JPH02298842A - Detecting method of defect - Google Patents

Abstract

PURPOSE:To detect a defect highly precisely without being affected by a processing by smoothing characteristic data, by estimating a change in a characteristic due to a processing applied to a material to be inspected, and by detecting the change on the basis of a differential between the estimated change and original data. CONSTITUTION:A density signal picked up by a TV camera 61 is inputted to an image processing unit 3 and a monitor TV 8 by a switching unit 7. When the camera 61 picks up an image of the surface of a steel plate having a flaw, the unit 3 prepares a differential image by using said image and a smoothed image and further subjects the differential image to binary-coding, so that a binary-coded image be obtained. Accordingly, data corresponding only to a flaw varying at a high frequency are obtained from the differential image, and by subjecting these data to binary-coding, data specifying the position of the flaw in a sub-scanning direction are obtained. By detecting a change in the physical characteristic of a material to be inspected, in this way, a defect can be detected highly precisely without being affected by a processing applied to the material to be inspected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a method for detecting defects in a material to be inspected. For example, the present invention is applied to detecting flaws on the surface of steel material after rolling.

[Prior art]

For example, on the surface of steel after rolling, there are
This may result in flaws that look like thin pieces of steel attached. Conventionally, such flaws were detected by visual inspection by an inspector waiting on the production line.

[Problem to be solved by the invention]

However, such visual inspection requires skill because it targets steel materials transported on a production line, and also imposes a heavy burden on the inspector.

Therefore, attempts have been made to image the surface of a steel material using a TV camera or the like and automatically detect surface defects through image processing. In other words, although high-temperature steel material emits light by itself, defects are detected by paying attention to the fact that heat recovery after cooling is slower in the outermost part than in other parts.

Although various problems arose when putting this type of defect detection device into practical use, the present applicant et al. filed an application and published Japanese Patent Publications Nos. 61-18694 and 57-52983.

No. 57-48735 and pending patent application 1986-26
Since these problems have been successively solved by the techniques disclosed in No. 1548 and the like, efforts are now being focused on further improving defect detection accuracy.

Therefore, an object of the present invention is to provide a highly accurate defect detection method that can automate the detection of defects in materials to be inspected.

[Means to solve the problem]

In order to achieve the above object, in the present invention, characteristic data indicating the physical characteristics of a predetermined detection range of a test material subjected to a predetermined process in a predetermined direction is placed at a two-dimensional position in the detection range. The feature data is extracted in correspondence, smoothed in a direction corresponding to the direction of processing, and the difference data between the smoothed data and the original feature data is obtained, and the data is applied to the test material based on the difference data. Detect defects that occur.

[Effect]

In this case, the test material is subjected to a predetermined process in a predetermined direction, so changes in the physical characteristics of the test material due to the process are almost continuous; When a defect occurs, the change becomes discontinuous. For example, when the material to be tested is a steel material that has been rolled and cooled, the temperature of the steel material changes gradually in the length direction, but the surface temperature changes in defective areas because heat recovery after cooling is delayed. becomes steep.

In other words, according to the present invention, the feature data is smoothed in the direction of processing, predicts changes in physical features that are planned to be applied to the test material, and calculates the difference between the prediction and the original feature data. Since the process detects changes in the physical characteristics of the test material that were not planned based on the process, defects can be detected with high accuracy without being affected by the processing applied to the test material. can.

Other objects and features of the present invention will become apparent from the following description of embodiments with reference to the drawings.

〔Example〕

FIG. 1 schematically shows a defect inspection apparatus for steel materials, which implements the present invention as an example. This device includes a system controller 1. Input/output device 22 image processing units 3-1 to 3-4 (
When there is no need to indicate each individually, use "3" to represent them. Same for everything else. ), camera controller 4-1
~4-4. Motor controllers 5-1 to 5-4. Imaging units 6-1 to 6-4. Switching unit 7, monitor TV
8. It consists of a steel material detection sensor BSEN, a conveyance speed sensor 5SEN, etc.

The imaging units 6-1 to 6-4 exit the heating furnace, rolling mill, etc. (not shown), are cooled with water from a water sprinkler, descaling device, etc., and are moved by arrows by a steel material conveying device (also not shown). Each of the six imaging units each captures an image of the left and right flange surfaces and the upper and lower web surfaces of a steel material (in this case, an H-beam) BM that is conveyed at a line speed V in the direction of the line speed V. Slit 62. ITV camera 61
and a case 63 that accommodates the slit 62. A rail 64 that movably supports the ITV camera 61 and the slit 62. It also includes a drive device or the like that changes the positions of the ITV camera 61 and the slit 62 independently.

