WO2020037803A1

WO2020037803A1 - Fillet joint penetration control method employing near-infrared binocular vision recognition

Info

Publication number: WO2020037803A1
Application number: PCT/CN2018/110523
Authority: WO
Inventors: 王克鸿; 宋正东; 黄勇; 王绿原; 吴统立; 叶丹
Original assignee: 南京理工大学
Priority date: 2018-08-20
Filing date: 2018-10-16
Publication date: 2020-02-27
Also published as: CN108856978A; CN108856978B

Abstract

A fillet joint penetration control method employing near-infrared binocular vision recognition uses front and rear near-infrared scanning CCD cameras (1, 2) and a computer image processing control system to perform penetration control on a fillet joint. The method comprises: firstly, selecting suitable near-infrared scanning CCD cameras (1, 2) on the basis of the principle of near-infrared technology; secondly, acquiring a weld pool image, the acquisition comprising using a front near-infrared vision sensor to acquire a feature parameter of a weld pool and using a rear near-infrared vision sensor to acquire a reverse melting width, and performing image processing and analysis on the acquired feature parameter of the weld pool and the reverse melting width so as to acquire penetration information of a weld seam; and finally, establishing a model on the basis of the acquired parameters so as to perform penetration control. In the method, two near-infrared scanning CCD cameras are used to acquire the feature parameter of the fillet weld seam and the reverse melting width, and derivation for the reverse melting width is eliminated, thereby eliminating the case in which a dimensional parameter and a characteristic parameter of a front weld pool image are used to represent penetration.

Description

Penetration control method of corner joint based on near-infrared binocular vision recognition

Technical field

The invention relates to the technical field of machine vision and the field of additive forming technology, in particular to a method for controlling penetration of an angle joint based on near-infrared binocular vision recognition.

Background technique

At present, most of the penetration control in the industry is still using monocular vision to collect images, use the feature information of the front molten pool captured by the camera, process the image to extract the features of the positive molten pool, and finally use the established relationship. The model derives the reverse melting width. The image captured in this way will be strongly disturbed by arc light, which will affect the derivation of the reverse melting width and will easily cause large errors. At the same time, there is relatively little research on the penetration control of fillet welds. The method of measuring temperature field using near-infrared method and visual sensing combined with the present invention can well control the penetration of fillet welds. It has great use value for the current development of welding technology.

Summary of the Invention

The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for controlling penetration of a corner joint based on near-infrared binocular vision recognition with high intelligence, high reliability, and strong adaptability.

The objective of the present invention is achieved by the following technical solutions:

The method includes the following steps:

S1: Fix the two workpieces to be welded on the positioner with a fixture at 90 °; install the welding torch at the end of the robot and place the two near-infrared scanning CCD cameras at 45 ° to the horizontal plane according to the normal line of the lens end surface- An angle of 60 ° is fixed on the welding gun with a clamp,

S2: Start the welding power source and perform welding.

S3: According to Planck's formula, in the near-infrared range, as the wavelength increases, the radiation intensity of the arc decreases rapidly. When the wavelength reaches around 2 μm, the relative radiation intensity of the molten metal pool is greater than the radiation intensity of the arc. With the increase of the wavelength, the relative radiation intensity of the molten metal pool does not change much, while the relative radiation intensity of the arc continues to decline. Since the radiation intensity of the molten metal pool is a useful signal and the radiation of the arc is interference, the near infrared can be used. CCD captures more valuable images.

S4: Two near-infrared vision sensors work simultaneously. Drive the near-infrared scanning CCD camera on the front to collect the surface of the diagonal joints to obtain the shape of the molten pool and its size parameters. At the same time, the back side of the near-infrared scanning CCD camera is driven to collect the image on the back of the diagonal joint, and the reverse side of the corner joint is mainly obtained to obtain the weld penetration information. The key technology of near-infrared scanning CCD for observing the back melting width is image processing.

S5: First filter the acquired image, then scan column by column to get the boundary, mark the boundary points, and then calculate the reverse melting width by Hough transform.

S6: Then perform image processing and analysis on the obtained molten pool and its size parameters, first filter and denoise the original image, then perform threshold segmentation, then extract the contour of the weld, and finally extract feature points to obtain geometric feature parameters .

S7: Then use the three characteristic parameters of the outer fusion width, the inner edge of the molten pool, and the difference between the outer fusion widths of adjacent images as the input of the neural network, and use the three states of unmelted, penetrated, and penetrated as the output of the neural network. To build a BP neural network model. Prediction of penetration based on established neural networks.

S8: The real-time welding characteristic parameters are compared with the established neural network model. If the weld is in the penetration state, the corresponding welding parameters are not changed.

S9: If the welding seam is in an un-penetrated or over-penetrated state, the computer adjusts the values of the current and voltage and the welding speed according to the results of the model established by the controller and sends them to the controller to achieve the effect of controlling penetration.

