JPH0663791B2 - Process diagnosis method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a process diagnosis method in a manufacturing process such as a steel process, and rapidly predicts and evaluates a process diagnosis characteristic value from the current operating conditions, and By taking measures, the process operation will be stabilized.

[Prior Art] In recent years, process diagnostic systems such as quality diagnostics have been developed by qualitative or fragmentary empirical rules based on knowledge industrial methods such as rule bases for process diagnostics in manufacturing processes such as steel processes.

However, no fixed method has been established for the diagnostic method used in this type of process diagnostic system,
The current situation is that each system is different.

On the other hand, the authors proposed in Japanese Patent Application No. 63-265678 a standard method for quality diagnosis based on empirical rules.

In the process diagnosis, the number of process performance data is large, and the intermediate state is often inferred once and inferred based on the result. Therefore, in Japanese Patent Application No. 1-145614, a process diagnosis method has been proposed which enables accurate diagnosis even when the number (kind) of input result data is large and which can predict an intermediate state.

[Problems to be Solved by the Invention] When a process diagnosis is performed based on the empirical rule of a specialist in the manufacturing process, due to the deviation from the target value of the operation result data,
In a routine method for diagnosing process characteristics and quality characteristics, by allowing vague expressions such as "a failure A occurs when the temperature is higher than the target value" in the expert's rule of thumb, The knowledge representation of can be made more flexible, and the acquisition and adjustment of knowledge can be made easier.

Therefore, the present invention is directed to Japanese Patent Application No. 63-265678 and Japanese Patent Application No.
In the diagnostic method of No. 145614, it is an object to make knowledge acquisition and adjustment easier.

[Means for Solving the Problems] In order to solve the above problems, according to the present invention, a plurality of types of operation record data obtained in a manufacturing process are diagnosed by a process diagnosis method according to deviations from their target values. For each of the multiple primary diagnostic characteristic items, the primary demerit value of the characteristic due to the deviation of each actual operating data from the target value is obtained using the fuzzy relational expression between the operational performance data and the primary demerit value, and then the The primary deduction values are summed to obtain the primary diagnostic characteristic evaluation value, and further, according to the deviation from the target value, for each secondary diagnostic characteristic item to be diagnosed, the deviation from the target value of the plurality of primary evaluation values The secondary deduction value caused by is calculated using the fuzzy relational expression between the primary evaluation value and the secondary evaluation value, and the final evaluation result is generated according to the secondary deduction value.

[Operation] In the process diagnosis method of the present invention, diagnosis is hierarchically executed a plurality of times. For example, in the primary diagnosis, steel strip temperature,
Input various operation results data such as hydration amount, adhesion amount, decarburization property, generate a primary deduction value from those values by fuzzy relational expression, and evaluate the primary deduction value according to this primary deduction value. And Then, in the secondary diagnosis, the primary evaluation value is input, the secondary deduction value is generated by the fuzzy relational expression similar to the above, and the final result is obtained based on the secondary deduction value.
Of course, not only the two-stage evaluation of the primary and the secondary, but a hierarchical diagnosis of three or more stages may be performed.

[Embodiment] Next, an embodiment of the present invention will be described with reference to the drawings.

In this embodiment, a system for performing quality diagnosis regarding quality characteristics (including coating characteristics) of the entire grain-oriented electrical steel sheet process will be described. Although not shown, this system includes, as hardware devices, a computer system main body including a central processing unit and a storage device, and a terminal device connected thereto. The terminal device includes a keyboard that is an input device and a display that is an output device.

For example, the actual quality diagnosis of coating properties is carried out according to the empirical rules of experts based on the operation result data of the decarburization line,
A track record of coating quality is obtained after about 10 days. Therefore, the occurrence of coating defects during that time is prevented by the quality prediction of the quality diagnostic system.

FIG. 1 shows a main part of a software program provided in the quality diagnostic system and a structure of data to be processed or referred to.

The outline of the processing of this system will be described with reference to FIG. The inference engine 20 processes the input operation result data 10 and evaluates the quality, that is, makes a diagnosis. Evaluation result,
That is, the diagnostic value 30 is easily displayed on the display.

The processing of the inference engine 20 is executed based on the data on the knowledge base 60. The data on the knowledge base 60 is a rule base described in a form suitable for the processing of the inference engine 20. Therefore, the data on the knowledge base 60 is difficult for a general operator to understand. However, to build a system with high diagnostic accuracy, updating the data on the knowledge base 60 is essential.

