- November 22, 2026
- Posted by: editor
- Category: RECD & Emission Control Devices
After a failure has been investigated, someone has to write it down. This is where many good investigations quietly go wrong. The engineers know that the pressure sensor tube was loose, the core was probably cracked, the fuel may have been wet and the regeneration had failed twice. What lands in the report is a smooth paragraph that reads: “The RECD failed due to excessive soot loading caused by poor fuel quality.” It sounds authoritative. It may even be right. But nobody can tell which parts were seen, which were measured and which were filled in to make the story flow.
A report like that is dangerous. A manager acts on it, a warranty team accepts or rejects a claim because of it, and the next failure is “explained” by the same sentence. If the guess was wrong, the real cause stays in place and the failure repeats.
This guide shows how to build a root-cause report for a retrofit emission control device (RECD) that does not guess. The method rests on four ideas: a claims ladder that tags every statement by how strongly it is supported, a report anatomy with a fixed order, a clear separation of root cause from contributing factors, and a fill-in template. We finish with a worked example. The approach is general. Where your company, supplier or regulator has a format, use that and apply the ladder inside it.
Why reports guess
Reports fill gaps for ordinary reasons. People want closure. Stakeholders want a single clear answer. Writing “we do not know” feels like failure. And the investigator has often seen so many similar faults that the pattern fills in by itself.
Root-cause guides warn against exactly this. One widely used guide advises against closing an analysis with “operator error” and says the question should continue until the process or system condition that allowed the error is found (Casrai guide to root cause analysis). A fleet-maintenance article on RCA for repeat failures makes a related point: fixing a symptom invites the same failure back (Fleet Maintenance on root cause analysis). Incident-report templates likewise draw a clear line between the root cause and the contributing factors that made the event possible or worse (Statuspal root cause analysis template).
The remedy is to make guessing visible. If every statement in the report carries a label, a reader can see at once where the report stands on solid ground and where it does not.
The claims ladder
Every statement in a root-cause report belongs on one of four rungs. Label each one.
| Rung | Label | Meaning | Example |
|---|---|---|---|
| 1 | Observed | Someone saw, heard or photographed it, and the record exists | “Soot streaks were present at the inlet flange on 12 Oct (photo 4).” |
| 2 | Measured | An instrument gave a value, and the instrument and conditions are known | “ΔP was 52 mbar at 60% load by portable gauge, against 31 mbar at commissioning.” |
| 3 | Inferred | Drawn from rungs 1 and 2 by reasoning, and the reasoning is stated | “The matching rise at the inlet tap and fall at the outlet suggests the restriction is in the core, since the exhaust pipe has no change.” |
| 4 | Assumed | Believed or reported but not checked | “The fuel was believed to be from the usual supplier.” |
Four rules keep the ladder honest.
- Only rungs 1 and 2 may be written as plain fact. Rung 3 uses words such as “indicates” or “is consistent with”. Rung 4 uses “was reported”, “is assumed” or “has not been verified”.
- A conclusion may rest on lower rungs only if it says so. If the root cause rests on an assumption, the report must say that the root cause is conditional.
- Every inference names its premises. “Because X and Y were observed, Z is inferred.”
- Every assumption becomes a to-do. Either verify it or leave it flagged. The next section’s template has a column for this.
The ladder is the heart of the method. It changes the report from a story into an evidence map. Readers can disagree with a rung-3 inference and still trust the rung-1 facts.
Report anatomy: ten parts in a fixed order
A fixed order helps the writer, who does not have to invent a structure, and the reader, who always knows where to look.
1. Summary (written last, placed first)
Three to six lines: what happened, the impact, the most probable root cause with its confidence level, the main contributing factors and the key corrective actions. Anyone reading only this part should not be misled.
2. Scope and boundaries
State which set, which RECD, which period and what the report does not cover. A clear scope prevents readers from stretching findings to other sites.
3. Event description
What the failure looked like, in neutral terms: the symptom, when it started, how it was noticed and what impact it had (downtime, compliance exposure, cost). Use rung 1 and 2 statements only. Leave out explanations here.
4. Timeline
A dated sequence of events from before the first symptom to the present, covering installation, service, fuel deliveries, weather, load changes, alarms and actions. If events are tracked on several lines, our guide on building a four-track timeline after an RECD fault shows how. Mark the source and rung of every entry.
