JPH06314199A - Process error cause estimation method - Google Patents

Abstract

(57) [Summary] (Modified) [Purpose] In water treatment plants, chemical plants, etc.
Accurately estimate the cause of the process abnormality. [Structure] The state in the process to be diagnosed is represented by some information items, and conditions for judging a measured value or the like corresponding to each information item as abnormal are set in advance. When estimating the cause of an abnormality, it is assumed that all information items that can be candidates for the cause of the abnormality are the causes of the abnormality, and whether or not a chain of several routes leading to the result is established. Search while checking the status of information items,
An information item that is assumed to be the cause of the abnormality and satisfies the condition that the entire route formed by a chain of several routes reaches all the information items indicating the abnormality is estimated as a candidate for the cause of abnormality. Further, among the information items, for the information item for which information indicating the information is not obtained, the state of the information item is assumed to be a certain state so that the chain of the results is not interrupted due to the assumed cause of the abnormality.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process abnormality cause estimating method applied to various process plants such as water treatment plants and chemical treatment plants.

[0002]

2. Description of the Related Art Generally, a computer-based automatic process abnormality diagnosis method uses a rule-based diagnosis method based on empirical knowledge of an expert in a target field and a structure information or design information of a target object. It is said that there are two ways of thinking of a logical diagnostic method that is directly used in.

In the rule-based diagnostic method, the patterns of occurrence of all abnormal states that can be assumed from past experience or current knowledge are sorted out, and the abnormal patterns actually occurring are identified. The cause is inferred by checking whether it is applicable. As a specific process, a table of patterns that appear due to an abnormality is created in advance and the table is searched.

The problem with this method is that if an abnormal pattern not prepared in advance occurs, it cannot be diagnosed, and if the scale of the target process becomes large, the consistency of the prepared pattern table will be maintained. It becomes difficult, for example, when the structure of the target is changed, it takes labor to correct the table (coordinating, etc.).

An attempt to solve such a problem is
The above-mentioned logical diagnosis method. As one method of this logical diagnosis method, the state in the target process (in the case of continuous process, for example, flow rate, tank level, valve opening, etc.) is used as several information items and between those information items. A model for diagnosis expressing a qualitative causal relationship of is prepared in advance, and a qualitative state of each information item determined based on information such as a measurement value obtained from the target process at the time of diagnosis, and the information described above. Collate the qualitative causal relationship between items and identify where the causal relationship is working (from an information item that can cause an anomaly to an information item that causes the anomaly as a result) There is a method of searching and paying attention to the cause → result and the cause → result chain obtained as a result of the search, and determining the most upstream information item of this route as a cause candidate of the abnormality.

[0006]

However, in the above-mentioned logical diagnosis method, the states of all the information items cannot be observed in the actual process, that is, even if the sensor cannot be attached for some reason. Because there is
For such information items, it is inevitable to make an assumption by assuming their qualitative state. However, when making such an assumption, there is a problem that the estimated cause of abnormality may change depending on the way of making the assumption.

As one method for solving this problem, the probability of abnormal occurrence at a plurality of locations at the same time is very small,
It is considered to assume a qualitative state of an unobservable information item on the assumption that the cause of the abnormality is unique. That is, by supposing the qualitative state of some unobservable information items to be plausible, the route formed by the chain of cause → result, cause → result can be observed starting from the only information item. Among them, all the information items that are the starting points that satisfy the condition that the qualitative state shows an abnormality but that reach all, that is, the condition that the occurring abnormality pattern can be explained are considered as candidates for the cause. .

At this time, if a non-policy assumption is made,
The number N of qualitative states that an information item can take is the number of unobservable information items. N trials (N ⁿ trials) are required. Therefore, in practical use, the number of cases should be minimized. The problem is figuring out a method.

[0009] Another problem is that if it is assumed that the cause of the abnormality is unique, it is impossible to diagnose an abnormality that happens to occur at a plurality of locations at the same time.

According to the present invention, when assuming the qualitative state of some unobservable information items and estimating the information item causing the abnormality, it is possible to omit meaningless assumptions to some extent, and it is impossible to observe. Even if the number of information items is large, it is possible to obtain the estimation result of the abnormality cause candidate in a short time, and it is also possible to reliably estimate the abnormality cause candidate even if multiple abnormal points occur at the same time. Has an aim.

