The idea of using electrical data analyzed by algorithms to assess the quality of the welds produced in robotic manufacturing emerged in 1995 from research by Associate Professor Stephen Simpson at the University of Sydney on the complex physical phenomena that occur in welding arcs. Simpson realized that a way of determining the quality of a weld could be developed without a definitive understanding of those phenomena. The development involved:
a method for handling sampled data blocks by treating them as phase-space portrait signatures with appropriate image processing. Typically, one second's worth of sampled welding voltage and current data are collected from GMAW pulse or short arc weldingprocesses. The data is converted to a 2D histogram, and signal-processing operations such as image smoothing are performed.
a technique for analyzing welding signatures based on statistical methods from the social sciences, such as principal component analysis. The relationship between the welding voltage and the current reflects the state of the welding process, and the signature image includes this information. Comparing signatures quantitatively using principal component analysis allows for the spread of signature images, enabling faults to be detected and identified The system includes algorithms and mathematics appropriate for real-time welding analysis on personal computers, and the multidimensional optimization of fault-detection performance using experimental welding data.Comparing signature images from moment to moment in a weld provides a useful estimate of how stable the welding process is. "Through-the-arc" sensing, by comparing signature images when the physical parameters of the process change, leads to quantitative estimates—for example, of the position of the weld bead.
Unlike systems that log information for later study or that use X-rays or ultrasound to check samples, SIP technology looks at the electrical signal and detects faults when they occur.Data blocks of 4,000 points of electrical data are collected four times a second and converted to signature images. After image processing operations, statistical analyses of the signatures provide quantitative assessment of the welding process, revealing its stability and reproducibility, and providing fault detection and process diagnostics. A similar approach, using voltage-current histograms and a simplified statistical measure of distance between signature images has been evaluated for tungsten inert gas (TIG) welding by researchers from Osaka University.