TW202343312A - A robotic welding method - Google Patents

Abstract

A robotic welding method includes a filming step, a learning step, a robotic welding operation step, an inspection step, and a feedback step. The method includes using a learning software to simulate and learn details shown in videos about a plurality of manual welding processes shot in the filming step and then using a robot equipment to carry out a welding work according to a result learned by the learning software to thereby form at least one weld caused by the welding work. The method further includes using a visual inspection software which acts to check whether there are weld deficiencies and revise the weld deficiencies to thereby obtain revised welding parameters. Then, the revised welding parameters are sent back to the robot equipment for adjusting the welding work of the robot equipment. The cooperation between the above software and the hardware helps increase the welding effect and the welding quality and also facilitates a decrease in welding costs and the welding failure.

Description

Robotic welding methods

The present invention relates to a welding method, in particular to a robotic welding method that utilizes robotic equipment to learn manual welding details.

The quality of welding technology is related to the quality of the overall product. Especially high-voltage and large-scale projects require good execution of the welding process. Manual welding is a skill that requires a lot of practice to master, and it involves many factors such as: The grip of the electrode holder Holding method, maintaining the angle between the welding rod and the welding piece, maintaining the distance between the welding rod and the welding piece, that is, the length of the welding arc, synchronized movements of the welder's eyes, head and holding hands, difficult welding positions, power adjustment, Personal protection measures and other factors make it easy for various human errors to occur during the manual welding process, making it impossible to ensure good welding quality. Furthermore, the cost of manual welding remains high. Currently, the number of certified high-level welding masters in Taiwan Welding wages are relatively high, and the welding failure rate of some special projects such as Taiwan's green energy wind power towers is also high. Even if we consider using foreign welding technology, the cost of transferring foreign welding technology is also quite expensive, so the welding process still needs to be strengthened. .

Existing patent documents such as CN103418942B, CN1056188983312, US6942139B2, etc. are welding-related technologies, which sequentially disclose welding robot intelligent control methods, intelligent welding seam methods, and robot cylindrical welding. However, although these patents use robots or intelligent methods for welding, they are not aimed at There is no clear solution to welding defects, and the welding operation of robots is not as flexible as manual welding. After all, the difficult manual welding skills mostly come from the countless practical experiences accumulated by senior welders and are varied. Compared with robots, There may be less variability in welding techniques, resulting in limited welding results.

Therefore, the purpose of the present invention is to provide a robot welding method that uses learning software to learn manual welding actions and details from videos, and further detects the welding operations performed by the robot equipment based on the learning results and corrects the welding defects, not only By improving the inspection level of welding practice, the corrected results can also be fed back to adjust the welding operation of the robot equipment, which will further improve the welding effect and welding quality of the robot equipment, and also help reduce welding costs and failure rates.

Therefore, the robot welding method of the present invention includes a shooting step, a learning step, a robot welding operation step, an inspection step, and a feedback step. In the shooting step, a shooting component is provided to shoot multiple manual welding processes into multiple videos. , especially these manual welding processes can be operated by one or more certified high-level professional welding masters, which is more conducive to promoting the learning software of this learning step to learn the skills of senior manual welding movements and details; in this learning step, prepare There is the aforementioned learning software. The learning software includes an imitation learning unit. The imitation learning unit is used to learn based on the videos to obtain a learning result. The aforementioned learning includes imitating and analyzing the welding presented in the manual welding process. The details of the welder's movements are obtained, and the original welding data related to the details of these movements is obtained. The original welding data includes changes in the welding angle, the movement speed of the welder when holding the electrode holder, etc., and then the learning will be The results are input into a robot equipment to perform the robot welding operation step; in the robot welding operation step, the aforementioned robot equipment is provided and the robot equipment includes a simulation unit and an operation unit, and the operation unit has interlocking mechanical components. To perform a welding operation and form at least one welding spot due to the welding operation, and before the mechanical component performs the welding operation, the simulation unit first simulates the welding action according to the learning results and obtains an operational welding parameter, The operation welding parameters are then input into the mechanical component to perform the welding operation, which not only helps to increase the accuracy of the welding operation, but also reduces costs; the inspection step is performed after the welding operation is performed. In the inspection step , equipped with a visual inspection software. The visual inspection software includes a test unit and a correction unit. The test unit is used to check whether there are welding defects in the welding operation, thereby analyzing the internal and external defects of the at least one welding joint to obtain A defect result is generated, and then the correction unit corrects the original welding data corresponding to the defect result to obtain a corrected welding parameter; in the feedback step, the corrected welding parameter is fed back to the robot equipment, So that the mechanical component of the operating unit adjusts the execution of the welding operation according to the corrected welding parameters.