[0011]

SUMMARY OF THE INVENTION The present invention expresses a state in a process such as a flow rate, a tank level, a valve opening degree, and the like as some information items, and expresses a qualitative causal relationship between these information items. Prepare a model for diagnosis in advance, qualitative state of each information item that is determined based on information such as measured values obtained from the process to be diagnosed when diagnosing, and qualitative between each information item To find the place where the causal relationship works,
It is applied to a process execution abnormality cause estimation method that estimates the information item existing in the most upstream of this route as a cause candidate of abnormality by paying attention to the chain of the resulting cause → result, cause → result.

In such an abnormality cause estimating method, according to the present invention, when the measured value corresponding to each information item becomes abnormal, the information item causing the abnormal situation is estimated. Do the following:

First, regarding each of the information items,
When the information item becomes abnormal, the abnormal state of the information item caused as a result is set in advance, and for each of the information items, the abnormality of the information item that may cause the abnormality of the information item. The state is set in advance, and further, for each of the information items, the condition for determining the measured value or the like corresponding to each information item as abnormal is set in advance.

Then, the state of each information item is checked at the time of estimating the cause of the abnormality, and for all information items that can be candidates for the cause of abnormality, the information item is sequentially used as a starting point (assuming that the information item is the cause of abnormality. The state of each information item is checked to determine whether or not a chain of paths from an assumed abnormal cause to a result based on the cause, that is, the above cause → result, cause → result chain is established. A condition that can be used to explain the abnormal pattern that occurs without conflict, that is, the cause → result, cause →
An information item that is assumed to be the cause that satisfies the condition that the entire path formed by the result chain reaches all the information items that indicate an abnormality is estimated as a candidate for the cause of abnormality.

For each information item for which information indicating its state cannot be obtained, the state of the information item is hypothetically set so that the chain of cause → result and cause → result is not interrupted. .

Furthermore, independent information items that do not affect each other are excluded from the range to be explained because each information item is determined to be the cause of the abnormality. The range to be explained means all the information items that can have a causal chain from the information item and all the information items that can have a causal chain to the information item.

[0017]

As described above, the present invention examines the state of each information item when estimating the cause of an abnormality, and for all information items that can be candidates for the cause of abnormality, it is assumed that the information item is the cause of the abnormality. Then, it is searched whether or not the chain of the cause → the result and the cause → the result is established while collating the states of the respective information items, and a condition capable of explaining the occurring abnormal pattern without contradiction, that is, , The above-mentioned cause → result, the route formed by the cause → result chain satisfies all the information items indicating an abnormality, and all the information items assumed to be the above-mentioned causes are assumed to be candidates for the abnormality cause. presume.

Here, in the present invention, of each information item,
If the information indicating the state cannot be obtained, the state of the information item is assumed so that the chain of the cause → the result and the cause → the result are not broken.

With this, it is possible to minimize meaningless assumptions, and it is possible to obtain the estimation result in a relatively short time even if the number of information items for which the information indicating the state cannot be obtained increases. it can.

Further, in the present invention, as described above, independent information items that do not affect each other are excluded from the range to be described because each information item is determined to be the cause of abnormality.

Therefore, even if a plurality of causes of abnormalities happen to occur at the same time, independent information items that do not affect each other should be excluded from the range to be explained because each information item is judged to be the cause of abnormality. By doing so, it is possible to enumerate a plurality of information items as candidate abnormal causes without being impossible to estimate.

[0022]

EXAMPLES Examples of the present invention will be described below. FIG. 1 is a diagram schematically showing an example of a tank system process. In the figure, F1 is the inflow amount of the fluid 2 flowing into the tank 1,
V1 is an inflow control valve for controlling the inflow amount, L is a tank 1
The level of the fluid 2 stored inside, LC is a level controller that detects the level of the fluid 2 stored inside the tank 1 and outputs a level adjustment output. The opening degree of the outflow control valve V2 is adjusted by this level controller LC, the outflow amount F2 from the tank 1 is controlled, and the level of the fluid 2 stored in the tank 1 is maintained at a constant value. V3 is a bypass valve, V4 is an outlet line valve, and V5 is a drain valve.

