A PHASE SPACE APPROACH TO DETECTING VOLUMETRIC DEFECTS IN FRICTION STIR WELDING

摘要

The paper reports on quantification approaches applied to phase space data generated from welding aluminium under varying process conditions. Previous work demonstrated the effectiveness of the Poincare map based method for detecting voids greater than 0.5 mm in diameter. However, to detect micro-voids additional processing appears to be required. This work describes a multi-step micro-wormhole detection algorithm that combines aspects of the phase space approach coupled with an estimate of the regularity of the material flow. The void-detection performance of the phase-space-based algorithm was compared with a previously developed neural network algorithm using a probability of detection analysis (POD). The phase-space-based algorithm provided slightly better detection than the previously developed algorithm. However, both evaluation algorithms could detect defects greater than 0.5 mm with 95% mean POD. These results are competitive with or better than those achieved with other conventional non-destructive methods that are done post welding.