Skilled Human Welder Intelligence Modeling and Control: Part 1 — Modeling Experiments using an innovative vision system based on the skill of an experienced welder were conducted to adjust the weld pool surface in real time Skilled human welders’ experiences and skills are critical for producing quality welds with the manual gas tungsten arc welding (GTAW) process. In this study, a skilled human welder’s response to 3D weld pool surface was modeled. To this end, an innovative vision system was utilized to measure in real time the specular 3D weld pool surface under strong arc interference in the GTAW process. Experiments were designed to produce random changes in the welding speed and voltage, resulting in fluctuations in the weld pool surface. A skilled human welder made adjustments on the welding current based on his/her observation of the weld pool and these adjustments were then recorded. Adaptive neuro-fuzzy inference system (ANFIS) was proposed to correlate a skilled human welder response to the fluctuating 3D weld pool surface and previous welding current adjustment made by the welder. It was found that the proposed ANFIS model can model the human welder intelligence with acceptable accuracy. The resultant FEBRUARY 2014 VOL. 93 46-s WELDING RESEARCH model will be compared with the model derived from a novice welder, analyzed, and utilized to control the GTAW process to achieve consistent complete joint penetration under different initial current and various disturbances in a future study. Introduction BY Y. K. LIU, Y. M. ZHANG, AND L. KVIDAHL ABSTRACT In the manual gas tungsten arc welding (GTAW) process (Ref. 1), skilled human welders can appraise the state of the weld joint penetration by observing the weld pool and adjusting the welding parameters accordingly to control the welding process to a desired penetration state. Because of their versatile sensory capabilities and experience based behavior, they are sometimes preferred over mechanized welding machines. However, inconsistent concentration, fatigue, and stress do build up such that welders’ capabilities may degrade during daily operations. On the other hand, the performance of automated welding machines can be maintained/guaranteed. The mechanism of a skilled welder’s experience-based behavior thus should be fully explored and utilized to develop intelligent robotic welding systems that combine a human welder’s intelligence and physical capabilities of the mechanized welding machines, which paves the foundation for next-generation manufacturing processes. Modeling skilled welders’ responses, i.e., how they respond to their sensed information, plays a fundamental role in facilitating such a development. In addition, the resultant welder response model may also be utilized to understand why less skilled welders are not performing as well as skilled welders. The welder training process can thus be accelerated in order to help resolve the skilled welder shortage issue the manufacturing industry is currently facing (Ref. 2). Extensive research has been performed to observe the weld pool using various sensing techniques (Refs. 3–10). Among these sensing methods, the vision-based sensing method has received considerable attention. The weld pool geometry is believed to provide abundant information about the state of the welding process (Refs. 11–13). At the University of Kentucky, a vision-based 3D weld pool sensing system for the GTAW process was recently developed and the weld pool was characterized by its width, length, and convexity (average height of the weld pool) (Ref. 14). However, despite the successes in monitoring and characterizing the weld pool, the interpreting and modeling of the mechanism of human welder behavior remains challenging (Refs. 15, 16). Neuro-fuzzy approach (i.e., the fusion of the neuro networks and fuzzy logic) determines the parameters in fuzzy models using learning techniques developed in neural networks, and has been successfully applied in various areas (Refs. 17–19). Jang (Ref. 20) developed the adaptive neuro-fuzzy inference system (ANFIS) by using a hybrid learning procedure. It possesses the advantages of adaptive rulechanging capability, fast convergence rate, and does not require extensive experiences about the process to construct the fuzzy rules. Recently, ANFIS was employed to model nonlinear functions, identify nonlinear components in control systems, and predict chaotic time series (Refs. 21–23). In this paper, a neuro-fuzzy model of skilled human welder intelligence is presented. Welding experiments were conducted by a skilled welder and the specular 3D weld pool surface was real-time measured by an innovative vision system (Ref. 14). A neuro-fuzzy model is constructed to correlate the welder’s adjustments to 3D weld pool characteristic parameters. In a future study, the proposed human welder model will be compared with that of the novice welder and further utilized to develop the intelligent welding machine that possesses human welder intelligence, yet free from human welder drawbacks. Background In this section, the principle of human welder’s behavior in performing a welding task and intelligent welding machines are briefly described. The diagram of the human welder’s behavior is shown in Fig. 1A. Given a certain welding task, a human welder starts with some initial estimation input I, including the current, arc length, welding speed, etc. After the initial input of the welding process, the welder perceives direct information Ω´ from the weld pool. This sensing process is perturbed by a noise representing the randomness of the human welder such that the perceived information slightly deviates from the actual information Ω. The human welder then compares the information observed from the welding process Ω′ and a certain goal in the welder's KEYWORDS Skilled Welder Intelligence Weld Pool ANFIS Modeling Machine Vision GTAW Y. K. LIU, and Y. M. ZHANG (yuming.zhang@ uky.edu) are with the Institute for Sustainable Manufacturing and Department of Electrical and Computer Engineering, University of Kentucky, Lexington, Ky. L. KVIDAHL is with Huntington Ingalls Industries, Pascagoula, Miss.
Welding Journal | February 2014
To see the actual publication please follow the link above