Further, in order to limit the interference of the welding arc light on the near-infrared thermal image shooting, the near-infrared sensitive monitoring device of the near-infrared scanning CCD camera device may be composed of a special photovoltaic cell. This photovoltaic cell is only sensitive to infrared light in the band of about 2 μm. This camera system can significantly reduce the interference of arc light.

Further, the parameters of the molten pool captured by the front near-infrared scanning CCD camera include the outer melting width, the inner edge width of the molten pool, the difference between the outer melting widths of adjacent images, and the like. The parameters captured by the back-side near-infrared scanning CCD camera are mainly the reverse melting width.

Compared with the prior art, the present invention has the following advantages:

(1) The present invention proposes to use two near-infrared scanning CCD cameras to obtain the characteristic parameters of the fillet welding seam and the reverse melting width, that is, the parameters of the molten pool captured by the front near-infrared scanning CCD camera include the outer melting width, The edge width, the difference between the outer fusion width of adjacent images, etc., the parameters captured by the reverse near-infrared scanning CCD camera are the reverse fusion width. It is no longer necessary to deduce the reverse melting width. It is avoided to use the characteristic size parameter and characteristic trait parameter of the front molten pool image to characterize the penetration.

(2) The near-infrared camera system used in the present invention is small in size and light in weight, has a wealth of acquired information, and has a clear image, and has advantages in acquiring a molten pool image.

(3) The present invention studies the penetration control of fillet welds, which has high practical application value for the current development of welding processes.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a schematic diagram of the overall system work.

FIG. 2 is a schematic diagram of the installation positions of the near-infrared scanning CCD camera on the front and the near-infrared scanning CCD camera on the back.

Figure 3 is a schematic diagram of the installation of two workpieces of a fillet weld.

FIG. 4 is a flowchart of reverse melting image processing.

FIG. 5 is a flowchart of the front molten pool image processing.

Figure 6 is a schematic diagram of the shape of the molten pool (A is the outer melting width, and part B is the inner edge width of the molten pool).

FIG. 7 is a graph of the back melting width corresponding to different penetration states.

Figure 8 is a statistical chart of the characteristic parameters of the molten pool and the penetration situation.

1 ----- Near-infrared scanning CCD camera 1; 2 ------ Angle-welded workpiece; 3 ----- welding gun; 4 ----- Near-infrared scanning CCD camera 2

detailed description

In order to make the objectives, technical solutions, and advantages of the present invention clearer and more specific, the present invention is further described below with reference to the accompanying drawings and specific embodiments. The following examples are used to illustrate the present invention, but are not limited to the present invention.

As shown in FIG. 1, the system working principle diagram of the present invention based on near-infrared binocular vision recognition of the fillet control method of the system includes: front and back two near-infrared

scanning CCD cameras

1 and 2, welding guns, to be performed Welded corner joints. Among them, the front and back two near-infrared scanning CCD cameras are fixed on the two sides of the workpiece to be angled with a clamp according to the normal of the lens end surface and the normal of the workpiece surface at an angle of 30 ° -45 °. The welding torch 2 is mounted on six axes At the end of the robot, the robot is connected to two left and right near-infrared

scanning CCD cameras

1 and 2, a six-axis robot, and a welding torch 3, respectively. Specific implementation: The other equipment used in the method for controlling the penetration of the corner joint based on near-infrared binocular vision recognition of the present invention mainly includes: two near-infrared scanning CCD cameras; a MIG welding gun from Pegasus in the United States; and a six-axis in the Japanese YSKAWA company Robots; controllers; welding machines and wire feeders from FRONIUS, Austria; two-axis tilting rotary positioners; MIG welding arc controllers; wire feeders and welding consumables, computers, stepper motor drivers, stepper motors, etc. .

Figure 2 is a schematic diagram of the installation position of the near-infrared scanning CCD camera on the front and the near-infrared scanning CCD camera on the back. In this example, the CCD camera is at an angle of 50 ° to the horizontal plane.

Figure 3 shows the installation of two pieces of fillet weld.

With reference to FIG. 1, FIG. 2 and FIG. 3, the present invention applies a near-infrared binocular visual recognition method for the penetration control of a corner joint. Taking an aluminum alloy of model 3003 as an example, the size of an aluminum plate is 200 * 100 * 6mm3. The steps are:

S1: The welding torch is installed at the end of the robot, and the two near-infrared scanning CCD cameras are fixed on the welding torch with a clamp at an angle of 50 ° from the normal line of the lens end surface to the horizontal plane. The two cameras are symmetrically distributed about the horizontal normal.

S2: clean the surface of the 3003 aluminum alloy substrate, remove surface impurities and oxides, open the protective gas cylinder, and prepare for welding;

S3: The two cleaned aluminum alloy substrates are fixed on the positioner with a fixture according to the method shown in FIG. 3, and the two are at an angle of 90 °;

S4: Determine welding process parameters. In this example, the wire feed speed is set to 7mm / min, and the welding speed is set to 6cm / min;

S5: Start the welding power source and perform welding.

