Predictive Maintenance for Pumps: Vibration & Data

Executive summary

A pump rarely fails without warning. Weeks or months before the stoppage it leaves measurable signals: vibration rises, the bearing temperature departs from its baseline, and the energy signature —the power drawn for the same operating point— begins to drift. Predictive maintenance means reading those signals in time and turning them into a scheduled intervention rather than an emergency. In this article you will see how they are measured (with the ISO 10816 / ISO 20816 standards for vibration), how predictive maintenance differs from preventive and reactive, and how to calculate the return on avoiding an unplanned stoppage with a verifiable worked example.

Who this is for

For maintenance, reliability and plant engineers who manage fleets of critical pumps and have to justify to management why they should install sensors or contract vibration-analysis routes. Also for anyone who works "run-to-failure" today and suspects they are overpaying in downtime, overtime and secondary damage, but needs to put a number on it. If you are asked to decide between continuing to react, doing calendar-based maintenance, or monitoring the machine's real condition, this article gives you the criteria and the arithmetic.

The real plant problem

The most expensive operating mode is reactive: run the equipment until it breaks. It sounds cheap because nothing is spent while it works, but the bill arrives all at once and at the worst moment. A pump that fails without warning drags in unplanned downtime, spare parts at emergency prices, overtime and, often, secondary damage: a bearing that moves out of tolerance ends up scoring the shaft or destroying the seal. Calendar-based (preventive) maintenance improves things, but it services healthy equipment and does not catch the fault that appears between two service intervals. The blind spot is the same in both cases: no one is looking at the machine's real condition.

Engineering fundamentals

Condition monitoring rests on three complementary signals.

1. Vibration. It is the earliest and richest indicator. You measure the root-mean-square (RMS) vibration velocity, in mm/s, on the bearing housings. The ISO 10816 standards and their successor ISO 20816 grade severity into four zones: A (newly commissioned equipment), B (acceptable for unrestricted continuous operation), C (unsatisfactory for the long term; operate for a limited time until corrected) and D (vibration severe enough to cause damage). The part specific to rotodynamic pumps is ISO 10816-7; the general part for machines above 15 kW is ISO 20816-3. For a medium pump (Group 2) on a flexible support, the reference limits are:

2. Temperature. Every bearing has a baseline at steady load. What matters is not an absolute value but the deviation (ΔT): a sustained rise in bearing temperature above its baseline, at the same load and ambient, betrays a lack of lubrication, misalignment or wear.

3. Energy signature. If the pump draws more power for the same flow and the same head, something changed: worn wear rings, recirculation, obstruction or a degraded impeller. Power is a mirror of efficiency.

The economic value of the method comes down to a single relationship:

Benefit = (Cost of the unplanned failure − Cost of the planned intervention) × expected frequency − Cost of monitoring

How to apply it step by step

• 1. Define criticality. Not every pump warrants permanent sensors; prioritize those critical to the process and to safety (ISO 10816-7 calls them Category 1).
• 2. Establish the baseline. Measure vibration, temperature and power with the machine healthy, at its normal operating point.
• 3. Set thresholds. Use the ISO zones for vibration (alert at B/C, action before C/D) and a ΔT alarm for temperature.
• 4. Track the trend. A single reading says little; the slope over time says everything.
• 5. Turn the signal into a work order. When the trend crosses the threshold, schedule the intervention in the next window, not when the equipment decides.
• 6. Close the loop. After repair, measure again and confirm the return to Zone A/B.

Worked example with numbers

Take a medium centrifugal pump of 45 kW, Group 2 on a flexible support, with monthly measurement routes. Its baseline is 2.1 mm/s RMS (Zone A/B). Over five months the trend evolves like this:

The month-5 reading (4.8 mm/s) crosses the B/C boundary of 4.5 mm/s: the alert threshold. More important than the value is the acceleration of the slope (0.3 → 0.6 → 0.9 → 0.9 mm/s per month): at that rate the pump would reach the C/D damage threshold of 7.1 mm/s in about two or three months. That is the window to schedule the correction while the equipment is still serviceable.

