Designing an AI-Powered Predictive Maintenance System for Industrial Equipment in Manufacturing Plants in Nnewi, Anambra State

Industrial equipment downtime at manufacturing plants in Nnewi, South East Nigeria's auto-parts manufacturing hub, results in significant production losses that erode competitiveness. This study designed an AI-powered predictive maintenance system for rotating machinery (electric motors and pumps) at a representative Nnewi manufacturing plant. A data acquisition architecture was designed using industrial IoT vibration accelerometers (ADXL355) and temperature sensors (PT100) mounted on 12 critical machine units, transmitting via Modbus RS-485 to an edge computing gateway. Feature extraction from time-domain and frequency-domain vibration signals (RMS, kurtosis, FFT peak amplitude) was implemented on the gateway using Python. A machine learning model combining Random Forest and an LSTM network was trained on 14 months of historical sensor data labelled with 47 documented failure events across bearing wear, imbalance, and misalignment fault categories. Model evaluation on a held-out test set achieved 91.4 percent fault detection accuracy with a false alarm rate of 6.2 percent. Remaining useful life predictions for bearings showed a mean absolute error of 3.8 days against actual failure dates. A Grafana monitoring dashboard visualised real-time health scores and maintenance scheduling recommendations. The study estimates maintenance cost savings of approximately 31 percent compared to time-based preventive maintenance schedules. Recommendations include expanding sensor coverage to all 47 production machines and integrating the system with the plant's SAP maintenance management module.

Keywords: predictive maintenance, industrial IoT, machine learning, manufacturing, Nnewi