JP2011013007A - Magnetic particle flaw inspection apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To determine a binarization threshold, coping with a brightness fluctuation in various magnetic particle flaw inspection images, reduce misdetections, and attain a high defect detection rate.SOLUTION: In a magnetic particle flaw inspection apparatus 1, a magnetic particle pattern is generated on the surface of a carried to-be-inspected material 2; an image of the magnetic particle pattern is captured by an imaging means 7; a surface defect K of the to-be-inspected material 2 is detected from the captured image; the binarization threshold Th is the sum of an average Vave of the luminance in the image and a multiple W of the standard deviation Vσ of the luminance; and the image is binarized, and the surface defect K is detected by using the threshold Th.

Description

  The present invention relates to a magnetic particle flaw detector that detects surface defects.

2. Description of the Related Art Conventionally, as a nondestructive inspection means for inspecting a surface defect of an inspection object that is a magnetic material, a surface flaw detection method that spreads fluorescent magnetic powder has been widely used. In this method, first, the inspection object is magnetized by various methods, a magnetic powder liquid in which magnetic powder having fluorescent properties is dispersed is sprayed, and the magnetic powder is attracted by the leakage magnetic flux leaking from the surface defect. After that, the magnetic particles aggregated around the surface defects are irradiated with an ultraviolet light source to emit fluorescent light, and the surface defects are detected by observing the magnetic powder indicating pattern.
This method is very sensitive and effective as a method for flaw detection on the surface of an inspection object because a clear magnetic powder indication pattern can be observed when the leakage magnetic flux increases even for a minute defect.

The magnetic particle indication pattern that appears by the magnetic particle flaw detection method using fluorescent magnetic particles is often visually confirmed for inspection objects with complex shapes, but the magnetic particle indication pattern is imaged with a camera for inspection objects with simple shapes. However, there are cases where a magnetic particle flaw detector that automatically extracts a magnetic particle indication pattern is used, which is widely used by material manufacturers in the steel industry and the like.
In such a magnetic particle flaw detector, it is important to maintain highly accurate and stable flaw detection performance, and a method for that purpose has been proposed. A method of effectively magnetizing the inspection object, a method of spraying the magnetic powder liquid uniformly and uniformly, a method of irradiating the ultraviolet light source uniformly, a method of imaging the magnetic particle flaw detection surface of the inspection object with high sensitivity and high resolution, An image processing method for extracting a defective portion from the captured image is required.

The image processing of the magnetic particle flaw detection image is basically performed by extracting a bright part in the captured image as a part that is aggregated into a defect and fluorescently emitted. However, in addition to defects, magnetic particles are simply accumulated as a mass, or when the surface roughness of the object to be inspected is rough, a continuous magnetic particle indication pattern is generated. Extraction increases false detection.
Therefore, paying attention to the shape of the magnetic particle indicating pattern, a shape recognizing means has been proposed and put into practical use, in which a bright portion solidified in a spherical shape is excluded as a lump of magnetic powder, and a long linear portion is extracted as a defect. ing.

For example, in a magnetic particle flaw detection method for a prismatic steel material, two types of product-sum filters are applied to the image so that two types of defects extending in the longitudinal direction of the prism and defects extending in the width direction can be detected at the same time. A method is disclosed in which a part is extracted (binarized), then a shape is extracted (labeled), and a defective part is discriminated by an evaluation value (see Patent Document 1).
What is important here is the threshold for binarization of luminance when judging the shape. If an appropriate threshold for binarization is not selected, many defects are missed and false detection occurs. . A method of selecting an appropriate fixed value after various trials and errors (fixed threshold method) has been used as a reasonable method for setting a threshold value for binarization. As the calculation speed increases, a method of dynamically calculating and determining an appropriate threshold value for each captured image has begun to be applied.

In addition, a method is proposed in which an average of the luminance of the entire image is obtained and a value that is two to three times the average value is used as a threshold for binarization (see Patent Document 2).
Further, a histogram of the luminance of each pixel of the image is calculated, and the minimum luminance value at which the luminance histogram frequency becomes 0 is assumed as the threshold value for binarization assuming that the background portion and the defect-corresponding portion are clearly separated on the luminance histogram. A technique to be adopted is presented (see Patent Document 3).

