JP2682112B2 - Automatic magnetic particle flaw detector - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial application] The present invention relates to an automatic magnetic particle flaw detector, in particular, various image analysis is performed by binarizing image data of an inspection object,
The present invention relates to a device for determining a defect of an inspection object.

[Prior Art] Conventionally, inspection of defects such as cracks and scratches that occur during machining of a metal workpiece has been performed by a magnetic particle flaw detector. However, since the determination of the defect depends on human eyes, there are problems such as oversight of the defect and lack of objectivity in evaluation.

Therefore, in order to solve such a problem, a fluorescent pattern of magnetic particles adhering to the surface of the inspection object that appears under the illumination of the ultraviolet lamp is photographed using an industrial television camera, and various images are taken based on the image data. Various automatic magnetic particle flaw detectors have been devised for performing various analyses.

In such an automatic magnetic particle flaw detector, in many cases, the obtained image data is binarized, cracks, scratches, etc. are extracted, and the quality is judged. More specifically, for example, the brightness of each pixel obtained by dividing the screen into minute areas is calculated (digitized) from the analog signal output from the TV camera, and the brightness of each pixel is higher or lower than a predetermined value (threshold value). Pixels are divided into two types depending on whether (bright or dark) (binarization). In the binarized image, the portion corresponding to the defect is recognized and its feature is extracted (for example, the area is measured. , Measuring the longest diameter, taking the ratio of long and short diameters, etc.) It is carried out in steps such as judging the quality of the inspection object from the extracted features.

[Problems to be Solved by the Invention] In an automatic magnetic particle flaw detector, a binarized image, which is a basis for determination, depends on the amount of light of an ultraviolet lamp that irradiates the inspection target and the amount and concentration of the magnetic powder liquid attached to the inspection target. Greatly affected. That is, if the threshold value is kept constant, even if the inspection object is the same, if those conditions change, the shape and area of the portion corresponding to the defect in the binarized image change, and the determination result different for each inspection There is a problem that occurs. Among them, it has been devised to make the irradiation amount of the ultraviolet lamp constant (actual application 63).
No. 39342), it is difficult to avoid the effects of fluctuations in the concentration and amount of magnetic powder liquid, and it was desired to take measures at the image processing stage.

Various methods have already been proposed for the binarization performed by the automatic magnetic particle flaw detector, but these conventional methods have problems. Therefore, in order to solve such a problem, the inventor of the present application previously invented a device for performing binarization by setting a threshold value based on the peak brightness in the brightness distribution (Japanese Patent Application No. 63-207728). . The present invention is effective when the image obtained from the television camera is directly binarized, but another binarization method should be considered in the following cases. That is, this is a case where the image obtained from the television camera is subjected to edge enhancement processing in order to clearly show the defect to be detected. By this processing, an image is obtained in which only the boundaries of the parts with different brightness in the screen are extracted, and the brightness distribution of this image data shows many peaks, unlike when no such edge enhancement processing is performed. In some cases, the maximum peak may appear at the point of zero brightness. In such a case, the present invention has been made because a threshold value determination method different from the above invention is required.

[Means for Solving the Problems] The present invention, which has been made to solve the above problems, has a magnetic powder liquid deposited on the surface of a magnetized test object W, as shown in the conceptual configuration of FIG. Then, the image data of the surface of the inspection object obtained by the television camera C is subjected to contour enhancement processing, and then binarized and analyzed to analyze the inspection object W.
The automatic magnetic particle flaw detector for performing flaw detection is characterized by including the following means.

(M1) Brightness distribution creating means for obtaining a distribution curve of the number of pixels for each brightness from the image data subjected to the edge enhancement processing.

(M2) Differentiation means for differentiating the distribution curve of the number of pixels for each luminance.

(M3) Binarizing means for binarizing the contour-enhanced image data with a threshold value determined based on a minimum value of a differential luminance distribution obtained by differentiating the distribution curve.