Case 63 is a heat-resistant dark box with transparent glass.
It has $631 and a reflecting mirror 632.

The window 631 is provided on the flange surface on the right side (in the conveyance direction) of the H-shaped steel BM in the case of the unit 6-1, and in the case of the unit 6-2.
In the case of the unit 6-3, the opening is parallel to the left flange surface, in the case of the unit 6-3, the opening is in the upper web surface, and in the case of the unit 6-4, the opening is parallel to the lower web surface. deflects the light incident through the window 631 toward the ITV camera 61 without twisting it.

The ITV camera 61 consists of a focal plane shutter and a row of light receiving elements, and each light receiving element receives an electric signal (hereinafter referred to as a density signal) according to the received light energy.

“High concentration” corresponds to “high energy”), 64
It is output in 0x480 pixel divisions (in this, a rask scan is performed in which main scanning is performed in the horizontal direction and sub-scanning is performed in the vertical direction, and the sub-scanning direction is parallel to the conveyance direction of the steel material@).

The slit 62 corrects the bias in the basic energy distribution on each surface of the steel material BM. For example, in H-beam steel, the accumulated energy at the boundary between the web and the flange is high, so the heat recovery after cooling occurs faster in that area than in other areas. Therefore, the web surface has a temperature distribution as shown at the left end of FIGS. 2a and 2b, and the flange surface has a temperature distribution as shown in FIGS. 2c and 2b.
It has a temperature distribution as shown at the left end of Figure d.

Therefore, in the unit 6-3.6-4 that images the web surface, the radiant energy from both ends of the web is attenuated by the slit 62-3.62-4, and the ITV camera 61-4 is
In the unit 6-1.6-2 which averages the light-receiving energy of the light-receiving element row of 61-4 and images the flange surface, the radiant energy from the center of the flange is attenuated by the slit 62-1.62-2. ITV camera 614.61-
The light received energy of the two light receiving element rows is averaged. These slits are driven by a drive device (not shown) and are positioned sufficiently close to the ITV camera to effectively average the received light energy of the array of light receiving elements.

In this example, the slit used in the unit for imaging the web surface is constructed of two slit plates shaped as shown in FIG. It consists of a single slit plate with the shape shown in Figure 3b (replace it as necessary if the steel material to be measured is different).For example, in the inspection of defects in steel sheet piles, it is used as a unit to image the flange surface. (Sometimes two slit plates are used for the slit.)

These slit plates are moved up and down by a drive device (not shown).

The motor controller 5 receives a command from the system controller 1 and operates a motor for changing the position of the ITV camera 61, a motor for changing the position of the slit 62, and a motor for changing the position of the slit 62.
controls the motor for changing the position of the slit plates, and positions each of them at the designated position.

The camera controller 4 controls the shutter of the ITV camera 61 in response to instructions from the system controller 1, and drives the shutter at the instructed timing and at the instructed speed.

The density signal output by the ITV camera 61 is branched into two directions by the switching unit 7, one of which is sent to the image processing unit 3.
and the other is selectively provided to monitor TV 8 .

The image processing unit 3 includes a dedicated microcomputer 301 and various arithmetic circuits, as shown in FIG. Here, for example, the ITV camera 61 is
Assuming that a steel material surface with flaws as shown in the figure is imaged, a difference image as shown in Fig. 7a is created using this image and the smoothed image shown in Fig. 6a, and then The image is digitized to obtain a binary image as shown in FIG. 8a.

In other words, if we focus on the sub-scanning line Qv parallel to the conveyance direction of the steel material passing through the flaw in the image and examine the density change in that direction, we can see that the low frequency changes from the image shown in Figure 5a to the one shown in Figure 5b Data is obtained by combining the waveform corresponding to the temperature unevenness that changes at high frequency and the waveform corresponding to the flaw that changes at high frequency, and from the image shown in Figure 6a, the temperature changing at low frequency as shown in Figure 6b is obtained. Data corresponding to unevenness can be obtained. Therefore, from these differential images shown in Figure 7a, data corresponding only to the flaws that change at high frequencies is obtained, and this is
By converting the data into values, data for specifying the position of the flaw in the sub-scanning direction as shown in FIG. 8b can be obtained.

