Original Research into the Application of Machine Learning for Predictive Maintenance of Gas Compressor Trains in Nigeria

Unplanned compressor failures are a primary driver of gas supply interruptions and production deferment in Nigerian gas infrastructure, yet the potential of machine learning (ML) approaches for predictive maintenance of compressor trains in the Nigerian operating context has not been systematically investigated. This study conducts original research into the development and validation of machine learning models for predictive maintenance of gas compressor trains, drawing on operational sensor data from compressor stations on the Trans-Forcados System and NGC compression stations in the Southwest zone. The research employs a systematic ML model development methodology, constructing a labelled training dataset from four years of vibration, temperature, pressure, and flow sensor data time-stamped against maintenance records and failure event logs for six compressor trains. A comparative evaluation framework assesses the predictive performance of five ML algorithms: random forest, gradient boosted trees, long short-term memory (LSTM) recurrent neural network, convolutional neural network (CNN), and an ensemble hybrid model combining CNN-LSTM architectures. Feature engineering incorporates domain-specific gas compressor operating parameters derived from thermodynamic first principles. The study's original contributions include: a Nigeria-specific compressor failure taxonomy derived from 8 years of historical maintenance records, a novel multi-scale temporal feature extraction approach for rotating equipment vibration signals, and a compressor health index formulation integrating multi-sensor degradation indicators. The CNN-LSTM ensemble model achieves a fault detection lead time of 7 to 14 days ahead of failure with 87 percent precision and 82 percent recall on the test dataset. Recommendations include a phased industry deployment roadmap for ML-based compressor monitoring and establishment of a centralised compressor failure data repository across Nigerian gas infrastructure operators. Keywords: machine learning, predictive maintenance, gas compressor, Nigeria, LSTM.