PSI - Issue 5

Alejandro Carvajal-Castrillón et al. / Procedia Structural Integrity 5 (2017) 729–736 Alejandro Carvajal-Castrillón/ Structural Integrity Procedia 00 (2017) 000 – 000

732

4

reported by Jardine (2006), pattern recognition is achieved by classification of signals based on features extracted from them, this technique is divided into two approaches: statistical and artificial intelligent approaches. Sohn et al. (2001), presented a study to evaluate the feasibility of using statistical pattern recognition for SHM in a bridge structure. The authors argued that this approach has the advantage of allowing noise from strain and acceleration data; however, in vibration-based damage detection, it is necessary to have clean data. However, pattern recognition techniques based on artificial intelligent algorithms is the most widely used. These techniques are artificial neural networks, evolution strategy, fuzzy logic systems, fuzzy neural networks, and evolutionary algorithms, as explained by Jardine et al. (2006). Figueiredo et al. (2014) developed a damage detection methodology for bridges based on vibration measurements. They implemented a two-step methodology consisting of a Bayesian approach, which is based on Markov-Chain Monte Carlo method, for overcoming the problem related with operational and environmental effects on data measurements and a Mahalanobis square-distance algorithm is used in conjunction to accomplish damage detection. This problem, related with isolating operational and environmental effects in data with those related with damage is a common paradigm in pattern recognition techniques. Sierra-Pérez et al. (2015) used unsupervised learning algorithms for uncoupling these effects in conjunction with principal component analysis (PCA) to perform damage detection based on strain field estimation by using FBGs. An electrically propelled UAV with a wingspan of 3.5 m and a take-off weight of 9 kg was designed as a testing bench to gather structural strain information under different realistic operational conditions. The main beams of the airplane’s wing have a rectangular box cross-section and are composed of CFRP skins and a balsa wood core. The right-wing beam was instrumented with FBG sensors, attached with cyanoacrylate to the balsa core and the skins were laminated over, leaving the sensors embedded. A total of 20 FBGs of different wavelength and reflectivity were embedded. The localization and distribution are described as follows: five FBGs in the beam’s top face with an average separation of 160 mm leaving an 180-mm gap from root to the first sensor for obtaining compression strains, five FBGs in the bottom for obtaining tension strains arranged in an analogous way that compression strain line, five sensors located at - 45° in beam’s left face for obtaining torsion strains with the same average separation and leaving 270 mm from root and finally, four sensors located at 45° in beam’s right face for obtaining torsion strains with the same average separation and leaving 270 mm from root. Additionally, one FBG intended to temperature measurement was added in order to perform a thermal compensation. Since wavelength localization is a key issue for avoiding sensor overlapping, a genetic algorithm was developed for optimizing the wavelength distance and obtaining the most suitable combination of sensors. The algorithm has a cost function the wavelength average distance difference which depends on the sensor set configuration, this parameter was optimized giving a value of 2.151 nm. The beam’s skins were manufactured with aerospace-grade carbon fiber and epoxy resin, performing hand lay-up as follows: one unidirectional layer was placed on each of the sides where tension and compression are the main loads, with the fibers parallel to the beam’s longitudinal axis ; and one bidirectional layer of CFRP located on each of the other two sides, which are mainly subjected to torsion loads, positioning the fibers at ± 45° with respect to the beam’s longitudinal axis. The optical data acquisition system mounted on the aircraft ’s payload compartment comprises a 1550.1-nm centered light source device using a super-luminescent diode, a 6-kHz, one-channel miniaturized interrogator with a 1525-to-1570-nm wavelength range, an optical fiber circulator linking the FBG sensors with the light source and the interrogator, and a 1x4 splitter. A light spectrum is emitted by the light source and passes through the circulator and the splitter to finally reach the FBGs embedded on the wing’s main b eam, where part of the spectrum is reflected correspondingly to the actual strain at each sensor. These optical features are collected and transformed into electronic signals by the interrogator, which sends it to a 1.44 GHz, Windows-based onboard minicomputer via USB. The wavelength data is received by a customized LabVIEW® program based on the one provided by the interrogator manufacturer. This software receives interrogator’s temperature and light spectrum, analyzes it and delivers the location and amplitude of each sensor reflected wavelengths on every measurement. This information is then temperature-corrected and concatenated in a string with its correspondent timestamp. Then, this string is streamed 3. Experimental Setup