FIG. 2 shows the concept of a diagnostic model for the process shown in FIG. That is, V1, F1, and V1 are information items that represent the state in the target process.
L, V2, F2, V3, V4 and V5 are prepared, and a qualitative causal relationship between the respective information items is expressed by an arrow “→” from the cause side to the result side. Here, the qualitative states that each information item can have are three states: oversized, normal, and undersized. Express the oversized as "+", the normal as "0", and the undersized as "-". To Also ○
Information items enclosed in are to be able to observe the status.

On the other hand, it is assumed that there are two kinds of qualitative causal relations, and causal relations of promoting each (if the cause side information item increases, the result side information item also increases and the cause side information item decreases). For example, if the information items on the result side decrease, the causal relation of suppression (the information items on the cause side increases,
The relationship is such that the result side information items decrease and the cause side information items increase if the result side information items decrease). Here, the causal relationship of promotion is represented by adding the symbol "+" to the arrow, and the causal relationship of inhibition is represented by adding the symbol "-" to the arrow.

However, when the present invention is carried out, the cause of abnormality can be estimated using a computer, and when the computer is used in such a manner, the arrow “→” or the type of qualitative causal relationship is used. The symbols “+”, “−” and the symbols “+”, “0”, “−” indicating qualitative states are merely arithmetic factors, and can be replaced with arbitrary symbols as a matter of course.

Next, with reference to FIG. 3, a process of estimating the cause of the abnormality when a certain abnormality pattern appears using the model described in FIG. 2 will be described. First, in FIG. 2, F, which is an information item whose state can be observed
F1, V with reference to the measured values of 1, V2, F2 and the judgment conditions for each information item to attain a qualitative state
2, the qualitative state of F2 is determined, and any one of the corresponding symbols "+", "0", and "-" is assigned. For other information items V1 and L whose state cannot be observed, "?"
Will be attached.

As an example, FIG. 3A shows a case where the qualitative states of F1, V2 and F2 show an abnormal pattern in which F1 is normal, V2 is excessive and F2 is excessively small. Therefore, in FIG. 3A, “0” is assigned to F1, “+” is assigned to V2, and “−” is assigned to F2.
"?" Is attached to 1, L.

Next, for all information items that can be candidates for anomaly cause, it is assumed that they are anomalous causes in sequence (assuming that they are the origin of the path formed by the chain of cause → result, cause → result) The causal relationship is searched, and it is compared with the abnormal pattern that has occurred to determine whether or not the assumption is valid.

By the way, as described above, when the qualitative state that information items can take and the kind of qualitative causal relationship are arranged, the fact that the causal relationship is working means that the arrow indicating the causal relationship is present. The product of the sign of "+" or "-" attached to "→" and the sign of "+", "-", or "0" that represents the qualitative state of the information items at both ends of the sign Will be positive.

In the example of FIG. 1, it is assumed that the information items that can be abnormality cause candidates are only V1, V2, V3, V4, and V5, and each information item that can actually be an abnormality cause candidate is started from that point. Let's follow up assuming that.

First, V1 is an information item whose state cannot be observed, so it is assumed that V1 is "+" or "-", but in any case, an arrow comes out from V1. The state of F1 ahead is "0"
Since it is (normal), there is no anomalous propagation of V1 → F1, and at this point the chain is broken and V2 is “+”, and F2
Since it cannot explain the situation where "-" is "-", it is determined that V1 is not the cause of the abnormality.

Next, regarding V2, since the qualitative state is known, it is assumed that the cause of the abnormality is that V2 is "+" or (excessive). But V2
Although there is a causal relationship of promoting from F2 to F2, V2 is too large but F2 shows a situation of being too small, which does not make sense, and the causal relationship of V2 → F2 is not working, At this point, the chain is broken and the abnormal pattern that is occurring cannot be explained, so it is determined that V2 is not the cause of the abnormality.

