Maintenance strategy
Preventive vs predictive vs condition-based maintenance: which one, for which machine
Preventive maintenance services a machine on a fixed interval of time or usage. Condition-based maintenance waits until a measured reading, such as vibration or temperature, crosses a limit. Predictive maintenance goes one step further and trends those readings to forecast when the limit will be reached, so you can plan the job. Most plants need all three, chosen machine by machine.
Three ways to decide when to maintain
Every maintenance strategy answers one question: when should this machine get attention? The three common answers differ only in what triggers the work.
Preventive maintenance uses a fixed interval. Change the compressor oil every 4,000 running hours. Grease the crane every week. Check the press hydraulics every month. The interval is set from the maker's manual and your own history. The job happens whether the part needed it or not.
Condition-based maintenance uses a measured reading. You set a warning limit on a bearing's vibration, a motor's winding temperature or a filter's pressure drop. When the reading crosses the limit, you raise the work. The machine tells you when, but only once the problem is already visible.
Predictive maintenance uses the trend of those same readings. Instead of waiting for the limit, you look at how fast the reading is moving and estimate when it will get there. That turns a warning into a date, so the bearing is ordered and the job sits in a planned stop.
There is a fourth option that people forget: run to failure. For a cheap, non-critical part with a spare on the shelf, letting it fail and replacing it can be the right choice. It should be a decision you write down, not an accident.
Why predictive maintenance is not always the better choice
Predictive maintenance sounds like the most advanced option, so plants often assume it is always better. It is not. It only works when three things are true.
First, the failure must give a warning you can measure. A bearing usually gets noisier and hotter before it seizes. In reliability-centred maintenance, the time between the first detectable sign and the actual failure is called the P-F interval. An electronic card or a fuse often gives no measurable warning at all, so there is nothing to trend.
Second, the warning time must be long enough to act. If a gearbox gives you three weeks of rising vibration, a weekly reading is enough. If a spindle goes from first sign to seizure in a day, a weekly round will miss it.
Third, the readings must come often and reliably. A trend drawn through four readings taken by different people with different meters is a guess. Readings from a sensor every few minutes, or a disciplined round on the same point with the same instrument, give you a trend you can trust.
If any of the three is missing, a well set preventive interval is usually cheaper and more dependable.
Four questions to pick the strategy for each machine
- Does a failure matter? If it does not stop production, hurt anyone, damage quality or cost much to fix, run to failure is a fair choice. Keep a spare and move on.
- Is there a warning sign you can measure? Vibration, temperature, current, pressure, oil condition or a visual check. If yes, set a warning limit and use condition-based maintenance.
- Do readings arrive often enough to trend? If the sign develops over days or weeks and your readings come more often than that, project the trend ahead. That is predictive maintenance.
- If there is no sign, does wear follow hours, cycles or time? Seals, belts, filters and lubricant usually do. Use preventive maintenance by time or by usage. If none of the four fits, look at a design or operating change instead.
Which machine gets which: a table to copy
This is a starting point for common machines in Indian plants. Your duty, environment and history may push a machine into a different row, so change the last column to suit.
| Machine and failure mode | Warning sign you can measure | Suggested strategy | Your choice (fill in) |
|---|---|---|---|
| Screw compressor: oil and filter wear | Not reliably | Preventive by running hours | |
| Screw compressor: air end bearings | Vibration, discharge temperature | Condition-based, predictive if readings are frequent | |
| Centrifugal pump: bearings and seal | Vibration, bearing temperature, leakage | Condition-based on a weekly round | |
| Hydraulic press: oil condition and pump | Oil temperature, pressure, contamination | Preventive oil change plus condition checks | |
| CNC machining centre: spindle bearings | Vibration, spindle temperature | Predictive where a sensor exists, else condition-based | |
| EOT crane: wire rope, brakes, limit switches | Visual wear, brake test | Preventive inspection by time, statutory checks | |
| Induction furnace: refractory lining | Wear readings, heats per campaign | Usage-based with condition checks | |
| Conveyor: belt and idlers | Visual wear, noise | Condition-based on a walk-down round | |
| Electronic cards, fuses, indicator lamps | Usually none | Run to failure with spares on the shelf |
Statutory inspections, such as those for lifting equipment and pressure vessels, stay on their fixed schedule whatever your strategy says.
