JP2011085468A - Defect inspection method for sheet-like material - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To correct accurate in-line shading, even if an inspection object is defective, immediately after the impregnation of resin and the surface condition changes over time, by eliminating the influence. <P>SOLUTION: The inspection method for moving sheet-like continuous body includes a first step of obtaining a plurality of lines of pixel data which are arranged in the width direction with fixed gaps in the longitudinal direction; a second step of extracting a minimum value from each of the pixel data lines in the longitudinal direction, by comparing the luminance value of the pixels in the data line for the plurality of data lines arranged in the width direction; a third step of performing smoothing treatment on each of the data lines in the width direction; a fourth step of preparing an image for shading correction, by arranging the smoothed data line in the longitudinal direction; and a fifth step of preparing the shading corrected image, by subtracting the image for shading correction from an inspection image cut out after an image P is cut out in the first step. The shading corrected images are sequentially obtained from the images cut out by the first to the fifth steps. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a shading correction technique and a defect inspection method for a continuous body of moving sheet-like material having unevenness of formation.

  It is necessary to accurately detect defects that impair the quality of products such as wrinkles, foreign matter, fluff, and resin transfer defects that occur in the manufacturing process of sheet-like objects such as steel sheets, films, and CFRP, and prevent outflow. There is. Further, by finding these defects early in the manufacturing stage, investigating the cause, and changing the manufacturing conditions and the like, the yield of subsequent products can be improved. That is, it is very important to detect the defects of the sheet-like material in-line with high accuracy.

  In an image inspection apparatus using an image sensor, luminance unevenness occurs in an inspection image due to unevenness in illumination light illumination, variation in image pickup system, and the like. This uneven brightness is called shading, and correcting shading so that the entire image has uniform brightness is called shading correction. However, if the surface of the object to be inspected has unevenness of formation, the brightness value of the normal part having no defect also changes over time due to the unevenness of formation, so that shading correction becomes difficult, and at the same time, minute defects and contrast differences It is also difficult to detect small defects. In addition, it means the uneven brightness of the surface that occurs discontinuously and irregularly.

  Patent Document 1 describes a method for accurately detecting fuzz, cracks, and foreign matter defects in a prepreg. However, in the inspection method described in Patent Document 1, since the prepreg is irradiated with light and only the scattered reflected light is compared with a preset threshold level, shading caused by uneven illumination or uneven formation of illumination is performed. Because it is affected, accurate detection is not possible.

  In Patent Document 2, when the luminance distribution of a reference sample is acquired in advance before inspection and shading correction data is generated, dirt on the reference sample and adhesion of dust to the reference sample do not affect the luminance distribution. Describes a method of performing shading correction with high accuracy by moving an acquisition position to acquire a plurality of luminance distributions and performing a comparison operation. However, since the method described in Patent Document 2 is based on the premise that a reference sample is used, it cannot be applied when the surface state is different between an in-line state and a sample whose time has elapsed after manufacture. In addition, this method cannot be applied when there is uneven formation and the luminance distribution of the subject in-line varies with time.

  In Patent Document 3, when a surface image of a material such as a steel plate or a slab is disturbed by dirt or the like, the dirt or the like is covered in order to replace the disturbed portion due to the dirt or the like of the image with a luminance corresponding to the texture of the material. If the surface of the material surface is bright, the maximum luminance value is extracted from the rectangular region.If the surface of the material surface is dark, the minimum luminance value is extracted from the rectangular region. Describes a method of correcting the luminance due to the contamination of the image or the like by using as a representative pixel of the rectangular area. However, since the method of Patent Document 3 is a method of replacing brightness values such as image stains with the brightness values of the formation of the material, the influence of formation unevenness cannot be removed.

JP-A-9-225939 JP 2006-270334 A JP 2002-214148 A

  It is an object of the present invention to provide a defect inspection method capable of reducing the influence and accurately correcting shading in-line even when a sheet-like object to be inspected has uneven formation.

