Operating hours, not calendar
Standalone by design
A ranked action
Warn, never promiseWhat gets measured, and what each signal is good for
| Signal | Finds | Note |
|---|---|---|
| Triaxial vibration | Imbalance, misalignment, looseness, bearing defect frequencies, gear mesh | The backbone. Needs a consistent mounting location to be comparable over time. |
| Ultrasound | Early bearing distress, lubrication condition, leaks, electrical discharge | Responds earlier than vibration on slow-speed and lubrication problems. |
| Surface temperature | Friction, cooling loss, lubrication failure in progress | Cheap, late. Confirming rather than predicting. |
| Motor current signature | Rotor bar, eccentricity, load anomalies | Reads the machine through the cable, no shaft access needed. |
Roughly four in five bearing failures trace back to lubrication rather than to the bearing itself, which is why ultrasound and a lubrication regime usually deliver more in the first year than any amount of spectral analysis.
Three rules that decide whether it works
Report on operating hours, not on the calendar
Duty and standby machines rotate. A monthly report on a calendar basis shows the standby machine as healthy when it has simply not been running, and it disguises how hard the duty machine has been worked. Trending on operating hours is the difference between a report that reflects the plant and one that flatters it.
Standalone, off the control network
Monitoring should not require a connection into plant control or IT networks. Separate acquisition, no path into the control system, nothing that can perturb a running process. This removes the objection that stops most monitoring projects inside plant IT, and it means the system can be installed without a shutdown.
Warn, do not promise
Nobody can honestly guarantee that a machine will not fail. What can be committed to is that the system will raise a warning on a developing fault within its detectable envelope, that availability is maintained, and that a missed warning on an asset reported healthy carries a stated consequence. Guarantee the warning, never the outcome. Any vendor guaranteeing the outcome has either not read their own detection limits or is not intending to be held to it.
The part that turns data into a decision
Detection is the easy half. The gap between a detected fault and a completed repair is where value is actually lost, and it closes with four things:
- A ranked action list across the fleet, not per-sensor alarms. Priority, estimated downtime if it is left, and what will be needed to fix it.
- A task pack the fitter carries to the machine. Diagnostics on one side, a checklist on the other. Confirm the noise audibly at the bearing housing, check lubricant level and condition, verify mounting bolt tightness. Diagnosis that does not reach the person holding the spanner has not been delivered.
- Scheduling that knows the production plan. A four-hour job recommended into a window that collides with a shipment is a recommendation nobody executes. Knowing the plan, and proposing the next slot that actually fits, is what turns a recommendation into a work order.
- A closed loop. Work order raised, executed, and the result fed back against the original diagnosis, so the thresholds improve instead of drifting.
How we start
One asset, proven, before anything scales. A nominated machine, instrumented, baselined over a defined period, and reported against agreed criteria. At the end of it you have evidence from your own plant rather than a case study from someone else's.
- Asset criticality. Which machines, if they stop, actually stop production or breach a permit. Monitoring everything is how monitoring programmes die.
- Baseline. What normal looks like on this machine, at this duty, in this plant. Without it, every reading is an opinion.
- Thresholds and escalation. Agreed with maintenance, written down, and owned by a named person.
- Prove it. Run through a defined trial. Review what it called, what it missed, and what it called wrongly.
- Scale on the same spine. Additional points on a proven architecture, not a second pilot.
Questions we get asked
How early does condition monitoring detect a failure?
It depends on the failure mode and the signal, and a specific lead time should not be promised in advance. Lubrication and early bearing distress often appear well before there is any audible or thermal symptom. Some failures are genuinely sudden and no monitoring detects them. An honest programme states its detectable envelope rather than a headline number of weeks.
Does it need to connect to our PLC or SCADA?
No. The acquisition is standalone by design, which means no path into plant control or IT networks and no shutdown to install. Where a plant does want operating state or production context brought in, that is a deliberate, separately agreed integration.
How many measurement points does a machine need?
Driven by the machine, not by a package. A typical motor and driven-equipment pair is instrumented at the bearing positions on both, with motor points where electrical signature is included. What matters more than the count is that the positions are consistent, so readings stay comparable over time.
Can you monitor equipment we bought from someone else?
Yes. The sensors go on the machine, not into the OEM's system, so the make and vintage of the equipment does not constrain it.
What happens if it misses a failure?
That should be written into the agreement rather than left to goodwill. Our position is that the commitment is on the warning, with a stated consequence for a missed warning on an asset reported healthy, after a proving period during which thresholds are set. What is never committed to is that a machine will not fail.