JP 2000-111527 A Japanese Patent No. 3225660 Japanese Patent No. 2564737

As described above, various methods have been proposed for determining a threshold value for binarization. However, when it is applied to an actual image, it is still a problem that false detection or defect oversight often occurs. .
The automatic threshold value determination means according to the conventional method described in Patent Document 1 is a factor that affects the luminance distribution of the magnetic particle flaw detection image (illuminance fluctuation of the ultraviolet light source, deterioration of the magnetic powder liquid, luminance fluctuation due to a change in the amount of application, ultraviolet light The influence of the intensity distribution on the surface of the inspection target of the light source, the magnetization intensity distribution in the inspection target, the surface roughness distribution of the inspection target, etc.) is large, and it is difficult to operate properly in the actual site.

In the method of determining the threshold value from the average luminance of the entire image disclosed in Patent Document 2, when the luminance distribution changes as a whole, such as illuminance fluctuations of the ultraviolet light source, magnetic powder liquid deterioration, luminance fluctuations due to dispersion amount fluctuations, etc. However, it is not effective for other fluctuation factors because of variations in the inspection object and changes in the distribution of the luminance histogram.
In the method of setting the minimum luminance value at which the luminance histogram frequency of 0 in Patent Document 3 is a threshold value for binarization, the luminance distribution of the defect and the luminance distribution of the background are rarely separated, and there are not a few In general, an overlap occurs, and it is often impossible to obtain the minimum luminance value at which the luminance histogram frequency is zero.

  In view of the above-described problems, the present invention determines a binarization threshold value that can cope with luminance fluctuations of various magnetic particle flaw detection images, thereby reducing false detection and realizing a high defect detection rate. A flaw detection apparatus is provided.

In order to achieve the above object, the present invention employs the following technical means.
In the magnetic particle flaw detector according to the present invention, the magnetic particle pattern generated on the surface of the object to be inspected is imaged by an imaging unit, and the surface defect of the object to be inspected is detected from the imaged image. The sum of the average luminance in the image and a constant multiple of the standard deviation of luminance is used as a binarization threshold, and the image is binarized using this threshold to detect surface defects. To do.
The inventors of the present application have considered the case where the variation in luminance at the normal portion (background) is large due to rough skin or the like when flaw detection is performed on the surface of the material to be inspected.

As shown in FIG. 5B, in the portion where the skirt of the luminance distribution of the normal portion overlaps with the luminance of the pixel in the surface defect K (the region indicated by L in FIG. 5B), it is normal. The luminance of some pixels in the portion is substantially the same as the luminance of some pixels in the surface defect K. Therefore, strictly speaking, it is difficult to discriminate a defect from a normal part only by luminance, and it is necessary to determine an appropriate threshold value for reducing the number of false detections and binarizing the surface defect K reliably.
However, when Th-A (twice the average brightness Vave) in FIG. 5B is used as the threshold for binarization, the brightness variation of the normal part (background) is not considered at all. As a result, false detection occurs considerably.

Further, if the minimum luminance value at which the luminance histogram frequency is 0 (the luminance value indicated by the point C at which the number of pixels is 0 in FIG. 5B) is used as the threshold for binarization, the maximum due to the surface defect K is obtained. Brightness higher than the brightness is set as a threshold for binarization, and the surface defect K is missed (see Th-C in FIG. 5B).
Therefore, the surface defect K is only a small part in the material to be inspected, and not only the average luminance Vave in the captured image but also the normal part so as to correspond to the distribution of the luminance distribution in the normal part occupying most of the surface. By using Th-B that also incorporates the standard deviation Vσ that can take into account variations in brightness, even if the material to be inspected has rough skin, there are few false detections and a more appropriate threshold value can be set.

As can be seen from this, a portion having a luminance higher than the sum of the average luminance Vave and the constant multiple W of the standard deviation Vσ of the captured image is binarized and extracted as a defect, thereby inspecting the material to be inspected. Even if the brightness of the normal part in the magnetic powder pattern image is large due to rough skin, etc., the defect part in the magnetic powder pattern image is only a part of the whole in the magnetic particle inspection of the inspection material Therefore, surface defects can be detected without leakage regardless of the material to be inspected and flaw detection conditions, and false detection can be reduced.
The image is divided into a plurality of block images, and the sum of the average luminance and a constant multiple of the standard deviation of luminance in each block image is set as a threshold value for binarization, and each block is used by using these threshold values. Preferably, the image is binarized to detect surface defects.