In the defect inspection method for a continuous body of sheet-like material having unevenness of formation of the present invention, the surface of the moving sheet-like continuous body is irradiated with light by line illumination, and the surface of the sheet-like continuous body is With a line sensor installed in the width direction, a continuous image composed of pixel data in the width direction of the continuum is acquired over time and stored in the storage means, and the continuous image stored in the storage means is stored. An inspection method for processing and detecting and identifying defects,
Read out the continuous images stored in the storage means, cut out the continuous images at a specific length in the longitudinal direction, from the cut out image P (n-1) at a certain interval in the longitudinal direction, or continuously, A first step of acquiring a plurality of columns of pixel data arranged in the width direction;
For the data rows arranged in the width direction obtained in the first step, the luminance values of the pixels in the longitudinal pixel data row configured in the width direction are compared, and the pixel data rows in each longitudinal direction are compared. A second step of extracting a minimum value of
A third step of performing a smoothing process on the data string in the width direction configured by the minimum value of each column extracted in the second step;
A fourth step of creating a shading correction image Q (n−1) by arranging the data sequence smoothed in the third step in the longitudinal direction;
The inspection image P (n) cut out after the image P (n−1) cut out in the first step is subtracted by the shading correction image Q (n−1) created in the fourth step, thereby A fifth step of creating a shading corrected image Z (n) of the image P (n),
This is an inspection method for a moving sheet-like continuum.

  In the shading correction technique and defect inspection method for a sheet-like material having unevenness of the present invention, threshold processing is performed to detect a defect in the shading correction image Z (n) created in the fifth step. The defect is identified by comparing the feature amount of the defect candidate with a preset reference value for pass / fail judgment, and the preset reference value for pass / fail judgment is the size and shape of the defect. Features.

  The shading correction image Q is obtained by sequentially repeating the first step to the fifth step on the nth image P (n) cut out in the defect inspection method for a sheet-like object having formation unevenness according to the present invention. (N-1) is updated, a shading correction image Z (n) is created for each inspection image P (n), and the obtained shading correction image Z (n) is compared with the feature amount of each defect. The processing is performed for each inspection image.

  In the shading correction technique and defect inspection method for a sheet-like material having uneven formation according to the present invention, the sheet-like continuum is a resin-impregnated sheet obtained by impregnating a reinforced fiber with a resin, and has uneven surfaces. Or unevenness of the impregnation degree immediately after impregnation with the resin, components of the resin used, etc., and uneven formation exists.

  In the shading correction technique and defect inspection method for a sheet-like material having unevenness of formation according to the present invention, the line-shaped illumination is a method of the surface of the sheet-like continuum having an incident angle with respect to the surface of the sheet-like continuum. When it is based on the line direction, it is provided so as to be 5 to 15 degrees from the normal direction, the line sensor is installed in the normal direction, and the optical system according to claim 8 or 9 is used. The image of the surface of the sheet-like continuum of 1 and 2 shows that the unevenness of the ground is on the bright side and the defect is darker than the darkness or the unevenness of the ground, compared to the brightness of the ground, which is less affected by the unevenness of the ground. It is the bright side.

  The object to be inspected has instability in the degree of impregnation immediately after impregnation of the resin, unevenness in the surface condition, unevenness of the surface due to the characteristics of the resin used, etc., and the surface condition such as size and shape changes over time Even in this case, it is possible to reduce or eliminate the influence, to accurately perform shading correction in-line, and to detect defects.

(A) In the 1st step of the defect detection method of this invention, it is a conceptual diagram of the pixel row | line | column arranged in the width direction acquired with the line sensor and memorize | stored in the memory | storage means. (B) In the first step of the present invention, FIG. 1 (a) is a conceptual diagram of an image in which the pixel columns of FIG. 1 (a) are sequentially stored in the storage means and the length in the longitudinal direction has reached a specific length. (C-1) It is the conceptual diagram which showed acquiring the pixel row | line | column arranged in the width direction at fixed intervals in the longitudinal direction from the image of FIG.1 (b) in the 1st step of this invention. . (C-2) It is the conceptual diagram which showed acquiring the pixel row | line | column arranged in the width direction continuously from the image of FIG.1 (b) in the 1st step of this invention. In the second step of the present invention, acquiring pixel rows arranged in the width direction continuously in the longitudinal direction from the image obtained in FIG. 1 (c-1) or FIG. 1 (c-2). It is the conceptual diagram shown. FIG. 6 is a conceptual diagram showing that a pixel row in which a minimum value is extracted and smoothed from the image obtained in FIG. 2 is created in the third step of the present invention. It is the conceptual diagram which showed arranging the image obtained in FIG. 3 in the longitudinal direction in the 4th step of this invention.

