PSI - Issue 80

Haomiao Fang et al. / Procedia Structural Integrity 80 (2026) 53–64 H. Fang et al./ Structural Integrity Procedia 00 (2025) 000 – 000

56 4

3. Methodology 3.1. Signal Post processing

In this section, a standard signal post processing pipeline is proposed prior to the implementation of the damage identification algorithm to enhance the quality of signal data, ultimately improving identification accuracies. The pipeline to be followed is illustrated in Figure 2. The post processing pipeline primarily consists of four steps, including noise removal, tone burst reconstruction, dispersion removal, and first wavepacket extraction.

Figure 2: Illustration of the standard signal pre-processing pipeline. In this study, a broadband chirp signal is selected as excitation signal, with an initial frequency 1 =10 , a final frequency 2 =600 , allowing for the reconstruction of any narrowband toneburst signal within this frequency range. The excitation chirp signal is defined by (2). ℎ ( ) = 1 (2 1 + 2 − 1 2 ) [ ( ) − ( − ℎ )] (2) Where = 200μs is the duration window, 1 =6V is the amplitude. For data collection, 10 consecutive measurements are performed, and the acquired signals are averaged to enhance the signal-to-noise ratio and eliminate the environmental noise. Subsequently, toneburst reconstruction is conducted to obtain the response of a narrowband toneburst excitation at any central frequency within the bandwidth. In this study, a central frequency of 50 kHz is selected, since A0 mode exhibits dominant responses at low frequencies and can be well separated from other modes. To eliminate the boundary reflection and mode conversion, the dispersion removal is implemented by replacing the original dispersion relation with the first-order approximation with linear dispersion term, which is referred as linear mapping. Refer to [18, 26] for more details of dispersion removal technique. Ultrasonic guided wave propagation depends on the properties of the media through which they travel. The presence of damage — such as cracks, corrosion, or delamination — alters the original wave propagation characteristics and leads to scattering. The damage scatter can then be used to define damage indexes and support damage characterization. Following the previous steps, the final step involves the extraction of the first wave packet to capture features for identifying damage presence and states. Since the plate is a quasi-isotropic laminate, the group velocity is approximately uniform in all directions, which allows for a faster identification of the first wave packet peak, as is defined in (3). = + 1 2 (3) 3.2. K-SVD baseline-free detection approach based on analytical excitation signal. A shapelet is a time-series subsequence that is used to capture local, discriminative patterns in Lamb wave pristine data, providing a means to identify and utilize distinctive features to differentiate between classes. In this study, it is assumed that a Lamb wave time series can be decomposed into a set of shapelets that represent the most universal and characteristic information shared by all pristine signals. These shapelets are trained and learned from candidates