JP5494566B2 - Defect detection method for steel - Google Patents

Description

  The present invention relates to a method for detecting defects in a steel surface and / or surface layer.

  Taking the case where the steel material is a steel plate as an example, the surface and surface layer of the steel plate include, for example, an indentation by a roll, a dross that is formed by dripping molten zinc bath into the galvanized layer, and argon gas during casting. Various defects may occur, such as blow hole defects scattered in the steel sheet formed by being captured in the slab, or surface defects due to non-uniform thickness of the galvanized layer.

  Among these defects in the steel sheet, defects that appear to have optically different colors compared to other normal sites have been detected by visual inspection by the operator. When visual inspection is performed by an operator, the production line speed has to be reduced, so that productivity is lowered. Further, the accuracy of visual inspection varies depending on the operator, and quality may vary. In recent years, as a high quality control level is required, detection of minute defects that are difficult to detect visually has been required.

  Patent Document 1 discloses a defect detection device that automatically performs defect detection based on image data of a steel plate surface obtained by photographing the steel plate surface with a CCD camera. Have difficulty.

According to Patent Document 2, the inventors heated a steel material surface, acquired the temperature distribution of the steel material surface during heating or cooling after the heating by an infrared thermography camera, and obtained Laplacian of the obtained temperature distribution data. A method to detect the defect by calculating is proposed. In Patent Document 2, using thermal image temperature distribution data obtained on the (x, y) plane with a pixel interval h, the Laplacian Δ _xy T of the temperature T (x, y) at coordinates (x, y) is It is represented by the formula (1).
Δ _xy T = {T (x + 1, y) + T (x−1, y) + T (x, y + 1) + T (x, y−1) −4 × T (x, y)} / h ² (1)

As shown in FIG. 6, the right side of Equation (1) is the surface temperature T (x, y) at the pixel based on the thermal image data acquired by the infrared thermography camera, and the position on the top, bottom, left, and right of the pixel, respectively. The temperature of each pixel to be calculated, that is, T (x + 1, y), T (x, y + 1), T (x-1, y), T (x, y-1) can be used for calculation. it can. Here, h is the size of the pixel (interval between adjacent pixels). In the above example, the Laplacian is calculated for each pixel. The reason why defects can be detected by calculating the Laplacian of the temperature distribution data is considered to be that Laplacian Δ _xy T is greatly related to the balance of heat transfer amount as shown in Equation (2). .
Δ _xy T = (1 / α) · (∂T / ∂t) −∂ ² T / ∂z ² (2)

  The value of the Laplacian of a portion where the surface of the steel material and / or the surface layer is not defective (hereinafter referred to as “healthy part”) is ideally “0”. However, since there is actually a slight temperature distribution, it does not become “0”. Therefore, when the Laplacian value falls within a range of values centered on 0 (for example, −1 to +1), the region is determined as a healthy part, and when the Laplacian value exceeds the range, the region is It is determined that the region has a defect (hereinafter referred to as “defect portion”).

JP 2004-219177 A International Publication WO2010-033113A1

Modern Mathematical Science Encyclopedia, Osaka Book, 1991, p. 950

  However, the method for recognizing defects by calculating the Laplacian of the surface temperature data proposed in Patent Document 2 has the following problems.

  The first problem is that temperature unevenness due to heating is erroneously recognized as a surface defect. When heating a steel material, it is difficult to heat the steel material uniformly, and temperature unevenness is generated in the heating process, and the temperature unevenness due to the heating is erroneously recognized as a surface defect.

  The second problem is that a so-called dross soot generated when hot-dip galvanizing is performed cannot be correctly recognized.

  The present invention detects a defect on the surface of a steel material without being affected by temperature unevenness that occurs when the steel material is heated, and detects a defect in the steel material that correctly recognizes a so-called dross flaw that occurs when hot dip galvanizing is performed. To provide a system.

  The inventors have found a correction method for reducing the influence of temperature unevenness that occurs when a steel material is heated in a method for detecting a defect in the steel material based on the Laplacian of the steel material surface temperature data.

