Compressive sensing based structural damage detection and localization using theoretical and metaheuristic statistics

Abstract

Accurate structural damage identification calls for dense sensor networks, which are becoming more feasible as the price of electronic sensing systems reduces. To transmit and process data from all nodes of a dense network is a computationally expensive BIG DATA problem; therefore scalable algorithms are needed so that inferences about the current state of the structure can be made efficiently. In this paper, an iterative spatial compressive sensing scheme for damage existence identification and localization is proposed and investigated. At each iteration, damage existence is identified from randomly collected sparse samples and damage localization is iteratively detected via sensing-processing cycles with metaheuristic sampling distribution updating. Specifically, simulated annealing and ant colony analogy are used for guidance in future selection of sensing locations. This framework is subsequently validated by numerical and experimental implementations for gusset plate crack identification. SHM systems [25,26], the transmission cost and energy consumption of centralized data processing becomes inhibitive as the sensor network expands in size [3,[27][28][29]. To overcome this obstacle, decentralized information extraction techniques and communication procedures have been proposed and tried in several studies [30][31][32][33], demonstrating increased energy efficiency and better flow of operation.In the meantime, options to compress the data for wireless transmission are also explored. This can be done by applying the off-the-shelf compression schemes via adaptive coding, and a number of studies that exploit the temporal information redundancy of structural monitoring signals for data condensation are also under way. Examples include compressive sensing from random selection [34][35][36], sparse matrix representation[37-39], bio-mimicry based signal interpretation [40], and wavelet analysis [41,42]. Such methods reduce the amount of data transmitted per node, yet the central station still needs to communicate directly/indirectly with all sensor nodes within network. But because damage is intrinsically a local phenomenon, even higher performance could be realized if within a dense network only a subset of sensors in the vicinity of possible damage communicates their data with the main repository given proper damage detection features and effective compressive data communication procedures.In this paper an iterative compact sensing framework on subset identification within dense sensor network for damage detection is introduced. It complements the existing temporal compressive sensing methods, where time histories are compressed at individual sensor nodes. In this scheme, damage existence is first identified via hypothesis testing on the determinant of sample correlation matrix, as a damage feature, among several random locations. Then damage is localized as the location where the maximum Damage Location Indicator (DLI) from iterative random sampling occurs. At each iteration damage indices are computed from the n...

Remaining Useful Life Estimation of Wind Turbine Blades under Variable Wind Speed Conditions Using Particle Filters

“…Condition monitoring and diagnosis is the fundamental step in the RUL estimation of wind turbine blades. With many advances in structural damage detection and diagnosis (Yao & Pakzad, 2012;Shahidi et al, 2015;Yao, Pakzad, & Venkitasubramaniam, 2017;Gulgec, Takáč, & Pakzad, 2017), SHM system should be capable enough to detect and localize the faults at a minimum required stage of evolution such that sufficient time is available to achieve economic CBM. Numerous SHM strategies (Kirikera, Schulz, & Sundaresan, 2007;Rumsey & Paquette, 2008;Furong Zhang, Yongqian Li, Zhi Yang, & Liping Zhang, 2009;Kim et al, 2014;LeBlanc, Niezrecki, Avitabile, Chen, & Sherwood, 2013;Yang, Peng, Wei, & Tian, 2017) for wind turbine blades have been proposed, studied and tested (Dutton et al, 2003;Kirikera et al, 2008;Ou, Chatzi, Dertimanis, & Spiridonakos, 2016).…”

An Unsupervised Learning Approach for Early Damage Detection by Time Series Analysis and Deep Neural Network to Deal with Output-Only (Big) Data

“…Big Data analytics was performed by Kim and Queiroz [19] for the condition evaluation of highway bridges by considering roughly one million data samples. Yao et al [20] presented an iterative spatial compressive sensing scheme for damage identification and localization by handling the Big Data problem.…”

Compressive-sensing data reconstruction for structural health monitoring: a machine-learning approach

“…21 CS has been widely used in many fields, including consumer camera imaging, 22,23 medical magnetic resonance imaging, 13 remote sensing, 24 seismic exploration, 25 and communications, especially for wireless sensor networks (WSN). 26,27 In SHM, the applications of CS theory have also been investigated in structural vibration data acquisition, 28 wireless data transmission and lost data recovery, [29][30][31][32][33][34] structural modal identification, 35,36 structural sparse damage identification, [37][38][39][40][41][42] and sparse heavy-vehicle-loads identification of longspan bridges. 43 As mentioned above, the nature of CS is to solve an ill-posed inverse problem.…”