JPH07262014A - Diagnostic type expert system - Google Patents

Abstract

PURPOSE:To efficiently reduce the amount of input data to an inferring mecha nism by providing an adjusting mechanism for adjusting the confidence degree of diagnostic rules to be used for the inferring mechanism while using the in ferred result stored in a storage means. CONSTITUTION:An inferring mechanism 43 decides diagnostic contents by executing inference while referring to the analyzed result stored in a pattern analyzed data storage part 42 and using the diagnostic rules stored in a diagnostic rule storage device 3. The decided diagnostic contents are stored in a diagnosed result storage device 5. At such a time, an inference record corresponding to an inference record storage device 6 is stored as well. Concerning the real fault cause of the newly stored inference record, an input is received from a terminal equipment 8 and while considering the item of the inference record completed as a result, a diagnostic rule adjusting device 7 adjusts the value of the confidence degree of diagnostic rules stored in the diagnostic rule storage device 3. Thus, processing time required for inference can be shortened.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a diagnostic type expert system for inputting history data concerning the state of a mechanism and an apparatus to be diagnosed and diagnosing the state of the object to be diagnosed by using an inference mechanism. The present invention relates to a diagnostic expert system that handles history data that has a meaning as a series of data strings or patterns rather than the history data itself.

[0002]

2. Description of the Related Art Conventionally, a system for diagnosing whether or not there is an abnormality in a target and how to handle the target based on various numerical values and indexes indicating the state of the target is typified by a device failure diagnosis system or a corporate management diagnosis system. , Built in a wide variety of fields.

This type of system is intended to realize the same diagnosis as that carried out by a specialist by embedding the criteria of judgment at the time of diagnosis made by a specialist in the target field into the program of the diagnostic system. , Collects data such as numerical values and indexes indicating the condition of the diagnosis target from the response to the system user or the sensor attached to the device, and based on the collected data, the failure diagnosis of the device to be diagnosed and the management of a certain company It diagnoses the condition.

In such a system, in order to realize a judgment similar to that of an expert, it is general to construct a rule-based system having an inference mechanism. Also,
Rule-based reasoning has the property that the processing time required for reasoning increases significantly with the increase in the amount of data given as input. Therefore, for a diagnosis target having a large amount of data indicating the state, only the data necessary for the inference processing is extracted and used as the input to the inference mechanism. As a result, the amount of input data to the inference mechanism can be reduced and the inference processing time can be shortened.

As a publicly known example related to this, for example, Japanese Patent Laid-Open No. 2-213943 describes a method of making a diagnosis by using a mechanism for extracting the characteristics of the measured quantity related to the diagnosis target.

On the other hand, the concept of certainty factor is used to flexibly perform rule-based inference. This is a value that indicates the certainty given to the facts described in the conditional part and the execution part of the rule. By using this, it is possible to take inferences by incorporating the relation between facts that is not absolute. . However, it is difficult to set this properly when constructing the system. Therefore, the method of adjusting the certainty factor may be used by utilizing the result of the inference.

As a known example related to this, for example, in Japanese Patent Laid-Open No. 2-284538, a history of failure analysis processing is stored for the purpose of network failure diagnosis, and the confidence factor calculation method is changed from the correct answer rate. The method is described.

[0008]

Conventionally, one of the following processing methods has been adopted as a method for extracting the characteristics of data relating to a diagnosis target.

(1) Extract necessary items as they are from the data.

(2) Some items are picked up from the data, and the characteristic amount is obtained by numerical calculation based on them.

The above-mentioned processing method is also adopted in the method described in the above-mentioned Japanese Patent Laid-Open No. 2-213943. The method using the neuro computer used here can be considered as a kind of the processing method of (2).

However, this method is not effective in the case of diagnosing the operating state from the history data recording the operation of the device operating in a certain procedure. This is because the individual data in the historical data often has more meaning as a series or sequence of data than as each.

In the above method using a neuro computer, it is possible to obtain a feature quantity by using a data string as an input, but it is difficult to improve the accuracy of the diagnosis result, and learning is required to improve the accuracy. The cost is also high.

Furthermore, in the case of making a diagnosis using history data, if an attempt is made to adjust the certainty factor by using the method described in Japanese Patent Laid-Open No. 2-284538, a rule which is a premise for deriving a result (that is, certainty factor) It is difficult to narrow down the rules that should be adjusted, and it is difficult to apply.