  Further, the inventors have found a method for determining whether or not the area is a dross defect from the area of the defect area, the eccentricity of the area, and the Laplacian after the correction of the area.

The gist of the invention is as follows.
(1) In detecting defects on both the surface and the surface of the steel material,
The step of heating the surface of the defect-free steel material to different temperatures in increments of 5 to 10 ° C. and the surface temperature distribution of the defect-free steel material heated to different temperatures in increments of 5 to 10 ° C. by using an infrared thermography camera. Step to get as
Calculating the average value of each defect-free steel surface temperature data;
Calculating Laplacian data for each defect-free steel surface temperature data;
Linking each defect-free steel surface temperature data and the Laplacian data;
Calculating a Laplacian standard deviation associated with a temperature at which the difference between the average value and the surface temperature is within ± 2.5 ° C. in each defect-free steel surface temperature data;
Obtaining a linear regression line of the average value of the obtained surface temperature and Laplacian standard deviation;
Heating the surface of the steel material to be inspected and obtaining the surface temperature distribution of the steel material to be inspected as the surface temperature data of the steel material to be inspected by an infrared thermography camera;
Calculating Laplacian data for the inspected steel surface temperature data;
Linking the steel material surface temperature data to be inspected and the Laplacian;
Calculating a Laplacian standard deviation of a defect-free steel material corresponding to the surface temperature of the steel material to be inspected from the regression line, and performing a correction to divide the Laplacian corresponding to the surface temperature by the Laplacian standard deviation;
A defect detection method for a steel material, wherein a defect site is detected using the Laplacian data after correction.
(2) Of the parts recognized as defects, the area is 0.25 mm ² or less, the eccentricity is 0.5 or less, the area is A (mm ² ), and the corrected Laplacian maximum value is L _max The method for detecting a defect in a steel material according to (1), wherein a portion where the relationship of “L _max ≧ 1.15 × (12.020√A + 1.4694)” is established is determined as Dross 疵.

  By the method of the present invention, there is a remarkable effect that it is possible to detect a defect in the steel material without being affected by the temperature unevenness of the steel material caused by heating. In addition, there is a remarkable effect of detecting dross traps with a high probability.

It is a figure which shows how to think about the relationship between the area | region which has a defect, and eccentricity. It is a figure which shows the relationship between a steel plate surface temperature and a Laplacian standard deviation. It is a figure which shows the regression line computed from the relationship between a steel plate surface temperature and a Laplacian standard deviation. The detection results when using the corrected Laplacian, (a) is a Laplacian image with correction, (b) is a binarization process of the image data of (a), and the detection area of the present invention is detected. It is the image which left the area | region determined with the method as the minimum dross 疵. The detection results when using an uncorrected Laplacian, (a) is an uncorrected Laplacian image, and (b) is an image obtained by binarizing the image data of (a). It is a figure which shows the view which calculates Laplacian from thermal image temperature distribution data. Is a diagram showing the relationship between the square root √A maximum value L _max and the detection area of the Laplacian after correction in the detection area of the dross defects. It is a perspective view which shows the condition which is detecting the defect of steel materials.

In the two-dimensional plane of x and y, for example, for temperature T ∇ ² T = ∂ ² T / ∂x ² + ∂ ² T / ∂y ² (3)
In the mathematical world, ∇ ² is called Laplacian (Laplace operator) in two variables. In the world of image processing, when digital image processing is performed on the obtained image data and the calculation of equation (3) is performed, ∇ ² T itself is sometimes referred to as T Laplacian and may be expressed as Δ _xy T.

When the equation (3) is to be solved by numerical analysis using the difference method, the data intervals in the x and y directions are set as h _x and h _y , respectively, and x and y are integer values in the temperature data T (x, y). If the simplest difference formula is used,
∇ ² T = ∂ ² T / ∂x ² + ∂ ² T / ∂y ²
≈ {T (x + 1, y) + T (x-1, y) -2 × T (x, y)} / h _x ²
+ {T (x, y + 1) + T (x, y-1) -2 × T (x, y)} / h _y ²
Can be described. If h _x = h _y = h,
∇ ² T
≈Δ _xy T = {T (x + 1, y) + T (x-1, y) + T (x, y + 1) + T (x, y-1) −4 × T (x, y)} / h ² (1)
Is obtained.