The method for inspecting a defect of a moving sheet-like continuum according to the present invention is used for inspecting a defect of the continuum during movement in a process including a step of moving a continuum having a certain width in a longitudinal direction. It is what The sheet shape here does not need to be continuous in the width direction as long as it has a certain width. For example, even in the case of a linear continuous body, it is aligned and aligned in the longitudinal direction. It is sufficient if the surface is formed as a whole. Forming a surface means a region of a surface to be formed when a linear (or tape-shaped) object has a gap in the width direction represented by being aligned in the longitudinal direction. The gap is 20% or less.
The width of the continuum to which the inspection method of the present invention can be applied is not particularly limited from the viewpoint of data processing. That is, when the width of the continuous body to be inspected is wide, it can be dealt with by adding a line sensor according to the width. Although not the essence of the present invention, from the viewpoint of handling the sheet, a width of 0.3 m to 3 m or less is preferable because the sheet travels stably and the disturbance during data acquisition is reduced.

As a sheet-like continuous body to which the inspection method of the present invention can be preferably applied, for example, a resin-impregnated sheet such as a steel plate, a film, and a prepreg can be raised. Therefore, it is effective when applied to the prepreg shown below.
A prepreg is a knitted fabric of reinforcing fibers such as carbon fiber, glass fiber, or aramid fiber, woven fabric, or a group of fibers arranged in one direction, which is sandwiched by a release film previously coated with a matrix resin, and pressed to form a matrix resin. It is a sheet-like intermediate substrate that is transferred and impregnated thereon. As a feature of the prepreg, there is thickness unevenness based on the density of the fibers. This is because the pressure during resin impregnation becomes non-uniform in the width direction due to unevenness of fiber spreading obtained during prepreg preparation, and unevenness is generated as the pressure is released and relaxed. Moreover, since such relaxation occurs immediately after the resin impregnation step, it is considered that the uneven formation of the resin changes with time. Thus, formation unevenness occurs on the surface of the prepreg, and the size and shape of the formation unevenness changes with time after the pressure during impregnation is released. For this reason, the state of the sheet surface immediately after the resin impregnation is unstable, and the luminance value changes with time even in the normal part. For these reasons, it is very difficult to detect defects by correcting shading with the conventional method, and the present invention has been accomplished by earnestly studying to overcome such difficulties.

  The line illumination used in the inspection method of the present invention means that the light emitting part has a shape extending linearly in a one-dimensional direction. Specifically, such line-shaped illumination is a straight-tube fluorescent lamp, or a plurality of optical fibers arranged in a line to guide light from a light source such as a halogen or LED with a light guide, or on the entire surface of the line-shaped illumination. Examples of the illumination include providing a cylindrical lens. The light amount can be easily changed, and LED line illumination is preferable from the viewpoint of cost and maintainability. The length of the light emitting part of the line-shaped illumination device is not particularly limited as long as it can illuminate the entire width of the moving sheet-like continuum, but covers the outer side of the continuum from 50 to 150 mm on one side. It is preferable to do. For example, if the width of the continuum is 1000 mm, the length of the light-emitting portion of the line illumination may be 1200 mm, and illumination may be performed to the outside by 100 mm.

  In such an inspection method of the present invention, data is acquired by a line sensor installed in the width direction of the sheet-like continuous body. The width direction here may be a direction that intersects with the longitudinal direction of the continuum, and does not necessarily need to be orthogonal, but from the viewpoint of specifying position information, It is preferable to intersect at an angle of 100 °.