Therefore, in this embodiment, of the data provided on the knowledge base 60, a knowledge table 40 is provided in which only various parameters required for the data are described in a table format. The data on the knowledge table 40 is relatively easy to understand and can be understood by general operators.

When the operator performs a predetermined operation from the keyboard of the terminal device, the program of the editor 70 is activated and the data in the knowledge table 40 is sequentially displayed on the display. By moving the cursor displayed on the display to specify the position of the data and further performing an input operation, editing processing such as correction, deletion and addition is performed on the data at that position. By this editing process, the knowledge table
40 data updated.

When the processing of the editor 70 is completed, the program of the data conversion 50 is activated, and the data held in the knowledge base 60 is regenerated based on the updated data in the knowledge table 40.

In addition, when the result (actual evaluation value) of the actual quality diagnosis by the expert is obtained, the program of the learning process 80 is started, and the knowledge table is based on the output of the inference engine 20 and the input actual evaluation value. The weight data existing on 40 is updated. The contents of the knowledge base 60 are updated as the knowledge table is learned.

An outline of the processing flow of the inference engine 20 of FIG. 1 is shown in FIG. The inference processing of the inference engine 20 is hierarchically configured, and in this example, there are two layers of primary diagnosis and secondary diagnosis. In FIG. 2, steps 1 to 3 correspond to the primary diagnosis processing, and steps 4 to 6 correspond to the secondary diagnosis processing.

In step 1 of FIG. 2, the input operation record is set.

In this example, 10 types of operation record data of dp1, dp2, dp3, dp4, steel strip temperature 1, steel strip temperature 2, steel strip temperature 3, hydration property, adhesion amount and decarburization property are input.

In step 2 of FIG. 2, a deduction value is calculated for each diagnostic item. The diagnostic items of this primary diagnosis include "peroxidation A", "peroxidation B", "peroxidation C", "insufficient oxidation A", and "insufficient oxidation B", as shown in the intermediate characteristic items in FIG. 3b. There are eight items: "," insufficient oxidation C "," decarburization abnormality A ", and" decarburization abnormality B ".

Figure 3a shows the knowledge (data in the knowledge table and knowledge base) used to calculate the deduction value for "peroxidation A" in the primary diagnosis. Specifically, each operation performance item The value of the weight of the deduction value and the characteristic pattern that adversely affects the diagnosis result are indicated by thick frames. That is, in the example of FIG. 3a, the weight of each item of dp1, dp2, dp3, dp4, ... Is 0, 70, 5, 10, .. The total is set to 100%.

The data on the knowledge table 40 that corresponds to the data in Figure 3a is shown in Figure 3b. In FIG. 3b, first, a list of primary diagnostic items is shown, followed by detailed data on each of the primary diagnostic items. Focusing on the column showing the contents of the peroxide A item, the first line shows the range in which the value of the evaluation result changes (80 to 40 points), and the second to eleventh lines show each operation. Information about actual items is shown. For example, in the numerical sequence following dp2, the first "1" is the number of patterns and the subsequent "1 1 2 1 7".
As shown in Fig. 3c, 0.0 "is correct if {if dp2 is large, take a deduction value that maximizes (80-40) x 70/100"}. It shows the parameter when expressed by the membership value. The first "1" is the membership function of the antecedent part. Is shown. Using “0” instead of “1” indicates that there is no rule, and using “−1” It is a membership function of. Next "1 2"
Shows the parameters shown in FIG. 3c of membership.
The next "1" indicates the power order of the membership function x'representing the truth value of the rule. The higher the order, the more correct the rule. "70" is the weight shown above.

Now, assuming that dp2 is 1.75%, the demerit value at that time is 0.75 from the goodness of fit of the antecedent part of the rule, and then 0.75 = X ¹ = a a = 0.75 from the membership function of the truth value of the rule. , The value of a in FIG. 3c is obtained as 0.75. this is,
Indicates the conformance of the consequent part of the rule. The maximum deduction value is
It is 28 from the formula of (80-40) x 70/100, and it is 21 when the value that the membership becomes 0.75 is calculated. When the rule is very correct, the function of membership in the truth value of the rule, such as x ² or x ³ , is used. When the correctness of the rule is a little ^weak, x ^{- 1,} x ^{- 2} membership functions, such as is used.

