JPH0518904A - Method for detecting welded part of steel pipe - Google Patents

Abstract

PURPOSE:To obtain a method for detecting the welded part of a steel pipe utilized for the nondestructive inspection of the welded part, marking and the like. CONSTITUTION:With a steel pipe 1 being rotated, the image signal of the inner surface of the pipe is extracted by using a TV camera 6. The amount of the features of the images of the inner surface of the pipe having the intrinsic value of the kind of the pipe is extracted based on the obtained image signal. The welded part and the base material are identified with a neural net 8, wherein the amount of the features of the images of the inner surface of the kind of the pipe to be detected is learned and stored beforehand. Thus, the welded part is detected. The size of the field of view is changed by adjusting the brightness of lighting, the lens of the camera and the like. The workings such as nondestructive inspection, marking and the like can be performed highly efficiently and highly accurately by nonlinear discrimination and pattern recognition based on neural net work learning.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to nondestructive inspection of welded parts,
The present invention relates to a method for detecting a welded portion of a steel pipe for use in marking or the like.

[0002]

2. Description of the Related Art As a method for detecting a welded portion of a steel pipe or a detection device, (1). A technique utilizing electromagnetic induction such as vortex flow disclosed in JP-A-54-27492, (2). 50
-38186, 56-39460, 56-1480.
No. 54, 59-58359, a technique utilizing the reflected echo detection of ultrasonic waves, which is disclosed in (3).
Technologies applying electromagnetic ultrasonic waves disclosed in Japanese Patent Publication Nos. 41347 and 58-96247, (4).
There is a technique disclosed in Japanese Patent No. 650, in which a welding portion or an outer cutting portion is captured as an image by a television camera and signal processing is performed.

[0003]

[Problems to be Solved by the Invention] By the way, the above (1), (2),
With regard to the technique of (3), it is easily affected by the deviation in wall thickness of the steel pipe, and the S / N value deteriorates due to residual stress and strain during pipe forming, resulting in false detection. Further, the technique (4) can be applied to a pipe making line, but cannot be applied to a UV-painted pipe or a refinement line having a flaw during transportation.

Further, in any of the techniques (1) to (4), since the linear discrimination signal processing is used to discriminate between the welded portion and the base metal portion, it is vulnerable to disturbance and noise and causes erroneous detection. ing.

As described above, the conventional detecting method or detecting apparatus is not affected by the deviation of the wall thickness of the steel pipe, the influence of residual stress or strain during pipe forming, the outer surface coating, the outer surface stain, the flaw during transportation, and the like. However, there is a problem of causing false detection.

The present invention provides a method for detecting a welded portion of a steel pipe which advantageously solves the above problems.

[0007]

The gist of the present invention was obtained in a method for detecting a welded portion of a steel pipe by extracting a video signal on the inner surface of the pipe using a television camera while rotating the steel pipe in the circumferential direction. A pipe inner surface image feature amount having a value specific to the pipe type is extracted from the video signal, and the welding part and the base metal part are identified by a neural net in which the inner surface image feature amount of the pipe type to be detected is learned and stored. A method for detecting a welded portion of a steel pipe, which is characterized by detecting a welded portion.

That is, the inner surface of the rotating steel pipe is illuminated with illumination without changing the position, and the inner surface of the pipe is projected using a television camera. The image signal is taken into the signal processing device and the image is processed to quantify the weld or the weld inner surface cutting trace as a feature quantity, and the weld is judged from the weld and the base metal by the weld determination neural net. Thus, the welded portion can be accurately detected without being affected by the above-mentioned disturbance.

The welded part determination neural net is for learning and storing the feature amount, which has been image-processed for each pipe type and pipe size in advance, for both the welded part and the base metal part.

[0010]

The signal waveform obtained from the television camera has a characteristic feature of the signal waveform of the weld portion which is unique to the pipe type and pipe size. Therefore, these characteristic quantities are collected in advance as sample data of the signal waveform characteristic quantities of both the welded part and the base metal part for all and / or part of the steel pipe type and the steel pipe size for which the weld is to be detected. .

