Root Cause Failure Analysis (RCFA) for Repeat Failures: From Preserving Evidence to Preventing Recurrence

What is root cause failure analysis?

Root cause analysis examines an event that has already occurred to understand its underlying causes rather than the immediately obvious symptoms, and its techniques are not designed to assign responsibility or liability. ^[1] The US Department of Energy’s guidance defines the root cause as the cause that, if corrected, would prevent recurrence of this and similar occurrences. ^[2]

That is where the practical difference lies. Replacing a failed bearing gets the machine running again, and failure analysis may establish that the bearing failed by fatigue. Root cause failure analysis (RCFA) keeps asking: what created the physical condition that caused the fatigue? Who made the decision that created it? And what in the system allowed that decision? ^[6]

Why does the same failure return after every repair?

When maintenance stops at the direct cause, it enters a fail-fix-fail loop. ^[5] HSE’s guidance distinguishes three levels: the immediate cause, which is the most obvious reason for the event; the underlying cause, the less obvious system or organisational failing; and the root cause, the initiating failing from which all the others spring, generally a management, planning or organisational failing. ^[3]

The Department of Energy guidance adds another useful distinction: the direct cause, which directly produced the occurrence, and the contributing cause, which added to it but would not have caused it alone. ^[2] Repeat failures usually have several contributing causes, so do not settle for the first one that appears.

The PROACT methodology classifies roots into three types: the physical root, the observable physical consequence of a decision; the human root, the decision error; and the latent root, the reasoning or rationale behind the decision. ^[6]

Which failures deserve a full analysis?

A full analysis takes up a multidisciplinary team’s time, so it is not done for every failure. Equipment that repeatedly causes problems for safety, production and cost, known as “bad actors”, is the first candidate. ^[5]

In practice, identify them like this:

1. Initial screening: interview operations and maintenance staff, who know the troublesome equipment before it shows up in reports. ^[5]
2. Verify with data: pull maintenance costs and outage counts by equipment type from the maintenance management system, and estimate production losses. ^[5]
3. Pareto charts: three charts, for outage frequency, repair cost and production loss, show whether the problem lies in one item or across a whole equipment class. ^[5]
4. Drill down: break the results down by failure type, then component, then equipment tag number. ^[5]

One case described in the Becht article in Inspectioneering is an olefins producer whose inspection focused on internal corrosion, until Pareto analysis showed that over 90 per cent of leaks came from external corrosion; it moved to a prioritised corrosion-under-insulation programme, which significantly reduced leaks, outages and fires. ^[5] In other words, data can overturn priorities a team assumes are obvious.

The same article estimates that unplanned and emergency work costs roughly 3 to 9 times as much as planned work; that is a consultant’s estimate rather than a general rule, but a useful argument when asking for resources for analysis. ^[5]

The steps: from preserving evidence to verifying the effect

IEC 62740 sets out the stages of the analysis process: initiation, establishing facts, analysis, validation and presentation of results. ^[1] The Department of Energy guidance adds corrective actions and follow-up phases. ^[2]

Who should be on the team?

The analysis needs a broad, knowledgeable team and good failure data. ^[5] PROACT stresses an unbiased team with defined roles and resources. ^[6] It helps to include people who operate the equipment, people who maintain it and an engineer who knows its design, with a facilitator who does not own the equipment and so has no earlier decision to defend.

Preserve the evidence before it is lost

Start by making the area safe and then preserving the scene, because evidence can be corrupted if items are moved or guards are replaced. ^[3] Collect data as soon as the failure occurs, and preserve failed components despite the pressure to restore service. ^[2] A part that is cleaned or thrown away takes the most important evidence with it. A useful practice is to capture failed parts in their “as failed” condition before anyone works on them. ^[5] PROACT organises data collection under five headings: parts, position, people, paper and paradigms. ^[6]

Diagnose the failure mode first. An analysis may miss the true root cause if the failure mode has not been established by studying the failed component itself, because each mode needs a different solution; bolt fatigue, for example, may result from insufficient preload at installation or loss of preload in service. ^[4] Interviews are for finding facts, not finding fault. ^[2]

Widen the search beyond maintenance. IEC 62740 notes that a cause may arise at any stage: specification, design, manufacture, installation, operation or maintenance. ^[1] A bearing that keeps failing may trace back to a wrong choice at the purchasing stage rather than a maintenance error.

Test every hypothesis

The analysis reconstructs the event backwards in a logic tree and puts forward hypotheses, each tested against evidence. Verification ranges from weak, such as someone saying it happened, to strong, such as reconstructing and testing each possibility. ^[6] If you use repeated “Why?” questions, keep going until the answers are no longer meaningful. ^[3]

Which analysis tool should you choose?

IEC 62740 lists a wide range of techniques, including the “why” method, the fishbone (Ishikawa) diagram, fault tree analysis and events and causal factors charting, with an annex on analysing human performance. ^[1] The Department of Energy guidance suggests which tool to use in which situation: ^[2]

Hypothetical example: a pump that keeps destroying its bearings

A cooling water pump at an industrial plant tripped three times in four months, and each time the drive-end bearing was replaced. The team kept the last bearing, and inspection showed a wear pattern consistent with shaft misalignment. It emerged that alignment was set after each installation with simple tools and no measurement.

• Physical root: misalignment between motor and pump.
• Human root: restarting without measuring alignment.
• Latent root: the maintenance procedure specifies neither a measurement method nor a tolerance, no suitable alignment tool is available, and the belief has taken hold that aligning by eye is good enough.

Actions: revise the procedure with a clear method and tolerance, provide the equipment, train the technicians, then monitor vibration and verify that the failure does not recur.

Corrective actions: when does the analysis succeed?

Test each proposed action with specific questions: will it prevent recurrence? Is it feasible? Does it introduce new risks? Can its implementation be measured? ^[2] Turn the actions into a plan in which every item has an owner and a timescale, and monitor progress. ^[3]

Do not accept “technician error” as the final answer. An investigation that concludes operator error was the sole cause is rarely acceptable, because human error usually sits on top of deeper causes such as training, supervision or equipment design. ^[3]

Most importantly, analysis alone does not improve equipment performance; performance improves only when the solution is executed in the field, and evaluating the outcome is essential, especially where the root lies in procedures or human factors. ^[5] So close every case by verifying that the actions are effective and the problem has not recurred. ^[2] Brief management and the people involved on the findings, and share the lessons with sites that have similar equipment. ^[2]

Conclusion

Root cause failure analysis breaks the fail-fix loop when it reaches the decision and the system that produced the failure. Select cases with data, preserve the evidence, test your hypotheses, and do not close a case until you have shown the failure has not come back.