality diagnosis device,
Acquire the relationship between multiple measured values acquired from the equipment in the first operating cycle before a predetermined event,
Acquiring a relationship between a plurality of measured values acquired from the facility at an early stage of the second operation cycle after the predetermined event;
The process of specifying the measurement value whose relationship has changed before and after the event as the first group measurement value is executed,
For the reference data processing unit of the abnormality diagnosis device,
The first group measurement value of the plurality of measurement values acquired in the first operation cycle is based on the first group measurement value of the plurality of measurement values acquired in the initial stage of the second operation cycle. Update process to update,
The updated first group measurement value and a measurement value other than the updated first group measurement value among a plurality of measurement values acquired in the first operation cycle are used as the equipment in the second operation cycle. A reference creation process as reference data for diagnosing
An abnormality diagnosis program for causing an abnormality diagnosis device characterized by the above to function.

Images