Other diagnostic items are also calculated by referring to the data on the knowledge base.

In step 3 of FIG. 2, the deduction values for each operation performance item obtained in step 2 are summed for each primary diagnosis item, and the deduction values for each primary diagnosis item are summed for each actual operation item. . An example of the result of this processing is shown in FIG. This information can be displayed on the display in the format shown in FIG. 6 if desired. The larger the deduction value displayed, the more important it is to improve the process characteristics. By referring to the display of the deduction value on the display, it is possible to instantly know the item that adversely affects the process characteristics.

Steps 4, 5 and 6 in FIG. 2 are basically the same as the above-mentioned steps 1, 2 and 3, respectively, and the results obtained in the primary diagnosis of steps 1 to 3, that is, 8 kinds for each intermediate characteristic item Data (peroxide A, peroxide B, peroxide C, peroxide deficiency A, oxidation deficiency B, oxidation deficiency B, decarburization abnormality A and decarburization abnormality B) as input data, coating defects, decarburization defects, magnetic properties Diagnosis is performed on four diagnostic items: defective and gloss.

In step 4, the evaluation value of the intermediate characteristic item obtained up to step 3 is set.

In the calculation processing of the deduction points in step 5, the film defect,
Deduction points are obtained for each of the four diagnostic items of decarburization failure, magnetism failure, and gloss failure. For example, for the coating failure diagnosis item, the deduction value is obtained based on the knowledge shown in FIG. 4a. A portion of the data on the knowledge table 40 that corresponds to the data referenced in this process is shown in Figure 4b.

Note that the meanings of the information described in each item of FIGS. 4a, 4b, and 4c are the same as those of FIGS. 3a, 3b, and 3c already described, respectively. Therefore, their description will be omitted.

In step 6 of FIG. 2, the deduction values for each primary diagnostic item (intermediate characteristic item) obtained in step 5 are summed up for each secondary diagnostic item, and the deduction values for each secondary diagnostic item are calculated for each secondary diagnostic item. Sum for each primary diagnostic item. The result of this processing can be displayed on the display in a tabular format, if necessary, as in the case of FIG. 5, which is the result of the intermediate diagnosis processing.

In step 7 of FIG. 2, the deduction value for each diagnostic item obtained in step 6 is subtracted from the full score value to obtain an evaluation value for each diagnostic item, and the total deduction points for each operation result item are arranged in descending order. They are rearranged, displayed on the display, and the operating conditions to be corrected are displayed to inform the operator.

Part of the data on the knowledge base 60 of FIG. 1 is shown in FIG.
Shown in Figures 6b and 6c.

First, refer to FIG. 6a. "FI marked on the top
The content of "NDINGS" is a list of operation performance item names entered in the primary diagnosis. The content of "HYPOTHES" below is a list of primary diagnosis item names. The content of "FH-RULE" is a list of names of existing rules for calculation of deduction values and their total values.

Reference is now made to Figures 6b and 6c. These parts include
The definition of the rule corresponding to each of the rule names described in “FH-RULE” in FIG. 6a is described. For example, 6b
In the first part of the figure, a rule relating to "dp1-defect deduction" is defined, and here, when dp1 is within the valid range for the rule, the processing described in the "THEN" column is executed. For example, the deduction value of ([dp1] [oxidation deficiency]) is (func-call 1 (1) '(1)' (2) '
It is calculated as the processing result of (5) '(50.0)' (1.0) dp1). Here, "func-call" indicates a LISP language function that can be executed by the inference engine 20, and the data that follows the function indicates the function parameter.

Also, the contents of ## Peroxide A deduction total calculation ## in FIG.
This is the definition of the rule regarding the calculation of the total amount of deductions for peroxide A of the primary diagnostic item. Here, if the data of [dp2] [peroxidation A] is 0.0 or more, [dp1] [peroxidation A], [dp2 ) [Peroxidation A], [dp3] [Peroxidation A],
[Dp4] [peroxidation A], [steel temperature 1] [peroxidation A], [steel temperature 2] [peroxidation A], [steel temperature 3] [peroxidation A], [hydration amount] [peroxidation The sum of [A], [adhesion amount] [peroxidation A], [decarburization] [peroxidation A] is stored in the "peroxide A deduction total".

