Pengdi Zhang

High-fidelity finite element numerical simulations of guided wave propagation in structures like pipelines are computationally demanding, posing significant challenges for real-time structural health monitoring (SHM) and advanced signal processing frameworks. The prohibitive computational cost of full-order models makes them impractical for rapid assessment, extensive parametric studies, or integration into inverse analysis algorithms, where swift and continuous data interpretation is essential. The advent of high-dimensional spatiotemporal data, from systems like Distributed Acoustic Sensing (DAS), further necessitates efficient analysis methods. This paper presents a computational framework for constructing reduced-order models (ROMs) as a surrogate for complex wave propagation and scattering phenomena. The proposed framework can efficiently construct ROM surrogates by combining Tucker tensor low-rank approximation and spectral polynomials. The surrogate model consists of the tensor product of one-dimensional basis functions, where the coefficient tensor is approximated in the low-rank Tucker form. To this end, we sample nodal displacements or accelerations using finite element analysis with representative parameter sets at a set of collocation points. The data are decomposed into a set of mode matrices and one small core tensor using Tucker decomposition. Lagrange interpolation is then applied to the mode matrices to predict values within the data range. Finally, the full spatiotemporal time history responses with new parameter sets are generated by multiplying the expanded mode matrices and small core tensors. The performance of the proposed method is studied by constructing ROMs for guided wave pipeline simulations based on limited learning data. The proposed ROMs can accurately predict behavior even for new parameter sets using the Lagrange interpolation scheme. An extremely high data compression ratio (more than 200) and efficient predictions of the response surfaces (2000 times faster than that of full finite element analyses using all parameter sets) can be realized. Furthermore, the inherent dimensionality reduction offered by Tucker decomposition provides compact, physically meaningful features. The parametric factor matrix, extracted from this decomposition, directly encodes how wave propagation patterns vary with defect parameters. These tensor-derived features are then effectively utilized as inputs for automated defect characterization, successfully estimating critical defect parameters—including distance, length, and width—with high coefficients of determination.

Reliable defect measurement in industrial pipelines remains difficult under small-sample conditions, where experimental guided-wave data are limited and affected by environmental and instrumentation noise. Although physics-based simulation can provide scalable training data, discrepancies between simulated and measured signals introduce measurement uncertainty and limit the practical reliability of data-driven models. To address this problem, this study presents a small-sample fiber guided-wave defect measurement framework that combines reduced-order modeling, signal adaptation, and domain-aligned classification.In the proposed framework, a reduced-order model derived from high-fidelity wave-propagation simulations is used to efficiently generate physically meaningful guided-wave responses over a broad defect parameter space. To reduce the discrepancy between simulated and experimental data, a neural-style-transfer-based signal adaptation step is introduced to reproduce experimental background characteristics in the time-frequency representation while maintaining the main defect-relevant wave structures. The adapted synthetic signals are then used together with limited experimental measurements to support defect identification in the target domain.The framework is evaluated using experimental guided-wave data collected from two pipeline configurations with repeated measurements at 11 sensor positions. To reduce overlap between training and testing conditions, the target-domain data are partitioned using a position-wise holdout strategy, in which measurements from held-out sensor locations are excluded from training. Results show that the proposed framework achieves the best performance among the compared methods in this held-out setting and improves the practical consistency between simulation-based training data and experimental measurements. These findings suggest that the proposed approach provides a useful route for reducing measurement uncertainty in small-sample fiber guided-wave defect measurement.

This research focuses on enabling real-time monitoring of critical infrastructure including pipelines in oil and gas transportation and the various threats they face, including human error, environmental factors, and material corrosion. The study proposes a novel approach to Structural Health Monitoring (SHM) using ultrasonic in-line pipe inspection with affixed transducers. More specifically, it explores the use of optical fiber sensors in various configurations for continuous monitoring. A significant innovation is the combination of guided wave technology with fiber optic sensors for pipeline and other structural health monitoring applications. The research employs machine learning and artificial intelligence (AI) for corrosion classification through guided waves, using Finite Element Analysis (FEA) simulations to generate labeled datasets for AI training. However, the scarcity of labeled training data in industrial contexts is a challenge. The study also investigates domain adaptation as a tool for knowledge generalization, applying insights from a well-labeled source domain to a less-explored target domain. A novel framework is demonstrated that leverages domain adaptation AI models by integrating simulation data with real-world measurements, thereby enhancing system state estimation and laying the groundwork for future digital twin systems. The thesis presents an innovative approach to infrastructure monitoring by integrating distributed fiber optic sensing, acoustic non-destructive evaluation, and artificial intelligence. This includes corrosion monitoring, reduced-order modeling, integration of AI frameworks with physical-based simulation models, and strategic deployment of optimized acoustic sensing based upon optical fiber sensors.

