Assessment of Transfer Learning Capabilities for Fatigue Damage Classification and Detection in Aluminum Specimens with Different Notch Geometries

Fatigue damage detection and its classification in metallic materials are persistently challenging the structural health monitoring community. The mechanics of fatigue is difficult to analyze further complicated because presence notches different geometries. These act as possible crack-nucleation sites resulting failure mechanisms that drastically from one another. Often, sensor-based tools used monitor detect critical such aluminum alloys. Through deep neural networks (DNNs), a approach can be ubiquitously extended for variety geometries appropriate applications. To end, this paper presents DNN-based transfer learning framework classify across candidate notch DNNs built upon ultrasonic time-series data obtained during testing Al7075-T6 specimens with two types geometries, namely, U-notch V-notch. baseline DNN shown achieve an accuracy 96.1% while V-notch has 95.8%. Both are, thereafter, subjected process by keeping certain number layers frozen retraining only remaining small volume other geometry. When layer retrained just 10% total data, above 90% observed specimens. Similar results also when interrogated results, summary, demonstrate data-thrifty quality combining concepts made high-performance