Nigeria’s oil and gas pipeline network  spanning over 5,000 km of trunk lines and more than 3,000 km of flow lines loses an estimated one billion US dollars annually to pipeline failures, environmental incidents, and non-productive time. The dominant monitoring approach in operation today is threshold-based SCADA, which detects large anomalies well but struggles with gradual degradation, early-stage corrosion, and ambiguous multi-fault scenarios. This paper proposes a Hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model tailored for fault detection across six fault classes: normal operation, corrosion/wall thinning, leak/pinhole, blockage/wax deposition, mechanical fatigue/crack, and external impact. The architecture exploits CNNs for spatial and frequency domain feature extraction from multi sensor inputs (pressure, flow rate, acoustic emission, vibration, and temperature) and LSTMs for capturing temporal fault evolution patterns that threshold systems cannot resolve. An attention mechanism variant further prioritises the most discriminative time steps and sensor channels. Evaluated against four baseline models threshold SCADA, SVM, standalone CNN, and standalone LSTM, the proposed CNN-LSTM model achieves 97.6% overall classification accuracy (F1 = 96.9%), rising to 98.3% with the attention extension, representing a 26-percentage point improvement over threshold based baselines. The paper derives the key architectural equations, presents simulation results with charts and code, analyses the Nigerian pipeline operating context in depth, and proposes a deployment framework addressing the country’s specific connectivity, power, and data scarcity constraints.

Keywords: CNN-LSTM, fault detection, pipeline monitoring, Nigerian oil and gas, deep learning, acoustic emission, predictive maintenance, SCADA, IoT, Industry 4.0, transfer learning, edge computing.