JPH0225736A - Method of deciding flaw characteristic - Google Patents

Abstract

PURPOSE:To make online decision of the flaw characteristics of an object to be inspected with good accuracy by learning flaw information, comparing the same and making the set value for deciding the characteristics of the flaw to the value lower than the wrong decision. CONSTITUTION:Light is projected to the flow of a steel sheet and the reflected light is received from the opposite direction. The flaw kind of the steel sheet is discriminated by executing signal processing to determine the numerical value of the flaw information and executing processing with discrimination logic by a discrimination tree. Judgment is then made whether the coincidence ratio of the discriminated flaw kind and the flaw kind by a visual inspection satisfies the prescribed conditions or not. When the ratio satisfies the conditions, the set value is decided as the most optimized value. The flaw kind having the high probability of being erroneously discriminated is considered when the ratio does not satisfy the conditions. The learning is successively executed by using the obtd. flaw information and the set value is successively optimized. The exact flaw kind is obtd. when the respective set values of the discrimination tree are optimized.

Description

[Detailed description of the invention] [Industrial use possible]

The present invention relates to a method for determining low characteristics, regardless of the type of the object to be inspected such as a steel plate or a slab, and in particular, to obtain information on defects on the surface of the object to be inspected using a surface flaw inspection device, and to evaluate the flaws and information. The present invention relates to a method for determining eaves characteristics suitable for use in determining the type and grade of defects.

[Conventional technology]

BACKGROUND ART Conventionally, in factories manufacturing steel plates, slabs, etc., in order to guarantee the quality of these products, surface flaws have been inspected using surface flaw inspection equipment to determine the characteristics of the flaws, such as the flaw type and flaw grade. Some of these surface flaw inspection apparatuses use optical methods to inspect surface flaws on objects to be inspected, such as steel plates and slabs. In addition, discrimination and determination of the type and grade of defects using this type of surface defect inspection device has conventionally been carried out using, for example, the fifth
The procedure was followed in the flowchart shown in the figure. That is, in the procedure shown in the figure, first, a light emitter and a light receiver are installed on the front and back sides of the steel plate to be inspected, or on either one side, and the light emitter emits light from a direction having a predetermined angle with respect to the flow direction of the steel plate. or a laser beam, etc., and the light receiver receives reflected light from a direction opposite to the direction of the light projection, which shows the characteristics of the flaw or defect, to obtain information about the surface flaw (step 110), and this type of surface flaw inspection device is described, for example, on pages 185 to 188 of ``Recent Advances in Cold Strip Equipment and Manufacturing Technology in Japan'' published by the Iron and Steel Institute of Japan in 1978. There is a description in . Next, the reflected light received as described above is subjected to signal processing to determine the numerical value of the eaves/information indicating the characteristics of the eaves (step 120), and the numerical value of the flaw information obtained in this way is By processing with the discrimination logic based only on the discrimination tree as shown in Figure 6, it is possible to determine which type of flaw is on the steel plate (e.g., abrasion, pinhole).
or the flaw grade (step 130), and the flaw type and eaves grade are determined from the discrimination results (step 140). In determining the type of defect using the above-mentioned discrimination tree, the defect information is
a, b, and C), and each set value for distinguishing them is (i, n, n), and the type of surface flaw is determined depending on whether the flaw information is larger or smaller than each set value. A
, B, (, D. In other words, in the determination tree shown in the figure, it is first determined whether the flaw information a is larger than the set value a (step 210 ),
If it is determined that the flaw information a is larger than the set value n, it is determined whether the flaw information C is thicker than the set value n (step 220). If the judgment result is positive, that is, the flaw information C is larger than the set value n. If it is determined that the flaw type is larger, the flaw type of the steel plate is determined to be A. Conversely, if the determination result is negative, the flaw type of the steel plate is determined to be B. On the other hand, when the flaw information a is not larger than the set value, it is determined whether or not the flaw information is larger than the set value m (step 230).
If it is determined that the flaw type is larger, the flaw type of the steel plate is determined to be C. Conversely, if the determination result is negative, the flaw type of the steel plate is determined to be D.

[Problem U that the invention seeks to solve]

However, when determining the characteristics of flaws (flaw type, flaw grade, etc.) on steel plates etc. using the conventional method, the following (1) to (
3) problems arise. (1) Conventionally, the above-mentioned discrimination tree and its setting values are created offline, so when the surface flaw inspection device is used online as a quality assurance device, it may be impossible to change the setting values of the above-mentioned discrimination/tree. It was becoming difficult. Therefore, it has not been possible to quickly and adequately respond to changes in objects to be inspected on the line. (2) In most cases, even if the type of defect detected on the inspected object is the same, the numerical value of the defect information is different, but in the conventional method, the discrimination tree is created offline, so
It is difficult to determine the respective set values to be optimal values corresponding to the numerical values of the flaw information. Therefore, there has been a need for a technique that can accurately determine the low temporality by optimizing each set value. (3) As a result of (1) and (2) above, the flaw characteristics determined by the conventional technology do not correspond well with those obtained by visual inspection, and it is determined that the eaves are of the same type by visual inspection and inspection by the inspection device. The rate of matching (hereinafter referred to as matching rate) was not good. In other words, with the prior art, it has not been possible to accurately determine low temporality online.

