JPS6142222B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for automatically inspecting defects on the surface of cold- or hot-rolled steel sheets, and particularly to a method for determining the type of defects, particularly those that occur in conventional surface defect inspection equipment. The present invention provides a method for detecting and processing defects that enables the application of an automatic surface inspection device to replace visual inspection in an inspection process in which defect evaluation differs depending on the type of defect and appropriate evaluation is difficult.

First, to give an overview of conventional optical surface defect equipment, conventional equipment consists of a light projector that illuminates the surface of a material to be inspected, and a light that is reflected from the surface of the material to be inspected and adds information representing the characteristics of the defect. It consists of a light receiving section that receives the received signals and converts them into electrical signals, and a processing section that processes the electrical signals obtained from the light receiving section to determine the presence or absence of a defect and the size of the defect. Normally, inspection uses changes in light reflectance at minute portions of the steel plate surface as basic information, so either or both of the luminous flux of the light emitter or the field of view of the light receiver is sufficiently narrowed down, and instantaneous extremely It is configured to measure surface reflection in a narrow field of view, and is configured to sequentially move or scan the field of view in order to inspect the entire surface of the steel plate. The main abilities required of such an optical defect inspection apparatus are the ability to discover defects, the ability to determine the harmfulness of the discovered defects, that is, the defect level, and the ability to identify the type of defect.

As mentioned above, in conventional inspection equipment, the basis of information is based on the light intensity reflectance at minute portions on the surface of the steel sheet, and the presence of defects is estimated from changes in the obtained reflectance signal. Normally, it is possible to distinguish between noise components and defects on the surface by amplifying the output signal of the light receiving device with an appropriate preamplifier and then extracting it as is, or by extracting only the output change using a differentiator and level-discriminating it with an appropriate threshold value. and the degree of harmfulness (level) of the defect. However, the determination of the degree of harmfulness of defects differs for each type of flaw depending on the purpose of use of the steel sheet, and
Since standards established on the basis of human visual judgment are used, it is extremely difficult to discriminate the detection level of the change in surface reflectance of the defect. That is,
According to the level setting value of the judgment circuit based on the level setting value of the judgment circuit based on a certain defect, there is a very high possibility that the judgment will be erroneously made regarding other types of defects. Therefore, in order to determine the level of surface defects, it is important to recognize the type of defects. The type of defect is the nature of the defect,
They are empirically classified according to their causes, etc.
When the purpose is to determine the defect level, it is possible to group similar items together into several groups. It is clear that if it is possible to identify which of these groups a defect belongs to, it will be relatively easy to determine the level of the defect.

In identifying these types of defects, defects are determined instantaneously by visual observation, taking into account the shape, color, site of occurrence, etc. of the defects empirically, and this function belongs to the field of so-called pattern recognition. Visual inspection differs greatly from conventional optical defect inspection systems in that it recognizes defects not only by their partial reflection characteristics but also by their planar aggregation, ie, defect images.

The present invention detects defects as a two-dimensional image instead of the conventional one-dimensional detection method, performs signal processing, and captures the characteristics of the defect in order to provide a defect inspection device with performance close to that of the above-mentioned visual inspection. This method identifies the type of defect.

The present invention will be explained below with reference to the drawings. FIG. 5 shows an example of a basic configuration of a defect inspection device based on the present invention. In the figure, 1 is a steel plate which is a material to be inspected, and 2 indicates defects occurring thereon. 3 is a light source for illumination. 4 is an optical system that converts the reflected light from the illumination at the defect into an image such as a light image; 5 is an imaging unit that converts the image converted by the optical system 4 into an electrical signal; and 6 is an image pickup unit that converts the image converted by the optical system 4 into an electrical signal; a two-dimensional signal processing unit that performs processing for extracting the features;
is a feature parameter identification determination unit for determining the flaw type from the post-processing information obtained by the processing unit 6;
8 is a level determination unit that determines the magnitude of each characteristic parameter obtained; 9 is a defect determination unit that determines the type and degree of the defect from the level correlation of each characteristic parameter; 10 is a display unit that displays the determined defect; This is an output display device that notifies the operator and outputs defect information to an inspection information processing device (not shown in this figure).

