Page 110 - Welding Robots Technology, System Issues, and Applications

Welding Robots
96

A number of defects in the weld joint will only give rise to minor changes in the
weld voltage and current or some other process parameters. As an example, a butt-
weld is welded; after some distance, the weld torch is not following the weld joint,
see Figure 3.12. By only measuring the welding parameters, only small changes
can be observed. Modern monitoring systems are however designed to extract
features from the data that highlight weld faults. Another problem is that too
sensitive detection will generate false alarms. This can be handled by regarding
random fluctuations as noise and the monitoring system must be able to observe
and classify such signals as a random variable.

To select a detection threshold, two conflicting requirements must be considered.
First, the threshold should be low enough to ensure that the probability of detection
is not too small. Second, the threshold should be high enough to ensure that the
false alarm probability will not be too large. For example, in practice, the false
alarm probability must be low when welding hundreds of meters of tube. The
welding process must not be stopped every meter due to false alarms. A simple
detection algorithm can be realized by simply comparing the actual variance of the
weld voltage (AC power level) with a particular preset level referring to a normal
welding condition. In this way the difference between the preset level and the
actual level is measured. A small difference will indicate a normal welding
condition, while a large difference indicates a fault condition. For short welds the
situation is the opposite and the system must be more sensitive to process
disturbances which can produce quality problems. A general problem is that there
is a run-in phase of the weld process which is difficult to monitor with respect to
quality control. This means that a smaller proportion of short welds can be
monitored with respect to quality and also for counteracting control. In practice,
different strategies can be used, specifically if the same welding operation is made
many times. In such cases template matching techniques can be used. If the weld is
for low volume or one-off production, generic methods must be applied.

3.8 Design of a Monitoring System for Quality Control

The task of a monitoring system for quality control is to extract some features of
the welding process, filter these as necessary and make use of them to detect
changes in the process. The system consists of sensors, a signal conditioning unit, a
feature extraction algorithm and a fault detection algorithm, see Figure 3.13. For
control purpose, specific control algorithms must be included which should include
the capability to map the monitored data with respect to quality and productivity
specifications.

Sensors provide information about the system being monitored like the weld
current, the arc voltage, the weld speed, the electrode extension, the wire feed rate
and the shielded gas flow. They are all representing properties that affect the weld.