JPH06102197A - Surface flaw detection method - Google Patents

Abstract

PURPOSE:To clearly grasp the flaw on the surface of running steel separately from noise. CONSTITUTION:A casting pattern 4 of grid is irradiated from a pattern casting device 2 on the steel plate 1 surface and the surface image is taken with a photographing device 5 and stored in an image memory 11. Interimage correlation amount and intraimage correlation amount are calculated and stored in memories 13 and 14. A flaw judgment device 15 judges the existence and the kind of flaw on the steel plate 1 surface from these correlation amounts.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting surface defects of materials, particularly surface defects of hot rolled steel sheet.

[0002]

2. Description of the Related Art Although many methods for detecting defects on the surface of industrial products have been developed, the method for detecting defects on hot-rolled steel sheets in the hot-rolling process of steel requires high heat and the target heat. It is considered to be difficult to realize because the rolled steel sheet is moving at a high speed and there are many components such as water vapor that cause noise when observing the surface.

Conventionally, as a method for detecting a defect from the result of surface observation, there is a method disclosed in Japanese Patent Application Laid-Open No. 02-108948. This is because by controlling the shutter speed so that the brightness within a predetermined range to be imaged is constant, it is possible to detect continuous rolling material defects without the need for new cooling means for detection. Is the way.

In Japanese Patent Laid-Open No. 02-298842, a smooth image signal is generated from an original image signal captured by a television camera, and a differential image signal is generated from the original image signal and the smooth image signal. A method for detecting a surface defect with high accuracy by eliminating the influence of the treatment applied to the inspection material is disclosed.

[0005]

According to the above method of controlling the shutter speed, the brightness of the entire screen can be controlled, but it is difficult to separate the noise portion and the signal representing the defect. This is due to a lot of noise, uneven brightness (shading), etc. Another drawback is that a reproducible image cannot be obtained and a large error occurs in the measurement result because the moving distance of the test object differs within the shutter opening time.

In the method using the original image signal and the smoothed image signal, it is necessary to change the degree of smoothing depending on the size of the defect, so that the defect is detected in a situation where the size of the defect cannot be known in advance. Difficult to do. The present invention provides means for overcoming this even if there is a lot of noise, being able to deal with uneven brightness, and enabling defect detection even in situations where the size of the defect cannot be known in advance. The purpose is to provide.

[0007]

According to the present invention, when the surface of a steel sheet is irradiated with a light beam having an appropriate two-dimensional pattern, the regularity of the shape of the reflected pattern or the regularity of the reflection pattern in time series is obtained. However, it utilizes a phenomenon that changes depending on the presence or absence of a defect and a phenomenon that changes depending on the shape of the defect.

According to the present invention, a surface of a steel sheet is irradiated with a light beam having a two-dimensional pattern to be imaged, an intra-image correlation amount relating to the obtained surface image is calculated, and / or the obtained surface image is set as a time series. The inter-image correlation amount between the time-series surface images is stored and calculated, and from one or more of the three of the intra-image correlation amount, the inter-image correlation amount, and the time-series correlation amount of the intra-image correlation, A surface defect detection method characterized by determining the presence and type of a flaw.

In this case, the two-dimensional pattern is preferably in a grid pattern, and a two-dimensional TV camera may be used for image pickup. The brightness of the captured image is kept constant by adjusting the feedback sensitivity. A preferred apparatus for carrying out the method of the present invention is (1) means for irradiating the surface of a steel sheet with a light beam having a two-dimensional pattern (2) means for imaging the surface irradiated by the light beam (3) with the imaging means Means for calculating the intra-image correlation amount relating to the obtained surface image (4) Means for storing the surface image obtained by the imaging means as a time series (5) Calculation of the inter-image correlation amount between the time series surface images Means (6) An apparatus comprising a means for determining the presence / absence and type of a flaw from one or more of the three correlation amounts, namely, the intra-image correlation amount, the inter-image correlation amount, and the intra-image correlation time-series correlation amount And it is sufficient.

[0010]

The method according to the present invention will be described with reference to FIG. First, the surface of the hot rolled steel sheet is irradiated with a light beam having an appropriate two-dimensional pattern. At the same time, the surface illuminated by the light beam is imaged. The original brightness pattern of the surface and the illuminated two-dimensional reflection pattern are superimposed on the image captured at this time.

In the superposed image, calculation for obtaining the intra-image correlation amount is performed. This is to obtain the feature amount from the change in the two-dimensional pattern irradiated with the brightness pattern. on the other hand,
The superimposed images are stored as a time series. this is,
This is because it is expected that there is a correlation between images having different photographing times regarding a defect that extends long in the running direction of the steel sheet, so that the time change of the images is observed.

