Optimizing Drone-Based Power Line Inspection for Rural Utilities)

Thesis Abstract

The rapid modernization of rural electricity networks has increased reliance on aerial drone technologies to inspect power lines, yet persistent challenges in data fidelity, inspection efficiency, and maintenance prioritization impede timely fault detection and asset management. This study addresses the gap by optimizing drone-based inspection workflows to enhance reliability, safety, and cost-effectiveness for rural utility operations. The aim is to develop an integrated framework that (1) improves image-based fault detection accuracy, (2) reduces inspection time per kilometer, and (3) supports evidence-based maintenance decision-making under diverse terrain and weather conditions. Specific objectives are to evaluate flight planning strategies that maximize coverage and minimize battery consumption; to compare computer vision algorithms for defect classification using high-resolution imagery and thermal data; to quantify the impact of flight parameters on data quality and detection rates; to develop a priority-based maintenance model informed by probabilistic risk assessment; and to validate the framework in a pilot study with six rural utility corridors totaling 312 km of line length, including 1200 annotated defect instances. The methodology adopts a mixed-methods research design combining quantitative performance evaluation with qualitative expert input. The population comprises rural utility operators, drone pilots, and asset managers engaged in overhead line inspection. A purposive sample of 18 drone operation teams across three rural utilities will be recruited, with each team conducting standardized inspections along two representative corridors. Data collection instruments include (i) high-resolution visual and thermal imagery captured by multirotor platforms equipped with 4K cameras and infrared sensors, (ii) flight logs and environmental metadata, (iii) a validated defect annotation schema applied to a subset of 400 4K images, and (iv) semi-structured interviews with eleven technicians and asset managers. Analytical techniques consist of (a) regression analysis and ANOVA to assess the relationships between flight parameters (altitude, speed, overlap) and defect detection performance; (b) convolutional neural networks (CNN) and transfer learning for defect classification, benchmarked against traditional feature-based methods; (c) a probabilistic risk model to derive maintenance prioritization scores; and (d) thematic analysis of interview data to identify organizational and operational constraints. Validity and reliability are ensured through cross-validation of CNN models (k-fold with k=5), inter-rater agreement on defect labeling (Cohen’s kappa > 0.8), and pilot-retest reliability of the maintenance model. The analytical framework integrates a data fusion approach to combine optical, thermal, and contextual metadata, with model specification guided by structural equation modelling to test the influence of data quality on maintenance decision effectiveness. Expected findings include (i) quantitative evidence that optimized flight plans with systematic overlap reduce missed defect rates by 22% and cut inspection time by 28% per kilometer; (ii) CNN-based defect classifiers achieving precision and recall above 0.85 for common line faults (insulators, corona discharge, corrosion) when fused with thermal cues; (iii) a robust maintenance prioritization model that improves decision speed by 35% and aligns with risk-based resource allocation; and (iv) qualitative insights into organizational barriers such as data governance, standardization of defect labeling, and operator training needs. The study contributes to knowledge by presenting an integrated drone inspection framework that combines advanced analytics, risk-informed maintenance planning, and practical workflow design tailored to rural utility contexts. The expected theoretical contribution includes extending the diffusion of innovations theory within asset management adoption and applying a data-centric decision support paradigm to utility-scale field inspections. The main conclusion anticipated is that end-to-end optimization of data collection, defect detection, and maintenance prioritization substantially enhances reliability and cost-efficiency of rural power networks without compromising safety. Practical recommendations entail standardized acquisition protocols, a centralized defect taxonomy with open datasets for benchmarking, continuous model retraining with new fault types, and policy guidance for drone operations in rural terrains. Further research is recommended to explore real-time on-board processing, battery-aware autonomy under adverse weather, and long-term cost-benefit analyses across diverse utility configurations.