In the present invention, ∇ ² T in the above equation (3) is sometimes referred to as T Laplacian, and more specifically, Δ _xy T in equation (1) is sometimes referred to as T Laplacian.

The equation of 3D unsteady heat conduction is
(1 / α) · (∂T / ∂t) = ∂ ² T / ∂x ² + ∂ ² T / ∂y ² + ∂ ² T / ∂z ² (4)
Can be described. α is called temperature conductivity. From this equation, ∂ ² T / ∂x ² + ∂ ² T / ∂y ² = (1 / α) · (∂T / ∂t) −∂ ² T / ∂z ²
If the left side of the above equation is replaced with the equation (3) and expressed as ∇ ² T = Δ _xy T, the equation (2) can be derived.

[First Embodiment]
The first embodiment is a method for reducing the influence of temperature unevenness that occurs when a steel material is heated against Laplacian for the surface temperature expressed by thermal image data.

  According to Patent Document 2, when the surface of a material is heated, the surface temperature of the material is acquired as thermal image data using an infrared thermography camera, and the Laplacian of the surface temperature is calculated, the Laplacian of the normal part is almost zero. On the other hand, since the Laplacian at the defect site has a value different from that of the normal part, the presence of the defect can be detected.

  However, when the same defect part of the same steel material was evaluated with different measurement conditions, it was found that the Laplacian value at the defect part fluctuated when the measurement conditions were different. For this reason, even if a threshold value for detecting a defect under a certain measurement condition is set, if the measurement condition changes, a situation occurs in which the defect cannot be detected with the threshold value.

  As a result of further investigation, it was found that the size of the Laplacian at the defect site changes depending on the temperature of the steel surface at the time of measurement. In addition, when heating and cooling a steel material for measurement, heating and cooling unevenness occurs even in a defect-free portion of the steel material, and the Laplacian of the defect-free portion is not strictly zero and varies spatially. When the standard deviation of the Laplacian of the defect-free part was calculated, it was found to be a finite value. And when the temperature of the steel material at the time of taking thermal image data changed, it turned out that the standard deviation of the Laplacian of a defect-free part and the size of the Laplacian of a defective part change in a substantially proportional relationship.

  Therefore, if the standard deviation of the Laplacian of the defect-free part is determined as a function of the steel surface temperature, the Laplacian at the time of defect detection is divided by the standard deviation of the Laplacian of the defect-free part corresponding to the steel surface temperature at the time of the measurement. As a result of the correction, the corrected Laplacian is normalized, and it is possible to detect the presence or absence of defects using a certain threshold.

  In addition, both the size of the Laplacian of the defect site and the size of the Laplacian standard deviation of the non-defect site are linearly fluctuated due to the effect of the steel surface temperature, and are not affected much by other than the surface temperature, In addition, it is not greatly affected by whether the steel surface was in the temperature rising process or the temperature falling process at the time of temperature image data collection. Therefore, a linear regression line is obtained for the relationship between the steel surface temperature and the Laplacian standard deviation for defect-free parts, and then the Laplacian of the material to be inspected is corrected by the Laplacian standard deviation that is regressed from the surface temperature of the material to be inspected by the linear regression line By doing so, the defect detection accuracy can be improved.

(Laplacian correction)
The basic idea is that the absolute value of the Laplacian in the defective part is larger than the absolute value of the Laplacian in the healthy part. However, the absolute value of Laplacian is small when the steel material temperature is close to room temperature, and the absolute value of Laplacian is large when the steel material temperature is high. In this respect, the height of the Laplacian peak in the defective portion is the same as the variation of the Laplacian in the non-defective portion. As described above. From this, in the method described in Patent Document 2, if the Laplacian value threshold for determining the defective part and the healthy part is not changed by the steel material temperature, the defective part is determined as the healthy part, and the healthy part is determined as the defective part. Over-detection that is determined frequently occurs.

