Reliability of detection of small defects in noisy weldments by advanced signal processing and pattern recognition techniques

Abstract

Reliable detection of a defect (3 mm long × 1 mm deep) in the top mid-section of maraging steel weldment poses problems during conventional pulse-echo ultrasonic testing. Acoustic noise due to coarse and textured microstructure in the top mid-section adversely influences the signal-to-noise ratio and thus the detection. The reliable detection of a small defect in rocket motor components is essential to achieve the designed payload capacity of the space vehicle. Advanced signal analysis techniques such as autocorrelation, demodulation, etc., were applied to the detection of a simulated fatigue crack on the top mid-section of the weldment. Confidence levels of only 60 to 85% could be achieved. When a combination of cluster analysis of cross-power data and Gaussian distribution function were applied, a confidence level of 95% for defect detection could be achieved.