  The line sensor applied to the present invention obtains data related to the brightness or brightness of one line, or light receiving elements (pixels) such as photodiodes that receive light are linearly arranged in a line and received light. This refers to a device that converts light into an electrical signal and outputs data. Specifically, there are a CCD line sensor, a CMOS line sensor, an X-ray line sensor, and the like. In order to detect a defect, it is preferable that there is little noise, and a CCD line sensor is preferable from this viewpoint.

  The resolution of the data acquired by the line sensor is preferably in the range of 1/3 to 1/5 with respect to the size of the defect to be detected. In the width direction, the range in which the pixel size is the resolution, and in the vertical direction, 1/3 to 1/5 of the size of the defect to be detected due to the relationship between the movement speed of the inspection object and the data acquisition speed. It is preferable to determine the data acquisition speed so as to achieve a pixel resolution. For example, when detecting a defect of 0.3 mm or more, a line sensor of 60 μm to 100 μm / pixel is selected.

  In the present invention, pixel data acquired by the light receiving elements arranged in parallel with the width direction is defined as pixel data (in the width direction), and the pixel data is acquired over time to be two-dimensional data. Is defined as an image.

  In the present invention, the defect refers to a flaw on the surface of the sheet-like continuous body or a foreign matter attached thereto during the manufacturing process. As for the prepreg material, defects such as resin transfer failure, fuzz and pills of reinforcing fibers, resin transfer failure portions connected in a streak shape in the longitudinal direction, wrinkles generated on the prepreg surface, and the like are listed as defects. . When the method of the present invention is applied to such a prepreg, as an optical system, a sensor is arranged in the normal direction with respect to the sheet surface, and line-shaped illumination is arranged at a position inclined by 5 to 15 degrees from the normal direction. This is true. By adopting such an arrangement, it is possible to obtain an image in which the formation unevenness is on the bright side and the defects are on the dark side or on the brighter side than the formation unevenness, compared with a normal base portion without uneven formation of the resin. It is.

  In the first step of the inspection method of the present invention, a continuous image acquired by the line sensor is cut out at a specific length, and the cut-out image P (n−1) is continuously spaced at a certain interval in the longitudinal direction or continuously. This is a step of acquiring a plurality of longitudinal data rows of pixel data at each position in the width direction. At this time, the specific length can be set as appropriate, and is usually preferably in the range of 500 to 2000 pixels. Here, the cut-out image P (n) represents the n-th cut-out image, and P (n−1) represents the image immediately before the n-th cut-out image. Then, n and n-1 should be defined as natural numbers, but in actual manufacturing sites such as prepregs, production as a product starts from a stable line, so just before starting as a product. Therefore, the data immediately before n becomes 1 (that is, immediately before the product starts after the line is stabilized) is treated as P (0).

  Here, cutting out continuous images with a specific length means that pixel data in the width direction is sequentially accumulated in the longitudinal direction in the storage means to form two-dimensional data, and the two-dimensional data has a specific length in the longitudinal direction. At the time of accumulation, the accumulated two-dimensional pixel data group is read out as an image, and the length in the longitudinal direction of the image to be cut out is not particularly limited as long as it is always constant throughout the examination. Can be set. From the viewpoint of further improving the inspection accuracy, it is preferable that both a portion where the unevenness of formation is intense and a portion where it is not intense are imaged in one image. From this viewpoint, it is constant in the longitudinal direction at the time of cutting. If the interval is increased, the interval is longer and more data strings are acquired.If the interval is continuously acquired, it is preferable to acquire more data sequences continuously. If more data strings are acquired, the processing amount increases. Therefore, considering these balances, it is preferable that the acquired data strings are 10 to 20 columns.

  Here, a certain interval in the longitudinal direction means that if the cut-out image is composed of 300 pixel lines in the longitudinal direction, for example, the 50th, 150th, and 250th lines are constant. The intervals (here, 100 lines) are taken out, and acquiring continuously in the longitudinal direction means acquiring, for example, the 100th to 110th lines.

  In the second step in the inspection method of the present invention, the pixel data in the column (in this specification, the pixel data is used as the luminance value) for the column of longitudinal data arranged in the width direction obtained in the first step. The minimum value of each column of pixel data is extracted. Since the brightness value of the part where the unevenness of the formation is noticeable is higher than the brightness value of the part where the influence of the unevenness of formation is hardly affected, by extracting the minimum value of each column, the Information about the luminance value of the portion with the least unevenness is obtained.