Like V1, V3 is an information item whose state cannot be observed, so a qualitative abnormal state is assumed. However, as described in the explanation of FIG. 1, since this is a bypass valve, it is normally closed completely. "0" or "+" which is accidentally opened 2
I know that it can only be in the street. Therefore, in this case, it is assumed that V3 is "+", that is, it is caused by accidental opening. However, this is the same as in the case of V2 → F2, which means that the causal relationship of V3 → F2 does not work, and at this point the chain is broken, and the abnormal pattern that is occurring cannot be explained, so V3 is also the cause of the abnormality. Judge that it is not.

Since V4 is also an information item whose state cannot be observed, a qualitative abnormal state is assumed.
As described above, it is known that this is an outlet line valve and can normally take only two states of full open "0" or closed "-" for some reason. Therefore, in this case, it is assumed that V4 is "-", that is, the fact that it is blocked for some reason is an abnormal cause. In this case, V4 "-"-> F2 "-"-> L "+"-> V2 "+"cause-> result, cause-> a chain of results is formed, and the abnormal propagation path it represents is one of the observable information items. Since all of the qualitative states show abnormalities (in this case, V2 and F2) are spread (that is, the occurring abnormal pattern can be explained), the blockage of V4 is determined as one of the abnormal cause candidates. At this time, the state item L whose state cannot be observed
The qualitative state of is assumed to be "+" so that the chain of anomalous propagation is not interrupted. Figure 3 shows this situation.
It becomes like (b). As is apparent from FIG. 3 (b), the product of the sign of the arrow and the sign of the information items at both ends thereof is all positive, and a causal relationship of V4 → F2 → L → V2 is working, and F2 is " It is presumed that V4 is the cause of "+" for "-" and V2.

Finally, V5 is an information item whose state is not observable, so a qualitative abnormal state is assumed. However, as described in the explanation of FIG. 1, this is a drain valve. It is known that normally, there are only two possible states: fully closed “0” or accidentally opened “+”. Therefore, in this case, it is assumed that V5 is "+", that is, the fact that it is accidentally opened is the cause of the abnormality.
In this case, V5 "+" → L "-" is reached,
In this case, L is assumed to be "-" so that the chain of anomalous propagation is not interrupted, so there is no anomaly propagation of L → V2, and at this point the chain is broken and the anomaly pattern that is occurring Therefore, it is determined that V5 does not cause the abnormality.

As a result of performing the verification based on various assumptions as described above, when the qualitative states of F1, V2, and F2 show abnormal patterns of normal, excessive, and excessive, respectively, the abnormal cause is V4 for some reason. It is supposed that it was blocked by.

In the above example, the case where the abnormality occurs at a single location has been described. Next, the case where the abnormality occurs at a plurality of locations at the same time will be described as another embodiment using another process example. .

Similar to the case of FIG. 1, FIG. 4 schematically shows an example of the process of the tank system. In the figure,
11, 12, 13 are first, second, and third tanks, F1 is an inflow amount of the fluid 2 flowing into the first tank 11, F2 is an outflow amount from the first tank 11, and F3 is a second tank. Tank 12
Is an inflow amount of the fluid 2 that flows in, F4 is an outflow amount from the second tank 12, and F0 is an outflow amount from the third tank 13. L1 is the level of the fluid stored in the first tank 11, L2 is the level of the fluid stored in the second tank 12, and L0 is the fluid stored in the third tank 13. Shows the level of.

FIG. 5 shows the concept of a diagnostic model for the process shown in FIG. 4, and in the figure, the meanings of symbols such as arrows ".fwdarw.", "+", "-" Are as described above. Is. Here, the information items F0, F1, F2, F
3, F4, L0, L1, and L2 will be described as items for which the state can be observed.

