Page 111 - Welding Robots Technology, System Issues, and Applications
P. 111

Sensors for Welding Robots

                           All parameters except the electrode extension can be measured  directly, but the
                           electrode extension can be estimated from the weld current and the weld voltage.   97

                           The analog signal conditioning  unit carries out low-pass filtering of the analog
                           signals from the sensors in order to prevent aliasing after the A/D converter. As the
                           environment around the measurements of the weld process can be highly disturbed
                           by electrical noise, the filter has the additional task of suppressing such noise. The
                           bandwidth required for the monitoring system is, however, mainly determined by
                           the short-circuit time, which is in the order of milliseconds, or about 500 Hz. The
                           rise time of the whole system should not exceed 10 % of the short-circuit time and
                           thus, the bandwidth needed should be in the order of 5 kHz. This makes it possible
                           to extract features like peak pulse time, background pulse time and short- circuit
                           time with adequate accuracy. Furthermore, high  voltage transients caused by
                           spatter or short-circuitings can also be measured.


                                                                                 computable
                                                        signal   Analog signal   signal
                             Process         Sensors             conditioning
                                                                 and A/D








                                                                       Test
                                Feature                                quantity
                                              feature   Filtering S/N           Level
                                extraction
                                algorithm              enhancement              detector


                                          Figure 3.13. Block diagram for a monitoring system

                           Most modern commercial monitoring systems for GMAW divide the process into
                           time segments or windows and the sample mean for each segment is estimated.
                           These features are filtered and compared with a predefined threshold. As a result of
                           the mean value calculation, information about higher frequencies is lost since the
                           mean value of the signal has gone through a low-pass filtering process.

                           An appropriate feature which retains information about higher frequencies is the
                           variance. This is due to the fact that the variance (or the area below the curve of the
                           power spectral density) is an AC power estimate. A change in  variance due to
                           disturbance is an indication of changes in the power spectral density.

                           Another feature which may be relevant for monitoring purposes is the peak of the
                           weld voltage spectrum. When welding in either the spray arc or short-circuiting
   106   107   108   109   110   111   112   113   114   115   116