[Purpose of the invention]

The present invention has been made to solve the above-mentioned conventional problems, and it is possible to easily change the set value for low-temperature determination to one with high accuracy, and therefore, the low-temperature determination of the object to be inspected can be The purpose of the present invention is to provide a low temporality determination method that can accurately determine temporality online.

[Means to solve the problem]

In the low-temperature determination method of the present invention, when inspecting an object to be inspected for a flaw to obtain information about the flaw and comparing this flaw information with a predetermined set value to determine the characteristics of the flaw, The above objective is achieved by learning the flaw information and setting the set value to a value with fewer erroneous determinations.

[Operation and effects of the invention]

The present invention will be explained in detail below. A procedure for determining the type of flaw according to the present invention when inspecting the surface flaws of an object to be inspected, for example, surface flaws on a steel plate online using a surface flaw inspection device, and determining the flaw type, for example, the flaw type. is shown in Figure 1. That is, when the procedure shown in the figure is started, the type of flaw is determined in steps 310 to 330 in the same manner as steps 110 to 130 shown in FIG. 5 above. In this case, the tree for determining the type of flaw shown in FIG. Setting [(J, m, n
) is larger or smaller than each step 210-23.
0 to determine the type of flaw on the steel plate. Here, an example of the comparison result of the type a of the steel plate obtained by this discrimination and the flaw type obtained by visual inspection is shown in Fig. 2. In the figure, the comparison result xij is visually inspected to be the flaw type i. This is the number or percentage of steel plates judged to have flaw type J by the flaw inspection device. Originally, the type of flaw determined by the surface flaw inspection device and the type of flaw determined by visual inspection must match. Therefore, based on Fig. 2, we calculated the concordance rate of the flaw type (Ident+-cal Ra
tio; I R) is calculated and discussed using the following formulas <1) to <4). In this case, each matching rate is set as IR^ to IRQ for each flaw type A to D, respectively. IRA= (X AA/ (X AA+X A day+X
AO+X AD)l X100 (%) ・(
1) IRe= ix day e/ (X BA+X EIB
+X BC+X Mortar 0)l X100 (%)-(2
>IRc= (X cc/ (X CA+X CEl+
X cc+X co)l xlOO(%) ・(3
) IRo= fx oo/ (X 0AfX D day + X
oc+X oo)l xioo (%)・・
・ (4) At the stage where each match rate IR is calculated from the discrimination results as shown in Figure 2, focus on the parts with low match rates. For example, perform a flaw inspection based on the discrimination result of defect type A by visual inspection. When the device discrimination results are obtained as shown in FIG. 3, the matching rate IR^ is 60% from equation (1). Next, paying attention to the calculated match rate value, it is determined whether the match rate satisfies a predetermined condition, for example, 60% or more (step 340). As a result of the judgment, if the above-mentioned conditions are satisfied, it is judged that the set value n of the above-mentioned discrimination tree is the optimum setting value, and the flaw type discriminated by the above-mentioned discrimination tree is determined as the flaw type of the material to be inspected (step 360). On the other hand, if the above-mentioned conditions are not satisfied, it is considered whether the probability of erroneously identifying flaw type A as the same flaw type (hereinafter referred to as misclassification rate) is high (step 350). That is, in the case of Figure 3, X8 days, x days C, XA
Maximum value of [) 11ax (X AEI, X AC,
AO) is selected, and as a result of selection XAB has a high misclassification rate, what is determined to be defect type A by visual inspection is determined to be defect type B by the surface defect inspection device. It can be said that there are many. Therefore, when it is determined in this way, it can be said that the setting value n of the discrimination tree is not the optimal setting value, so it is decided to sequentially perform learning using the obtained pressure information C, and the setting value By optimizing n, if each setting value of the discrimination tree is optimized in this way, the determined flaw type will be accurate. Here, a method for optimizing the setting values will be explained. In this case, pressure information (a, b,
A learning method using a discriminant function based on the pressure information of the lick taken in c) will be described. In the following, a discriminant function will be described in the case where only C among the pressure information is used to optimize the set value n, that is, in the case of a univariate case. Now, the defects of the inspected object obtained as a sample (specimen) are divided into two groups for each of the flaw types A and B, and are called group A and group B, respectively. First, the probability distributions of groups A and B are tested for homoscedasticity of both groups by N^(μ
＾, σ(＾]2), N days (μB, σ(8)'), the null hypothesis Ho: σ(A)' for these groups A and B
