A novel data-driven K-SVD transferrable baseline method for multi-damage identification for composite fuselage panels

Abstract

This paper presents a novel, scalable, data-driven, and transferrable baseline framework for damage detection and localization in large, complex composite structures operating under in-service conditions, such as varying temperatures. The proposed approach addresses a key challenge in the adoption of ultrasonic guided wave Structural Health Monitoring (SHM) for damage detection in large structures: the need to collect baseline measurements.

The proposed framework detects damage in composite fuselage panels by creating a dictionary of pristine data from simpler structural components, i.e. flat composite coupon panels and mono-stringer elements. By utilizing shapelets generated at these lower structural levels, the methodology can be extended to more complex geometries, accommodating variations in temperature, material properties, sensor configurations, and specimen sizes, making it suitable for structures in service. The framework is built around the K-SVD algorithm, which efficiently captures the most representative shapelets while significantly reducing computational time. The reliability and accuracy of the algorithm are validated using pristine coupon data to minimize false alarms, and the approach is then scaled to address more complex curved fuselage panels. Furthermore, the framework incorporates a reliability-based threshold into the K-SVD algorithm for the first time, establishing a detectability threshold for this transferrable baseline method and enabling both damage detection and severity assessment. The effectiveness and robustness of the methodology are demonstrated experimentally using data from large 5-meter curved fuselage panels subjected to varying temperatures and multiple damage scenarios. The results show a true positive rate of 0.951 and a false negative rate of 0.944 for damage detection in the curved fuselage panels within a temperature range of 25°C to 50°C.