As a result, by dividing the image into block images, even when there is a variation in brightness in a part of the image, such as when partial skin roughness occurs on the surface of the material to be inspected, it is optimal. The binarization threshold value can be determined, and it is possible to absorb fluctuations in brightness due to the portion of the material to be inspected caused by uneven brightness of the light source.
Further, a block image is set around each pixel in the image, and the sum of the average luminance and a constant multiple of the standard deviation of luminance in each set block image is set as a threshold for binarization, and these thresholds are set. It is preferable to binarize each pixel using a value to detect a surface defect.

Thereby, the change of the binarization threshold value between each pixel and its periphery can be smoothed to determine a more suitable binarization threshold value.
Furthermore, it is preferable that the constant multiple is 3 or more and 5 or less.
As a constant multiple of the standard deviation of luminance, it is determined from experience that 3 to 5 is appropriate from various experiments.

  According to the present invention, it is possible to determine a binarization threshold value that can cope with luminance fluctuations of various magnetic particle flaw detection images, reduce false detections, and realize a high defect detection rate.

It is a block diagram of the magnetic particle inspection apparatus which concerns on this invention. It is a flowchart figure of the processing algorithm of an image. It is a flowchart figure of the processing algorithm which calculates the threshold value of binarization. It is a schematic diagram which shows the state which divided | segmented the image into the block image in 1st Embodiment. It is a luminance distribution diagram showing when the variation in luminance of pixels is small and large. It is a schematic diagram which shows the state which set the block image centering on each pixel of the image in 2nd Embodiment. It is a figure which shows a mode that the original image 1 containing a surface defect is binarized. It is a figure which shows a mode that the original image 2 containing a surface defect is binarized. It is a figure which shows a mode that the original image 3 containing a surface defect is binarized.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 shows a magnetic particle flaw detector 1 according to the present invention. This magnetic particle inspection apparatus 1 is an apparatus that continuously and automatically detects a surface defect K of a material to be inspected 2 such as a prismatic steel material (steel piece).
The magnetic particle flaw detector 1 includes a conveying unit 3 that conveys the material to be inspected 2, a magnetic particle dispersing unit 4 that spreads a magnetic powder liquid on the surface of the material to be inspected 2 that is conveyed by the conveying unit 3, and The magnetizing means 5 for magnetizing the material to be inspected 2 on the downstream side, the illumination means 6 for illuminating the surface of the material to be inspected 2 on the downstream side of the magnetizing means 5, and the magnetic powder pattern fluoresced by the illumination means 6 are imaged. The image pickup means 7 for detecting the surface defect K by processing the image picked up by the image pickup means 7 is provided.

The conveyance means 3 is continuously conveyed at a speed of 30 to 40 m per minute by a pair of upper and lower and left and right support rollers that support the material 2 to be inspected.
The magnetic powder spraying means 4 sprays an inspection liquid containing magnetic powder (iron powder with a phosphor attached) onto the surface of the inspection object 2 to be inspected for surface defects K. A spray nozzle 4 a facing the surface of the inspection material 2 is provided.
The magnetizing means 5 uses a quadrupole electromagnet in this embodiment, and is composed of two sets of electromagnets in which a coil 5b is wound around a U-shaped magnetic body 5a. The surface of the material to be inspected 2 positioned between the two poles by applying an alternating current to the coil 5b between the pair of tip portions (poles) of each U-shaped magnetic body 5a close to the material to be inspected 2 It is intended to apply a magnetic field.

When the inspection object 2 to which the magnetic field is applied by the magnetizing means 5 has a surface defect K, a magnetic field leaked in the vicinity of the surface defect K is present, so that a stronger magnetic field is generated than a portion without the surface defect K. Thus, the magnetic powder containing the phosphor aggregates in the vicinity of the surface defect K at a high density, and a pattern of magnetic powder appears so as to correspond to the surface defect K.
The illuminating means 6 is an ultraviolet lamp or the like that irradiates the magnetic powder pattern generated by the magnetic flux leaked near the surface defect K with ultraviolet rays. In addition, the illumination unit 6 uses a light source with less luminance unevenness so that setting of the threshold Th is not difficult when binarizing the captured image.

