Predictive Maintenance Data Checklist for Manufacturers

The minimum viable dataset for predictive maintenance is 90 days of continuous sensor data at 1Hz or higher sampling rate, plus at least 12 months of historical maintenance records with failure event labels, timestamps, and root causes. For assets without existing sensor coverage, iFactory can deploy wireless vibration and temperature sensors that begin generating usable data immediately. The most important factor is data quality — 90 days of high-quality, well-labelled data consistently outperforms 12 months of noisy, unlabelled data in iFactory's deployment benchmarks.

For standard rotating equipment operating below 3,600 RPM (60 Hz), a minimum sampling rate of 2 kHz is required to capture bearing fault frequencies up to the 10th harmonic. For high-speed equipment above 10,000 RPM, sampling rates of 10-20 kHz are recommended. iFactory's platform automatically adjusts sampling rate recommendations based on the asset type, operating speed range, and expected failure modes. When in doubt, sample at the highest rate your data infrastructure supports — the platform will downsample automatically, but data captured at too low a rate cannot be recovered.

Failure event labelling requires four pieces of information per event: the asset or component that failed, the timestamp of failure onset (when the asset began operating abnormally), the timestamp of failure detection (when the fault was identified), and the root cause or failure mode. Each event should also include the asset's operating state at the time of failure — production mode, speed, load, and product being produced. iFactory's platform can automatically label a portion of failure events by cross-referencing maintenance work orders with sensor data patterns, reducing manual labelling effort by up to 60%.

Yes. While vibration analysis is the most common PdM technique for rotating equipment, predictive maintenance can be performed using current draw, temperature, pressure, flow rate, torque, acoustic emissions, or any sensor signal that changes as equipment degrades. For example, current signature analysis detects bearing faults, misalignment, and electrical issues in motors without dedicated vibration sensors. Temperature trends detect coolant system degradation, insulation failure, and friction increases. iFactory's platform supports 40+ sensor types and automatically selects the most predictive signals for each asset and failure mode combination.

Model retraining frequency depends on the stability of the underlying equipment and operating conditions. For stable, continuous-operation assets with consistent production schedules, monthly retraining is typically sufficient. For assets with frequent production changeovers, seasonal operating patterns, or evolving failure modes, weekly retraining produces better results. iFactory's platform monitors model performance continuously and triggers automatic retraining when it detects data drift, concept drift, or a drop in prediction accuracy below configurable thresholds. This eliminates the need for manual retraining scheduling while ensuring models remain accurate over time.