Development of a Predictive Maintenance Framework Using Machine Learning for CNC Machining Operations in Nigerian Automotive Component Manufacturing Plants

Abstract

Computer Numerical Control (CNC) machining operations in sub-Saharan African automotive manufacturing facilities remain disproportionately reliant on reactive and time-based maintenance strategies, leading to unplanned downtime rates that exceed global benchmarks by a margin of 30–45%. In Nigeria's emerging automotive sector, anchored by Innoson Vehicle Manufacturing Company (IVM) in Nnewi, Anambra State, production losses attributable to CNC tool failure and unscheduled machine stoppages have been estimated at ₦2.3 billion annually, yet systematic predictive maintenance (PdM) frameworks calibrated for the operational realities of this environment remain absent from published literature. This study developed and validated a hybrid machine learning framework integrating Random Forest (RF) classification and Long Short-Term Memory (LSTM) neural network regression to predict cutting tool wear state and imminent machine failure events from multivariate IoT sensor data streams collected directly from CNC turning centres at IVM's Nnewi facility. A sensor array comprising vibration accelerometers, acoustic emission transducers, spindle current monitors, and thermocouple assemblies logged 14 feature variables at 1 kHz sampling frequency across 1,840 operational hours. The RF classifier achieved a tool-wear-state classification accuracy of 93.7% (F1-score: 0.924) on the held-out test set, outperforming a baseline Support Vector Machine (SVM) by 8.4 percentage points. The LSTM regression model predicted the remaining useful life (RUL) of cutting inserts with a Root Mean Square Error (RMSE) of 4.31 minutes and a Mean Absolute Percentage Error (MAPE) of 6.8%, enabling a proactive replacement window of 18–22 minutes before catastrophic failure. Simulated deployment of the framework projected a 34.2% reduction in unplanned downtime and an annual cost saving of approximately ₦786 million against baseline operations. These results demonstrate that context-specific, data-driven maintenance systems are technically feasible and economically compelling for African automotive manufacturing, and provide a replicable methodology for similar low-to-middle-income industrial environments.