JP3209297B2 - Surface flaw detection device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a surface flaw discriminating apparatus for judging the grade of a surface flaw generated on the surface of a strip-shaped sheet material such as a steel sheet.

[0002]

2. Description of the Related Art In general, the grade of a surface flaw generated on the surface of a steel strip is greatly affected not only by the flaw depth but also by the type of flaw. That is, since the degree of damage varies depending on the type of flaw, the type of flaw must be accurately determined. Therefore, JP-A-4-
As shown in Japanese Patent Publication No. 142412, a first neural network for determining a flaw type from a detection signal detected by an optical surface flaw detection device, a detection signal of the optical surface flaw detection device, and a first neural network. There is known a second neural network that determines a flaw grade from a flaw type determination signal from a neural network.

However , since a single neural network is used to determine which type of flaws are generated on the surface of a steel sheet based on a detection signal from an optical surface flaw detection device, a large-scale neural network is required. As a result, learning for discriminating the flaw type takes time, and in some cases, the learning does not converge. Also , when adding a new flaw type to the neural network, re-learning must be performed and it takes enormous learning time, and when trying to cover all flaw type discriminations with one large-scale neural network, in the unwarranted neural net that optimal solution can be obtained not converge the learning within a finite time, there is a problem that could lead to practical use impossible, yet to determine all of the flaws species by one of the neural network, computer However, there is also a problem that the load is large and it takes time to determine the type of flaw.

[0004]

The problem to be solved by the present invention is to eliminate the above-mentioned problems, shorten the time for flaw type discrimination, and ensure that learning of the flaw type discrimination neural network converges. An object of the present invention is to eliminate re-learning of a flaw type discrimination neural network accompanying addition of a new flaw type.

[0005]

SUMMARY OF THE INVENTION The present invention is connected to an optical surface flaw detection device for detecting a surface flaw and calculates a detection signal from the optical surface flaw detection device to output a flaw signal intensity. The arithmetic unit and a large number of flaw type discrimination neural nets for discriminating whether or not each type of flaw is connected to the optical surface flaw detection device and determining whether or not the flaw is the flaw type are connected to the flaw grade determination unit. It is characterized by the following.

[0006]

According to the present invention, the surface of a running strip is scanned by an optical surface flaw detection device, and when a surface flaw is detected on the strip, a large number of flaw type discrimination neural discriminates for each discrimination flaw type. net outputs the flaw grade determining unit discrimination signal by determining whether the detected surface flaws signal or identical to that learned flaw type patterns, respectively. At this time, the intensity of the flaw is calculated from the detection signal of the optical surface flaw detection device , and the flaw signal intensity is calculated.
The flaw signal strength is input to the flaw grade discriminating section by the output computing section. The flaw grade determining unit comprehensively determines the degree of damage and the size of the surface flaw according to the flaw type from the flaw type discrimination signal output from the flaw type discrimination neural network and the flaw signal intensity from the calculation unit. The determination of the flaw grade of W is performed.

[0007]

Next, the present invention will be described in detail with reference to the illustrated embodiment. Reference numeral 1 denotes an optical surface flaw detection device. The optical surface flaw detection device 1 includes a laser oscillator 2, a rotating mirror 4 that scans a laser beam from the laser oscillator 2 on the surface of the band-shaped plate member, and a band-shaped plate member W. The photoelectric conversion device 5 into which the reflected light is incident, a signal preprocessing unit 6 that performs a filtering process on the detection signal converted into an electric signal by the photoelectric conversion device 5, and an output from the signal preprocessing unit 6. Analog-to-digital converter 7 for digitizing the detection signal
And a memory 8 for taking in the digital signal of the analog-to-digital converter 7.