  In the subsequent third step, smoothing processing is performed on the data string in the lateral direction formed by the minimum value of each column. The smoothing process is a process of smoothing an image by using a moving average filter, a Gaussian filter, or the like, using a pixel of interest and its surrounding luminance values. If the brightness values in the column in the second step are all brightness values where the unevenness of formation is significant, the extracted minimum value is also the brightness value of the portion where the unevenness of formation is intense, so this processing is performed. Thus, it is possible to obtain a luminance distribution that reduces the influence of uneven formation.

  Subsequently to this, in a fourth step, a shading correction image Q (n-1) is created by arranging this in the longitudinal direction for the specific length described above. The shading correction image here is an image for shading correction of the next cut out image, and the shading correction image Q (n−1) is obtained from the cut out image P (n−1). It is assumed that the shading correction image is displayed. Therefore, Q (0) treats data immediately before starting as a product (that is, immediately before starting as a product after the line has been stabilized) as Q (0), similarly to P (0) described above. Shall.

  In the subsequent fifth step, the inspection image P (n) cut out after the image P (n−1) cut out in the first step is used as the shading correction image Q (n−1) created in the fourth step. By performing the difference, processing for creating a shading correction image Z (n) of the inspection image is performed. The difference is a process of comparing the luminance values of the same pixel positions of the inspection image and the shading correction image and subtracting the luminance value of the shading correction image from the luminance value of the inspection image. That is, Z (n) = P (n) −Q (n−1). Further, after taking the difference, it is preferable to perform a process of adding 128 if the luminance range of the image is 256 gradations and adding 512 if it is 1024 gradations. This is because when P (n) has a defect, the luminance value of the defect is lower than the luminance of the non-defect portion and Z (n) is a negative value, so that a negative value is not output. This is because increasing the overall luminance value of Z (n) is preferable for data processing.

  In this way, a shading correction image can be obtained from images sequentially cut out in the nth and n + 1th order.

  In such an inspection method of the present invention, the shading correction image created in the fifth step of claim 1 is subjected to threshold processing to detect defects, and the feature amount of the defect candidate and the criterion for pass / fail judgment set in advance. A process of identifying and identifying the defect by comparing the value is performed. The reference value for pass / fail judgment here is preferably a value corresponding to the size or shape of the defect, and specifically, such values include “the diameter of the smallest circle circumscribing the defect candidate”, “defect candidate”. The minimum rectangle size circumscribing ”,“ the number of defective candidate pixels ”, and the like. For example, for a defect candidate with a high degree of circularity such as fluff or fluff, the diameter of the smallest circle that circumscribes it may be based on the width of a rectangle that circumscribes the defect candidate with a high degree of linearity such as streaks. Can be mentioned. Further, the threshold value used in the present invention does not always have to be a constant value. When the sheet surface state is unstable and the luminance changes with time, the overall luminance level of the image Z (n) varies. Therefore, it is preferable to adjust the threshold value accordingly. As a method in such a case, for example, the threshold value can be adjusted based on a change in the average value M (n) of the shading correction image Q (n). That is, taking as an example the case of applying to the threshold value used for the determination of Z (n + 1), the average values M (n) and M (n n-1) is obtained, and M (n) -M (n-1) is added to the threshold used for the determination of Z (n). In this way, the inspection can be continued by sequentially adjusting the threshold value for each inspection image. Here, n ≧ 1. This is because it is not necessary to correct the threshold at the point of n = 0, and it is applied to the second and subsequent points after the start of inspection.

  In the inspection method of the present invention, the shading correction image is updated for each inspection by repeating the first step to the fifth step for the nth and n + 1th sequentially cut images, and for each inspection. A process of creating a shading correction image for the inspection image P (n) is performed. This is because shading correction is performed on the image P (n−1) and the inspection image P (n) cut out one sheet before the inspection image, and the first to fifth steps are performed on the next cut out image. By repeatedly generating the next shading correction image repeatedly, it is possible to obtain a shading correction image that is always appropriately subjected to shading correction. Further, it is possible to sequentially detect and identify defects in such a shading correction image.

