Development of an Intelligent Predictive Maintenance System for Critical Equipment in Nigeria’s Power Generation Sector

This dissertation develops an intelligent predictive maintenance system for critical rotating equipment in Nigeria's thermal power generation sector, making original contributions to both the methodology of intelligent maintenance systems design and the application of artificial intelligence to infrastructure reliability improvement in developing country energy systems. Nigeria's power generation infrastructure chronically underperforms its installed capacity, with equipment failures contributing significantly to the generation-consumption gap that constrains economic activity across all sectors and geopolitical zones. The research develops a multi-modal condition monitoring and fault prognosis architecture incorporating vibration signal processing, thermographic imaging analysis, oil quality monitoring, and operational parameter trending. A novel ensemble machine learning model combining convolutional neural networks for vibration feature extraction, gradient boosting for multi-parameter fault classification, and a long short-term memory network for remaining useful life prediction is designed, trained on a hybrid dataset comprising publicly available bearing fault datasets and original sensor data collected from four power plant facilities in Ondo, Kogi, Bauchi, and Kano states. The system achieves fault detection sensitivity of ninety-three-point-four percent and remaining useful life prediction error of eight-point-seven percent on the industrial test dataset, substantially outperforming single-model benchmarks. A human-machine interface design and implementation roadmap for utility adoption is provided. Keywords: predictive maintenance, intelligent maintenance systems, machine learning, power generation, Nigeria