JP2014044510A - Abnormality diagnostic device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device that determines whether a diagnosis target instrument is abnormal or not by evaluating not only the degrees of the respective deviations of the input data of the diagnosis target instrument and reference data but also the time difference between these data and that displays the parameters of a cause determined to be abnormal.SOLUTION: An abnormality diagnostic device comprises: an input data acquisition section 101 configured to display the states of diagnosis target in a time series and acquire diagnostic input data having time information; a reference data acquisition section 102 configured to acquire reference data including the time information and compared with the diagnostic input data; a deviation calculating section 104 configured to calculate the degrees of the respective deviations of the diagnostic input data and reference data on the basis of both the data; a time dependent component calculation section 105 configured to calculate the time dependent components of the degrees of the deviations on the basis of these degrees of the deviations; a determination section 106 configured to determine based on the degree of each deviation whether the diagnosis target is abnormal or not; and an abnormality cause diagnostic section 107 configured to determine the cause of the abnormality in the diagnosis target on the basis of the degrees of the deviations and the time dependent components of the degrees of the deviations when a determination is made that the diagnosis target is abnormal.

Description

  The present invention relates to an abnormality diagnosis apparatus for diagnosing a transient apparatus abnormality.

  An equipment abnormality is generally a state in which equipment constituting equipment, for example, a pump in a nuclear power plant, does not function normally. As an example of a state in which the device is not functioning normally, there are a state in which a part of the device is damaged and the device is stopped, a state in which the device does not perform the designed performance, and the like. Possible causes of such equipment abnormalities include disturbances, defects during manufacturing, construction and repair, aging due to wear and material deterioration, overload during operation, and the like.

  In response to such a device abnormality, a technique for attaching a sensor to a device and determining whether or not the device is operating normally, that is, functioning from its operation data is called abnormality diagnosis. As abnormality diagnosis, a threshold determination method for determining whether a sensor value is constant or falls within a predetermined range is widely used. This method has an advantage that abnormal signs such as a gradual increase in sensor value can be known.

  However, there are cases where the sensor value cannot be expected to be constant. For example, when the operation mode changes before the start of movement of the pump or the prime mover or the device is stabilized, for example, when the set output of the pump is significantly increased or decreased. At a nuclear power plant, there are pumps that are not used during normal operation, but are used when stopped or in an emergency, and this is called surveillance to check whether they work normally in an emergency by moving them periodically. Perform an inspection. In conventional surveillance, devices such as pumps are intended to check whether they can be activated and whether they have a predetermined performance, and not to detect abnormal signs.

  As a method of detecting a symptom when the sensor value does not become a constant value, there are a method of detecting from a steady state portion of time series data and a method of detecting from a transient portion. The present invention is directed to the latter. As a method for diagnosing an abnormality from time-series data in a transient state, there are an MTS (Maharanobis-Taguch System) method, an SBM (Similarity Based Modeling) method, a DP (Dynamic Programming) matching, and the like. In the present invention, since the verification data is minimized, a similarity-based diagnostic method such as DP matching (see, for example, Non-Patent Document 1) is used.

  Patent Document 1 describes the following plant diagnostic apparatus. Based on the signal indicating the operation state of the device input from the signal input unit and the logic stored in the logic database, the logic determination unit determines the plant state candidate case group, and the time-series waveform determination unit inputs from the signal input unit. The plant state is diagnosed by evaluating the similarity based on the distance between the time-series data of the signal representing the processed state and the case data of the candidate case group stored in the case database. In this plant diagnostic apparatus, the current time-series data transmitted from the signal input unit is used as input data, and past time-series data stored in the case database is used as reference data, and the similarity between the input data and the reference data is determined. It is described that it is calculated, and the similarity is evaluated and output by threshold processing.

JP 2002-99319 A

Seiichi Uchida, “DP Matching Overview: Basics and Various Extensions”, IEICE Technical Report, IEICE, December 2006, 106, 428, p. 31-36

  However, in the plant diagnosis apparatus of Patent Document 1, the similarity between the input data and the reference data is calculated based on the similarity of the data change with respect to the passage of time, and the time difference between the input data and the reference data is changed to obtain the highest similarity. The cause of the abnormality is diagnosed based on the reference data in the case of high. Therefore, it is not possible to diagnose a phenomenon in which a time difference occurs between the input data and the reference data as abnormal.

  The abnormality diagnosis device according to claim 1 represents a state of a diagnosis target in time series, an input data acquisition unit that acquires diagnosis input data having time information, and a reference that includes time information and is compared with the diagnosis input data A reference data acquisition unit that acquires data, a divergence degree calculation unit that calculates a divergence degree of both data based on diagnostic input data and reference data, and a time-dependent component of the divergence degree based on the divergence degree A time-dependent component calculation unit, and an abnormality diagnosis execution processing unit that executes an abnormality diagnosis process based on the degree of deviation and the time-dependent component of the degree of deviation are characterized.

  In the abnormality diagnosis apparatus according to the present invention, diagnosis is performed using a time-dependent component of the divergence degree, so that detailed diagnosis based on transient data is possible.