The smoothed data is density data CI of pixels on the sub-scanning line.
This is predicted data in which the density data of the pixels adjacent to the lower side is predicted by the first-order lag element and the second-order lag element while sequentially capturing data obtained by digitally converting the density signal of the TV camera 61 by the A/D converter 311. . In other words, the difference data is data that was not predicted from the upper continuous density data. Specifically, the number of main scanning pixels (6
Using line buffers 327 and 328 for 40 pixels), a constant (×α) multiplication circuit 322 and an addition circuit 323 calculate the first-order lag element, and an addition circuit 324 calculates the second-order lag element. However, the calculation is omitted here using the sampling period as a unit. ), constant (Xβ) multiplication circuit 325 and addition circuit 326
The predicted data is calculated by This prediction data is
Difference data is obtained by subtracting the density data of the pixel of interest (pixel of interest), and the calculation is performed in the subtraction circuit 321.

Note that if the density data of the pixel of interest is 0 (z) and the predicted data (smooth data) at that time is P (z), then (Z
indicates the sub-scanning address. ), first-order lag element S (z)
teeth.

5(z)6P(z)+α(0(z)-P(z))
”(1) The second-order lag element S' (z) is: S'(z)=S'(z-1)+(0(z)-P(z))
-12) Prediction data P of the next pixel adjacent to the bottom
(z) is P (z) = S (z) + β・S'(z
) =” (3).

The difference data is compared with binarization thresholds st and h given by the microcomputer 301 in a comparator circuit 331, and when the density is higher (brighter) than that, rz 1
'1. When it is low (dark), it is binarized as II O$7. This binarized data is input to the counter 332.

The counter 332 is reset by the sub-scanning synchronization signal and counts up at the falling edge of the binary data "ON." This count data is stored in the shift register 332 at the end of each main scan.
33. Therefore, when the processing of one frame is finished, the shift register 333 stores the binary image '''.
Data is obtained by projecting a 0" pixel onto one sub-scanning line. For example, a
t Opt 1. Assuming that a binary image including flaw 2 is obtained, the shift register 333 has the tth
As shown in the ob diagram, vertical projection data F (
z) is obtained (however, each vertical projection data is specified by a sub-scanning address Z). The microcomputer 301 compares this vertical projection data F (z) with a predetermined threshold value Wt, and continuously sets the threshold value Wt in the sub-scanning direction.
For each data group exceeding h (corresponding to the binary data in Figure 10c obtained by binarizing the histogram in Figure 10b), the maximum number of pixels in the main scanning direction is the "width", and the number of pixels in the sub-scanning direction is the "width". "height".

After extracting the feature amount whose "position" is the central sub-scanning address, this feature amount is further examined based on the flaw determination criteria given by the system computer 1, and each group is detected.

Further, in the image processing unit 3, the vertical projection data F
While detecting (z), average density data M is generated. This average concentration data M is calculated by I as shown in FIG. 9a.
Among the pixels on the main scanning lines Q1 and Q2 that are apart by a specified value P/2 above and below from the center of the imaging screen of the TV camera 61, the density data is averaged for those whose density data exceeds a predetermined value δ, It is used to adjust the shutter speed of the ITV camera 61. Here, δ is a value that divides the concentration distribution on line Q1 or Q2 into a background part and a steel part as shown in FIG. 9b when the shutter speed is appropriately set.

The average concentration data M is calculated by the gate circuit 341 shown in FIG.
．． Addition/subtraction circuit 342. Buffer 343. Comparison circuit 344
, the counter 345 and the microcomputer 301. The comparison circuit 344 compares the density data and the predetermined value δ.
The gate circuit 341 generates a gate signal that permits "gate opening" when the concentration data exceeds δ by comparison with δ.
give to The gate circuit 341 opens the gate and sends the concentration data to the addition/subtraction circuit 342 only when "gate opening" is permitted by this gate signal. Addition/subtraction circuit 3
42 and a buffer 343 accumulate the differences between the sent density data and the provisional average density data D to generate difference cumulative data D. During this time, the counter 345 counts the number of pixels for which "gate opening" is permitted, and generates count data N. Note that the provisional average density data D is calculated by the addition/subtraction circuit 3.
This is used to reduce the number of bits in 42 bits.

Buffer 343 and counter 345 are reset by the sub-scanning synchronization signal.

In the microcomputer 301, the Q1 line and 0
At the end of main scanning of 2 lines, read the cumulative difference data D and count data N, and calculate the average D/ of the cumulative difference data D.
N is determined and a provisional average density Do is added thereto to generate average density data M.

Next, the overall operation of the device of this embodiment will be explained.

Please refer to FIG.

