Roadmap › Industry 4.0 & IIoT — Smart Manufacturing Software › Module: Advanced Use Cases
Lessons in this course (10)
8. Predictive maintenance and AI in manufacturing
Advanced★ Must knowLesson 8 of 10
| Strategy | Trigger | Downside |
|---|---|---|
| Reactive | Fix after failure | Unplanned downtime, secondary damage |
| Preventive | Fixed schedule (hours, cycles) | Replaces healthy parts, still misses random failures |
| Condition-based | Threshold on a measurement (vibration > limit) | Late warning |
| Predictive | Model predicts failure/remaining useful life from trends | Needs data, labels and domain knowledge |
Data used: vibration (accelerometers, FFT spectra), temperature, current/power, pressure, acoustic, oil analysis, cycle counts, alarm history, maintenance records (labels).
Approach: collect and label history → feature engineering (RMS, kurtosis, spectral peaks, trends) → anomaly detection (isolation forest, autoencoders) when few failures exist, or classification/regression for remaining useful life → deploy at edge or cloud → alert + create CMMS work order → measure avoided downtime and false alarms.
Other AI use cases: visual inspection with computer vision (defect detection), process parameter optimisation, demand-driven scheduling, energy optimisation, generative-AI assistants that answer operators from manuals and work history.