In this device, a defect image is captured by an optical system 4 at an imaging unit 5.
The signal is connected to the electrical signal, converted into an electrical signal, and sent to the signal processing section 6. In the signal processing section 6, after performing necessary preprocessing such as removal of an average value and trend on the defect image information, two-dimensional frequency analysis processing and defect area and length calculation processing are performed in parallel. The output 13 of the signal processing unit 6 is an image of the defect image on the spatial frequency axis, that is, four parameters based on the two-dimensional frequency component distribution pattern, and three parameters based on the area and length calculation method.
There are two parameters.

The four parameters based on the two-dimensional frequency component distribution pattern will be explained below. First, FIG. 1 shows an example in which frequency analysis was performed on two-dimensional image data. The results of frequency analysis (two-dimensional Fourier transform or Hadamard transform) on the grid-shaped two-dimensional image data in Figure 1 a are shown in Figure 1 b.
It is a dimensional frequency component distribution pattern (two-dimensional power spectrum density). FIG. 1c shows the frequency component distribution of a standing wave in the transverse direction, and FIG. 1d shows it in the longitudinal direction. It is well known that frequency analysis of time series data extracts the periodicity of changes contained in time series data, but frequency analysis of image data extracts the periodicity of changes contained in image data on a two-dimensional plane. It extracts sexuality. In the example shown in Figure 1, the horizontal periodicity appears mostly on the axis where the vertical frequency is 0, and conversely, the vertical periodicity appears mostly on the axis where the horizontal frequency is 0. (This indicates that there is no vertical and horizontal correlation in the image data). These are called a horizontal standing wave frequency component and a vertical standing wave frequency component. When frequency analysis is performed on the surface of a steel plate as image data, most of the frequency components due to surface defects are generally 2.
The frequency analysis results of the steel plate surface where the high frequency components are concentrated in two directions, and where surface defects have occurred compared to the normal steel plate surface, show that the frequency components are concentrated in the horizontal or vertical direction (or both directions) depending on the change due to the surface defects. ) of the standing wave frequency components, the frequency components in a relatively high band become large. If the characteristics of each surface defect can be associated with the pattern of the two-dimensional frequency component distribution, it is possible to conversely identify the surface defect from this two-dimensional frequency component distribution pattern.

Now, the frequency components calculated by digitally performing two-dimensional frequency analysis on the image data are expressed as P(f ₁ ,
f ₂ ) (f ₁ : Horizontal frequency coordinate: −Nrf ₁
Nr, f ₂ : Vertical frequency coordinate: 0f ₂ Nw (Nr, Nw are determined from the Nyquist theorem in image data sampling). Four parameters are extracted using P(f ₁ , f ₂ ) to describe the two-dimensional frequency component distribution. Not all P(f ₁ , f ₂ ) are used for parameter extraction;
It is only necessary to use components near the standing wave frequency components in both directions ({P(0, f ₂ )} and {P(f ₁ , 0)}). Conversely, using neighboring components prevents ignoring frequency components due to surface defects that are buried in slow changes in both directions.

Specifically, the standing wave near frequency components Pr(f ₁ ) and Pw(f ₂ ) are defined from P(f ₁ , f ₂ ) using the following equations.

r and w correspond to slow changes in the original image data, and it has been experimentally found that it is appropriate to select 2 to 4.

Parameters DFr and DFw that express the directionality of surface defects using Pr (f ₁ ) and Pw (f ₂ ), and parameters that express the sharpness of surface defects in the horizontal and vertical directions.
Extract Rr and Rw. DFr, DFw, Rr, Rw are 2
There are four parameters that describe the dimensional frequency component distribution.

An example of an actual analysis result is shown in FIG. 2 to explain the four parameters. DFr and DFw are given by the following definition formulas.

