The Design of Piezoelectric Sensor Networks for Integration in Structural Health Monitoring Systems

Abstract

Structural Health Monitoring (SHM) systems play a vital role in maintaining the safety and reliability of engineering structures by enabling the early detection and localization of damage. Central to the effectiveness of these systems is the strategic placement and design of sensor networks, which must not only provide comprehensive coverage, but also ensure robustness in the presence of sensor faults. The objective of this dissertation is to advance the design of piezoelectric sensor networks for integration into SHM systems by addressing key challenges related to sensor placement, network resilience, and the incorporation of physical wave propagation characteristics. Sensor placement is a fundamental aspect of SHM as it directly influences the system's ability to detect and localize damage. Inadequate placement can lead to insufficient coverage, poor sensitivity to damage, or redundancy that does not contribute to detection performance. Effective sensor deployment ensures that critical regions of the structure are monitored and that damage-induced wave interactions can be reliably captured. To that end, a sensor network—defined as a set of sensors working collaboratively through configurations such as pulse-echo and pitch-catch—is designed to achieve optimal spatial coverage and high-fidelity monitoring. This dissertation addresses three core research objectives. The first is to develop an optimal sensor placement methodology tailored to structures with closed and curved geometries. The second objective is to design multiple, robust sensor networks that are resilient to sensor failures while maintaining performance under such conditions. The third objective is to incorporate the physics of guided wave (GW) propagation into the placement model, ensuring that sensor deployment aligns with the fundamental behaviors of wave-material and wave-damage interactions. The first major contribution of this work focuses on the challenge of sensor placement on closed sections, such as fuselage structures or pipelines, which present geometric and topological complexities. A transformation technique is introduced to map any closed or arbitrarily shaped surface into an equivalent flat plate domain. This transformation is accompanied by a carefully defined set of boundary conditions to replicate the original wave propagation characteristics in the transformed space. By doing so, the problem becomes tractable using planar optimization techniques while still preserving the physical accuracy required for effective SHM. This approach enables sensor placement on geometrically complex structures without compromising the fidelity of wave-based damage detection. The second contribution advances the concept of robustness in SHM systems by introducing a novel model for the design of multiple, independent yet interacting sensor networks. These networks are configured to work synergistically, increasing the system’s resilience and maintaining performance under fault scenarios. Each sensor network is capable of operating autonomously, yet they are designed to collectively maximize detection reliability through shared coverage. Both pulse-echo and pitch-catch modes are incorporated to enhance spatial reach and signal diversity. The optimization framework developed in this work seeks to maximize a weighted objective function consisting of two interdependent components: coverage, defined as the percentage of control points monitored by at least three sensing paths in a network, and robustness, quantified as the average number of networks covering each monitored point. This dual-objective formulation ensures that the resulting sensor configurations not only provide widespread monitoring but also offer redundancy to mitigate the effects of potential sensor failures. The third and final contribution integrates the underlying physics of wave propagation into the sensor placement process using a machine learning (ML) approach informed by simulation data. Numerical simulations are conducted using ABAQUS on an aluminum plate to capture wave propagation under both undamaged and damaged conditions. The simulations model various scenarios involving actuator-sensor pairs and localized damage events to generate a comprehensive dataset. This dataset is then used to train a five-layer Artificial Neural Network (ANN), which learns the complex relationships between damage, wave attenuation, and signal characteristics. The trained ANN is employed to predict whether damage at a given location would significantly affect the communication between a specific actuator-sensor pair. This prediction is binary and reflects whether the induced damage would be detectable based on wave interaction characteristics. By incorporating this physics-based information into the placement algorithm, sensors are only placed where they are physically capable of detecting damage, resulting in more meaningful and efficient sensor configurations. Each of the three contributions culminates in the proposal of a new sensor placement model that has been validated through both simulation and experimental testing. The models demonstrate strong performance in terms of detection capability, spatial coverage, and robustness, confirming their applicability in practical SHM settings. Experimental results align closely with simulation outcomes, further supporting the validity of the proposed methodologies. In summary, this dissertation presents a comprehensive framework for the design of piezoelectric sensor networks in SHM systems, emphasizing optimal placement on complex structures, robustness through redundant yet cooperative network architectures, and the integration of wave physics through machine learning. These contributions collectively address key limitations in current SHM practices and offer scalable, resilient, and physically informed solutions for the next generation of intelligent structural monitoring systems.