2. Through-the-Arc Sensing
2.7 Plant Identification and Control
Open-loop tests were performed on the welding process to establish a model that can be used to tune the PID controller. The derivative action was not used since the current measurement was noisy. The oscillation data, in terms of the width and the pitch of oscillation, had to be adjusted until the left and right peaks were distinctly detectable. A closer weaving of the welding torch resulted in distortion of current peaks due to previously deposited metal being detected by the current sensor. A typical width of oscillation of 6 mm and a pitch of oscillation of 4 mm (see Figure 3) were seen to give the best results. Needless to add that the other process
parameters, such as, wire feed rate, voltage, and the inert gas flow rate have to be adjusted for optimal operation during welding.
Open-loop tests were conducted by injecting a pulse of -2.5 volts at the controller output and carrying out welding on a given piece. The response of the robotic welding system at the output of the moving average filter is monitored and plotted. From the open loop response so obtained, one can easily calculate the first order plus dead time (time delay) response of the plant using standard techniques [12]. The pulse direction is reversed and the open loop tests were repeated.
Tuning formulae are then used to calculate the optimal tuning parameters for the PI controller [12] given by:
m=Kp[e + (l/Ti)/edtJ (4)
where m is the controller output, e (error) is the controller input, Kp is the proportional gain and Tt is the integral time.
The optimal settings were then set on the PI controller. The plant is put in a closed loop and the welding is carried out to verify the closed-loop system performance under a lateral disturbance, which was injected to the torch. Using the results, the PI controller settings were fine-tuned until satisfactory performance in the closed loop was obtained under disturbance.
(Seam)
Filtered peak current
difference X
PID
controller w
Robot direct
control w
3-Axis Robot
i
w
'
Filters Peak difference
Weld current detector
Weld process Figure 12 Control system for seam tracking using through-the-arc-sensing.
Figure 12 shows the block diagram of the control system used. The output of the system is the torch position which has a reference 'zero' corresponding to the seam position at any given time along the programmed welding path. An open loop test is performed by opening the loop at the point marked 'X' in Figure 12. A test signal of about Ap = 2.5 V is injected at the output of the PI controller. The result is shown in Figure
13.
Output (volts) 0.000005
Figure 13 Open-loop process reaction curve test.
From Figure 13, using the process reaction curve method, one can calculate:
L = 3 s, T= 4 s, ACp = 1.2 x 10~5 volts A r = A Q > / r = 3 x l ( r6v o l t s / s
Kp = 0.9 xApl{NL) = 2.5 x 105 l/T, =1/(3.331) = 0.1
where L is the time delay in seconds, and T the plant first-order time constant in seconds, and ACp the plant open-loop steady state response in volts to a step input of 2.5 volts.
It was found that a scale factor of 105, m&Kp = 1.0, 7]? = 0.1 provided the best performance for the controller. Figure 14(a) shows the result of open loop test with a pulse injected at position ' X ' . The weld can be seen to be veered off from the seam. Figure 14(b) shows the result of pulse test under the closed loop. It can be seen that the effect of the pulse has been corrected and the seam is followed by the weld.
\tl/ \~S£J\sU.-lAJKjy
(O) uosea-ioop Figure 14 Open-loop and closed-loop comparison with pulse injection.
Another test involved was to offset one of the end points off the seam by 5 mm. In the open loop, the torch followed the pre-programmed path
without making any correction resulting in a weld, which is away from the actual seam, as demonstrated in Figure 15(a). Under the closed loop, the PI controller corrected the torch position during welding continuously, resulting in a weld which is exactly on the seam as can be seen from Figure
15(b).
(a) Open-loop (close-up view)
(b) Closed-loop (close-up view)
Figure 15 Open-loop and closed-loop comparison (ramp input).
Figure 16 Angle weld.
Figure 15 shows the actual results on a 280 mm long fillet-joint. We tested our controller on a piece of angle weld by teaching only start and end points. The result showed that it followed the seam instead of joining these two points (see Figure 16).The results obtained through welding under the closed loop, as described above, are representative of many such trials conducted to verify the closed-loop design and implementation.