For example, in the analysis results shown in Figure 2, frequency components with high power are concentrated in the vertical direction compared to the horizontal direction (in the original image data, changes in the horizontal direction are slow and changes in the vertical direction are large). ), {Pr(f ₁ )}
has a smaller amplitude than {Pw(f ₂ )}. In other words, DFw=0, the frequency component of the standing wave in the vertical direction has a large power, and DFr<0 (DFr in Fig. 2).
= -1), and it can be quantitatively understood that the power of the frequency component of the horizontal standing wave is small.

Next, a parameter indicating whether or not there are frequency components due to surface defects in the vertical and horizontal directions is determined.
Rw, Rr. Vertical direction as shown in Figure 2 c and b,
Once the horizontal standing wave frequency components are extracted, it is necessary to determine in which frequency band the components due to surface defects are located. In general, surface defects such as point defects and linear defects produce large changes in the vertical or horizontal direction (or both), so relatively high frequency components in the relevant direction often become large ( (See Figure 2). The numerical values of this are Rw and Rr. The definition formulas for Rr and Rw are given by the following formulas.

n _r and n _w can be determined by calculating the spatial frequency from the size and spacing of the target surface defects.

Therefore, Pr(0), Pr(1),...Pr(n _r −
1) and Pw(0), Pw(1), ......Pw(n _w −
1) is considered to be a frequency component not caused by surface defects, for example, due to dirt on the surface of the steel plate, and Pr(n
_r ), Pr (n _r +1), ......Pr (Nr) and Pw
(n _w ), Pw (n _w +1), Pw (Nw) can be considered to be frequency components due to surface defects.

The two-dimensional frequency analysis method identifies defect types based on changes in image data, that is, the periodicity (repeatability) of surface defects, and is not necessarily an effective method for defects that occur only one time. do not have. The defect area and length calculation method compensates for the shortcomings of the two-dimensional frequency analysis method for identifying defects that occur singly.

This method directly uses electrical signals sent from the imaging device, and this will be explained with reference to Figure 3. Figure 3a shows a large number of video signals corresponding to each horizontal scanning line of a television camera. Figure b shows a signal that has been thresholded and ternarized using threshold levels indicated by dotted lines.
This ternary signal is used to open and close the gate and control the clock input to the counter to detect defects.
By knowing the positions where N ₁ and N ₂ exist, it is possible to determine whether or not they are the same defect, and to determine the length and area of the defect. The area and length calculation method is performed on the same image data in synchronization with the two-dimensional frequency analysis method. With this method, multiple area and length information can be obtained for one image data, but among them, the length L of the defect with the largest area and the positive value in ternarization are used to identify the defect type. Indicates negative (reflected light scattering defect, reflected light absorbing defect)
There are two ATRs.

By setting an image field of appropriate size on the steel plate surface, it is possible to evaluate the six parameters extracted by the two signal processing methods described above and whether the steel plate surface is dirty due to the occurrence of defects. It is possible to identify the type of defect present in the image using the parameter iRiS relating to the level of the image electrical signal. FIG. 4 specifically shows an example of a method of combining the seven parameters used in the present invention. This involves performing an off-line test using a defective sample, calculating seven parameters, organizing the data, determining boundary values for each parameter, and constructing identification logic. The physical meanings of the parameters used in FIG. 4 are as described above, and the characteristics of each surface defect used for identification are outlined below.

