JPS6340843A - Surface flaw detection - Google Patents

Abstract

PURPOSE:To achieve a highly accurate decision on the authenticity, type and harmfulness of a flaw, by extracting flaw images in a certain range of an area in preference centered on the flaw probably with the highest degree of harmfulness to perform a signal processing. CONSTITUTION:A writing controller 15 outputs a quantization signal to a device 13 synchronizing a scan start signal 11 of a scanner 7 together with a length measuring pulse 12 obtained according to the movement of a material 3 to be inspected through a touch roll 8 and a pulse generator 9 to quantize a scan signal 10 to be stored into a memory 14. On the other hand, the signal 10 is differentiated with a device 16 to be compared with multiple stages of flaw signal level by a device 17 and a flaw detection signal 18 is outputted into an image extraction controller 19. When a signal 18 generates at a point 24 within a range 21 defined by a scanner scanning range W and a desired length, the controller 19 is made to stand-by to extract a flow image of (n)X(m) pixels corresponding to an area 25 centered on the point 24 from the memory 14 up to the m'-th line following a line containing the point 24. When a signal 18 generates at a point 26 higher than at the point 24, it is made to standby to extract an area 27 centered on the point 26.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for detecting surface flaws on a moving inspected material, such as a steel piece or a steel plate.

[Conventional technology]

As shown in FIG. 5, a method for inspecting the surface of a moving inspected material, such as a rolled steel plate, involves directing the laser beam IA from a laser oscillator 1 in a direction perpendicular to the moving direction of the inspected material 3 using a rotating mirror 2. There is a method in which the reflected light from the surface of the material to be inspected 3 is focused by a lens 4 and converted into an electrical signal by a photoelectric converter 5 for inspection. A method using a -dimensional scanner is well known, such as a method in which the surface of the material 3 is irradiated with a constant intensity illumination 6 and inspected by scanning with a linear array sensor 7 whose field of view is set perpendicular to the direction of movement of the material 3 to be inspected. It is.

The surface flaw detection method used in the above detection method compares the scanning signal of the scanner or the differential signal of the scanning signal with a predetermined flaw signal level, and detects a flaw if it exceeds the flaw signal level. Then, the degree of harmfulness of the flaw is determined based on the magnitude of the scanning signal or its differential signal. One of the main purposes of product surface inspection
The first is quality assurance for consumers, and harmful defects (missed or undetected) are a major problem. Therefore, in order to increase the detection sensitivity, lowering the above flaw detection level may result in detecting flaws that are sufficiently low in harmfulness that do not need to be detected (overdetection), or detecting things that are not flaws as flaws (false detection). Become,
In the above detection method, it is difficult to suppress over-detection and erroneous detection by the detection device and increase the detection rate (number of flaws detected by the detection device/number of flaws to be detected). This is because the degree of damage to a flaw cannot be determined simply by the magnitude of the scanning signal or its differential signal, and furthermore, the criteria for determining this differ depending on the type of flaw.

Conventionally, surface flaw inspections have been performed visually by inspectors, so the quality of automatic surface flaw inspections is often evaluated by comparing the results with visual inspections. However, the examiner captures the flaws occurring on the surface of the inspected material as a two-dimensional image, intuitively performs pattern recognition to determine the type of flaw, and further determines the degree of harmfulness of each type of flaw. Under these circumstances, in automating surface flaw inspection, it is necessary to capture flaws as images and perform advanced signal processing such as image processing and pattern recognition processing to first determine the type of flaw and then determine its degree of harmfulness. Conceivable.

The problem with this case is that when attempting to process an image of the entire surface of the material to be inspected, the amount of data to be processed becomes enormous, and no matter how high-performance a signal processing device is used, it cannot be processed within a practical time.

Therefore, it is necessary to suppress the amount of data to be processed to the minimum necessary. As a method, the scanning signal of the scanner is quantized into several levels to create a flaw distribution pattern on the surface of the material to be inspected, and histograms are created in the length and width directions within a certain section length in the direction of movement of the material to be inspected. A method of determining the type and harmfulness of defects using the characteristic quantities extracted from the histogram (Japanese Unexamined Patent Publication No. 59-99238), or scanning the surface of the material to be inspected with a TV camera and determining the differential value of the scanning signal. An image of a limited area with the specified point on the screen where the signal exceeds a predetermined flaw signal level is stored in a storage device, and further image processing and pattern recognition processing are performed to determine the authenticity of the flaw and the type and size of the flaw. A detection method (Japanese Unexamined Patent Publication No. 59-138904) and the like have been proposed.

