Puplished at: Measurement

Abstract:
Non-destructive testing (NDT) of additively manufactured (AM) metal components typically relies on costly imaging or ultrasonic systems. We introduce a low-cost mechanical tapping device combined with a machine learning (ML)-based acoustic measurement workflow, and demonstrate its superiority as a measurement system compared to standard fundamental frequency analysis approaches. We treat the classification pipeline as a calibrated “soft sensor” that outputs a defect probability. While maintaining a very simple mechanical tapping system, we hypothesize that introducing Type A measurement uncertainty and fusing multiple probabilistic outputs significantly improves decision accuracy. Furthermore, we demonstrate that varying the tap location constitutes a distinct measurement modality, offering validation beyond the benefits accrued from mere repetition. Using a dataset of 80 Ti–6Al–4V specimens with defects validated by computed tomography (CT), we experimentally show that the proposed multiple mechanical tapping fusion method significantly reduces the probability of error.

By Oshrit Hoffer and Oz Golan 2/6/2026

Full article