  If a defect is detected in the inspection image P (n), the shading correction image Z (n) is not created from P (n) in the inspection of the inspection image P (n + 1) cut out next. Instead, a shading correction image is created using the shading correction image Q (n-1) created in the previous inspection, that is, if a defect is detected in the inspection image, the shading correction is performed in the next inspection. It is preferable not to update the image.

  In addition, when shading correction is performed when the luminance value of the normal portion changes with time like a prepreg, it is preferable to update the shading correction image for each inspection in order to follow the change with time as much as possible. If the change over time is slow, etc., it may be updated every few to several hundreds.

  In the inspection method of the present invention, the inspection object is not particularly limited as long as it is a sheet-like continuous body, but it is more effective for those having uneven formation on the surface of the continuous body. In particular, a resin-impregnated sheet in which a reinforcing fiber is impregnated with a resin like a prepreg is more preferable. Regarding the reason, as described above, the surface of the prepreg immediately after impregnating the resin has uneven formation due to unevenness of the resin impregnation due to the resin components used and the unevenness between the reinforcing fibers even in the part without defects. However, the size, shape, and luminance value are characterized by changing with time.

  In such an inspection method of the present invention, the line irradiation is provided so as to be 5 to 15 degrees from the normal direction with respect to the normal direction of the surface of the sheet-like continuous body, and the line sensor is provided in the normal direction. It is preferable to carry out the treatment. When the sheet-like continuum here is a prepreg, the brightness of the ground (the original brightness excluding the influence of formation unevenness when the sheet-like continuum is illuminated) is determined by setting the angle of such illumination and line sensor. Value), the formation unevenness can be imaged on the bright side, and the defect can be imaged on the dark side or on the bright side of the formation unevenness, which is preferable.

  Hereinafter, preferred embodiments of the defect inspection method of the present invention will be described with reference to the drawings.

The first step will be described with reference to FIG.

  FIG. 1A shows a pixel row arranged in the width direction, which is obtained by the line sensor and stored in the storage unit. Although FIG. 1A shows 30 pixels, the number of pixels can be arbitrarily set according to the pixel size of one pixel set according to the defect to be detected and the width of the sheet-like continuous body to be inspected. The pixel rows are sequentially stored in the storage means. As shown in FIG. 1B, when the length in the longitudinal direction reaches a certain length, this is cut out to form one image. In FIG. 1B, the length in the longitudinal direction is 10 pixels, but it is preferable to set the length as long as possible in balance with the data processing amount.

  With respect to the image having a specific length in the longitudinal direction shown in FIG. 1 (b), as shown in FIG. 1 (c-1), in the width direction with a certain interval in the longitudinal direction. A plurality of columns 1 of aligned pixel data are acquired. In the example of FIG. 1 (c-1), five columns are acquired at intervals of two columns, but the interval between columns to be acquired and the number of columns are appropriately set in consideration of the range to be acquired and the amount of data that can be processed. can do. Further, instead of the method shown in FIG. 1C-1, the column 2 of pixel data arranged in the width direction continuously in the longitudinal direction may be extracted as shown in FIG. 1C-2. good. In the example of FIG. 1 (c-2), five columns at the top of the image are acquired. The position in the image to be acquired and the number of columns are appropriately determined in consideration of the range to be acquired and the amount of data that can be processed. Can be set.

  Next, the second step will be described with reference to FIG.

  As described above, from the column 1 or 2 of the acquired plurality of pixel data (this is indicated by the symbol A in FIG. 2), as shown in FIG. 2, the same position in the width direction (in the width direction) A pixel data string in the longitudinal direction composed of pixel data (which may be referred to as a position) is taken out, luminance values in the pixel data strings are compared, and a minimum value is extracted for each pixel data string. For example, in FIG. 2, the luminance value of the leftmost longitudinal pixel column in the width direction is “92”, “80”, “86”, “91”, “86”, and “89”. The value “80” is extracted as the minimum value. This is performed in the full width direction, and a new pixel row 3 arranged in the width direction constituted by the minimum value of the pixel rows in the longitudinal direction is newly created.