Worked example: a forging plant in Rajkot
Worked example (illustrative)
Take a forging and machining plant in Rajkot with 120 assets. Every machine is on a monthly preventive schedule, set up when the plant first moved off paper. All numbers below are illustrative.
The maintenance head runs the four questions over the 18 machines that stop a line when they fail, and over a sample of the rest. Here is what comes out:
| Group | Count | Before | After |
|---|---|---|---|
| Forging press main motors and gearboxes | 6 | Monthly PM | Weekly vibration and temperature round, trended; PM only for lubrication |
| Screw compressors | 3 | Monthly PM | PM every 2,000 running hours, plus a weekly discharge temperature reading |
| Cooling water pumps | 9 | Monthly PM | Condition-based on the weekly round with warning limits |
| Other machines | 84 | Monthly PM | Kept on PM, interval reviewed against breakdown history |
| Small fans, lamps, controller cards | 18 | Monthly PM | Run to failure, spares stocked |
What the example shows (illustrative). The plant drops 18 monthly jobs that were adding no value and moves 9 pumps to a round that takes less time than the old PM. The six press drives get the most attention because a failure there stops the whole forge. When a gearbox reading starts rising at about 0.2 mm/s a week and the warning limit is 1.0 mm/s away, the trend says about five weeks. That is enough to order a bearing and fit it in the monthly shutdown instead of during a night shift.
Common mistakes
- Buying sensors before choosing failure modes. A sensor on the wrong machine gives you a nice chart and no saved breakdowns. Pick the failures that matter first, then decide what to measure.
- Setting limits nobody owns. A warning limit with no named person and no action is a number on a screen. Decide who gets the alert and what they do.
- Trusting a trend built on bad readings. Different meters, different points on the casing and missed rounds make a trend jump around. Mark the measuring point and use the same instrument.
- Dropping preventive work too early. Keep the old interval running until the condition readings have proved themselves over a few months.
- Treating every machine the same. One strategy for the whole plant is easy to schedule and wrong for most machines.
- Never reviewing the choice. After every breakdown, ask whether the strategy for that machine should change. The record of that decision is what makes the next one better.
Sources
- Predictive maintenance (opens in a new tab), Wikipedia
- Condition monitoring (techniques and related ISO standards) (opens in a new tab), Wikipedia
- Reliability-centered maintenance (opens in a new tab), Wikipedia
- Maintenance of work equipment (opens in a new tab), Health and Safety Executive (UK)
Frequently asked questions
What is the difference between preventive and predictive maintenance?
Preventive maintenance is done on a fixed interval of time or usage. Predictive maintenance trends condition readings, such as vibration or temperature, to estimate when a part will reach its limit, and plans the work for that time.
Is condition-based maintenance the same as predictive maintenance?
They are close but not the same. Condition-based maintenance acts when a reading crosses a limit. Predictive maintenance uses the trend of the readings to forecast when the limit will be crossed, so the job can be planned earlier.
Do I need sensors for predictive maintenance?
Not always. A disciplined manual round with a handheld vibration meter or thermometer can be trended if readings are taken often enough, at the same point, with the same instrument. Sensors help when the warning time is short.
Which machines should stay on preventive maintenance?
Machines whose wear follows running hours, cycles or time, such as filters, belts, seals and lubricant, and anything with a statutory inspection. Also machines where no measurable warning sign exists.
When is run to failure acceptable?
When the failure does not stop production, does not create a safety or quality risk, is cheap to fix and a spare is on hand. Write the decision down so it is a choice, not an oversight.
Keep reading
Condition monitoring
Warning and critical limits on each parameter, alerts and a drift-to-limit projection.
ReadPreventive maintenance guide
Types, real examples and how to start a PM programme in one week.
ReadMTBF and MTTR calculator
Check whether a strategy change is cutting failures and repair time.
Read