The system controller 1 receives the type code and rolling number of the steel material to be measured by the operator via the input/output device 2.
etc. are input, the position of the ITV camera 61 and shutter speed (initial value) are determined by referring to the data stored in advance.
and imaging conditions regarding the position of the slit 62 and the position of the slit plate, as well as the binarization threshold sth and the flaw determination standard (W
I L, W2L.

HIL, H2L), etc. are set, and the motor controller 5. The data is transferred to the camera controller 4 or the image processing unit 3 and a standby mode is set. As a result, the motor controller 5 positions the ITV camera 61, the slit 62, and the slit plate at specified positions, and the camera controller 4 sets the specified shutter speed. Further, the microcomputer 301 of the image processing unit 3 initializes the input/output board, internal registers, etc., registers the image processing conditions, provides the binary positive value sth to the comparator circuit 331, and sets the standby mode.

After that, when the steel material detection sensor BSEN detects the tip of the steel material to be measured, the system controller 1 and the microcomputer 301 of the image processing unit 3 start up, and the system controller 1 takes an image according to the conveyance speed of the steel material detected by the speed sensor 5SEN. Interval (interval at which image processing is performed)
Set. This imaging interval is based on the conveyance speed of the steel material of 4 m.
When the speed exceeds /s, the frequency is set to 15Hz, and when the speed is less than 4m/s, the frequency is set to 10Hz. In other words, ITV camera 6
1 captures one screen every 1/30 seconds, so when the imaging interval is set to 15 Hz, the image is captured every other screen as shown in Figure 12a, and when the imaging interval is set to 10 Hz, the image is captured every other screen, and when the imaging interval is set to 10 Hz, the image is captured every 12th frame.
As shown in Figure b, image processing is performed for every second screen.

Furthermore, the system controller 1 calculates the effective area according to the conveyance speed of the steel material. This effective area limits the range of flaw determination.

The length Va in the sub-scanning direction is the conveyance speed of the detected steel material V, the shutter speed of the focal plane shutter v', the imaging period fs, the length of one pixel Cps, and the overlap rate ε. When Va=v −fs−”C
+ip-"v'/(v+v')・to (4)
is given by The effective area is Figure 12a and Figure 12b.
As shown in the figure, it is set for each sampling screen (screen that performs image processing), and the effective areas before and after overlap with an overlap rate of ε0, so any flaws on the imaged surface of the steel material are included in one of the effective areas. It will be done.

After setting the effective area based on the above equation (4), the system controller l registers the length Va in association with frame N02 (the number of the sampled screen), and also registers the sub-scanning address Vs of the upper end of the area and the lower end of the area. The sub-scanning address Ve is transferred to the microcomputer 301 of the image processing unit 3 to instruct the start of image processing.

When the microcomputer 301 receives an instruction to start image processing, the microcomputer 301 performs 1. At that time, image processing is performed on the screen imaged by the ITV camera 61. FIG. 13 is a flow chart showing an average density data detection subroutine performed at the beginning of image processing. This subroutine runs on line Q1 or Q.
It is activated at the end of the second main scan.

As described above, at the end of each main scan, the differential cumulative data D is provided from the output end of the buffer 343, and the count data N is provided from the output end of the counter 345. Therefore, the microcomputer 301 selects lines Q1 and Q2.
At the end of the main scan, the differential cumulative data D and count data N are read, and D/N + Do is calculated to obtain the average density data of each line, and then the average density data of each line is averaged. Once the average density data M of the screen is determined, it is transferred to the system controller 1.

If this average density data M is less than a predetermined value, the system controller 1 determines that the steel material has not reached the imaging position (the material detection sensor BSEN is installed upstream from the imaging position). The camera waits for the next screen to be captured, and if the captured image exceeds a predetermined value, it determines whether the shutter speed is appropriate. When determining whether the shutter speed is appropriate,
Average density data M and preset reference level D 1 p
D 2'vUx', tJ2 (however, Dl < Dl
<Ut <U2), when the average concentration data M is below the reference level D2, it is ll-2u, when it exceeds Dl and below D1, it is 'r 1 + D. When it exceeds D1 and below U1, it is 11Q. When exceeding H2U1 and below U2, re + 1'', when exceeding U2, It +
An evaluation value of 271 is set. Here, the sign of the evaluation value indicates overexposure or underexposure (minus corresponds to underexposure, plus corresponds to overexposure), and the number indicates the degree of overexposure or underexposure. Therefore, for example, when an evaluation value of 11 211 is set, the exposure is significantly insufficient, so the camera controller 4 is instructed to lower (slow down) the shutter speed by two steps.