BRIEF DESCRIPTION OF THE DRAWINGS It is the whole schematic diagram of the abnormality diagnosis system containing the diagnostic object apparatus using the abnormality diagnosis apparatus of 1st Embodiment by this invention. It is a functional block diagram of the abnormality diagnosis apparatus of 1st Embodiment by this invention. It is a figure which shows the whole flow of the process in the abnormality diagnosis apparatus of 1st Embodiment by this invention. It is a figure which shows the further detailed process flow in step S307 of the abnormality diagnosis process flow of FIG. It is the figure which materialized the processing flow of the abnormality cause diagnosis in 1st Embodiment. It is a data block diagram of the input time series data in 1st Embodiment. It is a data block diagram of the reference time series data in 1st Embodiment. It is a data block diagram of the weight table in 1st Embodiment. It is a data block diagram of the determination threshold value in 1st Embodiment. It is a data block diagram of the parameter-abnormality cause correspondence table in the first embodiment. It is a processing flow figure which occupies the calculation procedure of the threshold value in 1st Embodiment. It is a screen transition diagram of the abnormality diagnosis device in the first embodiment. It is a graph which shows input time series data and reference time series data when a deviation degree time independent part is large in 1st Embodiment. It is a graph which shows input time series data and reference time series data when a deviation degree time dependence component is large in 1st Embodiment. It is a functional block diagram of the abnormality diagnosis apparatus in 2nd Embodiment. It is a functional block diagram of the abnormality diagnosis apparatus in 3rd Embodiment. It is a figure explaining the preparation method of the reference data also including a transient state in the signal waveform at the time of a state change. It is a figure which shows the flow of the process which produces reference data from the signal waveform of FIG.

Hereinafter, an embodiment of an abnormality diagnosis apparatus according to the present invention will be described with reference to FIGS.
-First embodiment-
In the present embodiment, an example of the abnormality diagnosis apparatus 100 that performs abnormality diagnosis based on the degree of deviation between the input time series data 201 and the reference time series data 202 will be described.

FIG. 1 is a schematic diagram illustrating an overall configuration of an abnormality diagnosis system that diagnoses a diagnosis target device 1001 using the abnormality diagnosis apparatus 100 according to the first embodiment of the present invention.
The abnormality diagnosis system includes an abnormality diagnosis apparatus 100 that performs abnormality diagnosis of the diagnosis target device 1001, a sensor 1002, a controller 1003, an actuator 1004, a display 1005, and a speaker 1006. The sensor 1002 outputs a certain state of the diagnosis target device 1001 as a detection value. The controller 1003 calculates a control command using the detected value. The actuator 1004 gives a state change to the diagnosis target device 1001 in order to diagnose the diagnosis target device according to the control command.

  The sensor 1002 detects a signal corresponding to the state change of the diagnosis target device 1001 whose state has been changed by the actuator 1004. The actuator 1004 changes the state of the diagnosis target device 1001 and the diagnosis target, and differs depending on the state change of the diagnosis target device 1001 and the diagnosis target. Therefore, the sensor 1002 that detects the state change is also an appropriate sensor for detecting the state change of the diagnosis target device 1001 and the diagnosis target. For example, when the actuator 1004 is an actuator that adjusts the opening of the valve, the sensor is a sensor that detects the opening of the valve.

  The sensor 1002 and the controller 1003 are connected to the abnormality diagnosis apparatus 100 via a signal line that communicates a detection value and a control command, and the abnormality diagnosis apparatus 100 displays an abnormality diagnosis result on the display 1005 and the speaker 1006. Or output as audio.

FIG. 2 is a functional block diagram of the abnormality diagnosis apparatus 100 according to the first embodiment of the present invention.
The abnormality diagnosis apparatus 100 includes an input data acquisition unit 101, a reference data acquisition unit 102, an overall deviation calculation unit 103, a divergence degree calculation unit 104, a divergence degree time-dependent extraction unit 105, an evaluation unit 106, an abnormality A cause diagnosis unit 107.
Here, “degree of divergence” is used to mean “degree of dissimilarity”. In the calculation of “similarity” according to the conventional technique, the time lag is matched in the comparison between the reference data and the input data, and therefore the time lag is not determined as abnormal. In the present invention, this time lag is also evaluated as a component of “degree of deviation” between the reference data and the input data.

  The abnormality diagnosis apparatus 100 includes a data storage unit 200 that stores various data. The data storage unit 200 includes an input time-series data storage unit 201, a reference time-series data storage unit 202, a weight table storage unit 203, a determination threshold storage unit 204, and a parameter abnormality cause correspondence table storage unit 205. doing. In the following description, the data stored in each storage unit will be described with the number assigned to the storage unit. For example, the input time-series data is assigned the number 201 assigned to the storage unit.

  The output from the sensor 1002 provided in the diagnosis target device 1001 shown in FIG. 1 is added in real time together with the time information (time) of this output, and stored as input time series data 201 in the data storage unit 200. Sensor output data output from the sensor 1002 is referred to as control output data. In the control output data representing the state of the diagnosis target device 1001, for example, the valve opening, time information that is the time when the valve opening is detected by the sensor 1002 is added to the control output data, and this data is diagnostic input data. .