S6: Two near-infrared vision sensors work simultaneously. Drive the near-infrared scanning CCD camera on the front to collect the surface of the diagonal joints to obtain the shape of the molten pool and its size parameters. At the same time, the back side of the near-infrared CCD scanning camera was driven to collect the image on the back of the diagonal joint, mainly to obtain the fusion width of the reverse side of the corner joint, and perform image processing and analysis according to the process shown in FIG. 4 to obtain the weld penetration letter. According to Planck's formula, in the near-infrared range, as the wavelength increases, the radiation intensity of the arc decreases rapidly. When the wavelength reaches around 2 μm, the relative radiation intensity of the molten metal pool is greater than the radiation intensity of the arc. The relative radiation intensity of the molten metal pool does not change much, while the relative radiation intensity of the arc continues to decrease. Since the radiation intensity of the molten metal pool is a useful signal, and the radiation of the arc is an interference, it can be collected by the near-infrared CCD. To more valuable images.

S7: Next, image processing and analysis are performed on the obtained molten pool and its dimensional parameters and the reverse melting width according to the process shown in FIG. 5.

S8: Real-time image extraction of the geometric characteristics of the molten pool. As shown in Figures 6 and 7, the three characteristic parameters of the outer fusion width, the inner edge width of the fusion pool, and the difference between the outer fusion widths of adjacent images are used as the input of the neural network. The three states of penetration, penetration, and penetration are used as the output of the neural network. A 3-x-3 BP neural network including a hidden layer is constructed. The number of neurons in the input and output layers is 3, according to the established Neural network to predict penetration.

S9: The real-time welding characteristic parameters are compared with the established neural network model. As shown in FIG. 8, if the weld is in the penetration state, the corresponding welding parameters are not changed.

S10: If the weld is under- or over-penetrated, as shown in Figure 8, the computer adjusts the current and voltage and the welding speed according to the results of the model established by the controller, and sends them to the controller to achieve Controls the effect of penetration.

Claims

A method for controlling penetration of an angle joint based on near-infrared binocular vision recognition, which is characterized in that the method includes the following steps:

S1: Fix the two workpieces to be welded on the positioner with a fixture at 90 °; install the welding torch at the end of the robot and place the two near-infrared scanning CCD cameras at 45 ° to the horizontal plane according to the normal line of the lens end surface- An angle of 60 ° is fixed on the welding gun with a clamp;

S2: Start the welding power source and perform welding;

S3: Two near-infrared vision sensors work at the same time, driving the surface of the near-infrared scanning CCD camera diagonally on the front to collect images to obtain the shape of the molten pool and its size parameters; simultaneously driving the near-infrared scanning CCD camera diagonally on the opposite side Collect the image on the back of the camera to obtain the fusion width on the reverse side of the corner joint, and obtain the penetration information of the weld;

S4: Filter the acquired image, then scan column by column to get the boundary, mark the boundary points, and then calculate the reverse melting width by Hough transform;

S5: Perform image processing and analysis on the obtained molten pool and its size parameters, first filter and denoise the original image, then perform threshold segmentation, then extract the contour of the weld, and finally extract feature points to obtain geometric feature parameters;

S6: The three characteristic parameters of the outer fusion width, the inner edge of the molten pool, and the difference between the outer fusion widths of adjacent images are used as the input of the neural network, and the three states of unfused, fused, and over-transmitted are used as the output of the neural network. Establish a neural network model; predict penetration based on the established neural network;

S7: The real-time welding characteristic parameters are compared with the established neural network model. If the weld is in a permeable state, the corresponding welding parameters are not changed; if the weld is in an un- or over-penetrated state, the computer The feedback result of the model established by the controller is used to adjust the current, voltage and welding speed values to the controller to achieve the effect of controlling penetration.
The method for controlling penetration of a corner joint based on near-infrared binocular vision recognition according to claim 1, wherein step S3 is binocular vision for image acquisition.
The method for controlling penetration of a corner joint based on near-infrared binocular vision recognition according to claim 1, wherein the near-infrared scanning CCD camera device has near-infrared extraction of electromagnetic waves with a wavelength in the range of 780nm-2526nm.
The method for controlling penetration of a fillet joint based on near-infrared binocular vision according to claim 1, characterized in that before performing feature extraction on the weld, filtering and sharpening the image to enhance the image Therefore, the image is converted into a form more suitable for human or machine for analysis and processing, so that the edge of the weld and the base material are more easily distinguished, and it is convenient to extract the feature points of the weld.
The method for controlling penetration of a corner joint based on near-infrared binocular vision recognition according to claim 1, characterized in that, in step S6, a BP neural network intelligent control method is adopted to change the outer fusion width, the inner edge width of the molten pool, and the phase. The three characteristic parameters of the difference between the outer fusion width of the adjacent image are used as the input of the neural network, and the three states of unmelted, fused, and penetrated are used as the output of the neural network.