Smart Grid Fault Localization in Metro Rail Systems: A Case Study of Citylink Transit Authority

Thesis Abstract

This study addresses the challenge of rapid fault localization in smart grid-enabled metro rail systems, where intermittent power disturbances and signaling faults destabilize operations, degrade passenger experience, and increase maintenance costs. The aim is to develop and validate an integrated fault localization framework that leverages high-resolution telemetry, phasor measurement unit (PMU) data, and signaling system logs to pinpoint root causes within metropolitan rail networks. Specific objectives include (1) characterizing fault propagation pathways between traction power substations, third-rail/contact wire interfaces, and signaling subsystems; (2) designing a data-driven localization algorithm that fuses PMU-based voltage/current signatures with SCADA/logging events; (3) evaluating the diagnostic performance under diverse fault scenarios (temporary faults, permanent faults, and cyber-physical disturbances); (4) assessing the framework’s resilience to data gaps and communication delays; and (5) formulating operational guidelines for real-time fault containment and maintenance prioritization. Methodologically, the research adopts a mixed-methods design within Citylink Transit Authority, a large urban metro operator serving a daily ridership of approximately 1.2 million. The study population includes traction power networks, signaling controllers, and on-train systems integrated into the smart grid. A stratified sample of 12 substations, 8 interlocking locations, and 40 trains is selected to capture spatial and temporal diversity. Data collection combines quantitative streams from PMUs (voltage, current, frequency), substation fault records, SCADA event logs, and train-borne telemetry over a 24-month period, totaling an expected dataset of 2.4 terabytes. Complementary qualitative data are obtained through 20 semi-structured interviews with control center operators and maintenance engineers, and 6 focus groups with field technicians to contextualize anomalies. Instrumentation includes a high-resolution PMU deployment plan, standardized fault-report templates, and a structured interview guide, with pilot testing conducted to ensure reliability. The analytical framework integrates several techniques. Descriptive statistics summarize baseline electrical states and fault frequencies. Time-series analysis and dynamic regression quantify temporal dependencies between grid disturbances and signaling faults. A Bayesian network model is constructed to capture probabilistic causal relationships among substations, traction interfaces, and interlocking units, enabling real-time probabilistic fault localization. A machine learning component employs random forests and gradient boosting to classify fault types and prioritize likely root causes based on multi-modal features (phasor angles, magnitudes, SCADA events, and train-telemetry indicators). The model’s performance is evaluated with 10-fold cross-validation, reporting metrics including precision, recall, F1-score, and area under the ROC curve. To assess robustness, ablation tests simulate data outages and delayed communications. The theoretical underpinning draws on complex systems theory and distributed fault diagnosis, with the situational awareness framework guiding interpretation of operator-analytic alignment. Relevant theories include the Information Processing Theory of decision-making under uncertainty and the Causal Inference framework for predictive diagnostics. Expected findings indicate that the integrated framework achieves a diagnostic precision of 86–92% and a recall of 84–90% across major fault categories, outperforming baseline rule-based localization by approximately 25%. The Bayesian network is anticipated to reveal dominant causal pathways, such as traction substation disturbances propagating to interlocking logic via degraded signaling margins, while PMU-derived signatures effectively disambiguate electrical faults from cyber-physical disturbances. Sensitivity analyses are expected to identify critical data streams (PMU, interlocking logs) and quantify the impact of data latency on localization accuracy. The study contributes to knowledge by providing a replicable, data-driven fault localization approach tailored to smart grid-enabled rail systems, bridging electrical engineering, control engineering, and rail operations management. It also offers operational guidance for real-time fault containment, maintenance prioritization, and investment decisions in telemetry infrastructure. The main conclusion anticipated is that multi-modal data fusion, grounded in probabilistic causal modeling and machine learning, substantially enhances fault localization speed and accuracy in metro rail smart grids, enabling proactive mitigation and reduced mean time to repair. Recommendations include expanding PMU coverage, standardizing event logging, developing a real-time dashboard for control centers, and integrating the framework into preventive maintenance schedules, with further research proposed on scaling to multi-city networks and exploring anomaly detection under evolving signaling technologies.