  The input data acquisition unit 101 acquires the input time series data 201 from the data storage unit 200. The reference data acquisition unit 102 acquires the reference time series data 202 from the data storage unit 200. As will be described later, the reference time series data 202 may be data for comparison prepared in advance, or may be data in which control data from the controller 1003 is stored in real time.

  The overall deviation calculation unit 103 calculates how much the input time series data 201 (see FIG. 6) is translated relative to the reference time series data 202 to minimize the degree of deviation. The amount of translation is recognized as an offset and dead time (described later). The offset is a movement distance in the data axis direction when the divergence degree is minimized. The dead time is a movement distance in the time axis direction when the divergence degree is minimized.

  The divergence degree calculation unit 104 uses the time weight and the output weight of the weight table 203 (see FIG. 8) for the divergence degree between the input time series data 201 and the reference time series data 202 (see FIG. 7). It is calculated by.

The divergence degree time-dependent extraction unit 105 (see FIG. 2) extracts a time-dependent part, that is, a time-dependent component of the divergence degree, from the divergence degree obtained by the divergence degree calculation unit 104. Assuming that the divergence degree calculation method is DP matching as a method for extracting the time-dependent part, the calculation process is remembered when DP matching is performed, and the divergence degree time-dependent extraction unit 105 performs the divergence according to the calculation process. It is obtained by recalculating the time-dependent part of the degree.
Therefore, the divergence degree calculated by the divergence degree calculation unit 104 is divided into a time-dependent component and a time-independent component.

The evaluation unit 106 uses each threshold value of the determination threshold value 204 (see FIG. 9) to output an abnormality or a caution alarm when the deviation degree, the offset, and the dead time exceed the respective threshold values. When an abnormality alarm is output from the evaluation unit 106, the abnormality cause diagnosis unit 107 outputs an abnormality cause from the deviation degree and the time-dependent component of the deviation degree using the parameter-anomaly cause correspondence table 205 (see FIG. 10). To do. The display output unit 108 displays the output alarm and the cause of abnormality on the display 1005 or outputs the sound to the speaker 1006 as sound.
The method for calculating the threshold value will be described later.

(Difference calculation method using DP matching)
Here, a method of calculating a divergence degree using DP matching and a method of calculating a time-dependent component thereof in the abnormality diagnosis apparatus according to the present invention will be described.

DP matching refers to a pattern S between two one-dimensional patterns S = (s1, s2,..., Si,..., SI) and patterns R = (r1, r2,... Ri,..., RI). This is a technique for finding a combination of j = ui that minimizes the objective function F when there is a correspondence j = ui (i = 1,..., I) between the i-th element si and the j-th element rj of the pattern R.
In the first embodiment, the one-dimensional data strings s1 to sI and r1 to rI of each pattern are data strings related to time-series data output from the sensor 1002 that detects a change in the state of the diagnosis target device 1001. .

  Assuming that the distance between the pattern S and the pattern R is di (ui) = ｒsi−ri 総, the sum of di (ui) after DP matching can be regarded as the similarity between the pattern S and the pattern R. This similarity takes a value of 0 or more, and is 0 when the pattern S and the pattern R match. Therefore, the lower the similarity, the higher the matching degree of the waveform.

In the first embodiment, since the data strings of the two one-dimensional patterns S and R are evaluated by separating the time component of the divergence degree, the output value of each data in the data strings of the two patterns And a weighted distance in which a weight is provided for the difference (distance) between detection times.
Therefore, the two one-dimensional patterns S and R are respectively set as S = (s1, s2,..., Si,..., SI), R = (r1, r2,..., Rj,. The pattern divergence f (i, j) at tj is expressed as follows.

Here, ts (i) is the time at sj, tr (j) is the time at rj, cu is a weight for the distance between patterns, and cs is a weight for time. Equations (3), (4), and (5) indicate the initial value calculation method.
Assuming that the partial sum of the weighted distances of S and R is f (i, j), in Equation (1), the local distance is added to f (i−1, j−1) indicated by (a), Or, f (i-1, j) shown in (b) is added with local distance and weighted distance, or f (i, j-1) shown in (c) is added with local distance and weighted distance Select the smallest of the items.

Assuming that the total f (I, J) of the weighted distances thus obtained is F as shown in Equation (2), F can be regarded as the similarity between the pattern S and the pattern R as described above. . Or since F becomes large and a similarity degree becomes small, it can also be considered that F represents the deviation degree of two patterns.
A specific description using an example of time-series data will be described later with reference to FIGS.

(Weight setting method)
In Equation (1), the term of cu (cu‖si-rj‖) and the term of cs (cs {ts (i) -ts (i-1)}, or cs {tr (j) -tr (j-1)) }) Are preferably substantially equal. For example, if the term of cs is extremely large, only (a) is clearly selected. If the term of cu is extremely large, the influence of the term of cs becomes small, and as a result, only (a) is selected.

  As can be selected from (b) and (c), the difference between the average value savr of si and the average value ravr of rj is σ, the sampling period of si is ΔTs, the sampling period of rj is ΔTr, and cu is 1. Assuming that the weight ratio cu: cs is equal to the average value of the sampling period of si and rj and the difference σ, cu: cs = 1: cs = 1 / σ: 1 / {(ΔTs + ΔTr) / 2}, cu = 1, cs = 2σ / (ΔTs + ΔTr).