Pipelines are critical components of modern infrastructure, transporting resources such as oil, gas, and water securely and efficiently. Yet they face persistent threats—such as corrosion and cracking—that can lead to costly failures. To address these risks, guided wave techniques combined with fiber optic sensing have emerged as powerful tools. By enabling continuous, distributed measurements across long distances, these methods facilitate early detection of subtle defects, reducing maintenance costs and enhancing overall pipeline safety and reliability. However, interpreting the large and complex datasets generated by these systems can be challenging. Manual analysis is time-consuming and may miss subtle trends, prompting increased use of machine learning (ML) and artificial intelligence. While previous work has explored some of these aspects, our study is among the first to integrate fiber optic sensing, guided wave detection, and ML-driven domain adaptation into a unified framework for pipeline health monitoring. A key difficulty lies in bridging the gap between simulated and real-world data. Although simulations are easier to generate, they often lack the complexity and noise characteristics of actual measurements. Domain adaptation methods show promise, yet pipelines introduce unique obstacles, including varying wave propagation conditions, sensor placement differences, and limited experimental datasets. To tackle these challenges, we focus on modeling noise patterns that occur along the pipeline. Three primary noise factors—time-axial misalignment, coherent signal interference, and background fluctuations—shape the measured signals. We develop a position-specific noise model that learns from a single healthy pipeline, capturing how flexural waves evolve from sensor to sensor. By applying these learned noise characteristics to simulations, we create synthetic data that closely resemble real measurements. This approach not only helps overcome the scarcity of experimental datasets but also improves the robustness of ML models, enabling them to recognize and classify defects more accurately in varied pipeline scenarios. In doing so, we move beyond piecemeal solutions and present a comprehensive strategy that brings together fiber optics, guided waves, and ML-based domain adaptation. Our work sets the stage for more accurate, reliable, and data-efficient pipeline integrity assessments, ultimately improving infrastructure safety and reducing the need for extensive experimental data collection.

Natural gas pipeline integrity monitoring is crucial to detect potential leaks, find structural issues, and prevent environmental damage. This paper presents a system of natural gas pipeline monitoring that uses a specialized double Brillouin peak sensing fiber along with the Brillouin optical time domain analysis (BOTDA) technique. The calibrated sensing fiber coefficients for strain and temperature are 41.8 kHz/με and 0.9 MHz/°C for peak 1; and 47.2 kHz/με, and 1.11 MHz/°C for peak 2, respectively. Initially, lab tests were performed by installing a short section of double Brillouin peak fiber on a 1-inch steel pipe under pressure up to 1,000 psi at elevated temperatures. Simultaneous distributed measurements of temperature and pressure-induced hoop strain were successfully measured. Considering the long processing speed to extract Brillouin frequency shift (BFS), we employ a novel probabilistic deep neural network framework for rapid BFS prediction. Additionally, using the Finite Element Method, the effects of the pipeline pressure on hoop strain were modeled and compared to the experimental hoop strain under the same set of pipeline conditions. Finally, an actual 4-inch outer diameter steel natural gas pipeline was used for pilot-scale tests, where hoop strain was measured at various pressure levels. Leaks were simulated to demonstrate accurate pipeline integrity monitoring. At an internal pipe pressure of 1000 psi, hoop strain of approximately 300 με was observed, and the sensitivity was calculated as 0.28με/psi. The results of this pilot-scale study demonstrated that the system is capable of performing distributed monitoring sufficient to detect pipeline pressure and the presence of leaks to ensure the safe operation of gas pipelines in the field.

This study presents a framework for detecting mechanical damage in pipelines, focusing on generating simulated data and sampling to emulate distributed acoustic sensing (DAS) system responses. The workflow transforms simulated ultrasonic guided wave (UGW) responses into DAS or quasi-DAS system responses to create a physically robust dataset for pipeline event classification, including welds, clips, and corrosion defects. This investigation examines the effects of sensing systems and noise on classification performance, emphasizing the importance of selecting the appropriate sensing system for a specific application. The framework shows the robustness of different sensor number deployments to experimentally relevant noise levels, demonstrating its applicability in real-world scenarios where noise is present. Overall, this study contributes to the development of a more reliable and effective method for detecting mechanical damage to pipelines by emphasizing the generation and utilization of simulated DAS system responses for pipeline classification efforts. The results on the effects of sensing systems and noise on classification performance further enhance the robustness and reliability of the framework.

This study proposes a machine-learning-based framework for detecting mechanical damage in pipelines, utilizing physics-informed datasets collected from simulations for mechanical damage. The framework provides an effective workflow from dataset generation to damage detection and identification for three types of pipeline events: welds, clamps, and corrosion defects. While the study initially focused on optimizing the CNN structure using various advanced optimizers, it also investigated the impact of sensing systems on data classification and the effect of noise on classification performance. The study's analysis highlights the importance of selecting the appropriate sensing system for the specific application. The authors also found that the proposed framework is robust to experimentally relevant levels of noise, suggesting its applicability in real-world scenarios where noise is present. Overall, this study contributes to the development of a more reliable and effective method for detecting mechanical damage in pipelines. The proposed framework provides an effective workflow for damage detection and identification, and the findings on the impact of sensing systems and noise on classification performance add to its robustness and reliability.