Acoustic emissions based structural diagnosis technique for steel details

Abstract

Acoustic emissions (AE) are a non‐destructive testing technique used to detect micro‐cracks and defects in industrial structures, including areas that are difficult to inspect visually. In steel structures, welded joints are particularly critical, as they are prone to damage mechanisms such as yielding and high‐cycle fatigue in bridges. AE analysis is often based on simple signal features, such as peak amplitude and duration, which may be insufficient to describe damage evolution. To address this limitation, experimental tests were performed at the University of Trento on scaled, full‐penetration welded steel specimens representing a bridge deck weld. The specimens were instrumented with AE sensors and subjected to low‐cycle and high‐cycle loading to study yielding and fatigue damage. The acquired AE data were analyzed using a transfer‐learning approach based on convolutional neural networks and Mel spectrograms. The results show that the proposed approach performs well in detecting yielding‐related damage and provides interesting indications of fatigue damage evolution. This work is intended to lay the foundations for the development of an applicative AE‐based framework for steel structural diagnosis.