  This method of determining the weight is merely an example, and it is only necessary that the values of cu and cs take positive values. In addition, the value of cs may be increased when it is desired to provide sensitivity to time differences.

  FIG. 3 is a processing flow of the entire abnormality diagnosis apparatus in the first embodiment. After the start of processing (step S301), the input time series data 201 is first acquired by the input data acquisition unit 101 (step S302). Next, the reference time series data 202 is acquired by the reference data acquisition unit 102 (step S303).

  An overall offset calculation unit 103 (see FIG. 2) adds an appropriate offset and dead time to the input data 201, and an overall offset calculation unit 103 sets an offset value and a dead time at which the degree of deviation is minimized. The divergence degree between the input time-series data 201 and the reference time-series data 202 is calculated by the divergence degree calculation unit 104 using the offset value and the dead time at which the deviation degree is minimized (step S304).

  That is, in the overall deviation calculation unit 103, the offset that is the amount of movement in the data axis direction and the movement in the time axis direction when both data are translated so that the degree of divergence between the diagnostic input data and the reference data is minimized. A dead time, which is a quantity, is calculated.

  In step S305, the evaluation unit 106 determines whether the deviation degree, the offset, and the dead time have exceeded the respective threshold values. If the threshold is exceeded, the divergence degree time-dependent extraction unit 105 calculates the divergence degree time-dependent component (step S306), the abnormality cause diagnosis unit 107 performs abnormality cause diagnosis (step S307), and the display output unit 108 displays the display. An alarm and a cause of abnormality are output to 1005 and the speaker 1006 (step S308).

  FIG. 4 shows a more detailed processing flow of the content of the processing step S307 of the abnormality cause diagnosis of FIG. After starting the process (step S3071), the abnormality cause diagnosis unit 107 (see FIG. 2) reads, from the parameter-abnormal cause correspondence table 205 (see FIG. 10), the one that matches the conditional sentence among the plural conditional sentences (step S3071) S3072, S3073). The conditional statement is compared with the deviation degree, offset, dead time, and deviation degree time-dependent component, and if the condition is met, the cause of the abnormality is determined according to the condition. If not, check whether all conditional statements have been read. If not, read the next conditional statement. When all the conditional statements have been read, determine that the cause of the error is unknown or set a default error cause.

  FIG. 5 is a specific example of cause determination by the abnormality cause diagnosis unit 107 in the first embodiment. However, FIG. 5 shows the concept of cause determination in an easy-to-understand manner and does not show an abnormality diagnosis processing flow. The abnormality cause diagnosing unit 107 reads one conditional statement out of a plurality of conditional statements from the parameter-abnormal cause correspondence table 205. When the overall deviation, that is, the offset and the dead time do not exceed the threshold, the abnormality cause diagnosis unit 107 determines whether the time-dependent part of the deviation degree exceeds the threshold. If the threshold is exceeded, the abnormality cause explanation is regarded as a reaction delay, and if not, the abnormality cause explanation is regarded as excessive output and the process is terminated. If the offset or dead time exceeds the threshold, another conditional statement is read to determine the cause of the abnormality.

  6 is a data configuration diagram of the input time-series data 201 in the first embodiment, and FIG. 7 is a data configuration diagram of the reference time-series data 202 in the first embodiment. The input time-series data 201 and the reference time-series data 202 are each composed of a plurality of data elements, and the data elements are composed of time and actual data values at that time.

  FIG. 8 is a data configuration diagram of the weight table 203 in the first embodiment. The weight table is composed of time weights and output weights. The time weight is a weight to a part depending on the time of the divergence degree when the divergence degree is calculated by the divergence degree calculation unit 104, and the output weight is a weight to a part not depending on the time of the divergence degree.

  FIG. 9 is a data configuration diagram of the determination threshold value 204 in the first embodiment. The determination threshold 204 includes a divergence degree threshold 2041, an offset threshold 2042, and a dead time threshold 2043. Each threshold value indicates a minimum value when an abnormal alarm is output. That is, the determination threshold shown in FIG. 9 is used to determine whether or not to output an alarm.

FIG. 10 is a data configuration diagram of the parameter-abnormality cause correspondence table 205 in the first embodiment. Here, as parameters, a deviation degree, an offset, a dead time, and a time-dependent component of the deviation degree are used.
The parameter-abnormality cause correspondence table 205 includes a plurality of conditional statements 2051. The conditional statement 2051 includes a deviation degree threshold value 20511, a deviation degree inequality sign 20512, an offset threshold value 20513, an offset inequality sign 20514, a dead time threshold value 20515, a dead time inequality sign 20516, a time dependent part inequality sign 20517, a time dependent part inequality sign 20518, It consists of an abnormality cause explanation sentence 20519. When all the parameters such as the degree of divergence are expressed by the relationship between the threshold value and the inequality sign, the abnormality cause explanation 20519 is used to explain the cause of the abnormality.

