WELDING JOURNAL 61 of welding spots is desired. An ultrasonic sensor built into one of the weld gun electrodes is connected with the RIWA by coaxial cable. The RIWA unit has the fieldbus connection to the weld and robot controller. Once installed, the RIWA unit works as an unsupervised device automatically testing weld quality and sending feedback to the robot. A state-of-the-art algorithm has been developed for automated analysis of Mscans. It processes the weld “image” and recognizes the features of the nugget formation. Morphological analysis of extracted features allows the geometrical parameters of the liquid nugget to be determined and makes a decision about the weld quality. Figure 8 shows a user interface with multiple registered parts and one of the selected M-scans with automatically recognized features. Date/time stamps on the left mark every welded part. Each part has nine welds. As welding through the part progresses, welds are scanned and automatically characterized. At a certain part, the purpose failures were made by dropping the welding current. The system has successfully recognized undersized welds and stopped the robot. At its current state, the system performs unsupervised testing of weld quality and qualifies the results using a threelevel grading: acceptable, marginal, and unacceptable. Processing time of a single M-scan is about 150–250 ms for a 3 GHz Pentium D processor. It depends on the stack thickness and welding time, which determine the width and height of the M-scan. Special algorithms for efficient M-scan processing have been applied (Refs. 12, 13). The processing time requirements are strict since average cycle time is around 1.5–2.5 s/weld. The robot needs to receive feedback before it advances to the next weld. If needed, the operator can have access to the statistics of the production equipment performance — Fig. 9. Besides determining weld quality, the system proved to be useful in detecting nonstandard conditions such as cooling water tube failure shown in Fig. 10. The plot shows ultrasonic TOF through the stack at different production times. In Ref. 8, we have shown a strong correlation of this parameter with nugget diameter. At around 22 h, the cooling water tube was damaged (notice the sudden dip in average diameter) and production continued for a few hours with some welds being made undersized. The system was currently working in a passive mode, but it was able to track back every single weld and identify problematic products for fixing. As a robot makes hundreds of welds, the electrode tip surface experiences deformation and continuous contamination. This leads to excessive heat developed at the copper tip and could possibly lead to cooling water overheat and boiling. The ultrasonic system is capable of monitoring the cooling condition. Figure 11 shows abrupt improvement of the cooling after a tip dressing (cleaning) event. Usually, with bad tip conditions, the last three welds in every row are shown as grey. Grey stands for the welds that for some reason were not interpreted by the software. Additional analyses have shown that those last three greys are due to the water overheat. After tip dressing, those welds become recognizable and turn green. This information can be used to issue recommendations on the tip dressing frequency to optimize production process quality. The ultrasonic testing system communicates with the PLC using discrete I/O, DeviceNet, or other means of communication. The robot tells which part is being loaded, which weld on the part is welded, and when exactly to start ultrasonic scanning. In its turn, the ultrasonic system Fig. 8 — User interface. Fig. 9 — Statistics over the last 24 h. Fig. 10 — Weld quality dynamics over 24 h with cooling water tube failure at around 22 h.
Welding Journal | January 2014
To see the actual publication please follow the link above