Theses and Dissertations

ORCID

https://orcid.org/0009-0004-4416-2869

Advisor

Li, Gang

Committee Member

Whittington, Wil

Committee Member

Wang, Fei

Committee Member

Ibrahim, Hamdy

Date of Degree

5-15-2026

Original embargo terms

Visible MSU Only 1 year

Document Type

Dissertation - Campus Access Only

Major

Engineering (Mechanical Engineering)

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Michael W. Hall School of Mechanical Engineering

Abstract

Underground power cable networks are essential for maintaining reliable electric delivery in dense urban and industrial regions where overhead lines face space, safety, and right-of-way constraints. High-pressure fluid-filled (HPFF) pipe-type cable systems, in particular, offer very reliable high-capacity power transmission, but their buried installation and limited accessibility make fault detection and localization time-consuming and costly. Conventional approaches are often reactive and field-intensive, motivating retrofit-friendly monitoring strategies that can support both predictive maintenance and rapid fault localization. This dissertation develops and validates an acoustic monitoring and analytics framework for HPFF pipe-type cable infrastructure using non-intrusively mounted accelerometers installed at accessible locations (e.g., manholes), coupled with physics-guided signal processing and learning. First, the dissertation examines steel-borne acoustic pulse propagation experimentally in a representative HPFF steel-pipe configuration. Propagation characteristics are quantified under multiple embedment conditions to evaluate the effects of soil coupling on attenuation and to establish the feasibility of time difference of arrival (TDOA) based localization. The results demonstrate accurate fault pinpointing along the pipe using distributed accelerometer measurements. Second, building on these propagation insights, a physics-enhanced deep learning condition-monitoring approach is introduced for incipient fault detection using multi-sensor acoustic wave time series. The method combines TDOA-informed temporal windowing with a hybrid convolutional neural network–long short-term memory (CNN-LSTM) architecture to ensure the model receives temporally consistent input sequences that reflect wave-propagation dynamics in the HPFF steel pipe, improving interpretability and enabling reliable detection of subtle changes in sequential vibration patterns indicative of incipient faults. Finally, the dissertation introduces Echo-AE, a physics-informed sequence-to-sequence attention autoencoder that embeds wave-propagation constraints directly into the loss function and learns normal operating behavior for anomaly-based early-stage fault warning. This supports predictive maintenance in rare-event regimes by flagging deviations from baseline behavior while maintaining robustness under highly imbalanced data. Collectively, the proposed framework demonstrates the feasibility of using steel-borne acoustic sensing and physics-guided learning to enable proactive condition monitoring of HPFF cable systems, supporting earlier maintenance intervention and faster fault response than conventional reactive approaches.

Sponsorship (Optional)

The research presented in this dissertation was supported by the U.S. Department of Energy under Grant No. DE-SC0012070 and the U.S. National Science Foundation under Grant Nos. 2429540 and 2329791.

Download

Share

COinS