Fig. 4 — Weld pool characteristic parameters (width, length, convexity) and welding current adjusted by the skilled human welder in nine dynamic experiments. to maintain the same penetration state after the change of the welding speed. The arc length for these six experiments was set at 4 to 5.5 mm. During each individual experiment, the arc length was constant. In experiments 7 to 9, the arc length was varying from 4 to 5.5 mm, and the human welder adjusted the current in response to the fluctuations of the weld pool. The welding speeds in these three experiments were between 1 and 1.5 mm/s, which remains unchanged during each individual experiment. The other experimental parameters are shown in Table 1. The dynamic variation of weld pool geometry and the adjustment of current by the human welder were recorded, respectively. It was noticed that in these experiments the skilled human welder was able to control the welding process to desired penetration states. The obtained backside bead width in these experiments are within certain ranges (4 to 6 mm), which is considered acceptable in our applications. During the welding process, the human welder always scans the weld pool with a certain frequency regardless of the welder’s eye blink. The system’s sampling frequency is 2 Hz in this study. There is also a finite time delay existing in the linear model of the human welder’s behavior as a result of the neuromuscular and central nervous latencies (Refs. 16, 24). Based on the step response experiments, it was observed that the average time delay of the welder’s response was approximately 1.5 s, or three sampling periods. Figure 4 plots the measured input parameters (the weld pool width, length, convexity) and welding current adjusted by the skilled human welder. It can be seen from Fig. 4 that the tendency of current adjustment is roughly opposite to the length and width fluctuation. The trend of convexity basically coincides with the current variation. The variation of the geometry and the current indicates the human welder reduces the current as the weld pool length or width increases, and increases the current as the weld pool convexity increases. However, nonlinearity does occur in human welder adjustments corresponding to the weld pool characteristic parameters. Specifically, in sample number 1230 to 1280, the human welder increases the current as the weld pool width and length decreases, and the weld pool convexity increases. In sample numbers 1700 to 1800, on the other hand, the human welder increases the current while the width, length, and convexity all decrease. This indicates that the skilled welder response to the weld pool characteristic parameters are indeed nonlinearly correlated and may be better modeled by a nonlinear model rather than simple linear model. Human Response Dynamics It is apparent that a human welder makes adjustments based on multiple weld pools he/she observes, rather than a single pool at the current instant. However, the increase in the number of input parameters will complicate the nonlinear neurofuzzy model, and is thus not preferred. In this study, filtered weld pool characteristic parameters are utilized as the inputs of the neuro-fuzzy model, which combine the information from multiple weld pools. The designed filter can be expressed as CPf(k) = αfCPf(k–1) + (1 – αf) CP(k) (1) where CPf (k) = Wf (k), Lf(k), Cf (k)T and CPf (k–1) are the filtered weld pool characteristic parameters at instant k and k–1, with W, L, C representing the width, length, and convexity, respectively. CP(k) A B Fig. 5 — Modeling errors for both linear model and ANFIS model vs. filtering coefficient. is the measured weld pool characteristic parameters at instant k. αf is the coefficient of the filter, which should be a tradeoff between the filtering effect and response speed. It is shown in the following figure that αf = 0.6 gives a smaller FEBRUARY 2014 VOL. 93 48-s WELDING RESEARCH Table 1 — Experiment Parameters Welding Parameters Current/A Welding Speed/mm/s Arc Length/mm Argon Flow Rate/L/min 57~81 1~2 4-5.5 11.8 Monitoring Parameters Laser Projection Laser to Weld Pool Imaging Plane to Weld Pool angle/deg Distance/mm Distance/mm 35.5 24.7 101 Camera Parameters Shutter Frame Rate/fps Camera to Imaging Plane Speed/ms Distance/mm 4 30 57.8
Welding Journal | February 2014
To see the actual publication please follow the link above