The inter-image correlation amount is calculated for these time-series images. This is because the defect extending in the traveling direction is particularly detected as described above. Further, not only the time-series images but also the time-series correlation amount of the correlation amount in the image is obtained. This is because it is necessary to perform an enormous amount of calculation when the correlation amount between images of the time-series image is directly calculated. Therefore, for some, it is sufficient to calculate the time-series correlation amount of the intra-image correlation amount. This is because it is judged.

Finally, the presence or absence of a defect is determined from the above-mentioned calculation result of the intra-image correlation amount, the calculation result of the inter-image correlation amount, and the time-series correlation amount of the intra-image correlation amount. When it is determined that there is a defect, information regarding the defect is obtained from a plurality of feature quantities.

[0014]

[Embodiment] An embodiment will be shown, which is configured to implement the method of the present invention. FIG. 2 shows a flowchart of the present invention. A description will be given below with reference to this figure. The pattern projection device 2 is held on the upper part of the hot-rolled steel plate 1 and on the projection device stand 3.
A projected pattern 4 by this pattern projection device appears on the hot-rolled steel sheet 1. This projection pattern is captured as an image by the imaging device 5. The image pickup device 5 is supported by an image pickup device mount 6. The image signal captured by the imaging device 5 is stored in the image memory 11 while maintaining the order as a time series. The stored image memory 11 is converted by the image processing device 12 as a correlation pattern in the image. This correlation pattern is stored in the intra-image correlation memory 13. At the same time, the inter-image correlation amount is obtained by the image processing device 12 from the image memories 11 existing in the image memory as a time series for a plurality of times. This result is stored in the inter-image correlation memory 1
Store in 4. From the inter-image correlation memory 14, the defect determination recognition means 15 also determines whether or not there is a defect, and if there is a defect, recognizes information such as its position, size, and shape. The information obtained by the defect judgment recognizing means 15 is displayed on the display device 16, and the operating condition is changed according to the result, or more detailed surface inspection is performed in a post process. At the same time, the defect information is stored in the storage device 17 and used for analysis of products and operations.

Here, it is desirable that the projection pattern is regular for the sake of correlation. Therefore, it is found that it is preferable to form the projection pattern in a grid shape in order to obtain the defect information from the correlation. Further, in order to project the projection pattern and analyze the surface image obtained on the steel plate, it is desirable that a two-dimensional image can be directly obtained at high speed. Therefore, it is preferable to use a two-dimensional television camera as the imaging means.

Further, as a condition at the time of analysis, if the brightness changes drastically at each time, it becomes difficult to detect a defect even if correlation is obtained. Therefore, a method of feeding back the brightness to adjust the sensitivity of the camera can be considered. Currently, there are many solid-state image sensors that can easily change the sensitivity.
Compared with the method of changing the shutter speed, the distance that the object moves within the shutter time becomes constant, and a reproducible image can be obtained.

Here, the projection pattern 4 when the grid pattern is actually projected by the pattern projection device 2 will be described. A projected pattern when the hot-rolled steel sheet 1 has no defect is shown in FIG. Since it is a defect-free pattern, there is no noticeable contrast other than forming the entire grid.
On the other hand, the projected pattern when the hot-rolled steel sheet 1 has a defect is shown in FIG.
Listed. At this time, the defect 21 is superimposed on the projected pattern 4.
Has been observed. In reality, the defect 21 has almost no difference in brightness from the background, and it is difficult to discriminate the defect by a normal image processing method, for example, a binarization method.

Therefore, a defect can be revealed by using a method of correlating two-dimensional images. The two-dimensional image correlation is as follows. When the unit of the image represented by the coordinates (i, j) is called a pixel, the image is represented by f (i, j) as a function of the discrete integer coordinates (i, j). Here, f (i, j) represents the intensity of the pixel (i, j). The possible ranges of i and j are 0 ≦ i <I,
0 ≦ j <J. At this time, the two-dimensional autocorrelation image g
(M, n) is represented by the following equation.

[0019]

[Equation 1]

Such g (m, n) can also be represented as an image. This correlation image is shown in FIG. In addition,
Since g (m, n) = g (n, m), m <n in FIG.
The corresponding part is not shown. It can be seen that, in addition to the high correlation points 22 due to the grid pattern, the high correlation points 23 due to the defects occur. Highly correlated points with a grid pattern 2
Since the position of 2 can be specified from the projection pattern, it is easy to find the high correlation point 23 due to the defect. In the above method,
Although the method of shifting by one pixel for autocorrelation has been described, there is also a method of adjusting the shifting position so that the correlation of the lattice pattern is maximized.

Although the above is the result of performing only on a single image, it is also possible to detect and recognize defects by utilizing the correlation between images that change in time series. This method will be described below. FIG. 6 shows an image with minute defects in which a minute strain is generated in the lattice pattern by the minute defects 24a. This distortion is also very small from the overall background and difficult to detect from a single image. Therefore, we try to detect it from images taken at a later time. 7 and 8 show how the minute defects 24b and 24c are moving, respectively. Therefore, it is possible to recognize the small defects by correlating them.

