A simple K-SVD dictionary learning method for baseline-free damage identification purely based on analytical excitation

This paper presents a simple K-SVD dictionary learning method for transferable, baseline-free damage identification on composite coupon panels, relying solely on the analytical excitation. The analytical signal, encapsulating universal patterns of Lamb wave pristine signals, can be segmented, and reorganized as ‘shapelets’. These shapelets serve as a universal function to reconstruct baseline signals and further facilitate damage identification. The key innovation lies in the pure use of the analytical 5-cycle Hanning-windowed toneburst signal instead of requiring extensive experimental data, which minimizes data collection efforts and significantly lowers computational time. The proposed methodology is comprised of four steps: candidate generation, K-SVD dictionary learning, dictionary compilation and sparse coding, which has a wide application to different materials, temperature fields, specimen sizes, sensor configurations, etc. Prior to the implementation of the proposed method, a standard signal pre-processing pipeline is operated to enhance the data quality. In this study, experimental comparisons were conducted between the proposed baseline-free K-SVD, time reversal (TR), instantaneous baseline (IB), and reciprocity approach, leveraging data from composite coupon panels. Furthermore, it was extended to detect the barely visible impact damage (BVID) under a temperature range of [20: 5: 50] °C. Consequently, the effectiveness and robustness of the method have been validated based on the high accuracy and consistent localization performance across different damage scenarios.