CB oyster defects: Linear defects extending in the rolling direction Rough surface scale: Point defects that absorb reflected light Co-slip defects: Point defects or linear dense defects that scatter reflected light Normal surface: Uneven gloss (reflected light Surface roughness scale α: Typical rough skin scale (patterned) co-slip flaw α: Collective co-slip flaw CB oyster flaw α: Multiple oyster flaw (oyster flaw-like) co-slip flaw α :Collective co-slip flaw CB oyster flaw β: Multiple co-slip flaw oyster flaw-like co-slip flaw β: Collective co-slip flaw rough skin scale β: Weak rough skin scale in the width direction CB oyster flaw γ: Single Rough skin scale γ: Rough skin scale that is weak in the width direction For example, CB oyster scratch β is a multiple linear defect extending in the rolling direction, and the parameter
It has a two-dimensional frequency component distribution as shown in Figure 2, which is an example of the analysis results cited to define DFw, DFr, Rw, and Rr. That is, DFw=0, DFr=-1,
and Rw40 are required (calculation of Rr is unnecessary in the identification flow). Furthermore, since the electric signal of the defect is clear and has a high level, iRiS>8 holds true, and the length of the defect in the rolling direction is also long, so L40 holds true.
Therefore, classification can be performed correctly using the identification flow shown in FIG. Taking rough skin scale α as an example, since this defect is a point-like defect that absorbs reflected light, the electric signal changes drastically due to the defect in both the rolling direction and the width direction, and the required two-dimensional frequency component distribution Therefore, frequency components with high power are distributed in both the rolling direction and the width direction, and furthermore, frequency components with high power are distributed in both directions up to the high band. That is, DFw=DFr=0 and RwRr
40 is required. Furthermore, since it is clear that iRiS>8 and ATR1<0 hold true, this can also be correctly classified using the identification flow shown in FIG. Regarding ATR, the area and length calculation method is 1
The sum of multiple optical property parameters ATR obtained for one image data was used as the judgment parameter as ATR1, and the optical property parameter ATR of the defect with the largest area was directly determined as ATR2. This was unnecessary for the identification flow in FIG. And the fourth
As a result of testing approximately 50 samples using the identification flow shown in the figure, we obtained a 95.9% concordance rate between automatic flaw type identification and visual judgment.

It should be noted that there is a limit to the screen resolution of commonly used image-to-electrical signal converters, such as image pickup tubes such as vidicon used in TV cameras, or image conversion devices using solid-state image sensors that have a large array of photodiodes, charge transfer devices, CCDs, etc. In order to perform a full-width inspection of the surface of a steel plate, it is necessary to line up a plurality of image conversion devices or to sequentially move (scan) the field of view of the image conversion device in the amplitude direction.
Signal processing is performed by a normal digital computer system, but since signal processing takes a considerable amount of time, all signal processing is performed only on the image of the portion where the defect exists.

As described above, according to the present invention, it is possible to identify the types of surface defects, which was impossible with conventional surface defect inspection equipment, and the accuracy of determining defect levels is improved, as well as quality control before and after the process using classification information for each defect type. It has become possible to completely automate the conventional visual inspection for surface defects.

[Brief explanation of the drawing]

Figures 1 a to d are illustrations of two-dimensional frequency analysis, Figures 2 a to c are illustrations of the two-dimensional frequency component distribution pattern and parameters DFr and DFw, and Figure 3 is calculation of defect area and length parameters. A diagram explaining the procedure,
FIG. 4 is a flowchart explaining the procedure for defect type determination, and FIG. 5 is an explanatory diagram of the defect detection device. In the drawings, 1 is a steel plate to be inspected, 2 is a defect, 3 is a light source for illumination, 4 is an optical system for imaging the defect to an image conversion unit 5, 5 is an image conversion unit that converts the defect image into an electrical signal, 6 7 is a two-dimensional signal processing section, 7 is a feature parameter identification/judgment section, 8 is a level judgment section, 9 is a defect judgment section, and 10 is an output display device.

Claims (1)

[Claims] 1 The surface of the steel plate to be inspected is illuminated, the surface reflected light is imaged and photoelectrically converted, and 2
Perform dimensional frequency analysis to determine the vertical direction of frequency components,
A parameter that expresses the directionality of surface defects by evaluating the power level from the distribution in the lateral direction.
DFw and DFr, and the existence ratio of high frequency components to the whole band in the vertical and horizontal directions, respectively, are calculated to obtain parameters Rw and Rr that express the sharpness of surface defects in the vertical and horizontal directions, and the image electrical signal from the length L of the defect with the largest area in the image,
The ATR indicates the optical property of whether the defect is a reflected light scattering defect or a reflected light absorbing defect, depending on the positive or negative electrical signal level of the defect present in the image, and whether the steel plate surface is dirty due to the occurrence of the defect. A method for inspecting surface defects on a steel sheet, characterized in that three parameters of iRiS indicating an electrical signal level are obtained for evaluating whether or not the surface is damaged, and the type of defect is determined using these seven parameters.