= (IF CB) 'Alternative hypothesis H1: σ(^]2
Test ≠σ([3)2. In addition, μ^(=×〔＾))
, σ(^)2 (=S[^)2) are the mean and variance of group A, μ days (=Xc days]), σ(days)'' (=ScB)
') are the mean and variance of group B. The above test is performed using the variance S of groups A and B as shown in the following equation (5).
(^) 2, S (day) 2 ratio (dispersion ratio) F is the degree of freedom (n^
−1, nB 1), the following (8), <9
) Read the variables shown in the formula from the well-known F distribution table. F=S (＾)'/S(日)2...(5)P
r(F>F,';2: C(X/2)>=tx/
2... (6) Circle 11 Pr [F<F? 1.1 (1(α/2) ) ]=α
/2...(7) Mountain-1 F (α/2)...(8
)1s-1Feasy;2l(1-(α/2))n,-1=1/F(α)...(
9nn^-1 However, nA and nB are the numbers of samples in group A and group B, respectively. Also, P" is a symbol indicating probability. Then, as shown above, it is read from the F distribution table (8)
, the variable value shown in equation (9) is the following equation (1o) or (11)
If the formula is satisfied, the hypothesis Ho is rejected at the significance level α at that time, and the alternative hypothesis H+ is established, so A,
B The variances of both groups are not equal (σ(^)2≠σ(a)')
It can be thought of as a thing. F > F71°-1 (α/2)...
・ (10)7A-1 F<F (α/2) ... (11)71, -
1 On the other hand, if neither equation (10) nor equation (11) holds true, hypothesis Ho is not rejected at significance level α, so
The variances of both groups A and B are equal (σ(^)′2=σ(day)2
) is considered a thing. The homoscedasticity test for groups A and B is performed as described above, and after this test is completed, it is determined whether the sample variable X belongs to group A or group B. The case where the variances of are equal and the case where they are not equal are determined as follows. (i) When the variances of group A and group B are equal (σ^2=σB2
); The average μ^ of groups A and B, the standardized square distance D^2 from μ day to variable It can be determined that the variable X is included in the smaller group of DA'.
If <0 day 2, then X6A group, and if D^1>0 day 2, then x
It becomes group gB. By the way, each of the above square distances D A! , D a 2 difference day 2 1) AI is expressed as the following equation (14) based on equations (12) and (13). 0 day 2 pAt = ((X-μ day)'-(X-, μ^)')/σ2=-[
(2 (μ days - μ^) / σ beat) X (X - (μ^ + μ days) / 21] ... (1
4) Here, if μ days − μ^〉0 is assumed, then if the difference in the square distance is greater than O (D days 2D^2〉0), then the following formula (15) holds; conversely, If this difference is smaller than 0 (D day 2]) ^2 × O), then the following equation (16) holds true. x<(μ＾10μ days)/2=A8 days...(15)×
〉(μ^10μ days)/2=, u^ days...(16) And if formula <15) holds, x6A group, on the other hand,
If formula (16) holds true, it is determined that the group is X e B group. (ii) When the variances of groups A and B are not equal (σ＾2≠
σ days 2): If the value of the boundary that satisfies each square distance DA2.0 days 2 is φ, then the relationship of the following equation (17) is established from the above equations (12) and (13). (φ−μ^)2/σ(^)2 = (φ−μ days) f/σ・(days
) 2 (17) Also, from this equation (17), the following equation (18) is established. (φ−μ^)/σ(^) = (φ−μ days) / σ (days)
(18) Therefore, the boundary value φ can be obtained from the following equation (19). φ=((μ^・σce))+(μB・σ(^)))/(
σ (^) + σ (day)) ... (one) If this boundary value φ has a relationship x x φ with the variable X, then xg
If it is group A and there is a relationship of X>φ, it is determined to be group xeB. On the other hand, the misclassification rate when the variable X is determined as in (i) and (11) above can be determined as follows. In addition, the misclassification rate for classifying variable X that should be included in group A into group B is P^,
The misclassification rate for classifying variable X that should be included in group B as group A is 2.
day. (i) The variances of group A and group B are equal (σ(^)f==σC
days) 2 = σ2) case; The probability distribution of groups A and B is a normal distribution N^ (μ^. σ2.8 days (μ days, σ2), so each misclassification rate P^
, 2 days are calculated as in the following equations (20) and (21), and both are equal (P^=P day). PA=Pr [u = ((x − μA)/a)>(
(μΔ days - μ^) / σ)] P days = Pr [u = ((X - μ days) / σ) < ((
A^B-μday)/σ)ko... (21) (ii) The variances of groups A and B are not equal σ(^)2≠σ
In the case of (days) 2: The misclassification rate P^, 2 days is calculated as in the following equations (22) and (23), and both are equal (P^ = P days). P^ = pr [u = ((X-μ^)/σ^)〉(
(φ-μ^)/σ^) ] ... (22) PB=Pr [u = ((x-μ days)/σ days) <(
〈φ - μ days) / σ days)] ... (23) As described above, using the defect information C as a single variable, determine which of the groups of defect type A and defect type B the defect information belongs to. The misclassification rate P^ and Pe are calculated from equations (20) to (23), and the discrimination setting value C is set to <15) μ in equation (16).