The imaging means 7 employs a line sensor camera 7, and the line sensor camera 7 images fluorescence emitted from magnetic powder containing phosphors attached near the surface defect K of the material 2 to be inspected, and the imaging. The processed image is supplied to the defect detection unit 8.
[First Embodiment]
A magnetic particle flaw detector 1 according to a first embodiment of the present invention will be described.
FIG. 2 shows a processing algorithm in the defect detection unit 8 of the magnetic particle inspection apparatus 1.
When detecting the surface defect K existing on the surface of the inspection object 2, first, after acquiring the original image (step S1), as a pre-processing, removal of magnetic particles and defect enhancement processing by various image processing filters are performed. Perform (step S2).

Thereafter, a threshold value Th for binarization is calculated (step S3), and binarization processing is performed based on the threshold value Th (step S4).
The connected parts on the binarized image are extracted, and a labeling process for assigning individual labels is performed (step S5), and feature quantities such as area, maximum luminance, and aspect ratio are calculated for each label. An evaluation value is calculated from the feature amount (step S6). Then, a preset evaluation value threshold value is input (step S7), and it is finally determined whether or not there is a defect by comparison with the evaluation value threshold value (step S8). Those determined to be defects are added to the defect list (step S9), and this is repeated for all labels.

Specifically, in step S1, a one-dimensional line image captured by the line sensor camera 7 is sent to a defect detection unit 8 configured by the image processing apparatus, and is time-sequentially stored in a frame memory in the defect detection unit 8. The two-dimensional image is stored.
In step S2, if there is unevenness in the intensity of ultraviolet rays irradiated from the illumination means 6 or unevenness (magnetic powder accumulation) in the magnetic powder dispersed on the material 2 to be inspected, the brightness changes even in areas other than the surface defect K. The image processing filter performs a pre-process that makes the background luminance constant and enhances the defect portion luminance.

  In step S3, the threshold value Th for binarization is obtained by substituting the average Vave and standard deviation Vσ of the luminance in the pixels in the captured image into the equation (1).

In the present embodiment, the captured original image is cut into block images B (see FIG. 4), and a binarization threshold Th is calculated for each block image B.
In step S4, the original image is binarized by assigning 255 to pixels having a luminance higher than the threshold Th for each block image B and 0 to pixels having a low luminance. Therefore, the surface defect K candidate and the other portion are separated in the binarized image obtained by binarization.
In step S5, among the pixels determined to correspond to the surface defect K candidate in step S4, the mutually connected pixels are grouped into one group, and a different number is assigned to each pixel in each surface defect K candidate group. (Labeling process).

In step S6, an evaluation value is calculated for each labeled pixel group (region).
This evaluation value is calculated according to the formula (2), for example. That is, the evaluation value is set to be large when the luminance is bright even if the area of the pixel region to which the same number is assigned is small.

In step S7 and step S8, whether or not the pixel group is a surface defect K is determined based on whether or not the evaluation value of the labeled pixel group is larger than a predetermined evaluation value threshold value. .
In step S9, the pixel group determined to be the surface defect K is added to the defect list together with information such as the position, size, direction, and the like.
Steps S6 to S9 are repeated for all pixel groups determined to be surface defect K candidates in the binarized image.

The problem here is how to determine the threshold value Th for binarization in step S3. Various methods such as luminance fluctuations due to changes in the amount of application, magnetization intensity distribution and surface roughness distribution in the material 2 to be inspected, etc. In order to deal with the variation factors, the determination is made for each block image B divided in the width direction and the conveyance direction (longitudinal direction) of the material 2 to be inspected in the procedure as shown in FIG.
That is, in step S3, the block image B is cut out (step S31), and the average luminance Vave and the standard deviation Vσ of luminance are calculated for each block image B (step S32, step S33), and the above-described formula ( In 1), a binarization threshold Th is calculated (step S34).

In step S31, as shown in FIG. 4, 100 pixels (100 pixels) in the width direction of the inspection object 2 and 100 pixels (100 pixels) in the longitudinal direction are taken as one block image B by the line sensor camera 7. Cut out the image.
In step S32 and step S33, the average brightness Vave and the standard deviation Vσ are calculated using the data of the entire 100 × 1000 pixels in the cut-out block image B.
In step S34, the binarization threshold Th is determined for each block image B by using the linear sum (1).