[Operation] The inspection object W to which the magnetic powder is adhered is photographed by the television camera C, digitized into the luminance data for each pixel, and then the contour emphasis means BD performs the contour emphasis processing. The contour enhancement processing is processing for clarifying boundaries between portions of different brightness such as the background, the inspection object W, and defects in the screen by performing a special calculation on the image data. The image data subjected to the edge enhancement processing is sent to the brightness distribution creating means M1, and the brightness signal for each pixel is sorted by brightness to create a frequency (number of pixels) distribution for each brightness. In the differentiating means M2, the data of the distribution of the number of pixels for each brightness is differentiated by the brightness (or the difference between the adjacent frequency data in the case of a digital signal) to obtain the differential brightness distribution. The binarizing means M3 determines a threshold value (luminance) based on the minimum value of the differential luminance distribution, and binarizes the image data from the contour emphasizing means BD by the threshold value.
Here, as a method of determining the threshold value based on the minimum value of the differential brightness distribution, the value of the brightness corresponding to the second minimum point from the lowest brightness is used as the threshold value.

The above means may be directly connected, or a means for temporarily storing intermediate data may be provided between the means and devices as appropriate. For example, the edge-enhanced image data may be directly sent to the brightness distribution creating means M1, but the image data that has been photographed and edge-enhanced is temporarily recorded on a magnetic tape or the like, and the brightness is used at another occasion. It may be imported into the distribution creating means M1.

[Embodiment] An embodiment of the present invention will be described with reference to FIGS. As shown in FIG. 2, the automatic magnetic particle flaw detector according to the present embodiment has an ultraviolet lamp 4 for irradiating the inspection object 2 to which the magnetic particle liquid is attached,
An image processing device (IPU) 10 that inputs an image signal from the television camera 6 that captures an image of the inspection object 2 and the television camera 6 and performs various data processing described below, and displays images before and after data processing that is output from the IPU 10. It consists of a video monitor 20.

IPU10 is a computer, CPU11, ROM12, RAM13,
A video RAM 14, an A / D converter 15, and an external input / output circuit 16 are provided. The ROM 12 stores in advance a program for a process described below, predetermined constants, and the like, and the RAM 13 temporarily stores data in the middle of calculation in the following process. video
The RAM 14 stores the digitized image data. In addition, the IPU 10 may be provided with an external storage device for recording image data and the like.

The inspection object 2, which is considered to have a defect 1 such as a scratch or a crack, is magnetized, immersed in a fluorescent liquid containing magnetic particles, and then pulled up. When there is a defect 1 on the surface or in the vicinity of the surface, the magnetic flux leaks from this, so that the magnetic powder liquid adheres in a larger amount than other portions. Such a portion appears brighter than other portions due to the irradiation of the ultraviolet lamp 4.

The TV camera 6 captures such an image and sends an image signal to the IPU 10. The IPU 10 performs predetermined data processing on this signal to determine the presence or absence and the degree of the defect of the inspection object 2. The processing performed in the IPU 10 will be described below according to the flow chart of FIG.

When this routine starts, first in step 100, an image signal from the television camera 6 is input. TV camera 6
Since the image signal output from is an analog signal,
The D converter 15 quantizes the signal in a predetermined cycle and divides it into a predetermined number of pixels. This provides the digitized brightness of each pixel. The luminance data of each pixel is temporarily stored in the video RAM 14.

For the digital image signal obtained in this way,
In step 110, to remove noise (signal noise),
Performs smoothing processing. This can be performed, for example, by data processing of adding the luminance value of a certain pixel to the luminance value of eight pixels in the diagonally upper, lower, left, right, and peripheral directions and dividing it by 9 to obtain the value of that pixel. As a result, signal noise that appears sporadically on a pixel-by-pixel basis is almost eliminated.

Next, in step 120, contour enhancement processing is performed to enhance the defective portion. This can be done, for example, in the following manner. Of the eight surrounding pixels of a pixel, the value of the three pixels in the left column is -1 (or -1, -2,-).
1 or the like), and the values of the 3 pixels in the right one column are weighted by +1 (or +1, +2, +1 etc.) and added. Similarly, 3 pixels in the upper 1st row are weighted with minus, and 3 pixels in the lower 1st row are weighted with plus. The sum of these values is the value of that pixel. This value is zero when the pixel for which this calculation is performed is in a region where there is almost no change in brightness, but when that pixel is at the boundary, it appears as a positive or negative value.