The obtained signal waveform feature amount is input to the neural net as a teaching signal amount of the mark signal of the welded portion or the base metal portion, and the causal relationship between the signal waveform characteristic amount and the welded portion or the base metal portion is neural-processed.・ Data files related by online learning.

On the basis of the data file obtained by the above work, the signal waveform continuously input is recognized and judged for one or more rounds of the steel pipe.

[0013]

EXAMPLES Examples of the present invention will be described below with reference to FIG. 1 and FIG. 2 in the case of applying a hydraulic test facility for medium-diameter electric resistance welded pipe line welding portion to the welding portion detection and alignment. To do.

First, the time of storing the neural net learning will be described. In order to obtain the signal waveform characteristic amount that is the basis for the determination, the inner surface of the steel pipe 1 from which data is collected is illuminated by the lighting device 5 and photographed by the television camera 6, and this video signal is input to the signal processing device 7, where the signal waveform is obtained. The feature amount is extracted and transferred to the neural network learning device 8.

Here, as the feature quantity, the in-pipe image, which is two-dimensional data, is averaged in the pipe axis direction and averaged, and the signal waveform 12 compressed into one-dimensional data shows the maximum brightness value 9 representing the feature of the welded portion. An internal cutting width or welding width of 10 and contrast of 11 was used.

The operator selects the signal waveform characteristic amount to be learned, and further inputs the mark signal of the welded portion or the base material portion to the neural net learning device 8 to make it learn. Here, the sufficiently learned data file is transferred to the signal processing device 7.

The above operation is repeated for the steel pipe type for detecting the welded portion, if necessary.

Next, the automatic detection of the welded portion will be described.
When the steel pipe 1 is loaded onto the roller 2, the proximity switch 4 notifies the signal processing device 7 of the presence of the steel pipe 1. The signal processing device 7 controls the motor 3 so as to rotate a known outer circumference of the steel pipe 1. Here, the inner surface of the rotating steel pipe 1 is illuminated by the illumination device 5, the television camera 6 continuously captures the inner surface, and the video signal is sent to the signal processing device 7.

The signal processing device 7 extracts the feature amount from the transmitted video signal and prepares the learning data.
Based on the file, the welded part or base metal part is judged. The welding position is determined when the steel pipe makes one rotation, the motor 3 is controlled so as to rotate the detected welded portion to a designated position, and the steel pipe 1 is dispensed without rotating.

[0020]

The welded portion detecting method of the present invention changes the field of view size by adjusting the brightness of the illuminating device, the lens of the camera, the pipe diameter, wall thickness deviation, residual stress during pipe forming,
Accurate detection that is not affected by disturbance or noise is possible without being affected by distortion, regardless of the coating condition on the outer surface, and by non-linear discrimination and pattern recognition by neural net learning. Work such as nondestructive inspection and marking can be performed with high efficiency and accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing an example of a welding portion detection device suitable for implementing the present invention.

FIG. 2 is a diagram showing an example of a signal waveform characteristic amount in a pipe circumferential direction.

[Explanation of symbols]

 1 Medium-diameter ERW steel pipe 2 Rolling roller 3 Roller drive motor 4 Proximity switch 5 Lighting device inside pipe 6 Industrial TV camera 7 Signal processing device 8 Neural net learning device 9 Maximum brightness 10 Internal cutting width or welding width 11 Contrast 12 signal waveform

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Internal reference number FI Technical display location G06F 15/18 8945-5L 15/62 400 9287-5L

Claims (1)

Claim: What is claimed is: 1. A method for detecting a welded portion of a steel pipe, comprising:
While rotating the steel pipe in the circumferential direction, the video signal of the inner surface of the pipe is extracted using a TV camera, the image features of the inner surface of the pipe having values unique to the pipe type are extracted from the obtained video signal, and the pipe type to be detected in advance. Neural learning and memory of internal image features of
A method for detecting a welded part of a steel pipe, which comprises detecting the welded part by identifying the welded part and the base metal part with a net.