Also, the content of ## dp1 − total deduction calculation ## in FIG. 6c is the definition of the rule regarding the total calculation of the deduction value of dp1 of the operation result item, and in this example, the value of [dp1] [peroxide B] But
When it is 0.0 or more, [dp1] [peroxidation A], [dp1] [peroxidation B], [dp1] [oxidation deficiency A], [dp1] [peroxidation C], [dp1] [oxidation deficiency B] , [Dp1] [insufficient oxidation C], [dp1] [decarburization abnormality A], [dp1] [decarburization abnormality B]
Store the total sum of “dp1−total deduction”.

The data of this knowledge base is converted to the data shown in FIG.
Is automatically generated based on the data on the knowledge table 40. Outline of processing of data conversion 50
See the figure for reference.

FIG. 8 shows an outline of the processing in the learning processing 80 of FIG. In Fig. 8, x ₁ , ... xn are operation record data,
f ₁ , ... fn is a function for obtaining the probability a of the conclusion part of the rule from dp2 in Fig. 3c, ω ₁ , ... ωn is a weight to be learned, y is a quality evaluation prediction value, d is determined by the expert. The result of actual evaluation, that is, the actual evaluation value is shown. As a learning algorithm, an orthogonal learning method or an error backpropagation method can be used, but the algorithm of the orthogonal learning method is used here.

When the corrected value of ωi is Δωi, Δω = (Δω ₁ , Δω ₂ , ..., Δωn) ′ Δω = [k (d−y) / (λ + F′F)] · F where F = (f ₁ ( x ₁ ), f ₂ (x ₂ ), ... fn (x
n)) ′ That is, for each quality characteristic item, the defection contribution rate, that is, the amount of weight on the knowledge table is learned using the performance evaluation value d as a teacher signal. The knowledge base is updated as the knowledge table is learned.

In the above embodiment, the processing of only two layers of the primary diagnosis and the secondary diagnosis has been described, but the processing of the tertiary diagnosis and the fourth diagnosis is added by the same method, and the processing is changed to perform the multi-step process diagnosis. Good.

[Effect] As described above, according to the present invention, fuzzy (FUZZ
Y) The process diagnosis can be carried out hierarchically by using a relational expression and a more easily understood expression of an expert. Therefore, highly accurate process diagnosis is possible.

[Brief description of drawings]

FIG. 1 is a block diagram showing a software and data structure of a quality diagnostic system for carrying out the present invention. FIG. 2 is a flow chart showing an outline of processing of the inference engine 20 of FIG. FIGS. 3a, 3b and 3c are a map showing the contents of the intermediate diagnostic knowledge of the embodiment, a map showing a part of the knowledge table thereof and a graph showing the fuzzy relationship used therein. FIG. 4a, FIG. 4b and FIG. 4c are a map showing the contents of the knowledge of the secondary diagnosis of the embodiment, a map showing a part of the knowledge table and a graph showing the fuzzy relation used therein, respectively. FIG. 5 is a front view showing an example of information displayed on the display. 6a, 6b and 6c are maps showing the contents of the knowledge base created from the knowledge table. FIG. 7 is a flow chart showing the processing of the data conversion 50 of FIG. FIG. 8 is a block diagram showing the structure of the learning process 80 of FIG. 10: Operation data, 20: Inference engine 40: Knowledge table, 60: Knowledge base

 ─────────────────────────────────────────────────── ─── Continuation of front page (56) Reference JP-A-59-204707 (JP, A) JP-A-61-59900 (JP, A) JP-A-60-14303 (JP, A)

Claims (1)

[Claims] 1. A method of diagnosing a plurality of types of operation result data obtained in a manufacturing process according to deviations from the target values thereof, wherein a target of each operation result data is obtained for each of a plurality of primary diagnostic characteristic items to be diagnosed. The primary demerit value of the characteristic due to the deviation from the value is obtained by using the fuzzy relational expression between the operation result data and the primary demerit value, and then the primary demerit value is summed to obtain the primary diagnostic characteristic evaluation value, and further, According to the deviation from the target value, for each secondary diagnostic characteristic item to be diagnosed, the secondary deduction value resulting from the deviation from the target value of the plurality of primary evaluation values, the primary evaluation value and the secondary evaluation value. A method for diagnosing a process, characterized in that a final evaluation result is generated according to a secondary deduction value by using a fuzzy relational expression with and.