JPH09113467A - Method for deciding defect - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for deciding a linear defect surely through image processing in the defect inspection by fluorescent penetration. SOLUTION: An image processor determines a plurality of inspection windows and an approximate expression for the outline of work based on the images being picked up. In each inspection window, histogram of luminance is determined in the longitudinal and lateral direction of fluorescent pattern and a feature amount measuring window is set from the intersection of the peak waveform of each histogram and a preset threshold value. A defect is determined by calculating the feature amount of inner fluorescent pattern.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a visual inspection method for machine parts and the like (hereinafter referred to as a work), and more particularly to a defect discrimination method by image processing in a visual inspection by a fluorescent penetrant inspection method.

[0002]

2. Description of the Related Art Fluorescent penetrant inspection is one of the methods for detecting defects existing on the surface of a work. This is because a fluorescent solution is permeated into the defective part of the work surface, and ultraviolet light is emitted to this to make it emit light, thereby drawing out a fluorescent pattern with different brightness levels, and based on this fluorescent pattern, the defects existing on the surface of the work are detected. It is something to inspect. In such a fluorescent penetrant inspection method, a method of inspecting a defect existing on the surface of a work by image processing by capturing an image of a fluorescent pattern with an image capturing device such as a television camera has been proposed. The point of practical application of this method is dust and dirt,
He is involved in developing an image processing method that can reliably identify scratches.

Japanese Unexamined Patent Publication No. 6-160295 (hereinafter referred to as a known example) describes a flaw inspection method by image processing. In this method, it is assumed that the scratches have an elongated shape and are arranged in a single line, and the shape and directionality of the image to be inspected are determined to determine whether or not the scratches are present. For that purpose, the image data of the work is binarized with a certain threshold value to obtain the binary image group 11 shown in FIG. 8, and the pixels forming the binary image group 11 are divided for each binary image. The divided areas are created, and thereafter, the second moment is calculated for each divided area, the shape and directionality of the divided areas are calculated, and data for determining whether or not there is a scratch is created. This secondary moment indicates the degree to which the pixels forming the divided area are arbitrarily dispersed in the coordinate axis direction. The secondary moment in each direction of the divided area is calculated, and the secondary moment If the minimum second moment is the minor moment of second axis,
If this short axis second moment is smaller than the specified value,
It means that the divided region having the minor axis second moment has an elongated shape in a direction perpendicular to the axis of the minor axis second moment. Therefore, by determining the minor axis second moment, it is determined in which direction each of the divided regions extends long, and it is assumed that the two divided regions extending in the same direction constitute one scratch. The fused short-axis second moments are obtained from each divided region by merging two regions in which the axial directions of the short-axis second moments are the same and determining in which direction and how long the fused regions extend. A characteristic of the fusion area in which the fusion minor axis second moment is smaller than a predetermined value, the area of the fusion area, and the shape and directionality obtained by calculation from the area and the fusion minor axis second moment. The combination with the value determines whether or not it is a scratch.

[0004]

However, the known example has the following problems. That is, the work surface has serious defects that should be defective, defects that do not hinder practical use, stains that are not defects, and the like, and show the respective fluorescent patterns. And the directionality alone are not sufficient to identify serious defects. In addition, this known example targets a defect existing on a surface portion far away from the contour portion of the work, and discriminates between the binary image of the work boundary portion and the binary image from the fluorescent liquid penetrating the scratch. Is not listed at all. In the actual defect determination, the most important point is how to correctly distinguish the defect around the contour from the contour, and the object of the present invention is to distinguish the contour of the workpiece and An object of the present invention is to provide an image processing method capable of reliably discriminating between defects and other defects and stains.

[0005]

In the present invention, the surface of a work permeated with a fluorescent liquid is irradiated with ultraviolet light, the fluorescent pattern on the surface of the work is imaged and converted into an image signal, and the image signal is imaged. In a defect discriminating method for discriminating defects on the surface of a work by processing with a processing device, a contour of a work and a defect designating pattern are extracted from an image memory for storing image signals, and a plurality of inspection windows are based on the contour of the work. And a work contour approximation formula is calculated to determine the inspection range within the inspection window, and within the inspection range, a histogram of the sum of the brightness in the vertical direction of the fluorescent pattern is obtained, and the peak waveform of the histogram and a preset threshold value are set. From the intersection of, set a region having a width in the horizontal direction, obtain a histogram of the sum of the horizontal luminance in the region, from the intersection of the peak waveform of the histogram and a preset threshold An area having a width in the direction is set, this area is used as a feature amount measurement window of each defect, the physical feature amount of the defect designating pattern in the feature amount measurement window is calculated, and the area of the predetermined feature amount measurement window is determined. And a technical means for discriminating a defect in light of the characteristic amount discriminating value.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. In the present embodiment, a ferrite magnet having a shape having four curved sides as shown in FIG. 6 is targeted as the work. The point of the visual inspection of this work is to distinguish between a crack that is a serious defect and a defect that is a defect that does not cause any practical problems (hereinafter referred to as a pseudo defect), and reliably rejects a crack that is a serious defect. Is. FIG. 7 shows an example of a crack which is a serious defect and a pseudo defect. Here, the crack has a feature that it has a small area and is in the form of a thin line, and has a directionality orthogonal to the work contour line. On the other hand, among the pseudo defects, the chip and the unmachined have a relatively large area and exist in the contour portion of the work, and the dust, stains and pits have a relatively small area and the aspect ratio is close to 1, The feature is that it is irrelevant to the position of. Therefore, in the present embodiment, the main object is to discriminate elongated linear defects having a direction orthogonal to the workpiece contour line.