According to reports, the learning software is used to learn the actual manual welding movements and related details, and then the robot equipment performs the welding operation according to the learning results. This allows the welder's manual welding skills to be inherited and executed by the robot equipment. , not only avoids human errors caused by manpower, but also improves the welding effect and reduces welding-related production costs. For the welding joints produced after the welding operation is performed, the visual inspection software with deep learning can also be used to detect and Analyze to identify the quality of welding. If there are welding defects in the welding area, they will be corrected. Correcting the defects in the welding operation not only improves the inspection of welding practices, but also achieves the stability of welding quality and reduces the welding failure rate. Even the above-mentioned robot welding operation steps, inspection steps, and feedback steps can be implemented repeatedly until no welding defects occur, so that the robot equipment can achieve more ideal welding results.

The aforementioned and other technical contents, features and effects of the present invention will be clearly understood in the following detailed description of the preferred embodiments with reference to the drawings.

Referring to Figures 1 and 2, a preferred embodiment of the robot welding method 3 of the present invention includes a photographing step 31, a learning step 32, a robot welding operation step 33, an inspection step 34, and a feedback step 35; Among them, in the shooting step 31, it is equipped with a shooting component 311, and the shooting component 311 is used to shoot multiple videos of a plurality of manual welding processes. In this embodiment, it is better to shoot one or more videos. A certified high-level professional welding master P was filming the process of on-site welding, especially the changes in welding angles, the moving process of the electrode holder, the welding action process for difficult welding positions such as corners and curves, etc. The various basic welding skills and professional welding skills accumulated by the welding master P based on his own senior experience can be passed on to the robot equipment through these videos to serve as a basis for the movement changes of the robot equipment during welding operations and to avoid human errors. Problems such as affecting the welding effect may occur; after the shooting component 311 completes shooting, the learning step 32 can be executed according to the captured video.

In the learning step 32, it is equipped with a learning software 321. The learning software 321 includes an imitation learning unit 3211. The imitation learning unit 3211 is used to learn the videos to obtain a learning result, and then the learning result is Input to a robotic device 331; the aforementioned learning system includes the imitation learning unit 3211 imitating and analyzing the image details presented in the video, that is, imitating the welder P (i.e., the aforementioned high-level professional welding master) presented in the manual welding process. ), and obtain original welding data related to these action details. In other words, the imitation learning unit 3211 is suitable for analyzing, imitating and learning the action details of the welding master in the video, such as: the welding master's arms during welding. Movement changes, movement postures in difficult positions, etc., and welding data related to the details of these movements, such as: the welding angle between the welding rod and the weldment, the distance between the welding rod and the weldment, that is, the length of the welding arc (arc length), the moving speed of the welding master when holding the electrode holder, the movement angle of the welding master for difficult welding positions such as corners and curves, etc.; when the learning software 321 completes the learning, the learning results obtained based on the learning are: The robotic welding operation step 33 can be performed.

In the robot welding operation step 33, it is equipped with the robot equipment 331. In this step 33, the learning results learned by the learning software 321 are input into the robot equipment 331, so that the robot equipment 331 performs operations according to the learning results. Perform a welding action; wherein, the robot equipment 331 includes a simulation unit 3311 and an operating unit 3312. The operating unit 3312 has an interconnected mechanical component 33121, which is suitable for performing a welding operation, that is, a welding action operation based on the learning results. The mechanical component 33121 can be a robot arm that imitates the welding action of the welding master based on the learning results; of course, there may inevitably be deviations in angles, movement positions, etc. between the robot arm and the welding master's arm, but the deviation is still permissible. For example, the difference is less than 1~2% or lower within the range, or other parameters such as the gap between the welding rods can be adjusted to reduce the deviation value of the welding operation without affecting the execution of the welding operation; furthermore, in this article In the embodiment, the operating unit 3312 preferably also has a recording component 33122 for recording the time point when the welding is performed. That is, the detailed time when the mechanical component 33121 performs the welding process can be recorded by the recording component 33122 for subsequent use. Used when checking for defects.