When the evaluation value at+Ipr is set, the exposure is somewhat excessive, so the camera controller 4 is instructed to increase (fasten) the shutter speed by one step.

Thereafter, the system controller 1 instructs the microcomputer 301 of the image processing unit 3 to execute flaw detection processing.

FIGS. 14a and 14b are flowcharts showing a subroutine of flaw detection processing. This subroutine is activated when the system controller 1 issues a command to execute the flaw inspection process and screen capture is completed. This process will be explained step by step below.

As mentioned above, at the end of screen capture, the counter 33
Vertical projection data F (z) is output from the output terminal of No. 3, so this data is first read in step 2, and then stored in registers H, W, and G in step 3.

Clear (0) Z, i and 2.

Register G is a flag for detecting a group of vertical projection data F (z) that continuously exceeds the threshold value wth in the sub-scanning direction. If this value is O, register 2 (corresponding to the sub-scanning address) is sequentially incremented to search for vertical projection data F (z) that exceeds the threshold value wth (steps 4, 5, 13.14).

During this period, vertical projection data F (
z) is found, in step 6 the value of register G is set to 1 and register i is incremented by 1, and in step 7 the register W(i) specified by the value of register i (hereinafter referred to as register W(i) The value (initially O) of data F (z) is compared with the data F (z). At this time, the value of register W(i) is the data F (
z), the data F (z) is stored in register W(i) in step 8, the value of register H(i) is incremented by 1 in step 9, and the value of register 2 is incremented in step 13. Increment by 1 and return to step 4.

This time, the value of register G is 1, so step 10
The vertical projection data F (z) is compared with the threshold value wth, and steps 10, 7 to 9, 13, 14 and 4 are repeated until the vertical projection data F (z) becomes less than the threshold value wth. The projection data F(z) is the threshold wth
If the value is below, register G is cleared in step 11, and in step 12, the value obtained by subtracting 1/2 of the value of register H(i) from the value of register 2 is set to register Z (i
). That is, at this point, among the group of vertical projection data F (z) exceeding the threshold value wth, i
The width of the th group (maximum number of pixels in the group)" is stored in register W(i), the ``number of pixels in the sub-scanning direction of the height group)" is stored in register H(i), and the 9th place I! (sub-scanning address of the center pixel of the group)” is stored in register Z(1).
Each is stored.

The above processing is performed on all of the vertical projection data F (z), and the vertical projection data F (z) exceeding the thresholds wt and h are
The feature values of the groups are set in registers W(i), H(i
) and Z (i), each group is then examined closely by comparing the feature amounts and the flaw determination criteria.

The number of groups detected at this time is indicated by the value of register i, so in step 15 the value of register i is saved to register A, and in step 16 register Ay,
Clear (0) Aht By and Bh and register 1
．． Store 1 in y j and k.

As mentioned above, the range from the sub-scanning address Vs to the sub-scanning address Ve is set as the valid area by the system controller 1, so first, the value of register i is updated and stored in register Z (i). A group of vertical projection data F (z) whose position data falls within this range is searched for (step 17.26.27).

Vertical projection data F in which position data is included within the effective area
When the group (z) is found, the width data of the group stored in register W(i) is compared with the first width determination criterion WIL in step 18, and the width data is stored in register H(i) in step 19. The height data of that group is compared with the first height determination standard HIL. In these comparisons, the width data is the first width criterion WIL
or the height data exceeds the first height judgment standard HI
When it exceeds L, it is determined that the group of vertical projection data F (z) corresponds to a class A flaw, and in step 20, register W(i) is set in register Aw(j).
and the value of register H(i) in register Ah(j), respectively, and in step 21, the value of register j is incremented by one.

Also, the Hata data of the group being examined at this time is the first
Width judgment standard WIL or less, and height data is the first
If the height criterion is HIL, further step 2
2 and 23, the cap data of the group is compared with the second width criterion W2L, and the height data is compared with the second height criterion H2L (however, WIL>W2L, HIL
> H2L), and in this case, when the width data exceeds the second width judgment standard W2L and the height data exceeds the second height judgment standard H2L, the vertical projection data F
The group (z) is determined to correspond to class B defects, and in step 24, the value of register W(i) is set in register By(k), and the value of register H(
i), and in step 25, the value in the register is incremented by one.

The above examination is performed on all groups of vertical projection data F (z), and the feature values (width and height) of the detected class A flaws are stored in registers Ai/ and Ah.
The characteristic quantities (width and height) of the flaw are registered in registers BV and B.
After sorting the data into h, in step 28, those data are stored (uploaded) in association with the frame number.