With respect to the image data whose contour is emphasized in this way,
In step 130, the number of pixels is counted for each luminance to obtain the luminance distribution. An example of this brightness distribution is shown in FIGS. 4 (A), (B),
It is shown in (C). FIG. 4 (A) shows the inspection object 2 with no defect.
The brightness distribution curves of the image of (2) and (B) and (C) are examples of the curve of the inspection object 2 having a defect. (B) shows the case where the concentration of the magnetic powder liquid is high, and (C) shows the case of (C). A curve when the concentration of the magnetic powder liquid is low is shown. It is characteristic that some peaks and valleys appear in any of the curves after the edge enhancement processing.

These brightness distribution curves can be decomposed as shown in FIG. That is, the entire luminance distribution is i) the curve 50 due to the background, not the inspection object 2, ii)
Curve 52 due to the entire inspection object 2, and iii) Defect 1 on the inspection object 2 or a curve due to a partial adhesion of magnetic powder liquid, although not a defect (magnetic particle adhesion noise)
It is a superposition of three curves of 54 and 54. Therefore, when binarizing this image data, the second and third curves
It is appropriate to set the boundary (valley) 56 between 52 and 54 as the threshold value. When the background occupies a large part of the field of view of the television camera 6, as shown in FIG.
The peak of distribution 50 having the lowest brightness is the maximum peak. Therefore, determining the threshold value based on the brightness value of the maximum peak is not appropriate in the present case.

However, it is practically difficult to detect this valley 56. For example, the third curve 54 is 54a (defective / high magnetic powder liquid concentration), 54b depending on the presence or absence of defect 1 or the magnetic powder liquid concentration.
(Defects, low magnetic powder concentration), 54c (no defects). The second curve 52 also shifts to the left and right depending on the magnitude of the magnetic powder liquid concentration. On the other hand, the first curve 50 does not change under these conditions. Therefore, those curves
The valleys 56 and 58 between the points 50, 52, and 54 may or may not appear clearly depending on the conditions. In FIGS. 4A and 4C, the first valleys 60 and 62 do not appear.
Therefore, it is difficult to detect only the second valleys 65, 66, 67 as they are from the brightness distribution curve obtained from the edge-enhanced image.

Therefore, in order to accurately identify only the defect 1, in the present invention, the threshold value is determined by the following processing. First, step 14
At 0, the brightness distribution curve (FIGS. 4 (A), (B),
(C)) is differentiated and smoothed by the following equation.

y (i) = x (i) -x (i + 1) z (i) = {y (i-1) + y (i) + y (i + 1)}
/ 3 where x (i) is the value of each point of the original luminance distribution curve,
z (i) is the value of each point of the curve that has been subjected to the differential / smoothing process. The results of performing this calculation on the curves of FIGS. 4 (A), (B), and (C) are shown in FIGS. 6 (A), (B),
It is shown in (C).

By performing this differentiation / smoothing process, the first or second valleys 60, 61, 62, 65, 6 can be obtained with the original brightness distribution curve.
Even if 6 and 67 do not appear, the brightness values corresponding to those positions can be detected, and only defect 1 can be accurately represented.
It becomes possible to perform valuation. The reason is as follows. When two curves having independent smooth distributions as described above are superposed, even if no valley appears between them, an inflection point instead appears between both curves. Since this inflection point appears as the maximum or minimum point of the curve obtained by differentiating the original curve, the inflection point of the original curve can be known by taking the maximum or minimum point on the differential curve. Therefore, if the luminances corresponding to the second valleys 65, 66, 67 of the distribution curves of FIGS. 4 (A), (B), and (C) are to be obtained, FIGS. 6 (A) and 6 (B) are obtained. , (C), the second minimum point 75, 76, 77 from the lowest brightness of the differential curve should be taken. In step 150, the brightness of this second minimum point 75, 76, 77 is detected, and it is used as a threshold for binarization.
Th.

