Predictive maintenance
Predictive maintenance explained: what it is and what it costs to run
Predictive maintenance is maintenance triggered by the measured condition of an asset rather than by a calendar date or a runtime interval, so the work is scheduled from evidence of degradation instead of from a fixed plan. Measurements catch the onset of that degradation early enough to act, which means the repair happens just before failure rather than after it. It is also called condition-based maintenance, and the two terms are used interchangeably in most plants.
For field service
Know what the job needs before the van rolls
Venta Capture, a product of VentaVid, lets the customer show you the fault first, so the engineer arrives with the right part or does not need to arrive at all.
The difference from preventive maintenance is the trigger, not the technique. Preventive says the oil is due at 5,000 miles. Predictive analyses the oil and says it has another 5,000 in it.
How does predictive maintenance work?
Every predictive programme needs a signal that changes measurably before the failure does. The established ones:
- Vibration analysis. The workhorse for rotating equipment. Bearing and alignment faults show up in the spectrum weeks to months ahead.
- Infrared thermography. Electrical connections, motors, steam traps, refractory. Heat appears before smoke.
- Oil analysis. Wear particles and contamination reveal what is happening inside a gearbox without opening it.
- Ultrasound. Leaks, steam traps, early bearing wear and electrical discharge.
- Motor current signature analysis. Rotor bar and load faults read from the electrical supply.
- Structured visual inspection. The oldest one, and still the only condition signal available on most of the asset base.
A reading on its own is not a prediction. You need a baseline, a trend and an alarm threshold, which is why the first months of any programme produce data rather than decisions.
What predictive maintenance delivers
The most-cited figures come from the US Department of Energy's Operations and Maintenance Best Practices Guide, which reports independent survey averages for a functioning predictive maintenance programme:
- Reduction in maintenance costs: 25 to 30 percent.
- Elimination of breakdowns: 70 to 75 percent.
- Reduction in downtime: 35 to 45 percent.
- Increase in production: 20 to 25 percent.
- Return on investment: 10 times.
The same guide is more conservative about the incremental step: an estimated 8 to 12 percent saving over a preventive programme alone, rising to 30 to 40 percent where a facility is coming off heavy reactive maintenance. On the cost-per-horsepower comparison it reproduces, predictive runs at $9 per horsepower per year against $13 preventive and $18 reactive.
Predictive maintenance explained: a worked example
A 250 kW extraction fan is on a monthly vibration route. Overall velocity sits at 2.1 mm/s for a year, then climbs to 3.0, then 4.4 over six weeks, with the increase concentrated at the bearing defect frequency. The bearing is ordered and changed during a shutdown already planned for the following month. Under a preventive regime the same bearing would have been left until the annual overhaul, or found by the failure.
How predictive differs from preventive and reactive
Three strategies, three triggers: reactive waits for failure, preventive follows a schedule, predictive follows measured condition. No plant runs one of them alone, and the mix is the real decision. The DOE guide puts the average facility at more than 55 percent reactive, 31 percent preventive and 12 percent predictive, while continually top-performing facilities sit at under 10 percent reactive, 25 to 35 percent preventive and 45 to 55 percent predictive.
Read the top-performer numbers carefully. Even at the best sites, roughly half the programme is still schedules and inspection routes. Predictive is the dominant strategy there, not the only one.
What predictive maintenance needs that most equipment does not have
The savings figures assume a condition signal exists. On most installed equipment it does not. Retrofitting sensors, wiring or gateways to older assets, then building the history to know what normal looks like, is a project rather than a setting, and it is rarely justified on a low-criticality asset.
Cost is the documented blocker. NIST's report on the costs and benefits of advanced maintenance in manufacturing (Douglas S. Thomas, 2018) found cost was the most prevalent barrier to adopting advanced maintenance strategies, named by 92 percent of respondents, followed by technology support at 69 percent, human resources at 62 percent and organisational readiness at 23 percent. The DOE guide is equally blunt: much of the diagnostic equipment runs past $50,000, and training staff to use it well takes real funding.
Which is why visual inspection still carries so much of the load. For the large share of assets that will never be instrumented, a person looking at the machine is the condition signal, and the quality of the programme depends on whether that look is consistent, recorded and comparable to the last one.
That is a workflow problem rather than a sensor problem. Venta Capture, a product of VentaVid, sends a link to whoever is already standing at the asset, guides them through the exact views and questions a reliability engineer would ask, and returns a timestamped, structured case for review by someone who was not there. It does not diagnose anything: the assessment stays with the qualified person. What it changes is whether a manual condition check produces a comparable record, which is the difference between an inspection route and a trend. The related practices are guided capture, equipment verification and remote diagnostics.
