Condition monitoring is an activity, measuring signals that reveal machine health, such as vibration, temperature, oil condition, and current. Predictive maintenance is a strategy that consumes those measurements, adds trend analysis to forecast when failure will occur, and schedules the repair just before it. One is a data source; the other is a decision system built on top of it. They are not two products you pick between.
The confusion is understandable, because vendors sell both words on the same brochure. But treating them as rival options leads to real mistakes, buying a monitoring system and expecting predictions it was never set up to make, or budgeting for predictive maintenance without funding the monitoring that feeds it. This piece untangles the two: what each actually is, how they nest, where condition-based maintenance sits between them, and how to tell which you need.
What is the difference between condition monitoring and predictive maintenance?
The core difference is category: condition monitoring is what you measure predictive maintenance is how you decide. Condition monitoring produces health data. Predictive maintenance is a maintenance strategy that turns that data into a forecasted failure window and a planned work order. You can run condition monitoring purely to log readings; it becomes predictive maintenance only when someone trends the data, forecasts a failure date, and acts on it.
Put concretely: a technician who records bearing vibration every month is doing condition monitoring. When that vibration trend is projected forward to say the bearing will reach failure in about six weeks, so the replacement is scheduled into planned downtime three weeks out, that is predictive maintenance. Same reading, different altitude of decision.
How do the two relate on the maintenance ladder?
They sit on the same ladder, at different rungs, with condition-based maintenance between them. The ladder runs from doing nothing with data to forecasting with it:
- Condition monitoring you measure and record health signals. This alone changes no maintenance decision; it just makes the machine's state visible.
- Condition-based maintenance (CBM) you set thresholds on those signals and trigger work when one is crossed. This is the first rung where monitoring drives action. CBM reacts to condition but does not forecast a failure date.
- Predictive maintenance (PdM) you add trend analysis and, increasingly, models on top of the condition data to forecast when failure will occur, and plan the repair into that window.
So the honest framing is not "condition monitoring vs predictive maintenance" but "condition monitoring, then what you do with it." A plant can stop at monitoring (visibility only), stop at CBM (threshold-triggered work), or invest in PdM (forecasting), and it usually stops at different rungs for different assets. The per-asset logic is the whole subject of our equipment reliability guide.
| Condition monitoring | Predictive maintenance | |
|---|---|---|
| Category | Activity, measurement | Strategy, decision system |
| Question answered | What is the machine's condition now? | When will this machine fail, and when do we fix it? |
| Core output | Readings and trends | A forecast failure window and a planned work order |
| Added ingredient | Sensors, routes, baselines | Trend analysis, prognostics, planning |
| Skill needed | Data collection and diagnostics | Diagnostics plus prognosis and scheduling discipline |
| Can exist alone? | Yes, as pure visibility | No, it requires condition monitoring underneath |
Can you have condition monitoring without predictive maintenance?
Yes, and many plants do, without realizing it is a legitimate stopping point. A monthly vibration route with alert limits, an infrared survey of electrical panels, quarterly oil samples: all condition monitoring, none of it forecasting a failure date. It delivers real value, early warning, fewer surprise failures, as condition-based maintenance without ever becoming predictive maintenance. For most assets that is exactly the right ceiling, because forecasting adds cost and skill that only the most critical machines repay.
The reverse is impossible. Predictive maintenance without condition monitoring underneath is just a calendar with ambition. There is no failure to forecast without measured degradation to trend. Anyone selling "predictive maintenance" that does not rest on a working condition monitoring program is selling a dashboard, not a strategy.
How does condition monitoring become predictive maintenance?
Turning monitoring into forecasting is a defined progression, not a purchase. Follow the sequence and you climb the ladder deliberately instead of buying a tool and hoping.
- Get repeatable condition data first. Fixed measurement points, constant conditions, a consistent interval. A forecast is only as good as the trend under it, and a trend is only as good as measurement repeatability.
- Establish healthy-state baselines. You cannot detect degradation, let alone project it, without knowing normal. Capture baselines under standard operating load for every monitored point.
- Trend, do not just record. Plot each parameter over time. The moment you look at the slope rather than the latest number, you have left simple monitoring and started prognosis.
- Estimate the P-F interval per failure mode. Learn how much warning each mode gives between first detectability and functional failure. That interval is the planning window predictive maintenance sells.
- Project the trend to a failure window. Extend the slope to the action limit to forecast when the asset reaches functional failure, by analyst judgment first, by model later once you have enough data.
- Convert the forecast into planned work. Schedule the repair into planned downtime inside the P-F window, with parts staged. A forecast that does not become a dated, resourced work order is a prediction nobody acted on.
Notice how much of the climb is discipline rather than technology. The step from monitoring to predictive maintenance is mostly baselines, trending, and planning and scheduling not a magic algorithm.
The same bearing at three rungs
Take one gearbox bearing and watch what each rung does with the identical reading, so the distinction stops being abstract:
- Monitoring only. A technician records 3.2 mm/s of vibration this month, up from 2.8 last month. The number is logged. Nothing else happens, the plant now knows the bearing's state, and that is the entire deliverable. If a supervisor happens to glance at the log and worry, that is luck, not a system.
- Condition-based maintenance. The same 3.2 mm/s crosses a pre-set alert limit, which automatically opens an inspection task; a later reading crossing the action limit opens a work order. Work is now driven by condition, but nobody has said when the bearing will actually fail, only that it has degraded past a line someone drew.
- Predictive maintenance. The rising trend is projected forward to the action limit, forecasting functional failure in roughly five to seven weeks. The bearing replacement is scheduled into planned downtime three weeks out, the part is staged from spare-parts inventory and production is warned. The emergency became a calendar entry.
Same sensor, same physics, three very different amounts of value, and each higher rung is impossible without the one beneath it. That is why "versus" is the wrong word. The real question is how far up the ladder a given asset justifies climbing.
What do the standards and numbers say?
The primary references keep the two straight:
- ISO 17359 governs the condition monitoring activity, asset selection, failure-mode focus, parameters, baselines, and limits (ISO 17359:2018). Prognostics, the forecasting layer that turns monitoring into prediction, is a separate body of ISO work (the ISO 13381 series), which itself signals that measurement and forecasting are different jobs.
- The economics belong to the strategy, not the sensors: U.S. DOE FEMP guidance maintained by PNNL puts predictive/condition-driven savings at 8–12% over preventive-only programs and 30–40%+ over reactive ones (PNNL, O&M Best Practices). That gain comes from acting on the data, which is why monitoring alone captures only part of it.
- The BLS projects 13% growth from 2024 to 2034 for industrial machinery mechanics and maintenance workers, much faster than average (BLS). Scarcer technicians raise the payoff of forecasting, which concentrates hours on machines that genuinely need work.
Which one does your plant actually need?
Most plants need both, layered by asset. Every critical asset needs condition monitoring; only the subset where a forecast pays for itself needs full predictive maintenance on top. Non-critical equipment may need neither, deliberate run-to-failure or a preventive interval is often the right economic answer. The decision is per asset, driven by failure consequence and whether degradation is measurable, and the metric that proves either investment worked is the same: MTBF trending up and unplanned downtime trending down.
The real-world blocker is rarely the technique choice. It is that condition readings, machine monitoring data, work orders, and downtime logs sit in separate systems, so nobody can trend across them or turn a forecast into a scheduled job without manual stitching. Connecting those sources into one operational layer, no rip-and-replace, is what lets a plant climb from monitoring to predictive maintenance without buying a new platform for each rung; the CLS case study shows unified plant data in practice.