When the inspected material is required to have a beautiful surface quality, such as a stainless steel plate, the individual size of the surface flaws to be detected may be very small, and there are many types of flaws to be judged. The level of toxicity is also subdivided. On the other hand, there are various situations in which flaws occur, such as when they occur singly, when multiple flaws occur locally, or when they occur over a wide area. It may be decided. However, in the former method of the prior art described above, an image covering a relatively wide area determined by the scanning width of the scanner and a predetermined interval length is processed in real time, so the resolution (number of pixels) of the image is low, and the result obtained by processing the image is low. There is a problem in that the number of feature values that can be detected is limited, and therefore only a rough type of flaw can be determined. In addition, in the latter method, images in a limited area are processed, so it is possible to process high-resolution images in real time, but since it does not have a priority image cutting function, storage Even if it is possible to prevent some oversights by increasing the capacity of , real-time processing performance will inevitably deteriorate. Furthermore, there is a problem in that it is not possible to accurately determine the degree of harmfulness of a flaw using only an image of a relatively narrow limited area.

[Problem that the invention seeks to solve]

As mentioned above, in order to detect surface flaws with high precision, it is necessary to perform sophisticated signal processing in real time, but in order to reduce the amount of data to be processed, only general but general information is processed. Or if it is detailed (using only a limited range of information does not provide enough information to judge the type of flaw or its degree of toxicity), images of flaws that occur in chronological order may not be processed. This results in serious defects being overlooked and real-time processing performance being reduced, making it difficult to provide a practical surface flaw detection method.

The present invention solves these problems by scanning the entire surface of the strained material to be tested and detecting defects that are suspected to be defects. By preferentially extracting detailed images of defects in a certain area, mainly those that are considered to be highly harmful, the amount of data to be processed can be kept to the minimum necessary without overlooking serious defects. This is a surface flaw detection method that performs advanced signal processing to make it possible to determine the authenticity of flaws, the type of flaw, and the degree of harmfulness in real time. This is a surface flaw detection method that makes it possible to further improve the accuracy of determining the type and degree of harmfulness of flaws by constructing and processing a rough flaw distribution image that serves as the background.

Means 1 for Solving Problem C] FIG. 1 shows an example of the configuration of an apparatus for carrying out the present invention. - Scanning the surface of the inspected material 3 using the dimensional scanner 7; On the other hand, in synchronization with the length measurement pulse 12 corresponding to the amount of movement of the inspected material obtained via the touch roll 8 and pulse generator 9 and the scan start signal 11 of the scanner 7, the writing controller device 15 sends a signal to the quantization device 13. By outputting a quantized signal, the scanning signal 10 is quantized, and the write microcontroller 15 transfers the quantized data of the scanning signal 10 to the frame memory 14 while shifting line by line in synchronization with the length measurement pulse 12. Remember.

On the other hand, the scanning signal lO is differentiated by a differentiator 16 and compared with a predetermined multi-stage flaw signal level by a comparator 17, and a flaw is detected at a level corresponding to the flaw signal level exceeded by the differential signal of the scanning signal 10. By outputting the signal 18 to the image cutout controller device 19, the occurrence of something suspected to be a flaw is notified.

In response to the flaw detection signal 18, the image cutout controller 19 extracts n×m images centered on the position on the frame memory 14 corresponding to the position on the surface of the inspected material where a suspected flaw has occurred.
In order to cut out the pixel defect image, the system waits for a predetermined number of measurement pulses. While waiting to cut out the flaw image, a higher level flaw detection signal is sent to the image cutout controller device 1.
9, the cutout of the former flaw image is stopped, and in order to cut out the flaw image corresponding to the latter flaw detection signal, the system waits again for a predetermined number of length measurement pulses, and during that time, further processing is performed. If a high-level flaw detection signal is not input, the latter flaw image is cut out and transmitted to the signal processing device 20. The signal processing device 20 performs image processing and pattern recognition processing to determine the authenticity of the flaw, as well as the type and degree of harmfulness of the flaw.

Cutting out the flaw image will be further explained with reference to FIG.

When a flaw detection signal is generated at a point 24 in a range 21 surrounded by the scanning width W of the scanner and an arbitrary length of the surface to be inspected, the image cutting controller 19 detects n In order to extract a pixel defect image from the frame memory 14, the process waits for m' lines after the line including point 24. If a higher level flaw detection signal does not occur during that time, the limited area 25
For example, if a flaw detection signal of a higher level is generated at point 26 than at point 24,
The former cutting out of the flaw image is stopped, and the process waits again to cut out the flaw image corresponding to the limited area 27 centered on the point 26. By repeating these processes, it is possible to preferentially cut out an image of a suspected flaw with the highest level within a certain scanning section length.