The above process is repeated at each set imaging cycle until the steel material detection sensor BSEN detects the absence of steel material and the average density data M becomes equal to or less than a predetermined value (no steel material is present at the imaging position).

The system controller 1 controls the microcomputer 3 of the image processing unit 3 when the average density data M becomes equal to or less than a predetermined value following the steel detection sensor BSEN detecting the absence of steel.
01 to notify the end of the process. As a result, the microcomputer 301 transfers the data related to the flaw registered in association with the frame number, so the system controller 1 uses the effective area length Va stored in association with the frame number. Organize that data,
The input/output device 2 stores each group, its classification, and feature values in correspondence with the distance from the tip of the steel material and the distance from the rear end.
The image is displayed on a CRT display and also printed out via a printer.

In the above embodiments, the bias in the basic energy distribution on the imaging surface of the steel material is corrected by the slit, but this may also be done by masking the output signal of the ITV camera 61.

However, since the temperature difference that occurs in the optical part is small for the entire energy range of the imaging surface, care must be taken when selecting an ITV camera when correcting the bias in the energy distribution by signal processing.

In addition, although an example in which the present invention is applied to detecting flaws on the surface of steel materials using imaging with an ITV camera has been described, it is also possible to apply the present invention to defect detection using physical characteristic data of the test material collected by other means. can be similarly applied. For example, if data collected by an X-ray camera is processed according to the present invention, internal defects can be detected with high accuracy without being affected by processing applied to the test material.

〔Effect of the invention〕

As explained above, in the present invention, feature data indicating the physical characteristics of a predetermined detection range of a test material subjected to a predetermined process in a predetermined direction is associated with a two-dimensional position of the detection range. The feature data is smoothed in a direction corresponding to the direction of processing, the difference data between the smoothed data and the original feature data is obtained, and the difference data generated in the test material is calculated based on the difference data. Detecting defects.

In other words, the feature data is smoothed in the direction of processing,
Predict changes in the physical characteristics of the test material that are planned by the process applied to the test material, and predict changes in the physical characteristics of the test material that were not planned by the process based on the difference between the prediction and the original feature data. Therefore, defects can be detected with high accuracy without being affected by the processing applied to the test material.

Note that the device shown in the example has high reliability in actual flaw detection, with no oversights or overdetections of 1% or less (ratio to the number of products).

[Brief explanation of drawings]

FIG. 1 is a schematic diagram showing the configuration of a steel defect detection apparatus that implements one embodiment of the present invention. FIGS. 2a to 2d are explanatory diagrams for explaining correction of bias in energy distribution of H-shaped steel tapping. FIGS. 3a and 3b are plan views showing the specific shape of the slit 62 used in the embodiment device. FIG. 4 is a block diagram showing the detailed configuration of the image processing unit 3 included in the embodiment apparatus. Figures 5a, 5b, 6a, 6b, 7a,
FIG. 7b, FIG. 8a, and FIG. 8b are explanatory diagrams for explaining the outline of flaw detection processing performed by the image processing unit 3. FIGS. 9a and 9b are explanatory diagrams for explaining the average density data detection process performed by the image processing unit 3. FIG. 10a to 10c are explanatory diagrams for explaining extraction of flaw characteristic data performed by the image processing unit 3. FIG. FIG. 11 is a flowchart showing the overall operation of the embodiment device. FIG. 12a and FIG. 12b are explanatory diagrams for specifically explaining the setting of the imaging interval. FIG. 13 is a flowchart showing the average density data detection process performed in one image processing unit 3. 14a and 14b are flowcharts showing flaw detection processing performed by the image processing unit 3. FIG. 1 System controller 2: Input/output device 3: Image processing unit 4: Camera controller 5: Motor controller 6: Imaging unit 7: Switching unit 8: Monitor TV Applicant Nippon Steel Corporation Agent Patent attorney Nobuoki Sugi: ,'/ Door 3a figure ↓ East 3b figure 9a figure A#24 peak Atsuchoru 21 stuttering voice figure 11

Claims (1)

[Claims] Extract feature data indicating the physical characteristics of a predetermined detection range of a test material subjected to a predetermined process in a predetermined direction in association with the two-dimensional position of the detection range, and process the feature data. Smoothing in a direction corresponding to the direction of adding the data, and obtaining difference data between the smoothed data and the original feature data
A defect detection method for detecting a defect occurring in a test material based on the difference data.