  The inventors have measured the standard deviation of the Laplacian of the normal part corresponding to the surface temperature measured at each point on the steel material surface, and the Laplacian obtained from other surface temperature data is compared with the standard deviation previously measured. It has been found that by standardizing using the ratio, defects can be detected with one threshold even if there is an effect of uneven steel temperature.

  For each surface temperature data obtained when a steel material having only a healthy part is heated in advance at 60 ° C., 70 ° C., 80 ° C., 90 ° C. and 10 ° C. in advance, for example, as shown in FIG. Obtain the temperature, obtain the Laplacian and its standard deviation for the surface temperature data where the temperature difference between the surface temperature data and the average temperature is within ± 2.5 ° C, and the relationship between the surface temperature of the steel and the standard deviation of the Laplacian Create a linear regression equation. The reason for calculating the standard deviation of the Laplacian in the part where the surface temperature difference is within ± 2.5 ° C. is that the temperature can be basically regarded as the same temperature if the temperature difference is within ± 2.5 ° C.

Next, the surface temperature data T (x, y) is measured after heating the surface of the material to be inspected, and the Laplacian Δ _xy T (x, y) of the temperature data is calculated.

The standard deviation L _dfr (T (x, y)) of the Laplacian at the surface temperature T (x, y) given the corresponding Laplacian is calculated using the obtained linear regression equation, and the Laplacian Δ of the temperature data of the material to be inspected is calculated. Laplacian correction is performed by dividing _xy T (x, y) by Laplacian standard deviation L _dfr (T (x, y)), and defect detection is performed using the corrected Laplacian L _mod (x, y). Do. The corrected Laplacian L _mod (x, y) is normalized, and it is possible to detect the presence or absence of defects using a certain threshold value. This will be specifically described below.

In order to detect defects on both the surface and the surface of the steel material, the following procedure is performed.
First, the surface of the defect-free steel material is heated to different temperatures in increments of 5 to 10 ° C.
Second, the defect-free steel surface temperature distribution heated to different temperatures in increments of 5 to 10 ° C. is acquired as defect-free steel surface temperature data by an infrared thermography camera.
Thirdly, the average value of each defect-free steel surface temperature data is calculated.
Fourth, Laplacian data is calculated for each defect-free steel surface temperature data T (x, y). As a Laplacian calculation algorithm for temperature data, any method may be used as long as it is a numerical calculation method approximating the above equation (3). Most generally, Δ _xy T in the equation (1) can be used.
Fifth, the defect-free steel surface temperature data T (x, y) at each (x, y) coordinate and the Laplacian data Δ _xy T (x, y) are linked.
Sixth, in each defect-free steel surface temperature data, a standard deviation (hereinafter referred to as “Laplacian standard deviation”) is calculated for a Laplacian linked to a temperature at which the difference between the average value and the surface temperature is within ± 2.5 ° C. And the Laplacian standard deviation at that temperature.
Seventh, an average value of the surface temperature obtained at different temperatures in increments of 5 to 10 ° C. and a linear regression line of Laplacian standard deviation L _dfr (T) are obtained.
Eighth, the surface of the steel material to be inspected is heated, the surface temperature distribution of the steel material to be inspected is measured with an infrared thermography camera, and the surface temperature data T (x, y) to be inspected is obtained.
Ninth, Laplacian data Δ _xy T (x, y) is calculated for the steel material surface temperature data T (x, y). As for the Laplacian calculation algorithm, the same method as the algorithm for calculating the Laplacian from the surface temperature of the defect-free steel material can be employed.
Tenth, the surface temperature data T (x, y) to be inspected is linked to the Laplacian _Δxy T (x, y).
Eleventh, the Laplacian standard deviation L _dfr (T (x, y)) of the defect-free steel material corresponding to the surface temperature T (x, y) of the steel material to be inspected is calculated from the regression line and corresponds to the surface temperature. The Laplacian Δ _xy T (x, y) is divided by the Laplacian standard deviation L _dfr (T (x, y)),
L _mod (x, y) = Δ _xy T (x, y) / L _dfr (T (x, y)) (5)
Correct as follows.
Twelfth, a defect site is detected using corrected Laplacian data L _mod (x, y). As described above, the Laplacian at the time of defect detection is divided by the standard deviation of the Laplacian of the defect-free part corresponding to the steel surface temperature at the time of measurement, and the corrected Laplacian L _mod (x, y) is the standard. Therefore, it is possible to detect the presence or absence of defects using a certain threshold value.