  When the evaluation unit 106 determines that the diagnosis target is abnormal using the data of the parameter-abnormality cause correspondence table 205 shown in FIG. 10, the details of the content that is abnormal, that is, each of the above threshold values and The parameter determined to be abnormal by the inequality sign is selected and displayed on the display output unit 108.

(Threshold setting method)
Here, the above threshold value setting method will be described with reference to FIG.
First, a plurality of sets of time series data when normal and time series data when abnormality occurs are prepared in advance (step S502). Next, the divergence degree Fni, the offset Oni, and the dead time Tni are obtained for a plurality of sets of normal time series data (step S503), and the maximum values Fnmax, Onmax, Tnmax in the plurality of sets of time series data are obtained (step S503). S504).

  Next, the divergence degree Fai, the offset Oai, and the dead time Tai are obtained for a plurality of sets of time-series data in an abnormal case (step S505), and the minimum values Famin, Oamin, Tamin in the plurality of sets of time-series data are obtained ( Step S506).

The threshold values Fth, Oth, and Tth are obtained from the following equations from Fnmax, Onmax, Tnmax, and Famin, Oamin, and Tamin obtained from a plurality of normal and abnormal time series data as described above (step S507).
Fth = (Fnmax + Famin) / 2
Oth = (Onmax + Oamin) / 2
Tth = (Tnmax + Tamin) / 2

  When there is no abnormal data, pseudo abnormal time series data is created and used by adding noise to normal time series data.

  FIG. 12 is a screen transition of the abnormality diagnosis apparatus 100 according to the first embodiment. In the initial screen shown in (a), a caution alarm and a cause are displayed as an abnormal alarm display screen 401. In addition, a graph of the degree of deviation with respect to the surveillance cycle and a threshold line are also displayed. When an abnormal cause part or an abnormal next part is clicked, an abnormal cause determination graph 402 shown in (b) is displayed. The divergence degree time-independent component of the divergence degree calculated by the divergence degree calculation unit 104 and the time-dependent component of the divergence degree are displayed in a bar graph, and the threshold boundary line and the cause of the abnormality along the boundary line are displayed. To do.

  The bar graph of FIG. 12B shows the following causes. That is, when the time-dependent component of the divergence degree exceeds the threshold, it is an abnormality due to a reaction delay. If the time-dependent component of the divergence does not exceed the threshold, the output is excessive. As described above, the abnormality cause diagnosis unit 107 separately displays the time-dependent component and the time-independent component of the divergence degree. In other words, the abnormality cause diagnosis unit 107 calculates the ratio of the time-dependent component of the divergence degree to the entire divergence degree, and whether the cause of the large divergence degree, i.e., the cause of the decrease in the similarity, is a time system abnormality, It is possible to diagnose whether the output system is abnormal.

  When the bar graph of the deviation degree time-dependent component in FIG. 12B is clicked, a transition is made to the time-dependent part display screen 403 shown in FIG. The time-dependent part display screen 403 displays a graph of a deviation degree time-dependent component with respect to the surveillance cycle and a threshold line. When the bar graph of the divergence degree time-independent portion is clicked in the abnormality cause determination graph 402, a transition is made to the time-dependent portion display screen 404 shown in (d). The time-dependent part display screen 404 displays a graph of a deviation degree time-dependent component with respect to the surveillance cycle and a threshold line.

  FIG. 13 is a graph showing input time-series data 201 (see FIG. 6) and reference time-series data 202 (see FIG. 7) when the divergence degree time-independent portion is large in the first embodiment. The vertical axis represents the values of si and rj in the pattern S indicated by the input time series data 201 and the pattern R indicated by the reference time series data 202, and the horizontal axis represents the time ts (i) at si and the time tr (j) at rj. Each is shown.

A description will be given of how the calculation of the divergence degree F by the above formulas (1) to (5) is performed using the time series data shown in FIG. In DP matching, a combination of data values of i and j where s i and r j take the shortest distance is obtained, and a sum of distances of s i and r j is obtained according to the combination to obtain a divergence degree.
For example, time t5 (i), t5 (j) divergence f (2, 2) is
cu‖s2-r2‖ + f (1,1) (a)
cu‖s2-r2‖ + cs {tr (2) −tr (1)} + f (2,1) (c)
cu‖s2-r2‖ + cs {ts (2) −ts (1)} + f (1,2) (b)
The smallest one of the three divergence degrees is selected.
f (i, j) ≧ 0,
f (2,1) = f (1,1) + cu‖s2-r1‖ + cs {tr (2) −tr (1)},
f (1,2) = f (1,1) + cs {ts (2) -ts (1)} + cu‖s1-r2‖
Therefore, (a) is minimized. As a result of the above determination, if (a) is continuously selected, such as s1 and r1, s2 and r2,. Here, the values of s3 and r3 are larger than the other combinations, but are not shifted in terms of time because j = i, and therefore, the divergence degree time-independent portion is calculated to be large.

