Hybrid Models for Predictive Maintenance in the Oil and Gas Sector

Abstract

Nigeria’s oil and gas pipeline network is central to the country’s energy economy; however,
recurring failures due to corrosion, mechanical fatigue, and adverse environmental conditions
impose substantial financial and environmental costs on operators. This study proposes a
predictive maintenance framework for pipeline systems in Nigeria’s oil and gas sector,
employing a hybrid model that integrates a knowledge graph with a neural network to forecast
pipeline corrosion rates. The framework combines Graph Neural Networks (GNNs) with
Generalised Additive Models (GAMs) to predict pipeline lifespan and Remaining Useful Life
(RUL), enabling advanced risk modelling alongside interpretable outputs for decision-makers.
The workflow begins with multi-source data inputs—real-time sensor readings, historical
maintenance logs, and spatial network information—processed through specialised neural
network modules and integrated via a GAM layer that produces interpretable risk forecasts,
estimated lifespan values, and associated uncertainty intervals. The framework leverages a
large, representative dataset from multiple Nigerian operators and introduces explicit
uncertainty quantification via Bayesian inference. Validation against real-world failure records
from three regional operators, using a Bayesian statistical framework under the Weibull
distribution, estimated an expected failure rate of 0.008749 failures per hour (SE = 3.74×10−5)
with a 95% Bayesian prediction interval of [0.00866, 0.00893]. The hybrid model achieved an
F1-score of 0.882 on an independent test set, outperforming all standalone architectures and
traditional baselines. These results demonstrate the practical benefits of integrating advanced
machine learning with interpretable statistical methods for pipeline integrity management.
Keywords: Predictive Maintenance; Hybrid Models; Pipeline Lifespan Prediction; Oil and
Gas Sector; Neural Networks; Generalised Additive Models; Remaining Useful Life (RUL);
Operational Efficiency; Cost Reduction; Nigeria.