Weld energy reduction by using concurrent non-destructive evaluation

Abstract

The use of ultrasound to inspect multiple-pass welds on a pass-by-pass basis has the potential of reducing energy costs and improving productivity in industrial welding. For such a concurrent welding/NDE system to be feasible two problems must be solved: first, the inspection must be done in near real time since the benefits of the process require that any defects must be found and repaired before being covered by successive weld passes; second, the presence of benign geometrical reflectors at the corners of the weld preparation and at the top of the partially completed weld may mask flaws such as lack of side-wall fusion which will occur in the same physical location. Pattern recognition techniques solve both these problems. In these tests partially completed welds are inspected with an ultrasonic search head. The features are derived from the cues that humans use to discriminate the flawed from the good weld. These features are compared with those from a reference set which are statistical averages of the features of echo signals from known flaws and known good welds. Laboratory tests show a flaw detection rate of 90–95% with a false call rate of 8%.