In the next step 160, the brightness data of each pixel stored in the video RAM 14 in step 100 is binarized by comparing it with the threshold value Th. For example, the brightness B of a pixel is B
If it is <Th, the value 0 is given to the pixel, and if B ≧ Th, the value 1 is given to the pixel. In this way 2
The binarized image data is also stored in another area of the video RAM 14.

At step 170, feature extraction is performed from the binarized image of the inspection object 2. For feature extraction, an optimal method is selected according to the purpose of inspection.

For example, from the binarized data of the edge-enhanced image, each defect 1
Number the islands corresponding to (labeling). Then, calculate the area of each labeled island. This is determined by the number of pixels. The maximum area of these is extracted as a feature. Here, in addition to the area, the maximum length, the length-to-short ratio, and the like may be features.

In step 180, the quality of the inspection object 2 is determined according to the characteristics of the image thus extracted. For example, the inspection object 2 whose maximum area is equal to or larger than a predetermined value is made defective. This is the end of this routine. In addition, video RAM14
The original image data and the binarized image data stored in are properly displayed on the video monitor 20.

As described above, in the present embodiment, when binarizing the captured image data, the threshold value Th is determined based on the luminance value of the second minimum point of the differential curve of the luminance distribution. This point corresponds to the boundary point 56 between the curves 52 and 54 in FIG. 5, and by extracting only pixels having a higher brightness than that, only the defect 1 can be accurately extracted and binarized. Therefore, for example, even if the amount and concentration of the magnetic powder liquid adhering to the inspection object 2 increases and the overall brightness increases, the threshold value increases accordingly, so that the portion of the binarized image corresponding to the defect is increased. The shape and area hardly change, and the test result (judgment) is hardly affected.

[Effect of the Invention] In the present invention, the luminance distribution curve of the image data whose contour has been emphasized is differentiated, and the threshold value for binarization is determined based on the minimum value of the derivative curve. Therefore, it becomes possible to clearly determine the brightness threshold value of the boundary between the defect and the surface of the inspection object that is difficult to distinguish by the edge-enhanced brightness distribution curve, and an accurate binary image in which only the defect is extracted can be obtained. . For this reason, it is possible to obtain an image in which the contour of the defect clearly appears without being influenced by the concentration of the magnetic powder liquid or the amount of light from the light source, etc., so that automatic inspection of the inspection object can always be performed accurately. Become.

[Brief description of the drawings]

FIG. 1 is a conceptual block diagram for explaining the present invention, FIG. 2 is a block diagram showing the configuration of an automatic magnetic particle flaw detector which is an embodiment of the present invention, and FIG. 3 is a diagram showing processing performed by the flaw detector. Flow charts, FIGS. 4 (A), (B), and (C) are graphs showing the luminance distribution of the edge-enhanced image, FIG. 5 is an explanatory diagram in which the luminance distribution curve is decomposed, and FIGS. 6 (A) and (B). , (C) are graphs showing curves obtained by differentiating and smoothing the curves of FIGS. 4 (A), (B), and (C), respectively. 1 ... defect, 2 ... inspection object, 4 ... light source, 6 ... television camera, 10 ... image processing device

Claims (1)

(57) [Claims] 1. A magnetic powder liquid is attached to a magnetized surface of an inspection object, and image data of the surface of the inspection object obtained by a television camera is subjected to edge enhancement processing and then binarized and analyzed. In an automatic magnetic particle flaw detector that performs flaw detection on an inspection object, accuracy distribution creating means for obtaining a distribution curve of the number of pixels for each luminance from the image data subjected to edge enhancement processing, and differentiation for differentiating the distribution curve of the number of pixels for each luminance Means, and binarizing means for binarizing the contour-enhanced image data with a threshold value determined based on a minimum value of a differential luminance distribution obtained by differentiating the distribution curve. Automatic magnetic particle flaw detector.