[Second Embodiment]
The second embodiment is a method for determining whether or not a part determined to be a defective part is a minimal dross using the corrected Laplacian.

  Minimal dross has very small foreign objects in the surface, not on the surface. Therefore, when the steel material is heated and the surface temperature is measured with a thermographic camera, the change in surface temperature at the site where the minimal dross occurs is very small. Therefore, the method described in Patent Document 2 accurately indicates the presence or absence of the minimal dross. It was difficult to detect. On the other hand, when the variation of the Laplacian of the steel surface temperature was reduced by applying the first embodiment of the present invention, it was found that the presence or absence of minimal dross wrinkles can be detected with high accuracy by the following algorithm.

  A method for determining whether or not a defect detected from a Laplacian of the steel surface temperature is a minimal dross will be described with reference to FIG.

  In FIG. 1, reference numeral 1 denotes an area composed of pixels identified as a part having a defect based on the first embodiment of the present invention. 2 is an ellipse having the same second moment as the region 1. 5 and 6 are the focal points of the ellipse. 4 is the distance between the focal points, and 3 is the major axis of the ellipse.

The interfocal distance 4 and the major axis 3 of the ellipse 2 having the same second moment as the region 1 can be obtained as follows. The number of pixels constituting the defective area 1 is n, and (x _i , y _i ) (i = 1, 3,... N) is the coordinates of the pixel. At this time, the secondary moment M of the defect region 1 is defined by the following equation.
x ^~ = (Σ _{i = 1} ⁿ x _i ) / n, y ^~ = (Σ _{i = 1} ⁿ y _i ) / n
_{M 11 = (Σ i = 1} n (x i -x ~) 2) / n, M 22 = (Σ i = 1 n (y i -y ~) 2) / n
M ₁₂ = M ₂₁ = (Σ _{i = 1} ⁿ (x _i -x ^~ ) (y _i -y ^~ )) / n

The major axis of the ellipse where the second moment is equal to this matrix M and the focal distance can be obtained as follows. If eigenvalues of the matrix M are obtained and the values are λ ₁ and λ ₂ (λ ₁ ≧ λ ₂ ), the major axis a and the minor axis b of the ellipse can be obtained by the following equations.
a = (2√2 / π ^1/4 ) λ ₁ ^3/8 λ ₂ ^−1/8 , b = (2√2 / π ^1/4 ) λ ₁ ^-1/8 λ ₂ ^3/8
At this time, the interfocal distance L is obtained by the following equation.
L = 2√ (a ² −b ² )

  The eccentricity of the region 1 is a value obtained by dividing the interfocal distance 4 of the ellipse having the same second moment as the region 1 by the major axis 3. The eccentricity is an index representing the circularity of the region 1 and can take a value between 0 and 1. When the eccentricity is 0, it represents a circle, and when it is 1, it represents a line segment.

Among the parts recognized as defects, when the area is 0.25 mm ² or less, the eccentricity is 0.5 or less, the area is A (mm ² ), and the corrected Laplacian maximum value is L _max , “L _max A region where the relationship of ≧ 1.15 × (12.020√A + 1.4694) ”is determined to be a dross trap. Here, because of the feature that the minimal dross ridge is small and round, the upper limits of the area and the eccentricity were set to 0.25 mm ² and 0.5, respectively. In addition, at the location where Dross was generated, a relationship of “L _max <1.15 × (12.020√A + 1.4694)” was not found, so a relationship of “L _max ≧ 1.15 × (12.020√A + 1.4694)” was found. The thing was determined to be a Dross trap.

