JPS6091412A - Abnormality diagnosing method of system - Google Patents

Abstract

PURPOSE:To increase the range of diagnosis and to improve the accuracy by estimating the product characteristics by means of a partial numerical formula model and then estimating the final fault factor by means of the knowledge related to the fault propagation. CONSTITUTION:This diagnosis device consists of an electronic computer 2, a DISK2 which stores the knowledge related to the fault propagation, the knowledge of expression for a numerical formula and the product check data, and a CRT terminal 1. A plant fault diagnosis system is attained with execution functions of two steps. The step 1 is equal to an estimating function for a parameter which is hard to observe and is divided into the expression method for quantitative related knowledge and the application method for said knowledge. The quantitative knowledge is obtained by simplifying a logical formula, etc. For application of said knowledge, the value of another variable is estimated from the product check data and the quantitative related knowledge to decide the presence or absence of a fault. In the step 2, a production system is applied to study a factor having the possibility of a fault.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention is designed to quickly and efficiently identify the cause of an abnormality when an abnormality occurs in a system in which there are many parameters to be controlled and these parameters are in a complex relationship with each other. The present invention relates to a system abnormality diagnosis method suitable for accurate estimation. The present invention
Although it is applicable to systems in general, such as system equipment, information processing systems, and plants, the following explanation will focus on plants.

[Background of the invention]

Diagnosis of a manufacturing plant is to investigate the cause of an abnormality from observed abnormal phenomena. Typically, this diagnosis involves ■ developing a mathematical model between the abnormal phenomenon and its causes, ■
The procedure is to quantitatively estimate the cause of the abnormality using these mathematical models. However, recently, manufacturing plants have become significantly larger and more complex, and as a result, it has become increasingly difficult to develop mathematical models. Therefore, when diagnosing an abnormality, it is no longer possible to rely solely on theoretical mathematical models, and it has become important to utilize manufacturing technology or empirical knowledge regarding abnormality diagnosis.

Conventionally, attempts have been made to apply the results of artificial intelligence research to this type of problem. As a result, a method called ``Production System'' was devised and published.The idea here is ``■The abnormal phenomenon propagation relationship (the causal relationship between the phenomenon and the cause) obtained empirically by specialized engineers. )
``The rules are created into rules and stored in a file, and when an abnormal phenomenon is observed, the rules are combined in a syllogistic manner to infer the root cause.''

In such conventional methods, all rules are expressed in one-dimensional format (for example, "if A then B" and the rule indicate a causal relationship such as "If A is true, then B is also true"). Ta. However, abnormal propagation relationships such as simple cause-and-effect relationships can only express a portion of the knowledge possessed by specialized engineers. Therefore, there is a drawback that the range in which abnormalities can be diagnosed is narrow.

[Purpose of the invention]

The present invention facilitates the use of not only qualitative relational knowledge such as abnormal phenomenon propagation relations, but also quantitative relational knowledge such as experimental formulas/simplified theoretical formulas, and estimates types of abnormality causes. The purpose of the present invention is to provide a highly accurate system abnormality diagnosis method.

[Summary of the Invention] Diagnosis of manufacturing plants is usually carried out based on the following concept. In other words, when an abnormality occurs, the product characteristic values are selected so that it is easy to detect the abnormality, and it is easy to estimate which process in the plant the abnormality was caused. Diagnosis is made by examining the presence or absence of abnormalities. However, for technical or economic reasons, it is often difficult to directly measure such product characteristic values. In that case, it is necessary to estimate the value of the product characteristic of interest using the value of another measurable product characteristic and a mathematical model.

When the target plant is large-scale, it is difficult to develop a single global mathematical model as described above, and it may be difficult to estimate the product characteristic value of interest. However, even in large-scale plants, the number of variables that are actually controlled is relatively small, and the range of their fluctuations is usually not that wide. Therefore, even a partial model is considered to be useful as a numerical model necessary for estimating product characteristic values. Moreover, these mathematical models in a narrow partial region have the advantage of simplifying theoretical formulas or being easy to obtain as experimental formulas.

The present invention provides a hierarchical diagnosis method that 1) estimates the product characteristic values using the partial mathematical model as explained above, and 2) estimates the final cause of the abnormality using knowledge related to anomaly propagation. It provides a method.

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail according to examples.

FIG. 1 shows an example of a hardware configuration for realizing the present invention, which is the same as the hardware configuration for realizing a conventional production system method. Device 1
is an electronic computer, device 2 is a DI8, and device 3 is a CRT terminal. Device 2 has anomalous propagation related knowledge and mathematical expression knowledge,
and product inspection data. When the user requests analysis via the device 3, the device 1 starts abnormality diagnosis. In this diagnosis, if there is insufficient information, it is sent to the user via the CFLT terminal 3.

FIG. 2 is an example of the functional configuration of the plant abnormality diagnosis system according to the present invention. The diagnostic performance function can be implemented in two steps. The functions of each step and how to implement the functions will be described below.

Step 1 is a function of estimating parameters that are difficult to observe, and this embodiment will be divided into (1) a method of expressing quantitative relational knowledge, and (2) a method of utilizing that knowledge. First, I will describe the knowledge representation method. As mentioned above, quantitative knowledge can be obtained as a simplification of a theoretical formula or an experimental formula. This is a formula that expresses a quantitative relationship focusing on a small number of variables. Therefore, the following preconditions must be made in these formulas.