An object of the present invention is to efficiently reduce the amount of input data to the inference mechanism even when data such as history data whose meaning as a string of data is important is used, and further improve the accuracy of diagnosis. It is to provide a diagnostic expert system capable of performing.

[0016]

In order to achieve the above object, the present invention analyzes whether or not a history data includes a data sequence including a part of a plurality of predetermined data patterns. , If present, data pattern analysis means for inputting the analysis result consisting of the last history data in the data string and the number of the corresponding data pattern to the inference mechanism, and storage means for storing the diagnostic result by the inference mechanism, An adjusting mechanism for adjusting the certainty factor of the diagnostic rule used by the inference mechanism by using the inference result stored in the storage means.

[0017]

According to the above means, a plurality of data patterns which can be generated as history data are prepared in advance, and the data patterns are collated with the history data actually generated, whereby the history data is predetermined. Analyzes whether or not there is a data string that contains a part of multiple data patterns, and if there is, a reasoning mechanism based on the analysis result consisting of the last history data in the data string and the number of the corresponding data pattern. To enter.

Then, the inference result obtained by the inference mechanism is accumulated, and the accumulated content is applied to a predetermined calculation formula to adjust the certainty factor of the diagnostic rule used in the inference mechanism.

By replacing the history data with the data pattern number and supplying it to the inference mechanism in this manner, the amount of input data to the inference mechanism can be reduced and the processing time required for the inference can be shortened. Further, the inference record can be stored compactly, and the inference accuracy can be improved by adjusting the confidence factor at a higher speed.

[0020]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the diagnostic expert system of the present invention. The system of this embodiment includes a history data storage device 1, a history data pattern storage device 2, a diagnostic rule storage device 3, a diagnostic device 4, a diagnostic result storage device 5, an inference record storage device 6, a diagnostic rule adjustment device 7, and a terminal device. It is composed of 8.

Of these, the diagnostic device 4 includes a pattern analysis unit 4
1. Pattern analysis data storage unit 42 and inference mechanism 43
Made of.

As shown in FIG. 2, the history data storage device 1 stores the history codes 20 output by the device to be diagnosed in the order of occurrence time. For example, occurrence time 18:03:15, history code A2, or occurrence time 18:03
Information such as minute 17 seconds and history code C1 is stored in the order of occurrence time.

FIG. 3 shows the history data pattern storage device 2.
As described above, a plurality of correctly generated history code generation patterns 30 are divided by pattern numbers 31 and stored. For example, as “pattern number 1”, the pattern “A1 → C1orC2 → B1orB2 → C4 → D1or
The pattern 32 "D2" is stored. This pattern 32 means that “C1 or C2 occurs next to A1, then B1 or B2, then C4, and finally D1 or D2”.

On the other hand, in the diagnostic rule storage device 3, as shown in FIG. 4, a certainty factor 42 is added to a condition part 40 which is a combination of a pattern number 401 and an end code 402, and a conclusion part 41 which indicates the diagnostic content. Rule number 43
It is divided and stored in.

For example, diagnostic rule 4 of "rule number 1"
4, "pattern number 1", "end code C1",
"Diagnosis content" Unit A failure "", "Confidence level 0.7"
Information such as is stored.

Here, the diagnostic rule 4 of "rule number 1"
No. 4 indicates that the history data actually generated includes a data string including a part of the data string of "pattern number 1", and the end code of the data string is "C1", the diagnostic result is " It means that the unit "A unit is defective" and that the certainty factor of the diagnosis is "0.7".

The outline of the operation according to the above configuration is as follows.

First, the pattern analysis unit 41 of the diagnostic device 4
Reads the history data stored in the history data storage device 1, and does the history data sequence include a data sequence including a part of the history data pattern previously stored in the history data pattern storage device 2? It is checked whether or not the result is stored in the pattern analysis data storage unit 42.

For example, in the example of the history data shown in FIG.
Since the data strings of 2, C1, B1 and C3 include a part of the history data pattern of “pattern number 2”, the pattern analysis data storage unit 42 stores the corresponding history data as shown in FIG. “2” is stored as the pattern number 51 of the pattern, and “C3” is stored as the end code 52 at the end of the history data sequence including the pattern in the column of the pattern analysis data number 1.

Further, since the data sequence of A1, C1, B1, C4, D2 includes all the history data patterns of "Pattern No. 1", the pattern analysis data storage unit 42 of FIG. As described above, “1” is stored as the pattern number 51 of the corresponding history data pattern, and “D2” is stored as the end code 52 at the end of the history data sequence including the pattern in the column of the pattern analysis data number 2.