(Offline experiment)
In order to increase the reliability of the index by shooting a moving image with a frame rate of 30 Hz, the average value of 9 frames was adopted as the area and eccentricity of the detection area and the maximum value of the Laplacian in the detection area. The test steel after thermal image photography was subjected to a press test, and five dross ridges were confirmed.

  Table 1 shows the area and eccentricity of the detection region corresponding to five dross traps from the binarized image data of the Laplacian processed image.

With respect to the area where the defect area detected from the corrected Laplacian binarized image data is narrowed down under the condition that the area is 0.25 mm ² or less and the eccentricity is 0.5 or less, the Laplacian after correction in the detection area is corrected. FIG. 7 shows the relationship between the maximum value L _max and the square root √A of the detection area. Dross is plotted with a circle, and the others are plotted with a cross. A broken line in the graph indicates a relationship “L _max = 1.15 × (12.020√A + 1.4694)”. A distribution in which the Laplacian maximum value increases as the square root of the detection area increases can be read. In this distribution, Dross 疵 has a relationship of “L _max ≧ 1.15 × (12.020√A + 1.4694)”.

(Online verification)
Thermal image data on the surface of the steel material was collected using an infrared thermography camera while heating the steel material having only the healthy part in increments of 1 ° C. The number of pixels of the infrared thermography camera is 104 × 416, the pixel size is 0.17 mm, the distance between the infrared thermography camera and the steel material is 17 cm, and the infrared thermography camera is arranged perpendicular to the steel material surface. The Laplacian Δ _xy T (x, y) of the steel surface temperature T (x, y) for each heating temperature at a location where the temperature is within ± 2.5 ° C. for every 10 ° C. of the steel surface temperature is expressed as (1 ). Further, FIG. 3 shows a standard deviation of Laplacian Δ _xy T (x, y), which is plotted. The linear regression equation obtained from this is
L _dfr (T) = 0.00108T−0.01201 (6)
Given in.

  Next, as shown in FIG. 8, the steel material 11 was heated by the heater 12 disposed on the lower surface side of the steel material 11 while the steel material 11 was running. A temperature measuring device 13 is arranged on the upper surface side of the steel material 11, and the heating state of the heating device 12 is controlled by the temperature control unit 14 so that the temperature after heating of the steel material 11 is kept constant. An infrared thermography camera 15 for measuring the temperature on the upper surface side of the steel material 11 is disposed downstream of the heated portion of the steel material 11. The imaging data of the infrared thermography camera 15 is input to the input unit 17 of the detection device 16, the image processing unit 18 performs image processing, and the output unit 19 outputs the detection result.

Thermal image data of the surface temperature of the steel material was collected by the infrared thermography camera 15 while the steel material 11 was running. As the infrared thermography camera, the same camera as that used for measuring the steel material having only the healthy part is used. Here, the power of the heating apparatus 12 is adjusted so that the steel temperature is maintained at about 90 ° C. at the imaging position by the infrared thermography camera 15, and thermal image data is acquired by the infrared thermography camera, and is set to T (x, y). . Next, the Laplacian Δ _xy T (x, y) of T (x, y) was calculated by the above equation (1). Further, using the equation (6), the standard deviation L _dfr (T (x, y)) of the Laplacian corresponding to the relevant part temperature T (x, y) is calculated, and the Laplacian of the relevant part is expressed as in the equation (5). to correct.
L _mod (x, y) = Δ _xy T (x, y) / L _dfr (T (x, y)) (5)

First, a portion having a corrected Laplacian L _mod (x, y) of 3.8 or more was recognized as a defect. Next, when the area recognized as a defect is 0.25 mm ² or less, the eccentricity is 0.5 or less, the area is A (mm ² ), and the corrected Laplacian maximum value is L _max. A part where the relationship of “L _max ≧ 1.15 × (12.020√A + 1.4694)” was established was determined as a Dross trap.