[0008] 10 ₁ to 10 _n is an optical surface flaw detection device 1
The flaw type discrimination neural net 10 is connected to the memory 8 of the above. The flaw type discrimination neural nets 10 each learn and store a flaw type pattern corresponding to a different flaw type. Numeral 11 calculates the flaw intensity from the detection signal taken into the memory 8 of the optical surface flaw detection device 1 and outputs the flaw signal strength.
It is an arithmetic unit for inputting data. Reference numeral 12 denotes a flaw grade determination unit connected to the arithmetic unit 11 and the flaw type discrimination neural network 10. The flaw grade determination unit 12 receives flaws input from the memory 8 of the optical surface flaw detection device 1 via the calculation unit 11. The damage grade and the size of the flaw are comprehensively determined from the signal strength and the flaw type determination signal input from the flaw type determination neural network 10 to determine the flaw grade of the strip-shaped plate material. Reference numeral 13 denotes a motor for rotating the rotating mirror 4.

[0009] With the above-mentioned structure, the rotating strip 4 scans the traveling strip with laser light emitted from the laser oscillator 2, and the reflected light reflected from the strip is converted by the photoelectric conversion device 5. The signal pre-processing unit 6 converts the detection signal converted into an electric signal and converted into the electric signal.
Performs filtering processing. If there are surface flaws
The reflected light is disturbed, so this filtered
A specific waveform corresponding to the type of flaw appears in the detection signal. The surface flaw signal thus obtained is digitized by the analog / digital converter 7 and is taken into the memory 8 of the optical surface flaw detector 1. Then, if the surface flaw signal input to the memory 8 of the optical surface flaw detection device 1 is input to each flaw type discrimination neural net 10 in which the flaw type pattern is learned for each discrimination flaw type , Each of the flaw type discrimination neural nets 10 determines whether or not the flaw type pattern and the surface flaw signal pattern learned and stored in parallel are the same.

The flaw type is identified by the flaw type discrimination neural network 10, and a flaw type determination signal is input to the flaw grade determination unit 12. At this time, the detection signal taken into the memory 8 of the optical surface flaw detection device 1 is calculated by the calculation unit 11 and is input to the flaw grade determination unit 12 as the flaw signal intensity. From the flaw type discrimination signal from the neural network 10 and the flaw signal intensity from the arithmetic unit 11, the degree of damage and the size of the flaw according to the flaw type are comprehensively determined.

[0011]

As is apparent from the above description, the present invention provides a large number of flaw type discrimination neural nets for discriminating the detection signal of the optical surface flaw detection device for each discrimination flaw type. The net can be small-scale, the learning can be converged reliably, and when a new flaw type is added, a new flaw type discrimination neural net that determines the new flaw type is optical surface flaw detection Since it is only necessary to connect between the device and the flaw grade determination unit and to juxtapose it with the existing flaw type determination neural network and learn the flaw type added only to the newly added flaw type determination neural network, As a result, the learning time can be significantly reduced, and a highly flaw-free surface flaw discriminating apparatus can be obtained. Furthermore, since the flaw type discrimination neural nets simultaneously determine the flaw type in parallel, the discrimination processing time can be reduced. Moreover, since the flaw signal intensity that can be determined without using the flaw type discrimination neural network is performed by the calculation unit connected to the optical surface flaw detection device in parallel with the flaw type determination neural network, a time delay is caused by the calculation unit. In addition, since the load on the neural network for discriminating the flaw type can be reduced, the discrimination processing time can be further reduced. Therefore, the present invention is extremely large in bringing to the industry a surface flaw discriminating apparatus which solves the conventional problems.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing an embodiment of the present invention.

FIG. 2 is an explanatory view showing an optical surface flaw detection device of the present invention.

[Explanation of symbols]

 1 Optical surface flaw detection device 10 Flaw type discrimination neural network 11 Operation unit 12 Flaw grade judgment unit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G01N 21/84-21/958

Claims (1)

(57) [Claims] 1. An optical surface flaw detection device for detecting a surface flaw.
(1) is connected to the detection signal of the optical surface flaw detection device (1) .
Arithmetic unit for outputting a defect signal intensity calculated from No. (11), wherein the optical surface flaw detection apparatus (1) connected to the determined defect classification per
A surface flaw discriminating device , wherein a plurality of flaw type discriminating neural networks (10) for discriminating whether or not the flaw is of the type are connected to a flaw grade judging section (12).