PSI - Issue 37

Andrzej Katunin et al. / Procedia Structural Integrity 37 (2022) 292–298 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

293

2

1. Introduction With continuously growing demands to the structural safety and reliability, effective damage identification methods in structures are crucial to react on the identified damage timely and perform necessary repairing or replacement actions. From the variety of available non-destructive testing techniques applied nowadays for this purpose, vibration based damage identification is considered as effective and inexpensive approach widely used in mechanical and civil engineering applications (Mironov and Doronkin, 2021). However, the research results in the last decades clearly indicate that detection and identification of small damage in structures requires additional processing of raw results acquired from a modal analysis. Usually, mode shapes or modal curvatures are taken into account during such a procedure due to their high sensitivity to structural damage. This is due to the ability of detection and localization of local decrease of stiffness indicating damage. To-date, numerous approaches and algorithms have been developed for damage identification in various types of structures. These approaches can be grouped into two main categories: the approaches based on the damage index (DI) concept, involving various processing techniques, and the approaches based on direct application of various transforms. The first category is very large and diverse, and consists of numerous formulations of DI based on early developments, like modal assurance criterion (MAC) and its numerous derivatives (Farrar and James, 1997; Doebling et al., 1998; Görl and Link, 2003), strain energy-based formulations of DIs (Choi and Stubbs, 2004; Bayissa and Haritos, 2007; Hu and Wu, 2009), as well as numerous statistical features (Rezaei and Taheri, 2011; Rucevskis et al., 2009; Janeliukstis et al., 2017; Gorgin, 2020). The second category covers various types of wavelet transforms (Rucka and Wilde, 2006; Fan and Qiao, 2009; Katunin, 2015) as well as some alternative approaches, like time-frequency distributions (Katunin, 2020) or other transforms (Matt, 2013; Jianping et al., 2014; Pan et al., 2018; Katunin, 2021). These two principally different categories need to be assessed in order to evaluate the performance of the representative processing methods from both of them. Within this study, we compare a performance of the selected approaches from both of these groups, namely, the DIs based on smoothing polynomials and the curvelet transform (CT). Both approaches are reference-free, which means that only a damaged structure is considered in the damage identification procedure and no reference data is necessary for a comparison purpose, and thus, has a significant practical meaning. The approach based on DIs is selected due to its proven sensitivity to damage in previous studies (see e.g. Rucevskis et al., 2016; Janeliukstis et al., 2017), while the selection of CT for processing of mode shapes was chosen due to its exceptional filtering capabilities and some other useful properties (Bagheri et al., 2009; Nicknam et al., 2011). The studies were performed on experimental vibration data acquired for artificially damaged laminated composite structures. The advantages of the application of both approaches as well as their performance in terms of proper detection, localization, and identification of single and multiple damage sites were deeply analyzed and discussed. The results of this study show the superior filtering ability of both approaches, which makes it possible to identify small damage in composite structures using vibration data. 2. Experiments and processing methods 2.1. Acquisition of mode shapes The tested structures were manufactured and supplied by Izo-Erg S.A. (Gliwice, Poland). They were cut to the dimensions of 300×300 mm from the 1000×1000 mm sheets of the 12-layered glass fiber-reinforced polymer composite with the thickness of 2.5 mm. The structural damage was simulated using the milling machine. The schemes of the considered structures with spatial dimensions of simulated damage are presented in Fig. 1. In all cases the extrusion was made on the depth of 0.5 mm, which corresponds to 20% thickness reduction. The width of the simulated cracks was 1 mm. Before testing, the plates were clamped along the edges by steel frames connected with bolts and covered with the reflective powder to obtain fine signals from scanning using laser Doppler vibrometer (LDV). The connected frames with a specimen were mounted on the electrodynamic shaker, which provided mechanical excitation with the pseudo random signal. The shaker was connected to a PC through the power amplifier. Two LDVs were used during testing: the point LDV was focused on the frame to obtain the reference signal, while the scanning LDV was used for scanning