FIG. 3 shows another example of the configuration of an apparatus for carrying out the present invention. The difference from Fig. 1 is that in addition to the functions in Fig. 1, it has a function to extract and process rough flaw distribution information, and in parallel with cutting out the flaw image in Fig. 1, it is leveled. The flaw detection signal 18 is input to the pixel editing controller device 29. The pixel editing controller device 29 synchronizes with the pixel sampling signal of the flaw distribution image created by the pixel counting device 28 based on the scanning start signal 11 and length measurement pulse 12, and detects paper within the range corresponding to the pixel. The highest level among the signals is stored as the value of that pixel in the pixel memory 31 together with its position information. The pixel readout controller device 30 receives an image cutout signal input from the image cutout controller device 19, and reads flaw distribution data within a certain length of a certain section (FIG. 4) including the cutout image from the pixel memory 31. It is read out and transmitted to the signal processing device 20. The signal processing device 20 uses the flaw distribution data to construct a flaw distribution image of N×M pixels of a certain section length, performs image processing to obtain background information of flaws, and obtains the processing result of the eaves image in FIG. It is also used to determine the type of flaw and its degree of toxicity. FIG. 4 shows the relationship between the pressure area image and the flaw distribution image. Here, the size of the limited area 32 (width W, length /), the number of pixels of the eaves image (n x m), the length of the range 33 of a certain section length and the number of pixels of the flaw distribution image 33 (N x M) These parameters are determined based on the statistical properties of the defects that occur.

[Example] An example in which the present invention is applied to detecting surface flaws in the final process of producing stainless steel sheets will be shown below.

The flaw signal level in the comparison device is ±3 levels, the area of the limited area on the surface of the inspected material is 32" x 64 fins, the number of pixels of the corresponding eaves image is 64 x 64, and the density level is 256.
level, the number of waiting lines for image extraction is 1 line, the capacity of the frame memory is 512 pixels x 32 lines, the section length on the surface of the inspected material represented by the flaw distribution image is 1 m, the number of pixels in the image is 64 x 200, and the gray level is Set various quantities such as 3 levels, SUS 304.2B material, A feet width x 1
00m length and SUS 430, BA material, 1m width x 100m
As a result of long flaw detection, in the former case, the number of flaws detected was 47.
, number 36 determined to be flaws by signal processing, total bottom length 22m by visual inspection,
On the other hand, in the latter case, the number of defects detected was 193, the number of defects determined to be defects by signal processing was 63, and the total length of the bottom part was 29 m compared to the total length of the bottom part of 28 m, which showed good agreement with the visual test. Met.

〔Effect of the invention〕

By using the present invention for surface flaw detection, advanced signal processing can be executed in real time by minimizing the amount of data to be processed without overlooking serious flaws, making it practical for automatic detection. A significant improvement in accuracy can be expected while maintaining the same.

[Brief explanation of the drawing]

1 and 3 are diagrams showing an example of the configuration of an apparatus for carrying out the present invention, FIG. 2 is a diagram illustrating preferential extraction of eaves images in the present invention, and FIG. 4 is a diagram showing an example of the eaves part image in the present invention. FIGS. 5 and 6, which are diagrams explaining the relationship between images and flaw distribution images, are diagrams showing an example of a surface flaw detection method using a one-dimensional scanner.

Claims (2)

[Claims] (1) In a method of detecting surface flaws on a moving inspected material using a one-dimensional scanner, a flaw detection signal is generated by comparing the scanning signal of the scanner or the differential signal of the scanning signal with multilevel flaw signal levels. On the other hand, the scanning signal is quantized in synchronization with a length measurement pulse corresponding to the amount of movement of the material to be inspected, and is stored in a frame memory as an image of the surface of the material to be inspected with a certain section length while shifting one line at a time. Waiting for a predetermined number of length measurement pulses to cut out a flaw image of n×m pixels centered on the position on the frame memory corresponding to the position where the flaw detection signal is generated, and waiting to cut out the flaw image. To,
When a higher level flaw detection signal is generated, the process stops waiting to cut out the flaw image, waits again for the predetermined number of length measurement pulses to cut out the flaw image corresponding to the latter flaw detection signal, and then waits for the predetermined number of length measurement pulses. If a higher-level flaw detection signal is not generated during the waiting time, the latter flaw image is cut out, and the cut out flaw image is signal-processed to determine the authenticity of the flaw, the type of flaw, and the degree of harmfulness. A surface flaw detection method characterized by: (2) Among the flaw detection signals that are within the range corresponding to pixels larger than the pixels of the flaw image created based on the length measurement pulse corresponding to the amount of movement of the inspected material and the scanning start signal of the scanner. The highest level is stored as the value of that pixel in a pixel memory along with its position information, and a flaw distribution image of a certain section length on the surface of the inspected material including the flaw image is constructed from the values of the pixel memory, and the flaw distribution is 2. The surface flaw detection method according to claim 1, wherein information obtained by signal processing the image is used to determine the type and degree of harmfulness of the flaw.