Five Seal Failures in Fourteen Months on Cooling-Water Pumps
A refinery's cooling-water system uses six identical API 610 centrifugal pumps in parallel, three running and three on standby. Before a process debottlenecking 18 months ago, mechanical seals lasted about four years. Since then there have been five seal failures in 14 months, with seal life between 3,000 and 5,000 running hours. The current PM is a full rebuild at 8,000 running hours regardless of condition. Operations sometimes run pumps at roughly 30% of best efficiency point to control flow. Suction strainers are cleaned quarterly. A contractor performs a monthly vibration route, but there is no oil analysis and no thermography. Vibration readings are filed but not trended, and no one has reviewed them since the contract began. Production has asked maintenance to "do something about the seals," and the maintenance manager wants to shorten the rebuild interval to 4,000 hours. The plant runs a CMMS that records work orders, but failure codes are free text and inconsistent.
1. Apply RCM decision logic to this situation. What are the credible failure modes and what does the logic say about the current 8,000-hour rebuild?
Read the worked answer
The 8,000-hour rebuild is a scheduled restoration task, and RCM only accepts it if two things are true: the failure mode has a demonstrated age-reliability relationship, and the consequence justifies the task. Neither is established here.
Credible seal failure modes in this duty:
- Face wear or distress from running well below BEP (30% of BEP causes recirculation, hydraulic instability, and shaft deflection).
- Cavitation or contamination from a fouled suction strainer.
- Loss of flush/cooling, dry running on a standby pump that starts against a closed or restricted path.
- Elastomer degradation from water chemistry or temperature.
- Installation damage and misalignment at rebuild-which the current rebuild itself introduces.
- Bearing distress transmitting to the seal chamber.
What the evidence says. Five failures in 14 months, with lives spread 3,000–5,000 hours, is not a wear-out signature. It is a changed-rate signature that coincides with a process change. RCM would therefore not treat "seal wear" as the failure mode to manage by interval; it would target the operating-context causes (low-flow recirculation, strainer condition, flush plan) as the dominant modes and consider redesign or operating change. Scheduled restoration is only applicable once you can show a stable life and choose an interval below it.
If the seal failure is detectable before functional failure-leakage, temperature rise, vibration shift-an on-condition task is the technically applicable choice and is cheaper than repeated rebuilds.
2. Evaluate the maintenance manager's proposal to halve the rebuild interval to 4,000 hours. Under what conditions would you support it, and what would you do instead?
Read the worked answer
The argument for it. The current interval of 8,000 hours exceeds the observed failure-free life of 3,000–5,000 hours. In interval terms, the plant is rebuilding after failure, not before it. On that logic alone, 4,000 hours looks defensible.
Why it is still the wrong first move:
- It doubles the number of intrusive rebuilds on six pumps, doubling the opportunity for installation error, which is itself a credible failure mode. Infant mortality after rebuild can easily eat the gain.
- It does not touch the cause. If the pumps are being run at 30% of BEP since the debottlenecking, seals will keep failing at 3,000–5,000 hours and you will simply buy the same failures more often, at roughly double the annual seal and labour cost.
- The duty is uneven: three pumps run while three stand by. A fleet-wide running-hour interval applied uniformly ignores that running hours accrue at different rates per pump, and standby pumps have their own failure modes (seal faces sticking, stagnant water).
- No one has verified that the interval has any effect, because vibration is not trended and there is no teardown analysis.
What I would support. A documented interim task if a task is needed at all: set the interval below the shortest observed life with margin, around 2,000–2,500 running hours, applied only to the pumps with the worst duty, and only while the operating point is being corrected. Mark it as interim with a review date.
What I would do instead. Fix the operating point (minimum flow bypass or a variable-speed control), restore strainer ΔP monitoring, and confirm the flush plan against API 682-those address the mode rather than its symptom.
3. Define a 60-day data plan: what you would collect, how, and which measures would tell you whether the intervention worked.
Read the worked answer
Days 1–20, restore visibility
- Require the vibration contractor to deliver trended spectra with alarm limits and a written call-out, not a filed report nobody reads. Add monthly oil analysis (water content, ISO cleanliness) and thermography on couplings and seal chambers for the three running pumps.
- Log the operating point per pump: flow, discharge pressure, and calculated percentage of BEP. A clamp-on ultrasonic flow meter is sufficient if there is no permanent instrumentation.
- Log suction strainer differential pressure weekly instead of cleaning quarterly.
Days 20–60, fix the data at source
- Build a small, standard seal failure code set (face distress, thermal, elastomer, contamination, installation, dry run) and require every seal removal to carry a code plus photographs and a one-page teardown note. Free-text failure codes are the reason 18 months of evidence is unusable.
- Reconcile the five historical failures against that code set from memory and photos where available.
Measures that show the intervention worked
- Running hours to failure per pump, and MTBF across the fleet, tracked per pump rather than fleet-wide.
- Failure-mode Pareto: the share of failures attributable to low-flow operation.
- Percentage of running time below 50% of BEP (target: sharply reduced).
- Condition-detected seal distress versus in-service failure ratio; the goal is that the next seal problem is found by a route, not by a puddle.
- Number of rebuilds per year and seal cost per pump-year.
Set a review at day 60 to decide whether the interim 2,000–2,500 hour task stays or is deleted.