Predictive maintenance is worth doing when it produces enough lead time to act. A model that detects a failing bearing four hours before it seizes has told you something you already knew from the noise. One that flags it three weeks out, with a confidence you can plan around, changes how the plant is run. Faststream engineers the sensing, the analytics and the integration with the maintenance system that makes the prediction actionable.
| Modality | Detects | Typical lead time |
|---|---|---|
| Vibration (accelerometer) | Bearing wear, imbalance, misalignment, looseness, gear faults | Weeks to months for bearing degradation |
| Motor current signature analysis | Rotor bar faults, eccentricity, load anomalies, some bearing faults | Weeks; needs no sensor on the machine itself |
| Thermal | Bearing overheating, electrical connection faults, cooling failure | Days to weeks |
| Acoustic and ultrasonic | Compressed air and steam leaks, early bearing distress, electrical discharge | Immediate for leaks; weeks for bearing distress |
| Oil and particulate | Wear debris, contamination, lubricant degradation | Weeks to months |
| Process data | Efficiency loss, fouling, developing blockage | Varies by process |
Supervised fault classification requires labelled failures, and the whole point of a well-maintained plant is that failures are rare. A plant with three bearing failures in five years cannot train a supervised classifier on bearing failure.
So the practical framing is anomaly detection against a healthy baseline, combined with the physics: bearing defect frequencies are calculable from geometry and shaft speed, so a rising component at the ball-pass frequency of the outer race is interpretable without ever having seen that failure on that machine.
Physics gives interpretability. Learned models give sensitivity to patterns the physics does not enumerate. Used together they produce alerts a maintenance engineer will actually believe — which is the binding constraint on the whole application.
Predictive maintenance monitors equipment condition to forecast failure with enough lead time to schedule intervention, rather than repairing after failure or replacing on a fixed calendar. The value is in the lead time, not in the detection itself.
Most commonly vibration, motor current signature analysis, thermal, acoustic and ultrasonic, oil and particulate monitoring, and existing process data. Motor current is particularly useful because it needs no sensor mounted on the machine.
Less than expected, because the practical framing is anomaly detection against a healthy baseline rather than supervised fault classification. A baseline across the real duty cycle is usually enough to begin, and physics-based defect frequencies make alerts interpretable without prior failure examples.
Through CMMS or maintenance-system integration. An alert that generates a work order with supporting evidence gets acted on; an alert that generates an email competes with everything else in an inbox.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.