In FIG. 6, FIG. 7, and FIG. 8, absolute value difference images (not shown) are taken with respect to the projection pattern without defects (FIG. 3) to clarify the defects. At this time, the projection pattern changes subtly at each time due to the influence of surface irregularities, and therefore, if necessary, the calculation of the absolute value can be changed for each neighborhood by using the pattern. FIG. 9 is a diagram in which the respective difference images are superimposed and shown. It can be seen that the minute defects 25a, 25b, 25c which are revealed by the difference are regularly arranged in space. Then, correlation can be obtained between the difference images.

As a calculation means, it is conceivable that the regression calculation is performed by regarding FIG. 9 as a kind of graph. Further, although no particular method is shown, for example, the concept of optical flow known in the world of computer vision can be applied. By calculating this optical flow, information about the moving object (position, size,
The moving speed) can be obtained. The advantage of this method is that if the moving speed derived from the optical flow is significantly different from the moving speed measured on the line, then it is not a real defect but a false defect, so the reliability of the data is Is to increase.

The apparatus of FIG. 2 for carrying out the method was actually installed in a line and data was collected. At the same time, a visual inspection of the workers was performed offline at the same place, and compared with the data obtained by this method. The results are shown in Table 1. As a conventional method, a method based on binarization is adopted. Of the 50 cases confirmed to be defective by the operator, only 10 cases were matched by the conventional method and only 20 cases were smoothed by the smoothing method. On the other hand, in this method, up to 45 cases were matched. On the other hand, there were 30 cases of erroneously recognizing a non-defect surface location as a defect, while there were 30 cases with the binarization method and 20 cases with the smoothing method, but only 2 cases with this method. This shows that this method is excellent.

[0025]

[Table 1] ──────────────────────────────────── This method Binarization Smoothing method Detection position 45 10 20 Detects a position that is not a defect 2 30 20 Cannot detect 5 40 30 ──────────────────────────── ─────────

[0026]

As described above, according to the present invention, the regularity of the image is judged from the superposed image of the original image and the reflection image of the projection pattern, so that the defect signal can be easily separated from the noise. Further, since time series data of the superimposed image is stored and this correlation is taken, there is also an effect that noise and defect can be clearly distinguished.

From the above effects, the industrial contribution is extremely large.

[Brief description of drawings]

1 is a flow chart of the method of the present invention.

FIG. 2 is a block diagram of an apparatus according to an embodiment of the method of the present invention.

FIG. 3 is a projection pattern in a state where there are no surface defects.

FIG. 4 is a projection pattern in a state with surface defects.

FIG. 5 is a correlation image.

FIG. 6 is an image with small defects.

FIG. 7 is an image with small defects.

FIG. 8 is an image with small defects.

FIG. 9 is a superimposed image of difference images.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Hot-rolled steel plate 2 Pattern projection device 3 Projection device mount 4 Projection pattern 5 Imaging device 6 Imaging device mount 11 Image memory 12 Image processing device 13 Intra-image correlation memory 14 Inter-image correlation memory 15 Defect determination recognition means 16 Display device 17 Storage device 21 Defects 22 High correlation points due to lattice pattern 23 High correlation points due to defects 24a, 24b, 24c Small defects 25a, 25b, 25c Small defects due to difference

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Masakazu Yokoo 1 Kawasaki-cho, Chuo-ku, Chiba City Kawasaki Steel Co., Ltd. Technical Research Division (72) Inventor Hiroyuki Uchida 1 Kawasaki-cho, Chuo-ku, Chiba Kawasaki Steel Co., Ltd. Technical Research Headquarters (72) Inventor Susumu Moriya 1 Kawasaki-cho, Chuo-ku, Chiba City Kawasaki Steel Co., Ltd. Technical Research Headquarters

Claims (4)

[Claims] 1. A surface of a steel sheet is irradiated with a light beam having a two-dimensional pattern to be imaged, an intra-image correlation amount relating to the obtained surface image is calculated, and / or the obtained surface image is stored as a time series. Then, the inter-image correlation amount between the time-series surface images is calculated, and among the three of the intra-image correlation amount, the inter-image correlation amount, and the time-series correlation amount of the intra-image correlation, one or more correlation amount A method for detecting surface defects, characterized by determining the presence and type. 2. The surface defect detecting method according to claim 1, wherein the two-dimensional pattern has a lattice shape. 3. The surface defect detecting method according to claim 1, wherein a two-dimensional TV camera is used for image capturing. 4. The surface defect detection method according to claim 1, wherein the brightness of the captured image is kept constant by feedback sensitivity adjustment.