Compressive techniques for sub-Nyquist data acquisition & processing in vibration-based structural health monitoring of engineering structures

Gkoktsi, K. (2018). Compressive techniques for sub-Nyquist data acquisition & processing in vibration-based structural health monitoring of engineering structures. (Unpublished Doctoral thesis, City, University of London)

Text - Accepted Version
Download (11MB) | Preview


Vibration-based structural health monitoring (VSHM) is an automated method for assessing the integrity and performance of dynamically excited structures through processing of structural vibration response signals acquired by arrays of sensors. From a technological viewpoint, wireless sensor networks (WSNs) offer less obtrusive, more economical, and rapid VSHM deployments in civil structures compared to their tethered counterparts, especially in monitoring large-scale and geometrically complex structures. However, WSNs are constrained by certain practical issues related to local power supply at sensors and restrictions to the amount of wirelessly transmitted data due to increased power consumptions and bandwidth limitations in wireless communications.

The primary objective of this thesis is to resolve the above issues by considering sub-Nyquist data acquisition and processing techniques that involve simultaneous signal acquisition and compression before transmission. This drastically reduces the sampling and transmission requirements leading to reduced power consumptions up to 85-90% compared to conventional approaches at Nyquist rate. Within this context, the current state-of-the-art VSHM approaches exploits the theory of compressive sensing (CS) to acquire structural responses at non-uniform random sub-Nyquist sampling schemes. By exploiting the sparse structure of the analysed signals in a known vector basis (i.e., non-zero signal coefficients), the original time-domain signals are reconstructed at the uniform Nyquist grid by solving an underdetermined optimisation problem subject to signal sparsity constraints. However, the CS sparse recovery is a computationally intensive problem that strongly depends on and is limited by the sparsity attributes of the measured signals on a pre-defined expansion basis. This sparsity information, though, is unknown in real-time VSHM deployments while it is adversely affected by noisy environments encountered in practice.

To efficiently address the above limitations encountered in CS-based VSHM methods, this research study proposes three alternative approaches for energy-efficient VSHM using compressed structural response signals under ambient vibrations. The first approach aims to enhance the sparsity information of vibrating structural responses by considering their representation on the wavelet transform domain using various oscillatory functions with different frequency domain attributes. In this respect, a novel data-driven damage detection algorithm is developed herein, emerged as a fusion of the CS framework with the Relative Wavelet Entropy (RWE) damage index. By processing sparse signal coefficients on the harmonic wavelet transform for two comparative structural states (i.e., damage versus healthy state), CS-based RWE damage indices are retrieved from a significantly reduced number of wavelet coefficients without reconstructing structural responses in time-domain.

The second approach involves a novel signal-agnostic sub-Nyquist spectral estimation method free from sparsity constraints, which is proposed herein as a viable alternative for power-efficient WSNs in VSHM applications. The developed method relies on Power Spectrum Blind Sampling (PSBS) techniques together with a deterministic multi-coset sampling pattern, capable to acquire stationary structural responses at sub-Nyquist rates without imposing sparsity conditions. Based on a network of wireless sensors operating on the same sampling pattern, auto/cross power-spectral density estimates are computed directly from compressed data by solving an overdetermined optimisation problem; thus, by-passing the computationally intensive signal reconstruction operations in time-domain. This innovative approach can be fused with standard operational modal analysis algorithms to estimate the inherent resonant frequencies and modal deflected shapes of structures under low-amplitude ambient vibrations with the minimum power, computational and memory requirements at the sensor, while outperforming pertinent CS-based approaches. Based on the extracted modal in formation, numerous data-driven damage detection strategies can be further employed to evaluate the condition of the monitored structures.

The third approach of this thesis proposes a noise-immune damage detection method capable to capture small shifts in structural natural frequencies before and after a seismic event of low intensity using compressed acceleration data contaminated with broadband noise. This novel approach relies on a recently established sub-Nyquist pseudo-spectral estimation method which combines the deterministic co-prime sub-Nyquist sampling technique with the multiple signal classification (MUSIC) pseudo-spectrum estimator. This is also a signal-agnostic and signal reconstruction-free method that treats structural response signals as wide-sense stationary stochastic processes to retrieve, with very high resolution, auto-power spectral densities and structural natural frequency estimates directly from compressed data while filtering out additive broadband noise.

Item Type: Thesis (Doctoral)
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
Divisions: City University London PhD theses
School of Engineering & Mathematical Sciences > Engineering

Actions (login required)

View Item View Item


Downloads per month over past year

View more statistics