JPS6130685B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a signal processing method for detecting surface defects in hot steel materials. Recently, direct rolling (HDR) is being implemented in the blooming process as part of energy-saving measures, and along with this, instead of the conventional cold inspection, harmful surface defects are accurately caught and dealt with in the hot process. It has become necessary to do so. If it is possible to detect surface flaws on a steel billet that is in a red-hot state after blooming and rolling, it will be possible to adjust the amount of HS (hot scarf), remove hot spots, select quality, and provide feedback to the previous process. It has many potential applications and has great benefits in terms of yield, quality improvement, labor saving, and energy saving. Therefore, it has already been proposed to discriminate the surface flaws of a hot steel piece by the amount of radiant light from the surface, and it is said that this can be sufficiently discriminated. Generally speaking, when a high-temperature steel material with sufficient self-retained heat is cooled almost uniformly in the air compared to its surroundings, heat conduction differs depending on the conditions near the surface, and the surface temperature visible from the outside is caused by defects, scale, etc. This is because certain parts of the body appear different from normal parts. In other words, areas where there are openings such as cracks are bright because the inside of the material, which is at a higher temperature, can be seen, and areas where something has peeled off from the surface, such as scratches, scale, or deposits, are cooled and darken (black). )appear. However, when automatically recognizing the presence of a flaw in a certain part, it is necessary to detect that the part is different from a normal part in some way, but this is difficult to detect on the surface of a hot steel piece. In this case, there are two difficulties. First of all, an actual red-hot steel billet does not have a uniform surface temperature distribution, and the temperature is generally lower at the corners, and the amount of water applied during rolling varies.
Due to various factors such as the degree of HS application, there are quite large temperature irregularities on the surface. Second, even if the distribution of temperature spots can be detected and a reference pattern can be obtained, it is by no means possible to extract defects from the deviations. Due to the high temperature of the heated steel billet, scale is rapidly generated, and foreign matter such as deposits and dirt from the rolling process are also distributed, and these appear black like the sagging defects that should be detected. To deal with these difficulties, a water-cooled mechanical brushing pre-treatment device and a line hood to shut out the convection around the hot steel are installed just before the defect inspection location, and a standard In order to obtain a pattern (i.e., dynamic threshold), it is sufficient to adopt a signal processing method in which the input signal is peak-held two-dimensionally and then weighted averaged two-dimensionally. With equipment and line hoods,
It is possible to remove most of the foreign matter such as deposits and dirt, and it is also possible to supply the material to inspection with most of the relatively large scales removed.According to the latter signal processing method, scale and defects can be removed. Even if there are scratches, it is possible to obtain a normal pattern based on the temperature irregularities inherent in the material. Therefore, by adopting these methods, it became possible to treat the target as a deviation from the standard pattern with large scales and deposits removed, but in actual materials, there are also dot-like scales. and band-shaped scales remain,
They usually appear black rather than harmful bald spots. Furthermore, even if the same damage occurs, a harmless damage that is likely to fly away during the hot rolling process when it is about to peel off is more likely to occur due to heat conduction than a more harmful damage that has roots in the steel base. Looks black. Therefore, if a determination is made simply based on the magnitude of the deviation from the reference pattern, a large number of scales will be mixed in, and only a determination that has little correlation with the degree of harm will be output. In view of these conventional problems, the present invention captures radiant light from a high-temperature test material such as a hot steel piece with an imaging device, and detects scale and surface defects from the radiant light image in real time. This system is designed to distinguish between sagging defects and output defects, and to output judgments according to the degree of harmfulness of the defects. According to the experience of conventional hot visual inspection, if a person who is skilled in inspecting defects looks carefully, it is possible to roughly distinguish between scale and deformation, and in the case of deformation, whether it is harmful or harmless. This means that the radiant light image contains information that can distinguish between some kind of baldness flaw and scale, and determine the degree of harmfulness. Therefore, when we recorded the hot radiant light images and the distribution of actual defects in the cold for a large number of materials and put them together, we found the following. First, the signal waveform for the scale generally has a narrow convergence, as shown in FIG. There are some cases with wide convergence, but in that case the signal is deep (black).
There seems to be a correlation between width and depth, which is consistent with the physical consideration of the scale generation cooling process shown in FIG. That is, the scale is made up of layers of FeO, Fe ₃ O ₄ , Fe ₂ O ₃ , etc. in order from the one closest to the base iron. When the scale is small enough at the beginning, it has good adhesion to the steel base and does not cool down too much, but as it gradually grows and the area becomes larger, the unbalanced growth of each layer causes it to peel off from the base steel. happens.