FIG. 14 is a graph showing an example of the input time series data 201 and the reference time series data 202 when the deviation degree time-dependent component is large in the first embodiment. The vertical axis represents the values of si and rj in the pattern S indicated by the input time series data 201 and the pattern R indicated by the reference time series data 202, and the horizontal axis represents the time ts (i) at si and the time tr (j) at rj. Each is shown. In this graph, when i = 6,5, the shortest distance can be obtained by continuing to select (a) in equation (1). However, when i = 4, cu‖s4-r4‖ + f (3,3) (a)
cu‖s4-r4‖ + cs {ts (4) −ts (3)} + f (3,4) (b),
cu‖s4-r4‖ + cs {tr (4) −tr (3)} + f (4,3) (c)
The smallest one of the three divergence degrees is selected. In this example, (a) or (c) will be selected.
f (3,3) contains cu‖s3-r3‖,
It is larger than f (4,3) included in cu‖s4-r3‖ + cs {tr (4) −tr (3)}. Therefore, (c) is selected.

  Thus, when (c) is selected, a value obtained by adding the time-dependent components cs {tr (4) -tr (3)} is calculated as the divergence degree. On the other hand, when calculating f (3,2) from f (2,2), a value obtained by adding the time-dependent components cs {ts (3) -ts (2)} is calculated as the degree of divergence.

According to the abnormality diagnosis apparatus 100 of the first embodiment described above, the following operational effects can be achieved.
(1) Sensor output data representing the state of the diagnosis target device 1001 in time series and having time information is used as diagnostic input data. The diagnostic input data is compared with the reference data, and the divergence degree calculation unit 104 calculates the divergence degree between the two data. Based on the calculated deviation degree, the deviation degree time-dependent extraction unit 105 calculates a time-dependent component of the deviation degree. Based on the calculated divergence degree and the time-dependent component of the divergence degree, the abnormality diagnosis processing is executed by the evaluation unit 106 and the abnormality cause diagnosis unit 107. Since the diagnosis is performed using the time-dependent component of the divergence degree, a detailed diagnosis based on the transient data is possible.

(2) When the evaluation unit 106 determines that there is an abnormality based on the divergence degree, the abnormality cause diagnosis unit 107 identifies the cause of the abnormality using the time-dependent component of the divergence degree. Therefore, the cause of the abnormality can be specified in the abnormality diagnosis based on the transient data.

(3) The overall deviation calculation unit 103 causes an offset that is the amount of movement in the data axis direction when the data is translated so that the degree of divergence between the diagnostic input data and the reference data is minimized, and the time axis direction The dead time, which is the amount of movement, is calculated. The cause of the abnormality is determined by the abnormality cause diagnosis unit 107 based on the deviation degree, the deviation degree time-dependent component, the offset, and the dead time. Therefore, the cause of the abnormality can be specified in detail in the abnormality diagnosis based on the transient data.

(4) The abnormality cause diagnosing unit 107 calculates a ratio of the time-dependent component of the divergence degree to the entire divergence degree, and whether the cause of the large divergence degree, that is, the cause of the decrease in the similarity is an abnormality in the time system, It is possible to diagnose whether the output system is abnormal, and detailed diagnosis is possible.

-Second Embodiment-
The abnormality diagnosis apparatus 100 (see FIG. 1) of the present embodiment performs abnormality diagnosis based on the degree of deviation between the control output from the sensor 1002 and the control command from the controller 1003.

FIG. 15 is a functional block diagram of the abnormality diagnosis apparatus 100 according to the second embodiment.
In the abnormality diagnosis apparatus 100 of FIG. 2, the control output acquisition unit 1011 directly acquires the control output data from the sensor 1002 in real time, and as input time-series data 201 including time information (data acquisition time), that is, diagnosis input data. Use. In addition, the control command acquisition unit 1021 directly acquires the control command data from the controller 1003 in real time and uses it as reference time series data 102 including time information.

  In the second embodiment, in the comparison between the control command data input in real time and the control output data from the diagnosis target corresponding thereto, the relationship between these two data when the target diagnostic device is in a normal state ( It is assumed that the control amount or amplitude and time difference) are known in advance. Therefore, when the target diagnostic device is normal, the overall deviation calculation unit 103 and the divergence degree calculation unit 104 perform calculations by adjusting the difference in amplitude and time between the control command data and the control output data.

  In these operations, when new data is input, old data may be deleted so as to compare a predetermined number of data. Alternatively, all may be stored as in the first embodiment. Therefore, although omitted in FIG. 15, a data storage unit that holds at least a predetermined number of data with respect to control output data, ie, input time series data 101 and control command data, ie, reference time series data 102, is controlled. The output acquisition unit 1011 and the control command acquisition unit 1021 are provided.

  Therefore, this data storage unit may be provided in the data storage unit 200 as in the first embodiment. In this case, data that is temporarily or temporarily stored in the data storage unit is used as the data used in the calculation of the overall deviation calculation unit 103 and the divergence degree calculation unit 104 in FIG.

  According to the abnormality diagnosis apparatus 100 of the second embodiment described above, the same operational effects as those of the first embodiment can be achieved. Further, since the control command data from the controller 1003 is used as reference data and the degree of divergence between the control output data from the sensor 1002 and the reference data is calculated, abnormality diagnosis can be performed in real time.