The execution of the aforementioned robot welding operation step 33 is specifically: after learning by the learning software 321, the learning results are input into the simulation unit 3311. For example, the learning results can be directly input into the simulation unit 3311 by the learning software 321. , and the simulation unit 3311 can be a simulator to simulate welding according to the learning results, that is, simulate the details of the welding action of the welding master P (the graphic simulation shown in Figure 3), the simulation unit 3311 can also have a built-in adjustment mechanism And when it is judged that it is necessary, the simulated control parameters can also be adjusted and wait for them to stabilize, thereby obtaining an operating welding parameter; and after the simulation operation, the obtained operating welding parameters are then handed over to the mechanical component 33121. Actual execution of the welding operation (as shown in Figure 4), especially the operation of simulating first and then operating, is beneficial to promoting the accuracy of parameters to reduce errors in welding operations, and is also beneficial to reducing welding-related costs; and after executing the welding operation, it will also At least one welding point is formed, that is, the welding point can be located in one or more places.

The inspection step 34 is performed after the welding operation is performed; in the inspection step 34, a visual inspection software 341 is provided. The visual inspection software 341 includes a test unit 3411 and a correction unit 3412, so this step 34 is performed through The test unit 3411 performs inspection, that is, inspection and testing on the aforementioned welding joint. The detection is to confirm whether the welding operation of the robot equipment 331 will produce welding defects as shown in Figure 5, that is, whether the welding joint has: pores, Internal defects such as slag inclusions, cracks, incomplete welding, lack of fusion, etc., as well as appearance such as weld size does not meet the requirements, undercuts, weld nodules, arc craters, spatter, arc scratches on the surface of the weldment (base metal), etc. Defects, wherein the internal and external defects are scanned out through appropriate instruments. For example, the internal defects can preferably be scanned into images through X-rays, and the appearance defects can be captured into images using a camera. Therefore, The visual inspection software 341 can have the performance of deep learning to identify the quality of welding results, that is, analyze the internal defects and appearance defects of the welding joint, such as defect type, defect status and extent, defect location, defect quantification, etc. etc., thus obtaining a defect result (as shown in Figure 6), which not only achieves accurate analysis of welding defects, but also improves the inspection efficiency of welding practices.

When the test unit 3411 detects a welding defect, the defect result is transmitted to the correction unit 3412 so that the correction unit 3412 corrects the original welding data corresponding to the defect result, that is, corrects the defects in the welding operation. , and the correction can be based on the reasonable range preset by the software (for example, after using common sense to judge the severity of the defect, the experience value of experts or welding masters can be added as a preset reference for performing corrections) to the original The welding data is revised so that a corrected welding parameter can be obtained after the correction, and then the feedback step 35 is executed.

In the feedback step 35, the corrected welding parameters obtained by the correction unit 3411 are fed back to the robot equipment 331. For example, the correction unit 3411 can transmit the corrected welding parameters back to the robot equipment 331 to achieve a feedback effect, and The feedback refers to the feedback that can be fed back to the simulation unit 3311 for it to also perform simulation and control parameter adjustment and stabilization, and then to the mechanical component 33121 of the operating unit 3312 to perform welding, so the robot equipment 331 can still perform welding according to the The welding parameters are corrected to adjust the welding operation, so that after the original welding data corresponding to the welding defects (that is, the poor original welding data) has been corrected, the robot equipment 331 will no longer operate based on the poor data. The welding action avoids repeating the same mistakes, thereby allowing the robotic equipment 331 to obtain more ideal welding effects, stabilize welding quality, improve the welding effect, reduce the welding failure rate, and also help reduce related welding costs.

If there are no welding defects during the welding operation of the robot equipment 331, subsequent random inspections will be arranged based on actual needs or welding positions. For example, if it is a high-risk, high-voltage equipment that can generally conduct 100% random inspections in areas where welding is difficult, if it is flat, Part or all of the straight line parts can be inspected regularly or randomly to ensure the good welding quality and stability of the robot equipment 331.

Furthermore, the data learned or applied by the learning software 321 and the visual inspection software 334 can be stored in a system database (not shown in the figure), so that it can be provided to the software in real time when the above steps are implemented. Use or as a reference for correction, etc.; in particular, the above-mentioned robot welding operation steps, inspection steps, and feedback steps can be repeatedly implemented until no welding defects are generated, so that the robot equipment 331 can achieve a more ideal welding execution effect.

