Cluster-Aware Deep Sensing: A Self-optimising WSN Framework for Intelligent Pipeline Leak Detection

Abstract

Pipeline leak detection remains a critical challenge in oil and gas transportation systems because undetected leaks may create environmental, economic, and safety risks. Conventional monitoring approaches often have limited scalability, delayed response, high communication overhead, and reduced robustness in noisy industrial environments. This study proposes the Cluster-Aware Deep Sensing (CADS) framework, a self-optimising Wireless Sensor Network (WSN) architecture for intelligent and energy-efficient pipeline leak detection. The framework integrates Adaptive K-Medoid clustering, MapReduce-based distributed preprocessing, cluster-aware feature augmentation, and Convolutional Neural Network (CNN)-based anomaly detection within a unified architecture. Experimental evaluation was conducted using a multivariate pipeline monitoring dataset comprising 120,000 sensor samples containing pressure, acoustic emission, vibration, and temperature signals collected from simulated industrial pipeline environments and publicly available leak-monitoring repositories. Adaptive clustering improves energy balancing and robustness against noisy sensor readings, while distributed preprocessing reduces communication redundancy and enhances scalability. Structural cluster metadata are incorporated into CNN input representations to strengthen anomaly classification capability. The results showed that CADS achieved 96.8% detection accuracy, a 95.9% F1-score, 28% energy savings, a reduced detection latency of 1.6 s, and improved robustness under low signal-to-noise ratio conditions. The framework also maintained stable performance across large-scale deployments involving up to 2000 sensor nodes. These findings indicate that adaptive clustering, distributed intelligence, and deep sensing can support intelligent pipeline monitoring systems.