If the end code 52 does not match the end code of the predetermined history data pattern, it is considered that the device to be diagnosed has abnormally ended during the output of the history data. The code 52 can be a great clue for determining the diagnostic content. The diagnostic rule of FIG. 4 is described by utilizing this fact.

Next, the inference mechanism 43 makes an inference using the diagnostic rule stored in the diagnostic rule storage device 3 while referring to the analysis result stored in the pattern analysis data storage unit 42, and determines the diagnostic content. . The determined diagnosis content is stored in the diagnosis result storage device 5.

The diagnostic result storage device 5 stores diagnostic results as shown in FIG. For example, a diagnosis result “defective unit A” is stored as the diagnosis content 60.

At this time, the inference record is also stored in the inference record storage device 6. The inference record storage device 6 stores inference records as shown in FIG.

For example, in the column 71 of "Record No. 1", the rule number 71 is "3", the pattern number 72 is "2" and "1", the end code 73 is "C3" and "D2", and the diagnostic content. An inference record of “defective unit A” is stored as 74, and an inferential record of “defective unit A” is stored as the cause of actual failure 75. However, at the stage of storing the inference record at the time of executing the inference, the actual failure cause 75 is not yet stored.

The diagnostic rule adjusting device 7 receives an input from the terminal device 8 regarding the cause of the actual failure of the newly stored inference record, and stores it in the diagnostic rule storage device 3 in consideration of the items of the inference record completed as a result. The value of the certainty factor 42 of the present diagnostic rule is adjusted.

FIG. 8 shows the pattern analysis unit 41 of the diagnostic device 4.
5 is a flowchart showing a procedure of a pattern analysis process of history data in FIG.

The pattern analysis unit 41 initializes the data string stored in the buffer register (not shown) for collation to an empty string and the end code to empty, and sets all the history data patterns shown in FIG. Analysis is started as a candidate pattern of actual history data (step 801).

If there is no more history data (step 802), the combination of the candidate pattern and the end code at that time is stored in the pattern analysis data storage unit 42 (step 803), and the analysis process is ended.

If history data still exists (step 8)
(NO in 02), one history code is taken out from the beginning of the history data, the new data string is added to the end of the data string, and the candidate pattern is collated (step 804). here,
If there is no pattern that matches the new data string (the data string is included in the data string indicated by the pattern) in the candidate pattern set of FIG. 3 (NO in step 805),
The combination of the candidate pattern and the end code at that time is stored in the pattern analysis data storage unit 42 (step 80).
6) and returns to the initialization process (step 801).

If there is a pattern that matches the new data string in the candidate patterns (YES in step 805).
S), only the matched patterns are used as elements of the new candidate pattern set, and the history code at the end of the new data string is used as the end code (step 807).

At this time, if the end code is the end code of the pattern (YES in step 808), the combination of the candidate pattern and the end code at that time is stored in the pattern analysis data storage unit 42 (step 806). The process returns to the initialization process (step 801).

On the other hand, if the end code is not the end code of the pattern (NO in step 808), the process returns to the process of extracting the history code and collating the pattern (step 804).

FIG. 9 is a flow chart showing the procedure of inference processing in the inference mechanism 43 of the diagnostic device 4.

The inference mechanism 43 reads the analysis result by the pattern analysis unit 41 from the pattern analysis data storage unit 42 (step 901) and determines whether or not there is a diagnostic rule such that these pattern analysis data and the condition unit 40 match. Collation is performed (step 902).

If there is a matching diagnostic rule (step 9)
(YES in 03), the diagnosis content of the conclusion part 41 of the corresponding rule is stored as a diagnosis result (step 904), the set of the rule, the pattern analysis data and the diagnosis result is stored as an inference record (step 905), and the inference process is executed. finish.

If there is no matching diagnostic rule (NO in step 903), the inference process is ended as it is.

In this embodiment, there may be a plurality of combinations of diagnostic rules and pattern analysis data that satisfy the conditions. In that case, a plurality of diagnostic contents are output as a result, or conditions other than those shown here are output. By the rules regarding
Among them, the diagnostic content that makes the most probable diagnosis will be selected.

In such a case, not only the result of the pattern analysis but also a method of using a part of the original history data or the data extracted by a method other than the pattern analysis for inference can be considered.

FIG. 10 is a flow chart showing the procedure for adjusting the certainty factor of a rule in the diagnostic rule adjusting device 7.