Let us consider a case where an abnormality occurs in which the qualitative state of each information item has a pattern as shown in FIG. 6A in such a process. First, only valid causal relationships are left based on the qualitative state of each information item and the qualitative causal relationship between the information items, as shown in FIG. 6B. However, since it is premised that there is only one cause of abnormality, the path formed by the chain of cause → result, cause → result has an abnormal qualitative state among the observable information items, starting from the only information item. It is presumed that the starting point that satisfies the condition that all of the information items that indicate the above are satisfied is the anomaly cause candidate.However, as in this example, some information items that indicate anomalies cannot be traced by arrows. Then, there is no starting point that satisfies the above conditions. Therefore, as one method of avoiding this problem, information items whose causal chains do not reach each other (independent) are excluded from the range to be explained because each information item is determined to be the cause of abnormality. . The range to be described here means all the information items that can have a chain of causal relationships from the information item, and all the information items that can have a chain of causal relationships with the information item. In this example, this is shown in FIG.
This is equivalent to the independent consideration of the two systems A1 and A2 indicated by the two-dot chain line. The range to be explained is specifically listed in this example as shown in Table 1.

[0041]

[Table 1]

Considering this range, from FIG.
An underestimation of 2 and an underestimation of F4 are presumed to be abnormal cause candidates. In other words, it can be estimated that the pipes in the F2 and F4 portions may be accidentally blocked at the same time.

FIG. 7 shows a structure of means for automatically generating a program for executing the above-described abnormality cause estimation method on a computer, estimating it using the program, and displaying the result including the abnormality propagation path. Is shown.

In the figure, 21 is a man-machine interface, and this man-machine interface 21 is
The definition of each information item and the causal relationship between each information item (these are referred to as data definition 23) can be entered in the form of an adjacency matrix image via the interactive display device 22, and the state of each information item is shown. A condition for determining a value (measured value, etc.) as abnormal can be input. Further, the man-machine interface 21 has a function of displaying the result estimated as the cause of abnormality (abnormality cause estimation result 24) on the interactive display device 22.

The data definition 23 input by the man-machine interface 21 is used as model data 25, and the model data 25 is generated by the estimation program generation mechanism 26.
From this, a program incorporating the above-mentioned abnormality cause estimation method is automatically generated.

On the other hand, various data from the process 27 to be diagnosed is stored in the process data access interface 2
8 to be stored in the memory 29. The program generated by the estimation program generation mechanism 26 based on this process data is executed for each execution module 30, and the execution result 31 is obtained as the man-machine interface 21.
Is displayed as the cause of abnormality estimation result 24 on the interactive display device 22. These series of processes are performed by the CPU 32.

In FIG. 7, when the data definition 23 input from the interactive display device 22 includes, for example, information items A, information items B, information items C, information items D, and information items E, Each of these information items A, B,
When the information item becomes abnormal for each of C, D, and E, the abnormal state of the information item caused as a result is preset, and the abnormality of the information item is caused for each of the information item. An abnormal state of an information item that can be a cause is preset, and for each of the information items, a condition for determining a measured value or the like corresponding to each information item as an abnormality is preset. For example, when the information item A is excessively "+", the information item B is set to "+", and when the information item B is excessively "+", the information item C is set to "-".

Then, based on such data, a program in which an abnormal state estimating method is incorporated is generated, and the state of the process data (state item) from the process 27 to be diagnosed is checked by this program and all Regarding the information items that can be candidates for the abnormal cause of, starting from the information item in sequence (assuming that the information item is the cause of the abnormality), the chain of the above cause → result, cause → result is established. Whether or not the condition of each information item is searched while checking the condition, and the abnormal pattern that is occurring can be explained consistently, that is, the path formed by the chain of cause → result, cause → result indicates the abnormality. The information item that is assumed to be the cause and satisfies the condition that all the information items shown are distributed is estimated to be a candidate for an abnormal cause.

The abnormality cause estimation result 24 of FIG. 7 shows an example of the result assuming that the information item A is the cause of abnormality. Whether or not the path formed by the chain of cause → result, cause → result from the information item A as a starting point reaches all the information items indicating abnormality, that is, As described above, the sign "+" or "-" attached to the arrow "→" indicating the causal relationship and "+", "-" indicating the state of the information items at both ends thereof,
Whether or not a causal relationship is working is checked depending on whether or not the product of three signs of any of "0" is positive, and it is estimated whether or not the information item as the starting point is the cause of abnormality. . For example, when attention is paid to the information item A and the state item B, the symbol attached to “→” representing the causal relationship between them is “+”.
, The code indicating the state of the information item A is “+”, and the information item B is
If the sign indicating the state of “+” is “+”, it is possible to estimate an abnormal cause that the reason why the information item B becomes “+” (excessive) is because the information item A becomes excessive “+”. .