-Third embodiment-
FIG. 16 is a functional block diagram of the abnormality diagnosis apparatus 100 according to the third embodiment, and shows a configuration in the case where reference data is created from accumulated past input data.
In the abnormality diagnosis apparatus 100 of FIG. 2, the reference data acquisition unit 102 stores the past data storage unit 1022 that stores the input time series data 201 obtained from the input data acquisition unit 101 and the past data in the past data storage unit 1022. The reference data creation unit 1023 is used to create reference time series data 1023. The reference time series data creation unit 1023 sorts past data suitable for the reference data by DP matching with the model data, etc., and takes the moving average of the sorted past data as reference data.

  In FIG. 16, the input time series data 201 is described as being outside the data storage unit 200, but as described in the second embodiment, the input time series data 201 is a control output from the diagnosis target device. As described in the first embodiment, it may be control output data stored in the data storage unit 200.

(Reference data extraction example)
FIG. 17 shows a signal example (S) when the signal change speed is large and the increase / decrease is repeated among the past data used as the reference data. The signal sequence (S) is past data of the data S _{START to} S _END .
In the conventional abnormality diagnosis apparatus, data before only a change in state (for example, portion B) and data after a stable state (for example, portion C) are held as threshold values, as shown in FIG. It was not possible to determine whether or not there was an abnormality based on the change during the state change (transient state). Or, it was not possible to judge abnormality during the change.

  In the abnormality diagnosis apparatus according to the third embodiment, the abnormality of the signal to be diagnosed in the transient state is detected from the already measured signal data such as the signal sequence (S) in FIG. 17 by the processing flow as shown in FIG. Create reference data for accurate detection.

The processing flow of FIG. 18 extracts data S _{START to} S _END within the time range that is considered to be a transient state among the past data, that is, data D1 to DI, and the transient state among the past data within this time range. Is used as reference data by specifying a time range of data representing.

In step S601, 0 is set to the flag FS. In step S602, data D1 to DI are read. In step S603, 1 is set in the variable i, m is set in the variable j, and 1 is set in the variable k. The variable i is a number assigned to the data and takes a value in the range of 1 to I. In this processing flow, when creating reference data from past data, m data groups are picked up k times for data 1 to I, and the variance is calculated k times. Therefore, m is the number of data used for each distribution calculation, and k is the number of distributions.
In step 604, data D _{(k-1) j + 1 to Dkj} are picked up, and in step S605, the variance of these data is obtained.

For example, in the second round (k = 2), data D _{m + 1 to} D _2m are picked up at step 604. In this second round, if it is determined in step S606 that the variance is greater than or equal to a predetermined value, it is determined in step S607 whether or not the flag FS is zero. If the flag FS is 0, the variance becomes equal to or greater than a predetermined value for the first time. Therefore, in step S608, the time t _m corresponding to the _n-th sample data D _{m + 1-n} before the head data D _{m + 1. + 1-n} is stored as the reference data acquisition start time ts, and 1 is set to the flag FS in step S609. When indicating this in the general formula, data _{D (k-1) j +} 1 from the n-th previous sample data D _(k-1) time t _(k-1) corresponding to the _{j + 1-n j + 1- n} is stored as the reference data acquisition start time ts.

  In step S610, kj + 1 is set to the variable i, and k + 1 is set to the variable k. In step S611, it is determined whether or not the variable i is I. If step S611 is negative, the process proceeds to step S604.

For example, in step S604 of the third round (k = 3), data D _{2m + 1 to} D _3m are picked up. In step S605, the distribution of these data is obtained. If the variance is equal to or greater than the predetermined value, the process proceeds to step S607. However, since the flag FS is set to 1 in step S609 in the second round, step S607 is denied and the process proceeds to step S610. In step S610, kj + 1 is set to the variable i, k + 1 is set to the variable k, and the process proceeds to step S611. It is determined whether the variable i is equal to I. If the determination is negative, the process proceeds to step S604 and the same process is repeated.

Here, in step S606, for example, when it is determined that the variance for the m pieces of sample data of the fourth round (k = 4) data D _{3m + 1 to} D _4m is less than a predetermined value, in step S621, It is determined whether the flag FS is 0 or not. When step S609 has already been executed, the flag FS is 1, and the process proceeds from step S621 to step S622. In step S622, it stores the time t _{4m + p} corresponding to the sample data D _{4m + p} after p number from the data D _4m as reference data acquisition end time te. When indicating this in the general formula, it stores the time t _{kj + p} corresponding to the sample data D _{kj + p} after p number from the data D _kj as reference data acquisition end time te.
In step S623, data Ds to De corresponding to times ts to te among the data D1 to DI are acquired as reference data.

With the above processing flow, the reference data Ds to De from the start time ts to the end time te are extracted from the past data of the data S _{START to} S _END in FIG. In FIG. 17, the number of samples before the start of B is n, and the number of samples after the end of C is p. Then, the reference data from the start time ts to the end time te and the time series data for diagnosis are compared, and the divergence degree is calculated.

