Intelligentized Methodology for Arc Welding Dynamical Processes

In this chapter, an introduction is given on the development of welding handicraft, manufacturing technology and key technologies of welding automation and intelligentization. Recent twenty years have seen great development of welding robot in modern manufacturing industry, where arc welding is one of the mainstream technology. A large number of researches show that automatic control of the welding process requires not only good performance of the equipment, but also technologies, namely sensing, modeling and controlling of the welding process. None of the technologies is neglectable for welding process control, in which sensing is to monitor the process and extract characteristic information of the welding process; modeling is to identify the process based on acquired information; and controlling is to regulate the welding process based on the established models. The main part in controlling is to design controller for multi-variables coupled, nonlinear and time-varying situations.

Visual sensing technology is widely used in welding practices because visual devices are decreasing in price, increasing in reliability and improving in image processing hardware and software. As the most studied welding sensor, CCD(Charge Coupled Device) is more suitable for quality control of welding process than other means of sensing devices because it can obtain both two dimensional and three dimensional information of weld pool, which directly reflect the welding dynamics of molten metal. In this chapter, according to the analysis of arc spectrum and radiation of different materials, visual sensing systems with filters are described. Based on the filtering method, clear images of weld pool are obtained during pulsed GTAW. The first step of intelligentized arc welding is to imitate the visual system of a welder to extract weld pool size. Passive visual sensing technology has seen great progress in the past years for the abundant information it extracts. Brzakovic et al. [1] obtained the weld pool image in two directions and extracted its geometry information. Wang et al.[2] attempted to get aluminium alloy image for the first time, but it is not clear enough. Zhao et al. [3] developed a passive three-dimension visual sensing method through monocular image which is processed by Shape from Shadow (SFS) algorithm to get the three-dimension geometry of the pool. In this chapter, passive visual sensing system for both aluminium alloy and low carbon steel will be discussed.

Precise image processing algorithm is important for welding process control. Generally, original image cannot be directly used due to the disturbance from welding equipment. Moreover, fluctuation in welding current and arc light also lead to image degrading. All the above factors add difficulties to the image processing, and the image processing algorithms are required to be adaptive to different conditions. In this chapter, both 2D and 3D image processing methods are described. The 2D image processing methods used in this chapter include degrading image recovery, integral edge detection, projection, neural network edge identification and curve fitting to extract the length and width of the weld pool. 3D image processing methods include experimental and SFS(Shape-from-Shading) method to extract topside height of the weld pool. And image processing software exclusively for weld pool images is introduced at the end of the chapter. Real time control of weld pool dynamics is crucial for welding quality, which depends primarily on extracting and calculating geometric characteristics of the weld pool [1-4]. The weld pool contains abundant information about the welding process. Actually, in practice, a skilled welder can estimate the appearance of backside of weld pool by observing the shape, size and dynamic change of the topside of the weld pool and adjust accordingly. Image processing is aimed to obtain the relevant information by enhancing the necessary image features and suppressing undesired distortions. However, many disturbances, such as alternating magnetic field and the relative motion between CCD and weld pool, will affect the information acquirement. Therefore, image processing technology is necessary for the welding process.

GTAW is a thermal process during which the workpiece melts, solidifies and finally forms the welding seam. As is well known, arc welding is influenced by many complex factors, such as material metallurgy, heat conduction, physical chemistry reactions, etc. Due to its multi-variable coupling, nonlinear, time-varying, random and uncertain properties, GTAW dynamics is difficult to be modelled by classical linear system theory. In this chapter, analysis on the welding dynamics is made to understand the process of welding. Based on the analysis, both identification models and intelligent models, e.g. ANN, fuzzy rules model and RS-based model are discussed. ANN model is a “black box" and it is impossible to directly revise the model. For fuzzy rules model, the number of inputs, outputs and their linguistic variables cannot be too large, or it will lead to “rule explosion". RS model is promising for welding process modeling because compared with NN model, RS model is close in predictive ability; and however its complexity is much lower. GTAW is a thermal process during which the workpiece melts, solidifies and finally forms the welding seam. As is well known, arc welding is influenced by many complex factors, such as material metallurgy, heat conduction, physical chemistry reactions, etc. Due to its multi-variable coupling, nonlinear, time-varying, random and uncertain properties, it is very difficult to model welding dynamics by classical linear system theory. In recent years, some intelligent modeling methods have been introduced to welding. References [1-3] investigated fuzzy reasoning application in modeling, and Refs.[4-8] studied artificial neural networks for modeling. In this chapter, both identification models and intelligent models are discussed for the weld pool dynamics during pulsed GTAW.

As a complicated process with non-linearity, time varying and uncertainties, GTAW is difficult of modeling and control by classical linear system theory. Intelligent techniques are required to model and control such systems. Among many different “intelligent” approaches, neural network and fuzzy methodologies have emerged as powerful tools owing to their capabilities of emulating human learning processes. And model-free controller is another promising control strategy, which only needs the observed input output data, but can improve the performance of the controller. Based on the visual sensing technology and modeling methods mentioned in the above sections, several different intelligent controllers are described in this chapter, including PSD, Neural Network, model free controller, composite intelligent controller and they are also compared with the open-loop experiment

Current teaching play-back welding robot is not with real-time function for sensing and control of weld process. This chapter addresses the real-time vision sensing and intelligentized control techniques for robotic arc welding. Intelligentized robotic system includes a computer visual sensing system with image processing algorithms, weld pool penetration controller and seam tracking controller. Using composite filtering technology, a computer visual sensing system is able to capture clear weld pool images during robotic pulsed GTAW. Corresponding image processing algorithm is described to pick-up characteristic parameters of the weld pool in real time. Furthermore, intelligentized models and real time controller of weld pool dynamics during pulsed GTAW process are discussed in the robotic systems. Seam tracking is another key technology for welding robotic system. Image processing algorithms are presented to extract the seam trajectory and the offset of the torch to the seam in the weld pool images with grooves. An application of intelligentized welding robot systems is also described at the end of this chapter.