  Next, the third and fourth steps will be described with reference to FIGS.

  The pixel row 3 obtained as described above is smoothed by a moving average filter, a gas cyan filter or the like (third step), and the smoothed pixel row 4 has a length in the longitudinal direction. A shading correction image 4 is created by overlapping in the longitudinal direction so as to be the specific length (fourth step). In the example of FIG. 3, a moving average filter of 1 pixel in the longitudinal direction × 3 pixels in the lateral direction is used. However, the moving average filter can be set as appropriate according to data variation. FIG. 4 is a diagram in which 10 smoothed pixel rows 4 are stacked in the longitudinal direction. The shading correction of the inspection image is performed by taking the difference between the shading correction image 5 and the inspection image cut out next (fifth step).

Example 1
An embodiment of the present invention will be described. A prepreg production line running at a width of 1000 m and 5 m / min was used as an inspection target. As the line illumination, one high-intensity LED line illumination (HLND type manufactured by CCS) was used, and it was arranged so that light was incident on the surface of the prepreg. As an imaging means, one line sensor (NED e7450) having the number of elements of 7450, a drive frequency of 40 MHz, and a maximum inspection cycle of 192 μm is used, and the center position of the illumination and the center position of the line sensor are installed so as to overlap each other. did. At this time, the position was adjusted so that the resolution in the width direction of the prepreg of the captured image was 80 μm / pixel and the resolution in the longitudinal direction was 80 μm. The illumination incident angle was set at an angle of 10 degrees from the normal direction of the prepreg surface, and the amount of illumination light was adjusted so that the illuminance on the prepreg surface was 8000 lx. The line sensor was installed so that its optical axis was in the normal direction of the prepreg surface, and a light shielding plate was attached to the top and side surfaces so that external light did not directly hit the prepreg surface of the sensor imaging range.

  The measurement was started when the prepreg production line was started and stabilized.

  The image data picked up by the line sensor was temporarily stored on an image processing board attached on a personal computer, and when it was stored 1000 lines in the longitudinal direction, it was converted to an image of 7450 × 1000 pixels, and then read into a memory on the personal computer.

  The image captured at the start was defined as P (0), and 11 images up to P (10) were captured and captured continuously. As a result of visual confirmation at the time of image capture, a streak defect was found at a position corresponding to P (3). Further, the luminance depth of the acquired image is 255 gradations. First, P (0) is set as the first continuous image, and in the first step, 10 pixel data columns are acquired at 100 intervals, the second to fourth steps are performed, and the shading correction image Q is acquired. (0) was created. Next, the shading correction image Z (1) is created by performing the fifth step of subtracting the shading correction image Q (0) from the second image P (1) and adding the luminance value 128. . Next, for P (2), steps 1 to 4 are similarly performed on P (1) to create Q (1), and Q (1) is used as a shading correction image. The inspection was advanced after obtaining a shading corrected image Z (n). At this time, streaks, which are defects that occur due to continuous transfer of defective resin transfer parts, fuzz formed by cutting carbon fibers, and defects of resin transfer defects are lower in brightness than the surrounding areas. In the image after shading correction obtained by subtracting the images, the defect appears on the dark side. For each shading corrected image Z (n), defect detection processing was performed in three stages in the order of streak detection, fluff detection, and transfer defect detection. In the streak detection, the shading correction image is binarized with a threshold value 135, and a defect candidate having a width in the width direction of 3.5 mm or more and a length in the length direction of 60 mm or more is determined as a streak. In addition, there are many streaks that do not lead to defects of less than 3.5 mm, and in order not to falsely detect them as other defects, after determining the streaks, the luminance value of the streaks whose longitudinal length is 60 mm or more is determined. The mask was replaced with 0. Next, the fluff detection processing was binarized with a threshold value 138, and those having a minimum circumscribed circle of 15 mm or more were determined as fluff, and the luminance value of the determined fluff portion was replaced with 0, and mask processing was performed. Finally, the transfer failure detection process was binarized with a threshold value 145, and a transfer with a minimum circumscribed circle of 3.5 mm or more was determined as a transfer failure.