It can be obtained from ＾day or φ in equation (19). Here, we focused on the misclassification rate among these three numbers obtained, and compared the misclassification rate H1 in the past data with the misclassification rate H2 when new data was added to the past data. The smaller one of the different ratios is selected, and the set value for discrimination and the type of flaw are selected and determined according to the selection. That is, if the discrimination setting value n1 based only on past data, the discrimination result ° is Jl, the discrimination setting value when new data is added to the past data is 02, and the discrimination result J2 is,
When the misclassification rate is H1>H2, the discrimination setting value is changed to 02, and the flaw type is left unchanged at Jl.On the other hand, when the misclassification rate is H1<H
When it is 2, the set value remains nl and the flaw type is set to J2. Next, a specific example of the flaw type discrimination procedure when the variables are multivariate will be explained based on the flowchart of FIG. 4. First, if it is determined that there is a flaw by the surface flaw inspection device, based on the flaw information (a, b, c, etc.) mentioned above,
The type of flaw is determined using a determination tree as shown in the figure (step 410). Next, it is determined whether the match rate between the discrimination result and the visual inspection satisfies a predetermined match rate (step 420). In this case, if the match rate satisfies 60%, the set value is optimal. The flaw type is determined according to the matching rate (step 490). On the other hand, if it is determined that the matching rate is not satisfied, the defect information to be learned is determined from among the obtained defect information (a, b, c, etc.) (step 430), and then the determined (5) using p variables depending on the defect information.
-Discrimination is performed using a discriminant function based on equation (23) etc. (step 440). In this case, the number of variables p is I)=1.
Takes a value of 2... Next, the determination result J2, misclassification rate H2, and determination setting value (for example, 12) when the new data is added to the past data are calculated (step 450). Next, it is determined whether the misclassification rate H1 based only on past data is larger than the misclassification rate H2 (Hl>H2) (
Step 460), when the determination result is positive, the determination setting value is changed, for example, from °C1 to β2 (step 470); on the other hand, when the determination result is negative, the determination setting value is not changed (step 480). After the discrimination setting values as described above are determined, the type of flaw is determined (step 490).As described above, the type of flaw can be determined even when the variables are expanded to not only one variable but also multiple variables. This makes it highly practical. Note that the flaw characteristics to be determined are not limited to the type of flaw as described above, but may also be the low-temperature characteristics of the pond, such as the flaw grade.The procedure for determining low-temperature characteristics is shown in Figures 1 and 4. The procedure is not limited to the procedure shown in , but any procedure may be used as long as it can learn the flaw information and set the setting value to a value with fewer erroneous determinations. As explained above, according to the present invention, it is possible to learn the setting values for determining the flaw type online and change them to more accurate ones, so the following excellent effects can be obtained. It was done. (1) Since flaw information can be sequentially learned online and changes in the setting values of the discrimination tree can be easily and automatically executed in a computer, etc., work efficiency is improved without requiring any intervention by an operator or the like. (2) Since the set value is calculated using a discriminant function or the like, the set value can be easily and quickly calculated. (3) Since the optimal setting value can always be selected by (1) and (2), the low timeliness determination rate improves and it becomes possible to accurately determine the low timeliness. Therefore, it is advantageous in terms of quality assurance of the material to be inspected, for example, a steel plate. Moreover, complaints from users regarding steel plate surface flaws can be reduced.

[Brief explanation of the drawing]

Fig. 1 is a flowchart showing an example of the flaw type determination procedure when the present invention is carried out using a single variable, and Fig. 2 is a visual judgment result of the flaw type determined by the surface flaw inspection device to explain the present invention. FIG. 3 is a diagram showing an example of the discrimination result, FIG. 4 is a flowchart showing an example of the flaw type determination procedure according to the present invention in the case of multiple variables, and FIG. FIG. 6 is a flowchart showing an example of a type determination procedure. FIG. 6 is a flowchart showing an example of a flaw type discrimination tree.

Claims (1)

[Claims] (1) When inspecting the object to be inspected for flaws and acquiring information about the flaws, and comparing this flaw information with predetermined set values to determine the characteristics of the flaws, learn the flaw information and A flaw characteristic determination method characterized by reducing the number of false determinations of set values.