Sensor Fusion Checks Weld Quality
Robert C. Pini
COLUMBUS, Ohio -- When it comes to inspecting high-power laser welding, three sensors are better than one, according to engineers at the Edison Welding Institute. They found that optical, acoustic and charged-particle sensors can singly inspect welds with modest accuracy. But combining the methods online significantly improved accuracy.
Dubbed "sensor fusion," this technique of multiple measurement can be applied to any high-power "keyhole" laser welding process, explained Dave Farson, who conducted the tests at the institute.
Farson, an assistant professor of engineering at Ohio State University, said that automotive and aerospace industries are investigating the multiple sensing technique. Automakers are testing its application in production of outsourced items, such as for welding tailored blanks and air-bag inflators. They want to find ways to document the quality of outsourced work.
Currently, manufacturers periodically sample parts to inspect laser welds. They also use off-line nondestructive and destructive testing. Farson said optical sensing is the most commonly used testing method but pointed out that sensor fusion would allow parts makers to test all welds rather than just a sample.
To check the weld quality, automakers use optical sensors that consist of a silicon photodiode and
a transimpedance amplifier to gather information about the size and fluctuation of the laser plasma. The data correspond to weld quality characteristics, such as penetration depth or gaps. But they can also correspond to factors that do not affect weld quality, such as minor changes in shielding gas or the material being welded.
Accordingly, the challenge is to record a signal with low noise from extraneous factors. That, Farson said, is where sensor fusion excels. Using a statistical approach known as linear discriminant analysis, Farson can extract better information on weld quality by processing the three signals and distilling the most relevant information from them.
With sensors mounted, the researchers welded low-carbon steel sheet metal, deliberately creating difficult-to-measure defects such as poor fusion and gaps. Alone, the optical sensor correctly identified 61 percent of weld defects. The acoustic sensor identified 75 percent correctly, and the charged-particle sensorachieved 69 percent. Together, the sensors realized 83 percent accuracy.
Farson said he and his colleagues are working with manufacturers to integrate inspection systems that combine the signal analysis technique with commercially available sensing and computing products. The Big Three US automakers and many of their suppliers are members of the Edison Welding Institute.
MORE FROM PHOTONICS MEDIA