FIG. 1 is a conceptual diagram of the imaging method of the present invention.
A background in which the imaging device 4 and the ultraviolet irradiation device 3 are provided above the surface to be inspected of the workpiece 1 in which the concave portion is soaked in the fluorescent liquid and washed, and the fluorescent liquid is permeated to generate scattered light in the direction of the back surface of the surface to be inspected. The illuminating device 2 is provided, and each illuminating device images the fluorescent pattern of the surface to be inspected and the contour of the work. An image processing device (not shown) discriminates the imaged image signal from defects. Hereinafter, the processing procedure will be described based on the procedure of FIG.

1) Procedure This is a step for extracting a fluorescent pattern appearing as a work contour portion and a defect designating pattern from the input image signal by using a known binarization process.

2) Procedure This is an inspection window calculation step for determining the inspection area of the work. In this embodiment, five inspection areas as shown in FIG. 3D are provided on the basis of the appearance inspection data of the present work in the past, based on the defect type, the location where the defect occurred, and the detection resolution of the imaging device. A method for extracting the representative points of the contour of the work and creating the five inspection areas from the input image will be described with reference to FIG. First, the entire image range is scanned in the X direction, and steep change points x1s and x1e, x2s and x2e, x3s and x3 on each scanning line as shown in FIG. 3A.
, e, ..., The distance Lx1 between x1s and x1e, the distance Lx2 between x2s and x2e, the distance Lx3 between x3s and x3e ,.
Respectively. Similarly, as shown in FIG. 3B, scanning is performed in the Y direction over the entire image range, and steep change points y1s and y1e, y2s and y2e, y3s on each scanning line.
And y3e, ..., The distance Ly1, y between y1s and y1e
Distance Ly between 2s and y2e Ly2, Distance Ly between y3s and y3e
Ask for 3, ...

Next, Lx1, Lx2, Lx3, ... Are compared to find the maximum distance in the X direction. Similarly, Ly1, Ly2,
Ly3, ... Are compared to obtain the maximum distance in the Y direction. Here, x3e and y2e are abrupt change points due to the defect design pattern, and the distances Lx3 and Ly2 show values smaller than the work contour distance. Therefore, x3e and y2e are not adopted as the work contour points. Figure 4 shows the four points with the maximum distance.
As shown in (c), the representative points xs, xe, ys of each contour,
Let's say yes. Inspection windows W1, W2, W3, W4 and W5, which are five inspection areas as shown in FIG. 3D, are determined from the coordinates and offset values of the four representative points. As the offset value, a predetermined value is adopted based on the past data of the defect occurrence status of the target work.

3) Procedure FIGS. 4 (a) and 4 (b) are explanatory diagrams for calculating the outline approximation line of the workpiece by software in the previously determined inspection window. As shown in FIG. 4A, the inspection windows W4, W
In each of the sections 5, the X-direction scanning is sequentially performed as shown by the arrow to find a steep change point, and this is extracted as contour position data. Linear approximation is performed from the contour position data by the least square approximation method to obtain linear equations ax + by + c = 0, which are defined as contour lines. Similarly, as shown in FIG. 4B, in each of the inspection windows W2 and W3, Y-direction scanning is sequentially performed as indicated by arrows to obtain a steep change point, and this is extracted as contour position data. Here, the surface of the work in this portion is curved, and since it is not equidistant from the lens of the image pickup apparatus, the contour becomes a curve, and quadratic curve approximation is performed by the least square approximation method, and each quadratic curve expression y = ax. ²
+ Bx + c is obtained and used as the contour line. The hatched portions of the windows W2 and W5 in the figure show an example of the fluorescent pattern of the contour portion.

4) Procedure In the inspection window obtained in the procedure, the sum of the brightness of the work portion set based on the contour line calculated in the procedure is measured to calculate the characteristic amount measurement window of the fluorescent pattern. . FIG. 5A is a diagram for explaining a method of calculating the histogram of the sum of luminances using the inspection window W2 as an example. The case where there is a crack (linear in a direction orthogonal to the contour line) and a pseudo defect image having a relatively small area such as dust, stains, and pits is shown. In the inspection window, the approximate contour line calculated in the above procedure is shifted in the work area direction by a predetermined amount, and a new contour line R is set, and the contour line R is set.
The work part surrounded by is the inspection range. If the predetermined amount is set to such an amount that the pseudo defects existing along the contour line deviate from the inspection target range, it is possible to eliminate the fluorescent pattern which is obviously different from the crack at this point to some extent. Effective in terms of efficiency.