[0050]

According to the present invention, the state in the process to be diagnosed is represented by some information items, and some of measured values obtained from the process corresponding to each information item are abnormal values. In the method of estimating the information item that is the cause of the abnormal situation, when assuming the qualitative state of the unobservable information item, it is assumed that the chain of cause → result, cause → result is not interrupted. By determining the qualitative state from the upstream side of the chain by some method, meaningless assumptions can be omitted to some extent, and even if the number of unobservable information items increases, an estimation result can be obtained in a practical time. You can In addition, even if two or more abnormal causes happen to occur at the same time, each information item is judged to be the cause of the abnormality for the information items for which the predefined causal relationships do not reach each other (does not affect each other). By excluding it from the scope to be explained,
It is possible to enumerate a plurality of information items as candidate abnormal causes without falling into the unpredictable state. Furthermore, it is possible to generate a program for executing the above-mentioned abnormality cause estimation method on a computer and displaying the result including the propagation path of the abnormality by simply inputting in a tabular format. The effect of is obtained.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of a tank system process for explaining an embodiment of the present invention.

FIG. 2 is a diagram showing the concept of a diagnostic model for the tank system process shown in FIG.

FIG. 3 is a diagram illustrating an abnormality cause estimation process in the embodiment.

FIG. 4 is a schematic configuration diagram of a tank system process for explaining an embodiment in the case where abnormalities occur at a plurality of places at the same time.

5 is a diagram showing the concept of a diagnostic model for the tank system process shown in FIG.

FIG. 6 is a diagram illustrating an assumption of anomaly cause estimation in the embodiment when anomalies occur simultaneously at a plurality of locations.

FIG. 7 is a block diagram of means for carrying out the present invention.

[Explanation of symbols]

 V1 to V5, F0 to F4, L1 to L2 ... Information items

Claims (4)

[Claims] 1. A cause in which an abnormal state is caused when a state in a process to be diagnosed is represented by some information items and some of data obtained from a process corresponding to each information item has an abnormal value. In the process abnormality cause estimation method for estimating the information item that has become, for each of the information items, when the information item becomes abnormal, the abnormal state of the information item caused as a result is set in advance. In addition, for each of the information items, the abnormal state of the information item that may cause the abnormality of the information item is set in advance, and further, for each of the information items, the measurement value corresponding to each information item, etc. The conditions for judging that a failure is set in advance, the state of each information item is checked when estimating the cause of the failure, and all For information items that can be a regular cause candidate, it is assumed that the information items are the causes of abnormalities, and whether or not a chain of several routes from the assumed abnormal cause to the result is established. , While searching while checking the status of each information item, the entire route formed by the chain of several routes from the assumed cause of abnormality to the result is all of the information items indicating abnormality. A method for estimating a cause of a process abnormality, wherein the information item that is assumed to be the cause of the abnormality satisfying the condition of being distributed to the process is estimated as a candidate for the cause of the abnormality. 2. Among the information items, information items for which information indicating the state thereof is not obtained are those information items so that the chain of some paths from the assumed abnormal cause to the result is not interrupted. 2. The process abnormality cause estimation method according to claim 1, wherein the state is assumed to be a certain state. 3. Among the information items, independent information items that do not affect each other are excluded from the range that should be explained because each information item is determined to be the cause of abnormality. The process abnormality cause estimation method according to item 1. 4. As a means for estimating the cause of the process abnormality, when the certain information item becomes abnormal, data defining an abnormal state of the information item caused as a result, and causing an abnormality of the certain information item. A man-machine capable of inputting data defining abnormal information of information items which may be a cause and data indicating conditions for judging a measured value corresponding to each information item as an abnormality via an interactive display device (22). An interface (21) and an abnormality cause estimation program generation unit (26) for automatically generating a program for estimating an abnormality cause from the data input by the man-machine interface (21).
And, based on the process data from the process (27) to be diagnosed, the result estimated by the abnormality cause estimation program is displayed on the interactive display device (22) by the man-machine interface (21). The process abnormality cause estimating method according to claim 1, wherein