Extreme learning machine (ELM) is a learning algorithm for single-hidden layer feedforward neural networks. In theory, this algorithm is able to provide good generalization capability at extremely fast learning speed. Comparative studies of benchmark function approximation problems revealed that ELM can learn thousands of times faster than conventional neural network (NN) and can produce good generalization performance in most cases. Unfortunately, the research on damage localization using ELM is limited in the literature. In this chapter, the ELM is extended to the domain of damage localization of plate structures. Its effectiveness in comparison with typical neural networks such as back-propagation neural network (BPNN) and least squares support vector machine (LSSVM) is illustrated through experimental studies. Comparative investigations in terms of learning time and localization accuracy are carried out in detail. It is shown that ELM paves a new way in the domain of plate structure health monitoring. Both advantages and disadvantages of using ELM are discussed.
TopIntroduction
Plate structures have been applied extensively in aerospace, civil, and mechanical engineering, etc. The reason lies in that pate structures provide some advantages including light weight, high strength, and design flexibility. In practice, maintenance is needed to be performed by workers to guarantee the structure integrity for the safety reason, even though the structures are designed to tolerate some damage. It is known that one origin of structural damage is the impact exerted on the structures. Therefore, it is necessary to monitor the location and magnitude of the impact force (Wang & Deng, 1999). Such a monitoring process will be beneficial to the subsequent maintenance work of the structures (Haywood, Coverley, & Staszewski, 2005).
In the past few years, different techniques have been proposed to facilitate the damage localization of plate structures (Tibaduiza, Mujica, & Rodellar, 2013; Tibaduiza, Torres-Arredondo, & Mujica, 2013). For instance, Park, Sohn, & Olson (2012) developed an impact localization technique using scanning laser Doppler vibrometer (SLDV) and surface-mounted PZT transducers. The technique is able to locate the impact events simply by comparing an actual impact response with IRFs obtained from a grid of training points. The training data are collected by individually exciting PZT transducers and scanning the corresponding responses over the target area using the SLDV. Shin, Yang, & Lee (2013) proposed a group delay based location template matching (G-LTM) method for impact source localization on a plate. The method utilized a concept of the measure of similarity which is introduced by examining the phase structure of the fictitious frequency response function. Aiming at an on-line application, Yuan, Liu, & Qiu (2012) proposed a digital impact localization method targeting at finding a localized search area for further inspections for online monitoring of structural health, which is realized using a PZT sensor array. Qiu, Yuan, & Liu (2013) reported a digital sequence based impact area localization method combined with a de-noising method of the digital sequences dedicated to the task of on-line impact monitoring of large-scale composite structures in their whole service lifetime. By fusing the merits of triangulation method and the inverse analysis method, Liang, Yuan, & Liu (2013) presented a distributed coordination algorithm for improving the preciseness of real-time impact location of a composite structure. Ciampa & Meo (2010) reported a new in situ structural health monitoring method for locating an acoustic emission source in complex composite structures based on the differences of the stress waves gathered by six surface-attached acoustic emission piezoelectric sensors. Dehghan Niri, Farhidzadeh, & Salamone (2012) proposed an adaptive multisensor fusion algorithm for acoustic emission source location in isotropic plate-like structures in noisy environment. Based on the differences of the stress waves measured by six surface-attached acoustic emission piezoelectric sensors, Ciampa, Meo, & Barbieri (2012) proposed an in situ structural health monitoring method able to locate the impact source and to determine the flexural Lamb mode A0 velocity in composite structures with unknown lay-up and cross section. Jeong & Cho (2013) presented a structural health monitoring technique for locating impact position in a composite plate, which employed a single sensor and spatial focusing properties of time reversal and inverse filtering. Baravelli, Senesi, & Ruzzene (2013) fabricated a wavenumber frequency-steerable acoustic transducer and applied it to record the plate response. The source location was then determined through a time-frequency analysis of the received signal by exploiting the frequency selective response of the transducer which directly maps the dominant component of the received signal to the direction of arrival of the incoming wave. Perelli, De Marchi, & Marzani (2014) presented a novel impact localization algorithm based on the frequency warping unitary operator applied to E-spline wavelet multiresolution analysis to identify the location of acoustic emission (AE) sources due to impacts. The method applies a dispersion compensation procedure on the signals acquired by passive sensors, thus overcoming the difficulties associated with arrival time detection based on classical thresholding procedures.