With respect to the entire surface of the set inspection target range,
First, when calculating the histogram of the sum of the vertical luminance,
The result is as shown in FIG. Two brightness peaks,
The width X1 (corresponding to a pseudo defect in this example) and the width X2 (corresponding to a linear defect in this example) are obtained from the intersection of preset threshold values. Areas 1 and 2 in the vertical direction are set based on the widths X1 and X2. 5C and 5D show histograms of horizontal luminance sums calculated for the regions 1 and 2. Similar to FIG. 5B, the width Y1 (corresponding to the pseudo defect) and the width Y2 (corresponding to the linear defect) can be obtained from the luminance peak and the preset threshold value. From X1, X2, Y1, and Y2 thus obtained, a feature amount measurement window (hatched portion) surrounding each fluorescent pattern is calculated.

5) Procedure Finally, the detailed feature amount of the fluorescent pattern is measured for all the feature amount measurement windows calculated by the above method. First, the area S, the length L (the length of the principal axis of inertia), the width W (the length of the minor axis of inertia), the inclination θ (the inclination of the principal axis of inertia), and the inclination of the actual fluorescent pattern in the feature amount measurement window. Calculate the distance to the workpiece contour line R in the extension line direction, and then further length L
The aspect ratio, which is the ratio of the width W to the width W, is calculated. The actual fluorescent pattern in the window is not necessarily a continuous pattern but may be divided into a plurality of patterns. However, as described in the above-mentioned known example, one continuous pattern can be softly created. To process. With respect to the processed fluorescent pattern, the same calculation process as described above is performed again. Next, the calculated value is compared with a preset determination criterion of the characteristic amount peculiar to the inspection area for determining a pseudo defect and a crack, and it is determined whether or not there is a crack. This is performed for each of the above-mentioned five inspection areas, and the defect determination of the entire work is performed. Regarding the window itself, if the window area is smaller than the set amount, it is judged that the inherent fluorescent pattern is not a linear elongated shape, and it is excluded because it does not have the characteristics of cracks, and the processing time is Can be shortened.

[0015]

As described above, the present invention has the following effects. 1) Since the target range for defect detection is determined by setting the contour line by software, pseudo defects existing near the actual contour portion can be excluded in advance, and the processing time can be shortened. 2) Since the histogram of the sum of brightness is calculated and the feature quantity measurement window is set, even linear defects that are not detected as a continuous fluorescent pattern can be combined in one feature quantity measurement window, and line defects can be detected. There are fewer mistakes to miss. 3) At the time of setting the feature amount measurement window, if it is determined whether or not the target linear level, the non-target can be excluded in advance in the same manner as described above, and the processing time can be shortened.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram showing an imaging method of the present invention.

FIG. 2 is a diagram showing an image processing procedure of the present invention.

FIG. 3 is a diagram illustrating a method of creating a judgment window according to the present invention.

FIG. 4 is a diagram for explaining a work contour calculating method according to the present invention.

FIG. 5 is a diagram illustrating a feature amount measurement window of the present invention.

FIG. 6 is an explanatory view of the shape of the test material.

FIG. 7 is an explanatory diagram of a representative example of defects.

FIG. 8 is a diagram illustrating a binary image of a scratch in the related art.

[Explanation of symbols]

 1 Work 2 Background Illumination 3 Ultraviolet Illumination 4 Imaging Device W1-W5 Inspection Window

Claims (1)

[Claims] 1. A surface of a work permeated with a fluorescent liquid is irradiated with ultraviolet light, a fluorescent pattern on the surface of the work is imaged and converted into an image signal, and the image signal is processed by an image processing device to obtain the work. In a defect discriminating method for discriminating a surface defect, a contour of a work and a defect designating pattern are extracted from an image memory storing image signals, and a plurality of inspection windows and a work contour approximation formula are calculated based on the work contour. Then, the work part divided by the work contour approximation formula in the inspection window is set as the inspection range, and the histogram of the sum of the vertical luminance of the fluorescent pattern is obtained in the inspection range, and the peak waveform of the histogram is set in advance. An area having a width in the horizontal direction is set from the intersection with the threshold value, a histogram of the sum of the horizontal brightness in the area is obtained, and a peak waveform of the histogram and a preset threshold value are set. An area having a width in the vertical direction from the intersection with is set, and this area is used as the feature amount measurement window of each fluorescent pattern, and the length, width, inclination, and distance from the contour line of the fluorescent pattern in the feature amount measurement window are set. Is calculated, and a defect is discriminated based on the predetermined characteristic amount discriminant value.