These processes were repeated up to P (10). As a result, the defect in this example was a streak having a width of 4.0 mm, and the normal part was an image in which the formation unevenness appeared greatly, but it was present in P (6) without erroneously detecting the normal part. It was possible to determine and detect streak defects that had been confirmed visually.
(Example 2)
The same processing was performed on images P (100) to P (110) for the inspection object of Example 1 (P (99) was used as an image corresponding to P (0) in the example on the left). Except for the image to be inspected, defect inspection was performed under the same conditions as in Example 1. In this example, it was confirmed that P (107) had fluff by visual inspection, but by performing the inspection in the same steps as in Example 1, fluff having a minimum circumscribed circle diameter of 16 mm in the image of P (107). We were able to detect that there was. As in Example 1, the normal part was not erroneously detected.
(Example 3)
The same processing was performed on images P (200) to P (210) for the inspection target of Example 1 (P (199) was used as an image corresponding to P (0) in the example on the left). Except for the image to be inspected, defect inspection was performed under the same conditions as in Example 1. In this example, it was confirmed that there was a transfer failure visually at P (205). However, by performing inspection in the same steps as in Example 1, the minimum circumscribed circle diameter of the image of P (205) is 3. It was possible to detect that there was a transfer failure of 7 mm. As in Example 1, the normal part was not erroneously detected.

  According to the method of the present invention, even when the object to be inspected has uneven formation, and the surface condition such as the size and shape changes with time, the influence is removed and shading correction is accurately performed in-line. , Defects can be detected.

1a to 1e The widthwise pixel columns 2a to 2e that are extracted from the clipped image at a certain interval in the longitudinal direction. The widthwise pixel columns A1a to 1e that are continuously extracted from the clipped image in the longitudinal direction. Collective image, or collective images 2a to 2e 3 Width-direction pixel row 4 composed of the minimum values of the respective longitudinal columns 4 An image in which the pixel row composed of the minimum values is reconfigured to the same size as the cut-out image Brightness value

Claims (5)

The surface of the moving sheet-like continuum is irradiated with light by line-like illumination, and pixel data in the width direction of the continuum is obtained by a line sensor installed in the width direction of the surface of the sheet-like continuum. A continuous image composed of the following is acquired over time and stored in a storage means, and the continuous image stored in the storage means is processed to detect and identify defects,
Read out the continuous images stored in the storage means, cut out the continuous images at a specific length in the longitudinal direction, from the cut out image P (n-1), with a certain interval in the longitudinal direction, or continuously, A first step of acquiring a plurality of columns of pixel data arranged in the width direction;
For the data rows arranged in the width direction obtained in the first step, the luminance values of the pixels in the longitudinal pixel data row configured in the width direction are compared, and the pixel data rows in the respective longitudinal directions are compared. A second step of extracting a minimum value of
A third step of performing a smoothing process on the data string in the width direction configured by the minimum value of each column extracted in the second step;
A fourth step of creating a shading correction image Q (n−1) by arranging the data sequence smoothed in the third step in the longitudinal direction;
The inspection image P (n) cut out after the image P (n−1) cut out in the first step is subtracted by the shading correction image Q (n−1) created in the fourth step, thereby A shading correction image is sequentially obtained from images cut out by the fifth step of creating a shading correction image Z (n) of the image P (n),
Inspection method for moving sheet-like continuum. The threshold value process is performed on the shading correction image Z (n) created in the fifth step in order to detect a defect, and the feature amount of the defect candidate is compared with a preset reference value for quality determination. The method for inspecting a moving sheet-like continuum according to claim 1, wherein defects are identified and identified. The method for inspecting a moving sheet-like continuum according to claim 2, wherein a value corresponding to a defect size or shape is used as a preset reference value for quality determination. The method for inspecting a moving sheet-like continuum according to any one of claims 1 to 3, wherein the sheet-like continuum is a resin-impregnated sheet obtained by impregnating a reinforcing fiber with a resin. The method for inspecting a moving sheet-like continuum according to any one of claims 1 to 5, wherein the line sensor is installed in a direction normal to the surface of the sheet-like continuum.

Images