As the flaking progresses, heat conduction between the base iron and the scale deteriorates, making it difficult to recover heat and causing the scale to cool down. Therefore, the larger the area, the more exfoliation progresses, and the more the scale cools and becomes black. Next, the signal waveform corresponding to the bald spot generally has a wide width, as shown in FIG. Regarding the depth, highly harmful heges, that is, those with roots in the ground, do not grow very deep (do not turn black), as shown in Figure 4. In other words, it doesn't get very cold. On the other hand, less harmful bald spots that are about to peel off tend to get cold easily, so they become quite deep. Sufficiently large scratches generally have a considerable amount of pitch black areas. From the above, it seems appropriate to use the width and depth of the signal waveform as parameters for distinguishing between scabs and scales and determining the degree of harmfulness of scabs. Therefore, we looked at the distribution of about 500 scales and about 30 bald spots, with width on the horizontal axis and depth on the vertical axis [see Figure 5]. The results showed that, as expected, scale has a fairly strong correlation with width and depth, that the scale can be clearly separated from the sludge, and that the direction (area) in which the most harmful sludge is distributed, etc. It became clear. Therefore, by setting a certain functional relationship involving the width and depth of the signal, it is possible to obtain an output that takes adversely into account. In view of these points, the present invention focuses on the parameters of the width and depth of the signal in order to extract the baldness flaw from the beam radiation image, and the two parameters of the width and depth of the signal are This includes a function that corresponds to the degree of harmfulness of the target. That is, the feature of the present invention is that a video signal is obtained by capturing a radiant light image of a hot steel material, and from this video signal, parts below a preset reference level Po are individually extracted, and the signal depth is determined for each part. This is a signal processing method for surface defect detection of hot steel materials in which P and signal width W are calculated and whether or not there is a defect is determined based on a set of these parameters (P, W). At least one or more parts are taken out, and the image signal of the surface is compared with the actual flaw to measure the presence or absence of flaws or the degree of flaws in the part of the hot steel material corresponding to each of the parameter sets (P, W). By the way,
A defect data map corresponding to each parameter set (P, W) is created, and from this defect data map, the defect identification equation (P, W) = W/α + {β (|P-Po|)} = (P, W) When ≧1, it is determined to be a defect. When (P, W) < 1, it is determined that it is not a defect. However, α, β, and γ are determined by preliminary experiments using the defect data map, and then , the surface defects of the hot steel material supplied thereafter are determined by substituting the parameter set (P, W) calculated from the video signal of the material into the defect identification equation. The defect identification equation described above is basically derived based on the following empirical facts, as will be clear from the explanation below. The degree of harmfulness of a defect is proportional to the size of the defect.
Therefore, it is considered to be proportional to the width (W) corresponding to the size of the defect. The blackness of the flaw is related to the extent to which the flaw has peeled off from the base material. Furthermore, the greater the degree of peeling, the lower the degree of harm. That is, it is considered to be more ∝W. The relationship between the blackness (｜P-Po｜) and the degree of toxicity is due to thermal conduction,
Since nonlinear relationships are involved, such as the relationship between heat transfer, peeling, and degree of toxicity, it was assumed that the general formula ∝1/α+{β(|P−Po|)} can be approximated. That is, blackness (|P−
Po |) is assumed to be a variable, α: bias β: constant of proportionality γ: inversely proportional to a function whose power can be determined arbitrarily. Actually, when α, β, and γ in the above equation are set to appropriate values, the flaw data map (for example, Fig. 5, 12
It was found that (P, W) effectively separates scale and sludge defects, or ordinary sludge defects and highly harmful sludge defects with roots close to cracks. Next, an example of the signal processing method according to the present invention will be explained with reference to FIGS. 6 and 7. FIG. 6 is a block diagram showing the entire signal processing system, and FIG. 7 is a block diagram showing details of its main parts. As shown in FIG. 6, an appropriate optical system (lens, filter) is used to capture a radiant image of red-hot hot steel material 1, and an imaging device 2 such as an ITV camera or a solid-state imager is used.
A video signal VS ₁ is obtained by. Next, this video signal
After storing VT ₁ in a storage circuit 3 and converting it into a digital signal VS ₂ using an automatic sensitivity adjuster 4 and a sample-hold A/D converter 5, a line delay circuit 6 for performing line delay and a reference pattern are detected. The signals are sent to the reference pattern detection circuit 7 for the purpose of determining the raw video signal, and subtracted by the subtraction circuit 8 to obtain a deviation signal αVS of the raw video signal from the reference pattern. The hege extraction circuit 9 includes a comparator 10 and a signal depth detection circuit 1.