Condition 1: The variables appearing in the expression take values within a limited narrow range.

Condition 2: Even among variables that do not appear in the expression, there are variables that must take values within a narrow, limited range.

That is, quantitative relational knowledge can be expressed by a mathematical formula, and the above condition 1.2 may be added as the valid range of the mathematical formula. Specifically, for example, it can be expressed as follows.

Preconditions (1, 8) tr,≦tj≦t,, t, JεJ,, J, cJx,,
t≦X,≦xIaz−εl,, l, c l (1, b
) Relational Expression % Expression %) (2) Here, t is a measurement variable, and J represents the set thereof. XI
is an estimated variable, and its set is represented by 1. Therefore, equation (1,
8), (1, b) corresponds to the above condition IK. Further, Ci is a variable corresponding to the above condition 2, etc. are constants for indicating the range of each variable.

L represents a set of subscripts that identify equations for estimating the variable X.

Next, I will describe how to utilize the knowledge. The purpose of knowledge utilization is to estimate the values of other variables from product inspection data and quantitative related knowledge, and to determine the presence or absence of abnormalities. An example of a method for determining the presence or absence of an abnormality is shown in flowchart form in FIG. As shown in the figure, the determination of the presence or absence of an abnormality can be divided into the following two cases and carried out in the following manner. In FIG. 3, it is determined in block 31 whether the value of the estimated variable of interest can be determined, and if it is determined that the value cannot be determined, the entire process of block 32 is performed, and it is determined that the value can be determined. If so, the process of block 33 is performed.

:) When the value of the variable of interest, X, cannot be determined, there is generally a plurality of knowledge (formulas) for estimating the value of the variable of interest. In a simple case, if there is only one such knowledge (the number of elements of , as described above, is 1), if some of the preconditions of that knowledge do not hold, the xaO value I can't put it down. In this case, it can be determined that there is an abnormality in the variable that is not satisfied among the preconditions. That is, if a variable is found that does not satisfy the conditions expressed by equations (1, a), ke, and b), it can be determined that the value of that variable is abnormal.

In addition, when there are multiple knowledge R (formulas), if the ranking of the multiple values obtained using each knowledge is greater than a certain standard value, it is determined that the xaO value cannot be calculated. do.

In this case, it can be determined that there is an abnormality in the unconfirmed variable among the preconditions of each knowledge, that is, the variable in equation (1, c). For example, consider a case where there are two knowledge (formulas) for estimating the value of the variable of interest, X (the number of elements in
, , xa2. Let a certain infinitesimal value be 2, lXa, +
-X, +>a, the two estimation equations are expressed as (1,
It can be estimated that one of the conditions shown in c) is not satisfied (that is, there is an abnormality).

If) If you can find the value of the variable of interest In the above example, if I Compare the estimated value that can be claimed with the pre-registered reference range. As a result, if it is outside the standard range, it is determined that there is an abnormality. Here, (9) Even if a result is found to be valid when compared with the standard range, whether or not it is truly valid considering other measured values and estimated values requires qualitative knowledge. Verify when searching for the cause of the abnormality used. Rather than simply obtaining an estimated value and proceeding with diagnosis based on that value as an absolute value, the estimation calculation and production system are organically linked as described above.

The function of step 2 can basically be realized by applying a production system method researched in the field of knowledge engineering.

For example, assume that in step 1, it is found that the variables x,, x,' have an abnormality exceeding the upper limit of the standard. In step 2,
Apply production rules to investigate the possible causes of the above abnormality. Here, the production rule is the causal relationship f between the abnormal phenomenon and its cause.
It is expressed in the if then format, for example,
“Abnormal phenomenon Al.

The rule "When A2, As is observed, there is a high possibility that the cause is X" can be expressed as follows.

(10) if (abnormal phenomenon AI is observed) and (abnormal phenomenon A2 is observed) and (abnormal phenomenon A3 is observed) then (there is a high possibility that X is the cause), if this t
If there are rules for events i on the hen side and matters on the if side, those rules are applied to further investigate the fundamental cause.

Note that whether or not the estimation result of step 1 is truly valid can be checked by checking whether there is a contradiction in the stored production rules. Sunawachi, if then
If a contradiction occurs in the causal relationships between events expressed by the rules, it can be determined that the estimation result in step 1 is incorrect. However, this contradiction may also occur if the rule itself is unjust.

〔Effect of the invention〕

As explained above, according to the present invention, the following effects can be obtained.

(11) (1) The present invention provides a knowledge expression method and an inference method based on the knowledge that utilize abnormal phenomenon propagation related knowledge and mathematical expression knowledge such as experimental formulas as knowledge used for abnormality diagnosis. This expands the range of abnormalities that can be diagnosed and improves accuracy.

(2) Since knowledge can be easily changed, it is expected that the accuracy of plant abnormality diagnosis will improve.

[Brief explanation of the drawing]

Figure 1 is a block diagram for explaining the present invention and the conventional method, Figure 2 is a functional configuration diagram for explaining the functional configuration of the present invention, and Figure 3 is the main part of diagnosis (12).

Claims (1)

[Claims] The first step is to store knowledge related to abnormality diagnosis, the second step is to store product inspection data measured at the time of abnormality occurrence, and the data obtained in the first and second steps above are used to determine when an abnormality has occurred. A system abnormality diagnosis method consisting of a third step of estimating the cause of the problem.