Now the return. We compare the cost of letting it fail against that of intervening in time, with these stated assumptions:

If this class of pump produces, per the history, a failure of this type every two years (an expected frequency of 0.5 per year), the expected annual cost avoided is 0.5 × 22,000 = $11,000/year. Monitoring costs $3,000 in hardware per pump (vibration and temperature sensors plus a gateway) and $600/year in software and analysis. The result:

• Net first-year benefit: 11,000 − 3,000 − 600 = $7,400.
• Payback period: 3,000 ÷ (11,000 − 600) = 3,000 ÷ 10,400 = 0.29 years ≈ 3.5 months.

Monitoring pays for itself in a quarter and, from then on, every failure it turns from a surprise into a scheduled intervention is worth $22,000. Note that the largest saving is not in the spare part but in the production hours and the secondary damage avoided by not running the machine to breakage.

When it applies and when it does not

Continuous monitoring with permanent sensors is justified on critical pumps: those that stop the process, those handling hazardous fluids, or those with long spare-part lead times. On small, redundant or cheap-to-replace pumps the equation changes: a quarterly manual route may be enough, or even running to failure if there is immediate backup. Nor does it replace judgment: the ISO standard gives reference zones, not an automatic verdict; a pump may live in Zone C under watch if the manufacturer and the history support it. And the method loses value if the baseline is not well established or if you look at a loose reading instead of the trend.

Common mistakes

• Reading a value and not the trend: 3.9 mm/s is acceptable; 3.9 mm/s rising fast is an alarm.
• Comparing against the wrong standard: limits change between rigid and flexible support and between size groups; using the wrong table gives false positives or negatives.
• Not fixing the baseline: without a healthy reference, temperature and power say nothing.
• Measuring at the wrong operating point: comparing vibrations at different flows is not comparing.
• Monitoring everything equally: spending sensors on trivial pumps and leaving the critical one unwatched.
• Detecting and not acting: the signal is only worth something if it becomes a scheduled work order.

Decision checklist

• Did you rank your pumps by criticality before deciding what to monitor?
• Do you have a healthy baseline of vibration, temperature and power?
• Are you using the correct ISO table (part 7 for pumps; the right support and group)?
• Are you watching the slope of the trend, not just the last value?
• Did you define alert (B/C) and action (before C/D) thresholds?
• Does every alarm trigger a work order with a scheduled window?
• Did you estimate the cost of an unplanned stoppage to justify the investment?

Frequently asked questions

What is the vibration limit for a pump under ISO 10816?

It depends on size and mounting. For a medium pump (Group 2) on a flexible support, the reference RMS velocity boundaries are 2.3 mm/s (A/B), 4.5 mm/s (B/C, alert threshold) and 7.1 mm/s (C/D, damage threshold). The part specific to rotodynamic pumps is ISO 10816-7, and the current general standard for machines above 15 kW is ISO 20816-3.

What is the difference between predictive, preventive and reactive maintenance?

Reactive fixes the machine once it has already failed; preventive intervenes on a calendar even if the machine is healthy; predictive measures the real condition (vibration, temperature, power) and only intervenes when the data anticipates failure. Predictive avoids both the surprise of reactive and the wasted spend of preventive.

How much does predictive maintenance save versus reactive?

According to the U.S. Department of Energy (DOE) O&M Best Practices guide, a well-run predictive program saves 8% to 12% over a preventive one, and on the order of 30% to 40% over run-to-failure, with typical reductions in unplanned downtime of 35% to 45%. The bulk of the saving is in the production and secondary damage avoided.

SEMHYS tools

Before instrumenting a pump it is worth confirming that it operates near its best point: a poorly selected machine vibrates and wears more, and no sensor corrects that. Our free pump calculator helps you check the operating point, the total dynamic head and the NPSH margin, so you can separate a selection problem from a condition one. And if you are looking for engineering tools and guides to set up your reliability program, see the shop.

References

1. ISO (2022). ISO 20816-3: Mechanical vibration — Measurement and evaluation of machine vibration — Part 3: Industrial machinery with nominal power above 15 kW.
2. ISO (2009). ISO 10816-7: Mechanical vibration — Evaluation of machine vibration by measurements on non-rotating parts — Part 7: Rotodynamic pumps.
3. U.S. Department of Energy / PNNL (2010). Operations & Maintenance Best Practices — A Guide to Achieving Operational Efficiency, Release 3.0 (PNNL-14788).
4. Europump & Hydraulic Institute (2013). Guidelines on Pump Vibration.