This thesis develops a basic predictive maintenance strategy for rotating mechanical equipment using simple statistical analysis of vibration data. It uses the PU‑MAFAULDA imbalance dataset to extract interpretable time‑domain features, mainly RMS, standard deviation, and trend measures, and shows how these can indicate abnormal machine behavior and gradual degradation. A validation study on simulated data compares static thresholds, adaptive thresholds, and trend‑based indicators, demonstrating that combining thresholds with trend detection provides earlier and more reliable fault warnings than fixed limits alone. Building on these results, the work proposes a maintenance scheduling framework that classifies machine condition into three health states normal, warning, and critical and assigns clear maintenance actions and response times to each state. The method is designed to be transparent, low‑cost, and suitable for small and medium‑sized companies that lack large historical datasets or advanced machine‑learning infrastructure. Overall, the thesis provides a practical path for moving from purely preventive maintenance towards a simple but effective predictive maintenance approach.