Page 112 - Welding Robots Technology, System Issues, and Applications

Welding Robots
98
mode, the spectrum peak changes when going from the normal welding condition
to a fault welding condition.

When using an extended monitoring system, time information can also be used.
Some of the features refer to the parameters of pulsed GMAW like peak and
background current, peak and background voltage, and peak and background pulse
time. In short-circuiting welding the features are typically short-circuit time, arc
time, peak current or arc voltage.

However, for a simple monitoring system the primary feature to be monitored
when welding in the spray mode is the mean of the weld current. The variance of
the weld voltage can be used when welding in short-circuiting or pulsed mode.
Also, a short-circuiting detector algorithm can be useful when welding in pulsed
mode.

The fault detection algorithm provides variance reduction of the monitored features
as well as a limit detector. The principle of applying a variance reduction
technique, i.e. the filtering of the features, is that it results in increasing detection
reliability by increasing the signal-to-noise ratio for the resulting test quantity, i.e.
the filtered features. The algorithms developed and used (and partly exemplified
here) should be applied to each feature extracted from the signals measured. It is
important to note that data from the normal welding condition is assumed to be
available in order to train or validate the algorithms.

It should therefore be noted that a monitoring system for quality control purposes
should not only measure the defined welding data and compare these with the
nominal, but also be able to calculate the variance of measured data to retain
information about higher frequencies. To do this, algorithms must be adapted for
fast calculations and response. This is needed since the bandwidth of the
monitoring system basically is dependent and selected based on features from the
process to be monitored. Thus, an indication of the frequency of pulses in short-
circuit mode or pulsed GMAW is in the order of 100 Hz or 10ms. During
monitoring, a number of pulses or drop transfers should be monitored as some
could be outliers with respect to process stability criteria without affecting the
quality of the weld. A typical value could be 500 data samples representing about
five pulses or drop transfers. But given this, to define a threshold for correcting
measures of the process is a delicate decision. If one such quality estimate indicates
a change in the process stability, we might want to wait with any control actions
until we have the estimate from the next 500 data samples. The reasons for this are
simply that (i) such short disturbances do occur, (ii) more information is in general
needed to get an indication how to stabilize the process, and (iii) in most cases, the
weld processes do have some quite different characteristics with respect to time;
many parameters and phenomena are truly rapid in nature and need some special
attention to measure and monitor these, but the resulting weld is in general
somewhat forgiving if we are able to react on a disturbance and get the process
back to a stable arc again within a short time. How short this time is, is in principle
dependent on the size of the weld pool. It is of course better to react quickly, but