FIG. 4 shows an example in which minimal dross detection is performed on hot-dip galvanized steel. FIG. 4A shows an image of L _mod (x, y) obtained by applying Laplacian processing to the thermal image data, and FIG. 4B shows a result obtained by binarizing the detected Laplacian processed image data. It is the image which left the area | region determined with the method of this invention with respect to the area | region as the minimum dross 疵. As shown in FIG. 4B, a minimal dross ridge 11 was detected. In FIG. 4, “390, 400” on the horizontal axis and “410, 420” on the vertical axis mean coordinates on the steel sheet for associating with the positions of the dross ridges confirmed in the press test. The vertical axis value for the image density means a corrected Laplacian. In FIG. 5, the coordinate display is the same, and the Laplacian that has not been corrected for the density of the image is meant.

  When the test piece was pressed, a first star appeared, and the presence of dross could be confirmed at that location. The part where the star appeared was a part determined as a minimal dross ridge by FIG.

FIG. 5 shows an image when defect detection is performed without applying Laplacian correction using the same thermal image data as in FIG. FIG. 5A shows an image of Laplacian Δ _xy T (x, y) without correction, and FIG. 5B shows an image obtained by binarizing Laplacian image data without correction. It turns out that nothing is detected. The threshold for performing binarization processing was set to 0.33. For this value, Laplacian processing was performed on the thermal image data of the defect-free steel material heated to 90 ° C., and the maximum value of the Laplacian distribution was adopted as the threshold value.

1: Area 2: Ellipse having the same second moment as the area 3: Long diameter of ellipse 2 4: Focal length of ellipse 2 5: Focal point of ellipse 2 6: Focal point of ellipse 2 11: Steel material 12: Heating device 13: Measurement Heating device 14: Temperature control unit 15: Infrared thermography camera 16: Detection device 17: Input unit 18: Image processing unit 19: Output unit

Claims (2)

In detecting defects on both the surface and surface of steel,
The step of heating the surface of the same steel material as the steel material to be inspected and having no defects (hereinafter referred to as “defect-free steel material”) to different temperatures in increments of 5 to 10 ° C. and different temperatures in increments of 5 to 10 ° C. Acquiring the defect-free steel surface temperature distribution heated by the infrared thermography camera as defect-free steel surface temperature data; and
Calculating the average value of each defect-free steel surface temperature data;
Calculating Laplacian data for each defect-free steel surface temperature data;
Linking each defect-free steel surface temperature data and the Laplacian data;
Calculating a standard deviation (hereinafter referred to as “Laplacian standard deviation”) for a Laplacian associated with a temperature at which the difference between the average value and the surface temperature is within ± 2.5 ° C. in each defect-free steel surface temperature data; ,
Obtaining a linear regression line of the average value of the obtained surface temperature and Laplacian standard deviation;
Heating the surface of the steel material to be inspected and obtaining the surface temperature distribution of the steel material to be inspected as the surface temperature data of the steel material to be inspected by an infrared thermography camera;
Calculating Laplacian data for the inspected steel surface temperature data;
Linking the steel material surface temperature data to be inspected and the Laplacian;
Calculating a Laplacian standard deviation of a defect-free steel material corresponding to the surface temperature of the steel material to be inspected from the regression line, and performing a correction to divide the Laplacian corresponding to the surface temperature by the Laplacian standard deviation;
A defect detection method for a steel material, wherein a defect site is detected using the Laplacian data after correction. Of the parts recognized as defects, when the area is 0.25 mm ² or less, the eccentricity is 0.5 or less, the area is A (mm ² ), and the corrected Laplacian maximum value is L _max , “L _max 2. The method for detecting defects in steel materials according to claim 1, wherein a portion where the relationship of ≧ 1.15 × (12.020√A + 1.4694) is satisfied is determined as Dross 疵.