1, a signal width detection circuit 12, an arithmetic circuit 13, etc., and operates as follows. First, the deviation signal α
A comparator 10 compares VS with a set value Po. As shown in FIG. 7, the signal width detection circuit 12 has gates 14,
A width counter 15 and a width buffer 16 are provided, and the width counter 15 counts clock pulses only during the time when the deviation signal αVS is smaller than the set value Po.
temporarily memorized. As shown in FIG. 7, the signal depth detection circuit 11 includes a comparator 17, a buffer 18, a selector 19, a gate 20, and a depth buffer 21, and the deviation signal αVS from the comparator 10 is sent to the signal width detection circuit 12. At the same time, it is also sent to this signal depth detection circuit 11, and this signal depth detection circuit 11 successively calculates the minimum value P while the deviation signal αVS is smaller than the set value Po, and calculates the result. It is temporarily stored in the depth buffer 21. Now, when the deviation signal αVS becomes larger than the set value Po, the gate signal G becomes logic 0, and P and W are read out from the depth buffer 21 and the width buffer 16 at the timing of its fall, and the PROM (programmable read) of the arithmetic circuit 22 is read out. Only memory) address input terminal. On the other hand, the gate signal is directly input to the reading terminal of the PROM, and at the timing of the falling edge of the gate signal, the functions (P, W) listed in advance in the PROM are read out as defect data in the defect buffer 2.
It is stored in 3. , this determination is made as in Experimental Example 1.
This can be achieved by storing it in PROM, or by storing it in PLA (Program Logic Array).
It can also be realized by a multi-function operator (analog signal). In this way, if there is a signal such as a scuff mark or scale, it can be output in real time, that is, within one clock time after the signal ends, as a value comparing the scuff spot with its harmfulness, and the calculated result is 23
The data is sent to the CPU 24 via the CPU 24, and after high-level judgments such as calculation of the total flaw area and determination based on the defect position, the data is sent to the display unit 25, monitor 26, printer 27, etc., and is also sent to the external device (flaw removal device). used for control. Defects (scratching defects) in the radiant light image of heated steel materials are
They exist in various properties with unique halftones (gray) related to the various sizes and degrees of peeling caused by their occurrence, and they cannot be distinguished simply because the signal is large, but the real According to the invention, all such conventional problems can be solved. Below, we will look at the effects using specific examples. First, a reference pattern signal a' is created for a video signal of a radiation image as shown in FIG. 8a, and the deviation signal between the two is b. This signal includes one bald spot and many scales. Then, the result of performing the signal processing of the present invention is c.
As you can see immediately, the scale has been sufficiently suppressed and the bald spots have been emphasized. Next, in the case of a radiation image as shown in FIG. 9a, this signal includes many scales as well as deep-rooted harmful bald spots and less harmful bald spots that are about to peel off. If we focus only on blackness (signal depth), we may end up producing an output that is opposite to the degree of harmfulness, and scale will also be included. Further, by simply focusing on the width of the signal, the degree of harmfulness cannot be determined so clearly. However, according to the present invention, as can be seen from result c,
It is possible to obtain output commensurate with the degree of harmfulness of the sagging flaw. Next, an experimental example will be shown. Experimental example 1 Hot steel material: Slab (capped steel) Hot steel material temperature: 1150℃ (measured at the center of the surface of the material) Imaging device: 2048-bit solid-state imager ΓSpecific method Solid-state imager from the radiant light image of the hot slab The video signal obtained is converted into an 8-bit digital signal VS through an amplifier, an automatic sensitivity adjuster, and a high-speed sample-and-hold A/D converter. After that, on the one hand, a reference pattern signal that follows the temperature irregularities specific to the material is generated.
In order to extract NOR, peak hold processing is performed two-dimensionally, and then an average is taken over a certain fixed area. On the other hand, the digital signal VS is delayed by a shift register due to the time required to extract the reference pattern signal. A deviation signal αVS is obtained by subtracting the delayed signal and the reference pattern signal. From this deviation signal αVS, the parts below the reference level Po (=-0.1V) determined by considering the noise level (〓50mV) are extracted individually, and the signal depth P (V) is obtained.
The signal width W (mm) (value converted to width on the material surface) is calculated (see Fig. 10). First, attention is paid to several scales and parts that appear to be defects (sludge marks), and the harmfulness of the above parameter set (P, W) is investigated.
The flaw data map sampled in this manner is shown in FIG. In this Figure 11, 6 bald spots and 7 scales are sampled.
Therefore, α, β, and γ of the discrimination curve L ₁ W/α+{β(|P−Po|)}=1 (Equation 1.1) that separates the set of scab marks and the set of scales are determined. In this case, α=
2, β=7.5, γ=2. Therefore, the differential equation is (P, W) = W/2 + {7.5 (|P-0.1|)
}〓(Formula 1.2) When (P, W)≧1, it is a hegemonic state.When (P, W)<1, it is not a hegemonic state. Once the identification equation is determined in this way, the set of parameters (P, W) calculated from the video signal using the above method for the same steel type.
