Empirical Evaluation of Edge Computing Architectures for Real-Time Oil Pipeline Leak Detection in the Niger Delta

Oil pipeline leaks in the Niger Delta cause severe environmental and economic damage, and real-time automated leak detection through distributed edge computing and IoT sensors offers a proactive alternative to current reactive monitoring approaches, yet no empirical evaluation of edge architectures for this specific application exists. This study empirically evaluated edge computing architectures for real-time pipeline leak detection using acoustic emission and flow rate anomaly data from a simulated Niger Delta pipeline segment. A testbed was constructed with three edge node tiers: IoT sensor nodes (ESP32 with MEMS microphones and flow meters), edge inference nodes (NVIDIA Jetson Nano), and a gateway cloud tier (AWS IoT Greengrass). A time-series anomaly detection model (Autoencoder LSTM) was trained on 30 days of normal pipeline acoustic and flow data and deployed across three architectural configurations: cloud-only inference, gateway-only inference, and distributed edge inference. Leak detection latency was 480 milliseconds for cloud-only versus 85 milliseconds for distributed edge inference, a 5.6x improvement. Detection accuracy on a 200-event test set showed no statistically significant difference across architectures (F1 scores of 0.91 to 0.93), confirming that edge offloading did not degrade accuracy. Communication bandwidth consumption was reduced by 73 percent in the distributed edge architecture by local pre-processing. Reliability under simulated intermittent connectivity showed 100 percent event capture in edge configurations versus 34 percent event loss in cloud-only under the same connectivity interruption conditions. The study recommends the Nigerian National Petroleum Corporation adopt distributed edge inference for pipeline monitoring systems in areas with unreliable connectivity.