Cook's distance-based regression to enhance clustering of metal additive manufactured components through a resonance testing system

Resonance Testing (RT) represents a simple and fast pass/fail solution particularly suited for inspecting small mass-produced hard components without requiring radiation, scanning, or immersion in liquids, chemicals, abrasives, or other consumables. However, the variability introduced by the Additive Manufacturing (AM) process can mask the presence of defective parts. A set of 220 AM cylindrical Ti6Al4V specimens was inspected with an Acoustic Resonance Testing (ART) system to cluster compliant parts and specimens with a seeded defect, present in 18 of them with a threadlike shape. The dataset analysed consists of specimens build plate positions and response spectrum peak frequencies, captured by a set of detection windows through Simcenter Anovis software. To increase the performance of a clustering algorithm like HDBSCAN the systematic variability of resonance frequencies along the build plate positions has been compensated in the pre-processing step through a Cook’s distance-based linear regression.