Frequency-time characteristics of air-borne signals from laser welds

Abstract

The acoustic emissions from laser lap welds in stainless steel sheet were recorded and analyzed. The acoustic signals emanating from the weld were sensed with an instrument microphone and analyzed using short time Fourier spectra to characterize their time-frequency distributions. It was determined that the acoustic spectrum of good quality, full penetration could be differentiated from the spectra of poor quality welds, defined as either partial penetration welds or welds having a gap between the sheets being joined. A simple classifier based on total energy in the frequency band from 1 KHz to 2 KHz correctly discriminated full penetration from partial penetration over 90 percent of the time on average. Partial penetration welds had less energy in this range than did full penetration welds. A slightly more sophisticated approach incorporating time-averaging was found to be capable of predicting penetration with reliability approaching 100 percent. A classifier based on total signal energy was shown in preliminary trials to be capable of recognizing gapped lap welds from non-gapped lap welds.The acoustic emissions from laser lap welds in stainless steel sheet were recorded and analyzed. The acoustic signals emanating from the weld were sensed with an instrument microphone and analyzed using short time Fourier spectra to characterize their time-frequency distributions. It was determined that the acoustic spectrum of good quality, full penetration could be differentiated from the spectra of poor quality welds, defined as either partial penetration welds or welds having a gap between the sheets being joined. A simple classifier based on total energy in the frequency band from 1 KHz to 2 KHz correctly discriminated full penetration from partial penetration over 90 percent of the time on average. Partial penetration welds had less energy in this range than did full penetration welds. A slightly more sophisticated approach incorporating time-averaging was found to be capable of predicting penetration with reliability approaching 100 percent. A classifier based on total signal energy was shown in pr...