Page 97 - Welding Robots Technology, System Issues, and Applications

83
Sensors for Welding Robots

The feature recognition process includes the following basic tasks: (i) identification
and elimination of outliers, (ii) contour generation of the weld profile and
generation of line segments based on predefined templates, (iii) merging line
segments, and (iv) validation of joint parameters so that they are within predefined
tolerances and match the joint template, see Figure 3.5.

From a control point of view, seam tracking is usually performed with full
compensation of the position error measured. Seam tracking is typically only
performed using a nominal path. The nominal path is the assumed trajectory of the
weld joint and during tracking, the robot controller receives new target positions
from the sensor and the robot controller overrides the nominal path by changing
the position of the TCP while keeping a constant orientation. This has some
benefits and drawbacks. The benefits are that, given a nominal path, it is rather
straightforward to verify the ability of the robot to follow the path with some minor
changes while keeping the orientation constant. This means that issues related to
joint limits, singularities and possible collisions are minimized. Drawbacks are that
the user must define and program a nominal path.

If, instead, the robot is only instructed where to start and where to end, it must be
able to follow the path measured by the sensor on the fly. This put some additional
requirements on the robot system as it must be able to calculate the trajectory
including both target positions and orientations of the weld torch. In doing so, the
robot might easily get into control problems for the specific task that is related to
close to singular areas, joint limit and collisions with the work-piece. However, the
benefits high and more flexible control of seam tracking can be obtained using
available sensors.

In order to use the sensor data and let the robot follow the measured and generated
trajectory, the target data from the buffer of target positions must be filtered. A
suggested method is to use about five target positions and generate a polynomial
curve for x, y and z and a vector tangent at the current position. This is useful for
several reasons: (i) a generalized description of the weld joint makes it easy to
reuse the generated trajectory for calibration purposes, (ii) the generated trajectory
can be used as a nominal path for subsequent weld passes, (iii) the vector tangent
can be used for subsequent target positions even if target drop-outs occur such as at
tack welds and (iv) the vector can be compared with the current trajectory vector
and from that an optimal orientation of the weld torch can be calculated with
respect to the weld joint as measured by the laser scanner.

Typical operating data of a laser scanner is a scan sweep frequency of 10-50 Hz. If
we assume a welding speed of 20 mm/s, this means about one sweep per mm
during welding. This is in most cases more than sufficient. However, new welding
processes such as laser welding will increase the welding speed considerably and
for high requirements careful analysis and trials must be made. The accuracy of the
laser scanner is high and is better than 0.1 mm. However, it should be noted that
single scan sweeps may generate outliers as the weld joint and the environment
during welding generate severe disturbances.