If it is determined in step S611 that the variable i is I, an error display indicating that reference data acquisition has failed is displayed in step S631, and the process ends. This is because reference data acquisition failed for the following reasons.
That is, for example, when the variance of m data in the first round is equal to or greater than a predetermined value and the reference data acquisition start time ts is stored in step S608, the data up to the data DI is maintained without the variance for subsequent data being less than the predetermined value. This is because the reference data acquisition end time cannot be stored and reference data cannot be generated. Alternatively, when there is no signal pattern in which the variance is greater than or equal to the predetermined value in the data string from Di to DI, Step S606 is denied, and the process in which Step S621 is affirmed is executed up to the data DI until the reference data acquisition start time Since ts cannot be obtained, an error is displayed.

  According to the abnormality diagnosis apparatus 100 of the third embodiment described above, the same operational effects as those of the first embodiment can be obtained. Further, reference data is created by the reference data creation unit 1023 using past input time-series data in the past data storage unit 1022. Therefore, it is not necessary to prepare the reference data specially.

The first to third embodiments described above can be modified as follows.
(1) The evaluation unit 106 determines whether or not the diagnosis target is abnormal based on any one of the divergence degree, the offset, and the dead time, but even if the abnormality is determined using only the divergence degree. Good. The abnormality may be determined only by the offset and the dead time.
(2) The abnormality cause diagnosis unit 107 determines the cause determined to be abnormal based on the deviation degree, the deviation degree time-dependent component, the offset, and the dead time. The cause determined to be abnormal may be determined based on the component.

(3) In the above description, the similarity, that is, the degree of divergence between the diagnostic input data and the reference data is calculated by the DP matching method. Abnormality diagnosis may be performed by an algorithm for calculating similarity.

  The above description is examples of the embodiments and modified embodiments of the present invention, and the present invention is not limited to these embodiments and modified examples. Those skilled in the art can implement various modifications without impairing the features of the present invention.

DESCRIPTION OF SYMBOLS 100 ... Abnormality diagnosis apparatus 101 ... Input data acquisition part 102 ... Reference data acquisition part 103 ... Overall deviation calculation part 104 ... Deviation degree calculation part 105 ... Deviation degree time dependence extraction part 106 ... Evaluation part 107 ... Abnormal cause diagnosis part 108 ... Display output unit 200 ... Data storage unit 201 ... Input time series data 202 ... Reference time series data 203 ... Weight table 204 ... Determination threshold value 2041 ... Deviation degree threshold value 2042 ... Offset threshold value 2043 ... Dead time threshold value 205 ... Parameter-error cause correspondence table 2051 ... Conditional statement 20511 ... Deviation degree threshold 20512 ... Deviation degree inequality sign 20513 ... Offset threshold 20514 ... Offset inequality sign 20515 ... Dead time threshold value 20516 ... Dead time inequality sign 20517 ... Time dependent part inequality sign 20519 ... Abnormal cause explanation 401 ... Abnormal warning Display screen 402 ... abnormality cause determination graph 403 ... time-dependent part display screen 404 ... time-independent unit display screen 1001 ... diagnosis target device 1002 ... sensor 1003 ... controller 1004 ... actuator 1005 ... Display 1006 ... Speaker

Claims (6)

An input data acquisition unit that represents a diagnosis target state in time series and acquires diagnostic input data having time information;
A reference data acquisition unit including time information and acquiring reference data to be compared with the diagnostic input data;
Based on the diagnostic input data and the reference data, a divergence degree calculation unit that calculates the divergence degree of both data,
A time-dependent component calculation unit that calculates a time-dependent component of the divergence based on the divergence;
An abnormality diagnosis apparatus comprising: an abnormality diagnosis execution processing unit that executes abnormality diagnosis processing based on the degree of deviation and a time-dependent component of the degree of deviation. The abnormality diagnosis device according to claim 1,
The abnormality diagnosis execution processing unit
A determination unit that determines whether or not the diagnosis target is abnormal based on the degree of deviation;
An abnormality diagnosis apparatus comprising: an abnormality cause diagnosing unit that determines a cause determined to be abnormal based on the degree of deviation and a time-dependent component of the degree of deviation when the abnormality is determined. The abnormality diagnosis device according to claim 1,
An offset, which is a movement amount in the data axis direction, and a dead time, which is a movement amount in the time axis direction, when the data are moved in parallel so that the degree of divergence between the diagnostic input data and the reference data is minimized. And further comprising an overall deviation calculation unit,
The abnormality diagnosis execution processing unit
A determination unit that determines whether or not the diagnosis target is abnormal based on any one of the degree of deviation, the offset, and the dead time;
An abnormality cause diagnosing unit for determining a cause determined to be abnormal based on the deviation degree, the deviation degree time-dependent component, the offset, and the dead time when the abnormality is determined; Abnormality diagnosis device. The abnormality diagnosis device according to claim 1,
The diagnostic input data is time-series data related to a control output detected when a control command is given to the diagnosis target,
The abnormality diagnosis apparatus, wherein the reference data is time-series data of a control command given to the diagnosis target. The abnormality diagnosis device according to claim 1,
The abnormality diagnosis apparatus, wherein the reference data acquisition unit includes a past data accumulation unit that accumulates the diagnostic input data as past data, and a reference data creation unit that creates the reference data from the past data. The abnormality diagnosis device according to any one of claims 1 to 5,
The calculation of the divergence degree calculation unit uses DP matching, and the abnormality diagnosis device.