KR20020084869A - Automatic Recognition Method of Weld Joints - Google Patents

Abstract

PURPOSE: An automatic recognition method is provided to achieve improved efficiency of welding by eliminating the necessity of changing mode or correcting images even when the shape of fundamental welding joint changes. CONSTITUTION: An automatic recognition method, comprises an image acquisition step(50) of converting video signals of a visual sensor head into data; a neural net input step(51) of detecting specific data related to the shape of joint from the obtained image; a neural net output step(52) of searching the most similar welding joint pattern from the neural net learned by using the data detected in the previous step; an image processing step(53) of performing image processing for the shape of welding joint, on the basis of the pattern recognition determined by the neural net; and a welding line recognition step(54) of searching the welding line of joint through image processing.

Description

Automatic Recognition Method of Weld Joints

The present invention relates to a method for automatic recognition of the weld joint shape.

Methods for visual sensor systems for weld seam tracking are well known and specified. However, the automatic recognition of weld joint shape is still insufficient, and this method may contribute to welding automation and welding robot automation.

In general, as shown in FIG. 1, weld joint shapes are generally classified into five types. In general, vision systems for automatic weld line tracking configure the system to recognize one or two types of weld joint geometry. However, to determine the weld joint shape, the vision system is constructed in such a way that the operator directly selects the shape or switches modes. Therefore, despite the presence of various welded joint shapes, the limited vision system for certain members has contributed to the deterioration of utilization. Also, if the welded shape is changed, it is inconvenient because the operator has to manually operate the selection mode. If the shape is changed, it is inconvenient because the operator has to manually operate the selection mode.

According to the present invention, the weld joints generated in the actual field are divided into five types, and the laser beam is reflected on the divided shapes, and the reflected image according to the shape of the member is input through the camera and displayed on the monitoring screen. The image information of the laser strip thus obtained was learned through neural network to distinguish the pattern of each shape by output. By inserting the learned neural network into the front end of the image processing as shown in Fig. 3, even if various laser band-shaped image information is input, the closest joint shape is determined as the output. And the appropriate image processing is performed for each joint shape is a system that automatically finds the weld seam.

The present invention is to learn the shape information of each weld joint obtained by the visual sensor with a neural network and to automatically find the most suitable weld joint pattern by inputting arbitrary joint image information into the learned neural network and automatic recognition of the weld joint shape. Then, it is to provide a method in which image processing appropriate for the pattern of the recognized joint is automatically performed.

1 is a field view and an image view of the laser strip shape according to the weld joint shape

2 is a system site conceptual diagram of one embodiment of the present invention;

3 is a block diagram of a neural network used in the present invention

4 is a flowchart of a method for automatically recognizing a weld joint shape used in the present invention.

5 is a field view of an embodiment of the present invention;

<Description of the symbols for the main parts of the drawings>

(1) visual sensor head

(2) welding member

(3) laser beam

(4) monitoring screen

(5) laser strip

(6) welding line

10 image acquisition board

(11) industrial PC

(20) input of neural network

(21) output of neural network

(23) layers of neural networks

(24) weights between floors 23

50 image acquisition data

(51) Neural Network Input Data

(52) neural network output data

(53) image processing

(54) welding seam recognition

1 is divided into five representative shapes of the weld joint of the present invention. The visual sensor head 1 (camera + laser stripe), the welding member 2, the laser beam 3, the monitoring screen 4, and the weld joint shape (5) shaped as the laser stripe, which receive an image of the laser stripe in the figure. The weld line 6 recognized by the image processing is shown.

2 is a block diagram of the main parts of the system of the present invention. The visual sensor head 1 transmits an image of each member to the image acquisition board 10 through a camera. The image acquisition board 10 is a part for storing the image data so that the input image is converted into a digital value and image processing is possible. The industrial PC 11, which is the controller body, implements the neural network and performs program processing to enable automatic determination of the shape of the joint and recognition of the weld seam.

3 is a neural network used in the present invention. The weld joint shape data (laser band pixel information) is input as the neural network input data 51 to the input 20 side of the neural network, respectively, and the output 21 of the neural network has an ID corresponding to the weld joint shape corresponding to each case. To have a target value. The neural network adjusts the weight 24 between the layers 23 of the neural network in a direction in which the error between the given target value and the output is minimized. The neural network finally adjusted is inserted as shown in FIG. The shape of the weld joint is determined.

4 is an automatic welding joint shape recognition method used in the present invention. Image acquisition data 50 is a process of converting an image signal of the visual sensor head 1 into data. Neural network input data 51 process for detecting specific data related to the joint shape (laser band pixel information) in the acquired image, and the weld joint most similar in the neural network learned using the detected data as learning data. In the neural network output data 52 process of finding a pattern, image processing 53 is performed to perform image processing on the shape of the weld joint due to the pattern recognition determined by the neural network. Weld seam recognition 54 which finds the weld seam of the joint with respect to the identified joint by image processing.

5 is a field embodiment of the present invention. When the joint is automatically found by using a robot, if the user moves from time t1 to time t2 through t3, the image obtained between t1 and t2 is not matched with what is learned, but the welding joint is recognized by t3 This is an example. Since the welding point cannot be found by the information at t1, when the camera is located at the position of t3 through step t2, the welding joint for the current member is found, so the welding point can be easily found.

(Variations, applications and legal interpretations)

The present invention is not limited to the specific preferred embodiments described above, and various modifications can be made by those skilled in the art without departing from the gist of the invention as claimed in the claims. Of course, such changes are intended to fall within the scope of the claims.

The present invention has the following effects of the invention can be expected by the configuration and operation as described above.

The present invention automatically determines what the joint shape of the current member is from the image acquired by the visual sensor as described above, and is imaged to fit the selected member's characteristics so that the welding line can be accurately known and welded at the same time. The member gap information can be obtained to enable automatic welding tracking. Therefore, if the shape of the basic welding joint is changed, inefficient processing such as mode change or image processing correction for each can be reduced, so that the work efficiency and the use range of the visual sensor can be extended in welding work using the visual sensor.

Claims (3)

Weld characterized in that it consists of a method of automatically finding the most suitable weld joint pattern by learning the shape information for each weld joint obtained by the visual sensor (1) with a neural network and inputting arbitrary joint image information to the learned neural network. Automatic recognition of joint geometry. An automatic recognition method of a weld joint shape after the automatic recognition of the weld joint shape, the image processing corresponding to the recognized joint pattern is automatically performed. An image acquisition data step of converting the image signal of the visual sensor head into data, and thus a neural network input data step of detecting specific data related to the joint shape and a neural network output of finding a weld joint pattern from the learned neural network using the detected data. And an image processing step for the weld joint shape by the data step and the determined pattern recognition, and a welding line recognition step found from the image processing.