By substituting (Equation 1.2) into (Equation 1.2), it is possible to immediately determine whether or not there is a defect. Specifically, the contents of (Equation 1.2) are PROM when the values of P and W are used as addresses.
This is realized in the form of data output. The situation is shown in Figure 12. When each of the 62 pairs of (P, W) calculated from the newly input video signal was identified using (Equation 1.2), good results were obtained as shown in the following table. Therefore, a significant improvement in the detection rate can be expected compared to the conventional determination based only on signal depth or signal width.

[Table] Experimental example 2 Heated steel material: Slab (semi-killed steel) Heated steel material temperature: 1200℃ (measured at the center of the surface of the material) Imaging device: 2048-bit solid-state imager depth P
(V), and the signal width W (mm) is calculated. Then, a flaw data map is created by focusing on several scales and baldness (Fig. 13). In this Figure 13, there are 8 hege flaws (3 of them are R type, 4 are T type, and 1 is F type).
type) and the distribution of 10 scales are given.
From this distribution, α, β, and γ of the discrimination curve L ₂ W/α + {β(|P−Po|)}=1 (Equation 2.1) to separate the scale set and the hege set are determined. Ru. In this case, α=
7.5, β=7.5, and γ=2. Therefore, the discrimination equation is (P, W)=W/7.5+{7.5(|P-0.1
|)} ² (Formula 2.2) When (P, W) ≧ 1, it is a hegemonic state. When (P, W) < 1, it is not a hegemonic state. The situation is shown in Figure 14. In this case as well, as in Experimental Example 1, distributions that could not be separated by using only the signal depth or only the signal width as measurement parameters were successfully separated, and a high detection rate could be obtained. In addition, among the scabs, there are scabs that are closer to cracks and have roots that do not get cold easily and therefore do not turn black (R type), normal scabs (T type), and scabs that are about to peel off (F type). Yes, and if only the highly harmful R-type baldness is to be detected, another discrimination curve exists. That is L ₃ in Figure 13. That is, α
=7.5, β=7.5, and γ=3. When the discrimination curve is used in this way, it becomes possible to make a judgment according to the degree of harmfulness of the defect. As described in detail in the embodiments above, according to the signal processing method according to the present invention, composite parameters such as width and depth are extracted from the signal, and a function determined according to the actual state of the defect is used. By doing so, it is possible to obtain an output that corresponds to the degree of harmfulness, and its practical value is extremely significant.

[Brief explanation of the drawing]

Figure 1 is a signal waveform diagram of the scale, Figure 2 is a diagram showing the process of scale generation and cooling, Figure 3 is a signal waveform diagram of a bald spot, and Figure 4 is a signal waveform of a highly harmful bald spot. 5 is a distribution diagram of the signal width and depth, FIG. 6 is an overall block diagram showing an embodiment of the present invention, FIG. 7 is a block diagram of the hege extraction circuit, and FIGS. Figure 9 is a signal waveform diagram, Figure 10 is a signal waveform diagram in Experimental Example 1, Figure 11 is a flaw data map for determining the identification curve _L1 ,
Figure 2 shows the determination of defects using the identification curve L ₁ , and Figure 13 shows the identification curve L ₂ in Experimental Example 2.
FIG. 14 is a diagram showing defect determination using the identification curves L ₂ and L ₃ . DESCRIPTION OF SYMBOLS 1... Hot steel material, 2... Imaging device, 5... Sample hold A/D converter, 7... Reference pattern detection circuit, 8... Subtraction circuit, 9... Hege extraction circuit,
11...Signal depth detection circuit, 12...Signal width detection circuit, 13...Arithmetic circuit.

Claims (1)

[Claims] 1. Obtaining a video signal by capturing a radiant image of a hot steel material,
Based on this video signal, a preset reference level is
Signals in surface defect detection of hot steel materials where the parts below Po are individually extracted, the signal depth P and signal width W are calculated for each, and whether or not there is a defect is determined based on the set of these parameters (P, W). The processing method includes first treating at least one of the target hot steel materials in advance.
In order to measure the presence or absence of flaws or the degree of flaws in the parts of the hot steel material corresponding to each of the parameter sets (P, W), by comparing the video signal of the surface with the actual flaw. Therefore, create a flaw data map corresponding to each parameter set (P, W),
From this defect data map, the defect identification equation (P, W) = W/α + {β (|P-Po|)} = When (P, W) ≧ 1, determines a defect. When (P, W) < 1, it is not a defect. However, parameters α, β, and γ are determined in a preliminary experiment using a flaw data map, and then the surface defects of the hot steel material to be supplied are detected using the video signal of that material. 1. A signal processing method for detecting surface defects in hot steel materials, characterized in that discrimination is made by substituting a set of parameters (P, W) calculated from the above into the defect identification equation.