In addition, the luminance distribution of the original image may have various distributions depending on the state of the inspection object 2 and the magnetic particle flaw detector 1.
For example, in the case schematically shown in FIG. 5A, the variation in the luminance distribution of the normal portion (background) is small and clearly separated from the luminance of the surface defect K. In this case, as a method of determining the binarization threshold Th, the minimum luminance value at which the histogram frequency of luminance is several times the conventional average Vave of luminance or the binarization threshold Th is set as the binarization threshold Th. However, it is possible to appropriately extract the surface defect K.

However, for example, when the variation in luminance in the normal portion (background) as shown in FIG. 5B is large, the bottom of the luminance distribution in the normal portion overlaps the luminance of the pixels in the surface defect K. (In the region indicated by L in FIG. 5B, the luminance of some pixels in the normal portion is substantially the same as the luminance of some pixels in the surface defect K). Therefore, strictly speaking, it is difficult to discriminate between the defect and the normal part only by the luminance, and it is necessary to determine an appropriate threshold value Th for reducing the number of false detections and reliably binarizing the surface defect K. .
In FIG. 5B, the threshold value Th-A is set to twice the average brightness Vave, so that erroneous detection occurs considerably. On the other hand, the threshold value Th-B is a threshold value calculated using the equation (1). In this case, there are few false detections, and the threshold value is set more appropriately. In FIG. 5B, the constant W is set to 3 that is empirically valid.

Further, in the case of this example, the minimum luminance value at which the luminance histogram frequency is 0 (the luminance value indicated by the point C at which the number of pixels is 0 in FIG. 5B) is set as the binarization threshold Th. Then, a higher luminance than the maximum luminance due to the surface defect K is set as a threshold for binarization, and the surface defect K is missed (threshold Th in FIG. 5B). -See C).
As described above, in the magnetic particle flaw detector 1 that determines the binarization threshold Th by the expression (1) according to the present invention, the brightness of the flaw detection image is determined according to the flaw detection conditions such as rough skin of the material 2 to be inspected and uneven brightness of the light source. Even when the degree of variation of the above changes, it is possible to always determine an appropriate threshold value Th for binarization, reduce erroneous detection, and detect surface defects without omission.
[Second Embodiment]
A magnetic particle flaw detector 1 according to a second embodiment of the present invention will be described.

As shown in FIG. 6, the greatest difference in the second embodiment is that, for each pixel in the captured image stored in the frame memory, 100 pixels in the width direction of the material 2 to be inspected with the pixel as the center, the longitudinal direction A block image B of 100 pixels is set. In each of the set block images B, a binarization threshold Th is calculated based on the equation (1), and the pixel located at the center of each block image B is binarized using the threshold Th. It is a point.
That is, for all the pixels in the two-dimensional captured image stored in time series in the frame memory in the defect detection unit 8, the peripheral pixels of each pixel (block image B made up of 100 × 100 pixels) are considered, and the block image B The luminance average Vave and the standard deviation Vσ are substituted into the equation (1) to determine the binarization threshold Th of the pixel.

For example, the pixel G1 in FIG. 6 obtains the threshold value Th from the average brightness Vave and the standard deviation Vσ of the block image B1 composed of the peripheral pixels. Further, a pixel G2 shifted from the pixel G1 by several pixels in the width direction of the inspection object 2 obtains a threshold value Th based on the average Vave and the standard deviation Vσ of the surrounding block image B2, and further shifts in the width direction. G3 obtains the threshold value Th from the average Vave and the standard deviation Vσ of the block image B3.
Similarly, for the pixels G4 and G5 shifted in the longitudinal direction of the material 2 to be inspected, the threshold value Th is obtained from the block images B4 and B5 of the peripheral pixels.

  In this way, when the method of obtaining the threshold value Th for binarization that introduces the concept of moving average is adopted, the luminance change between each pixel and its surrounding pixels is smoothed, and a more suitable threshold for binarization is achieved. The value Th can be determined, which is advantageous for extracting an elongated surface defect K that extends over a plurality of block images B.

