| Intelligent behaviors of welder give inspirations to realize automatic welding. However, welding robot does not have the high adaptive abilities the same as a skilled welder. Similar to human sensing, welding sensors are required to realize automatic welding if welders are replaced by robots. Welding sensors can detect the inner and external results of welding, which are useful for the operation of robot or welding device. Except for welding sensors, automatic welding system must understand the welding process like welders. That means the system should have knowledge about the states of welding process. In other words, modeling of welding process is needed. Finally, a controller like human brain is required to establish the control scheme. Therefore, sensing, modeling and control of welding process are three key technologies to realize automatic welding.Vision sensing technology, especially of passive vision sensing, is similar to the observation behavior of welders and it has been a hot topic of the study on welding sensing methods. Robot has been an important approach for automatic welding because of its flexibility. Because of the hybrid characteristics of robotic welding process, a general model framework for robotic welding process can be derived through mixed logical dynamical (MLD) method. This study has very important significance to the modeling and control of robotic welding process. Therefore, this paper studied the visual information capturing and processing of welding process based on a binocular vision sensor for robotic aluminum alloy pulsed Tungsten Inert Gas (TIG) welding. Through careful analysis of the mixed logical dynamical characteristic of robotic welding process, two kinds of mixed logical dynamical models were established. One is for robotic welding system and another one is for weld pool dynamical process based on visual information. The results showed that robotic pulsed TIG welding process was a typical mixed logical dynamical system. This study not only could give a unified modeling framework to describe robotic welding system and weld pool dynamical process based on MLD modeling method, it also gave a good example to apply MLD theory to robotic welding process.The robotic welding experimental platform with the binocular vision sensor was established and the software system of vision information capturing and process for aluminum alloy pulsed TIG welding was designed. This platform not only could extract the vision features of the weld seam and weld pool, modeling and control of weld pool dynamical process also could be realized in this platform.In order to meet the requirements of the modeling of welding process, the geometrical characteristic parameter definitions were given. The arc light spectrum for aluminum alloy was measured during welding and the result showed that arc light intensity between 610nm and 690nm was lower than other ranges. As a result, the arc light between 610nm and 690nm not only gave a sufficient illumination for weld seam and weld pool, strong influence also could be attenuated. Thus, the optical wave pass filter between 610nm and 690nm was chosen and clear images of seam and weld pool were captured.The gray level gradient of seam image was mainly on the vertical coordinate throughout gray level analysis. Thus, the AR-SF (advanced roberts operator based on seam feature) algorithm was proposed and this algorithm could figure out the seam gap and center line of the seam quickly. The weld pool images were different for different penetration states so that it was very difficult to detect the weld pool edge by single method. As a result, the PL-PF (feature point location based on pool feature) algorithm was proposed, which could locate the up, down, left and right point of the weld pool for different penetration states. The topside width and topside length of the weld pool could be easily calculated according to these four points. Thus, a solid foundation for modeling and control of welding process was established. The hybrid characteristic analysis of robotic welding process based on mixed logical dynamical modeling framework was given. The continuous and discrete dynamics of the robot movement process and the welding device operation process were analyzed and then the WRMP-MLD (welding robot movement process mixed logical dynamical) model and the WDOP-MLD (welding device operation mixed logical dynamical) model were established. The validities of these two models are verified through collecting the distrete variables during welding. These two models gave a novel description of robotic welding system and they also gave a new approach for other studies, such as multi-robot system cooperation, circuit control of welding power, and so on.The misalignment and seam gap were two common phenomenons in welding process, which could be detected by visual sensing method. The misalignment and seam gap were considered as discrete variables in this paper. The influences over the weld pool dynamical process were studied and then the WPM-MLD (weld pool with misalignment-mixed logical dynamical) model and the WPG-MLD (weld pool with gap-mixed logical dynamical) model were established. The backside weld width could be estimated according to these two models during misalignment and gap.In order to meet the requirement of penetration control of welding process, two PID (proportional-integral-derivative) control experiments were designed to test the estimation ability of the WPM-MLD model and WPG-MLD model. The results showed that these two models could estimate the backside weld width well.
|
PDF Full Text Request