To summarize the foregoing, the robot welding method of the present invention captures the manual welding process into a video in a filming step, and then uses learning software to analyze the welder's movements and related welding data included in the filmed video in a learning step. The details are simulated and learned, and then in a robot welding operation step, a robot equipment is used to perform a welding operation and form a weld according to the learning results obtained in the learning step, and then in an inspection step, a visual inspection software is used To detect and analyze internal and external welding defects in the welding joint to achieve deep learning and correct the welding defects, and then in a feedback step, the corrected welding parameters obtained after correction are fed back to the robot equipment to correct the welding There is a lack of operation, so the overall cooperation of the aforementioned software (such as the learning software, visual inspection software, etc.) and hardware (such as the related equipment for the welding operation of the robot equipment, etc.) can promote the stability of the welding effect and welding quality. Improving the inspection of welding practices will also help reduce welding costs and failure rates.

However, the above description is only for illustrating the preferred embodiments of the present invention, and should not be used to limit the scope of the present invention, that is, any simple equivalent changes and modifications may be made based on the patent scope of the present invention and the contents of the description of the invention. , should still fall within the scope covered by the patent of this invention.

(this invention) 3: Robot welding method 31:Photography steps 32: Learning steps 33: Robot welding operation steps 34: Check steps 35: Feedback step 311: Shooting components 321:Learning software 331:Robotic equipment 341:Visual inspection software 3211:Imitation Learning Unit 3311:Analog unit 3312: Operating unit 3411:Test unit 3412: Correction unit 33121:Mechanical components 33122:Record component P:Welder

Figure 1 is a schematic step flow diagram of a preferred embodiment of the present invention. FIG. 2 is a schematic diagram showing the execution of the steps of the preferred embodiment of the present invention. Figure 3 is executed according to the steps of Figure 2, in which the simulation unit creates a welding simulation diagram based on the learning results. Figure 4 is executed according to the steps of Figure 2. After imitating and learning the manual welding movements of the welding master, the mechanical components of the robotic equipment (such as a robot arm) perform the welding operation. Figure 5 is performed according to the steps in Figure 2. After welding by the robot equipment, the welding defects detected by the visual inspection software are internal defects (the box on the left) and appearance defects (the box on the right). . Figure 6 is based on the steps of Figure 2, in which the visual inspection software conducts in-depth analysis of internal defects in the weld.

3: Robot welding method

31:Photography steps

32: Learning steps

33: Robot welding operation steps

34: Check steps

35: Feedback step

Claims (5)

A robot welding method, which includes: A shooting step, which is equipped with a shooting component and uses the shooting component to shoot multiple videos of multiple manual welding processes; A learning step, which is equipped with a learning software, the learning software includes an imitation learning unit, the imitation learning unit is used to learn for the videos to obtain a learning result, and then the learning result is input to a robot device, Among them, the learning of the imitation learning unit includes imitating and analyzing the details of the welder's movements during the manual welding process, and obtaining the original welding data related to the details of the movements, where the original welding data at least includes the welding angle. changes, the moving speed of the welder when holding the electrode holder; A robot welding operation step, which is equipped with the robot equipment. The robot equipment includes a simulation unit and an operation unit. The operation unit has mutually linked mechanical components and inputs the learning results learned by the learning software into the A simulation unit that first simulates the welding action based on the learning results and obtains an operational welding parameter, and then inputs the operational welding parameter to the mechanical component to perform a welding operation, thereby forming at least one weld; An inspection step, which is equipped with a visual inspection software. The visual inspection software includes a test unit and a correction unit. The test unit is used to check whether there are welding defects for the welding operation, thereby detecting and analyzing the at least one The internal defects and appearance defects of the welding joint are used to obtain a defect result, and the original welding data corresponding to the defect result is corrected by the correction unit to obtain a corrected welding parameter; and A feedback step is to feed back the corrected welding parameters to the robot equipment so that the mechanical component performs the welding operation based on the corrected welding parameters. The robot welding method according to claim 1, wherein the robot welding operation step, the inspection step, and the feedback step are repeatedly implemented until the welding defect no longer occurs. The robot welding method according to claim 1 or 2, wherein in the robot welding operation step, the operation unit further has a recording component to record the execution time of the welding operation. The robot welding method according to claim 1 or 2, wherein in the inspection step, the internal defects of the welding joint are obtained through X-ray scanning into images, and the appearance defects are obtained through images taken by a camera. inferred. The robot welding method according to claim 1 or 2, wherein the welder is a professional welding master.

Images