Hereinafter, a case where an original image including the surface defect K is binarized using the threshold value Th determination method according to the present invention will be described. Note that the constant W is 5 over the original image 1 to the original image 3.
When binarizing the original image 1 including the surface defect K extending over a plurality of block images B, the average brightness Vave and standard deviation Vσ of the pixels in the original image 1 are obtained, and Vave = 40.9. Vσ = 5.7. By substituting these into equation (1), the threshold Th for the original image 1 is calculated as 69.

As shown in FIG. 7, when binarization is performed based on the threshold Th, a binary image 1 is obtained.
If the image has a high S / N ratio like the original image 1, binarization can be performed with a fixed threshold value Th as in the prior art.
However, when the inspection object 2 has a portion such as the original image 2 in which the luminance of each pixel is generally low as shown in FIG. 8, a fixed threshold Th (for example, an original image to be described later). When the threshold 91) of the image 3 is used, there is a possibility that the existing surface defect K may be missed (see the binarized image 22).

Therefore, an average Vave and a standard deviation Vσ of the luminance of the pixels in the block image B including the surface defect K in the original image 2 are calculated (Vave = 35.5, Vσ = 5.6), and these are expressed by the equation (1). ), The threshold value Th suitable for the original image 2 can be derived as 63.
If the original image 2 is binarized using this threshold Th, even a surface defect K in a portion where the overall luminance is low can be detected without leakage (binarized image 21). reference).
Further, as shown in FIG. 9, when the surface defect K is detected in a portion where the surface (background) of the inspection object 2 is rough, a fixed threshold Th (for example, the above-described original image) is detected. When the threshold value 63) of 2 is used, even if there is no actual surface defect K, pixels exceeding the threshold value Th are generated, and as shown in the binarized image 32, Many false positives occur.

Therefore, in the original image 3 as well, the average luminance Vave and standard deviation Vσ (Vave = 40.6, Vσ = 10.1) of the pixels in the block image B including the surface defect K are substituted into the equation (1), The threshold value Th suitable for the image 3 can be derived as 91.
When the original image 3 is binarized using this threshold Th, even if there is a surface defect K at a site where rough skin occurs, only the actual surface defect K can be detected as in the binarized image 31. It becomes possible to prevent false detection.
With the above configuration, it is possible to determine the binarization threshold value Th that can cope with the luminance variation in the captured image of the magnetic powder pattern, and as a result, it is possible to reduce false detection and improve the detection rate of the surface defect K. .

In addition, the magnetic particle inspection apparatus concerning this invention is not limited to embodiment mentioned above.
Even if the weight constant W of the standard deviation Vσ of luminance is not 3 or 5, it is reasonable to set it to 3 or more and 5 or less empirically.
The imaging means 7 may be an area sensor camera instead of a line sensor camera.

  The present invention can be used as a magnetic particle scratching device for detecting surface defects.

DESCRIPTION OF SYMBOLS 1 Magnetic particle flaw detector 2 Inspected material 3 Conveyance means 4 Magnetic powder dispersion | distribution means 5 Magnetization means 6 Illumination means 7 Imaging means 8 Defect detection part K Surface defect Vave Average of brightness Vσ Standard deviation of brightness W Standard deviation weight constant Th Binary Threshold B block image

Claims (4)

In the magnetic particle flaw detector for imaging the magnetic powder pattern generated on the surface of the material being inspected by the imaging means and detecting the surface defect of the material to be inspected from the imaged image,
The sum of the average luminance in the image and a constant multiple of the standard deviation of luminance is used as a binarization threshold, and the image is binarized using this threshold to detect surface defects. Magnetic particle flaw detection device.   The image is divided into a plurality of block images, and the sum of the average luminance and a constant multiple of the standard deviation of luminance in each block image is set as a threshold value for binarization, and each block is used by using these threshold values. 2. The magnetic particle flaw detector according to claim 1, wherein a surface defect is detected by binarizing the image.   A block image is set around each pixel in the image, and the sum of the average luminance and a constant multiple of the standard deviation of the luminance in each set block image is set as a threshold value for binarization. The magnetic particle flaw detector according to claim 1, wherein each pixel is binarized to detect a surface defect.   4. The magnetic particle flaw detector according to claim 1, wherein the constant multiple is 3 or more and 5 or less.