The diagnostic rule adjusting device 7 receives the input of the cause of the actual failure from the terminal device 8 for the inference record in which the cause of the actual failure is not stored (step 1001), and diagnoses in consideration of the items of the inference record completed as a result. Rule storage device 3
The value of the certainty factor of the diagnostic rule stored in is added or subtracted (step 1002).

Here, in order to calculate a new confidence factor,
The formula of X1 = X0 + (P−X0) / 10 is used.

X1 is the new certainty factor, X0 is the old certainty factor,
P indicates whether the diagnosis is correct or not (1 if the diagnosis content and the cause of the actual failure match, 0 if they do not match).

Regarding the calculation of the addition or subtraction of the value of the certainty factor 42 of the diagnostic rule, if the new inference record indicates that the correct diagnosis is made, the certainty factor of the rule concerned is increased, and conversely the new inference record is obtained. The basic idea is to lower the certainty factor of the rules related to showing that a wrong diagnosis has been made.

In the present embodiment, the certainty factor is updated by the value of the formula as shown in step 1002. However, the value of the denominator of (P-X0) / 10, which is a part of the formula for adjusting the certainty factor, is " 1
If "0" is changed, the degree of influence of one adjustment can be changed.

In addition to this, a method of calculating a value to be adjusted by a formula for adding or subtracting a certain constant, a normal correct rate or a formula using it, or weighting by a pattern of the analysis result of the corresponding record is added or subtracted. A method of calculating the value to be used is also conceivable.

[0058]

As described above, according to the present invention, it is analyzed whether or not there is a data string including a part of a plurality of predetermined data patterns in the history data regarding the condition of the diagnosis object. , If it exists, the analysis result consisting of the last history data in the data string and the number of the corresponding data pattern is input to the inference mechanism so that the diagnosis by the diagnostic rule is executed, and the diagnosis result by the inference mechanism is displayed. Since the certainty factor of the diagnostic rule used by the inference mechanism is adjusted by using this, the amount of input data to the inference mechanism can be reduced and the processing time required for inference can be shortened.

Further, the inference record can be stored compactly, and the inference accuracy can be improved by adjusting the certainty factor at a higher speed.

Therefore, it is extremely effective for diagnosing a device that outputs data having a meaning as a data string or pattern rather than the data itself. For example, the operating status of network systems and the devices that make them up, exchanges, etc., fault diagnosis, or failure diagnosis of devices with stepwise procedures such as machine tools and process control devices, and work quality control (processing accuracy, etc.) Alternatively, it is extremely effective when diagnosing the sales situation based on POS data.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a diagnostic expert system of the present invention.

FIG. 2 is a diagram showing a specific example of history data.

FIG. 3 is a diagram showing a specific example of a history data pattern.

FIG. 4 is a diagram showing a specific example of a diagnostic rule.

FIG. 5 is a diagram showing a specific example of pattern analysis data.

FIG. 6 is a diagram showing a specific example of a diagnosis result.

FIG. 7 is a diagram showing a specific example of an inference result.

FIG. 8 is a flowchart illustrating a pattern analysis processing procedure of history data.

FIG. 9 is a flowchart showing an inference processing procedure in the inference mechanism.

FIG. 10 is a flowchart showing a procedure of adjusting a certainty factor of a rule in the diagnostic rule adjusting device.

[Explanation of symbols]

1 ... History data storage device, 2 ... History data pattern storage device, 3 ... Diagnostic rule storage device, 4 ... Diagnostic device, 5 ... Diagnostic result storage device, 6 ... Inference record storage device, 7 ... Diagnostic rule adjustment device, 41 ... Pattern analysis unit 42 ... Pattern analysis data storage unit 43 ... Inference mechanism.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Nobuhiro Onodera 6-81, Onoe-cho, Naka-ku, Yokohama-shi, Kanagawa Hitachi Software Engineering Co., Ltd. In-house

Claims (1)

[Claims] 1. In a diagnostic expert system for inputting history data relating to the state of a diagnosis target and performing diagnosis using an inference mechanism, data containing a part of a plurality of predetermined data patterns in the history data. A data pattern analyzing means for analyzing whether or not a column exists, and when it exists, inputting an analysis result consisting of the last history data in the data sequence and the number of the corresponding data pattern to the inference mechanism; A diagnostic expert comprising: storage means for storing a diagnosis result by the inference mechanism; and an adjustment mechanism for adjusting the certainty factor of the diagnostic rule used by the inference mechanism by using the inference result stored in the storage means. system.