5. Evidence register
A table of every item of evidence: photos, logs, instrument readings, samples, service reports, interviews and drawings. Give each an ID (E1, E2 and so on), the source, the date, who collected it and its condition. The rest of the report cites these IDs. If you need to know what to capture in the first place, see what to record in an RECD event log and planning intermittent-fault records.
6. Analysis
How you moved from evidence to cause. State the method used (see the next section) and show the reasoning. Include the alternatives you considered and rejected, with the evidence that rejected them. This part persuades a sceptical reader more than anything else.
7. Root cause and contributing factors
Written separately, as described below.
8. Confidence and open questions
List what remains unverified, what would change the conclusion and what test would settle it.
9. Corrective and preventive actions
Each action has an owner, a due date and a way of checking that it worked.
10. Appendices
Photos, raw logs, calculations, the controller’s settings export and supplier correspondence.
Choosing an analysis method
Different methods suit different jobs. A comparison article on the three most common ones describes 5 Whys as a quick way to dig down a single causal chain, fishbone diagrams as a way to organise many possible causes by category, and fault trees as a logical structure for complex, multi-condition failures (Casrai guide to root cause analysis).
| Method | Good for | Watch out for | RECD example |
|---|---|---|---|
| 5 Whys | A simple fault with one clear chain | Stopping early; hiding branches; reasoning from assumptions | Why did the alarm trip? Pressure was high. Why? Soot load. Why? Frequent light-load running. Why? A duty change nobody recorded. |
| Fishbone (cause-and-effect) | Brainstorming across categories when the cause is unclear | Becoming a wish-list with no evidence | Categories such as fuel, load, installation, sensing, maintenance, environment, with evidence ticks on each |
| Fault tree | Failures needing several conditions to occur together | Over-complexity; needs good data | Filter damage = thermal event AND weak core, where a thermal event = failed regeneration OR fuel in the exhaust |
| Hypothesis table | Comparing a short list of rival explanations | Weak if the predictions are vague | See the four-way fork for pressure falls and the twin sheet |
Whichever you pick, add the claims ladder. A “why” in a 5 Whys chain should carry a rung. A fishbone branch should carry evidence tick marks. A fault tree node should say how it was verified.
Root cause versus contributing factors
A common mistake is to name one cause when the failure needed several conditions. Separate the two.
- Root cause: the fundamental condition which, if removed, would have prevented this failure from happening (or from recurring in this way). It usually lies in a process, design, specification or decision, not in a part.
- Contributing factors: conditions that made the event more likely, faster or worse, but would not on their own have caused it.
- Triggers: the immediate event that set the failure off, such as a load step or a cold start.
- Detection gaps: reasons the failure was not caught earlier.
Test each candidate with a simple question: “If this had been different, would this failure have been avoided?” If yes, it is the root cause or a strong candidate. If it would have been milder, it is contributing. If nothing changes, it is background.
Be wary of causes that stop at a person or a part. “A technician loosened a tube” is a finding, not a root cause. Ask what allowed it: an unlabelled tube, a missing procedure, a drawing that was not updated, a missing check at sign-off. That is where the corrective action will bite. Our guide on auditing an RECD service report for missing information shows how gaps in the service record often sit behind such causes.
It is acceptable, and often honest, to write: “Most probable root cause: X (confidence: medium). Not yet excluded: Y.” A report that names two live candidates and the test that would separate them is more useful than one that names a single cause with no basis.
A fill-in template
Copy this into your document system and complete it.
Title and reference: _____________________ Date: __________ Author: __________
1. Summary
- What happened:
- Impact:
- Most probable root cause (confidence):
- Main contributing factors:
- Key actions:
2. Scope and boundaries
- Set / RECD / location:
- Period covered:
- Out of scope:
3. Event description (rung 1 and 2 only)
4. Timeline
| Date and time | Event | Source (E#) | Rung |
|---|---|---|---|
5. Evidence register
| ID | Item | Source | Date | Collected by | Condition or caveat |
|---|---|---|---|---|---|
| E1 |
6. Analysis
- Method used:
- Hypotheses considered, with predictions:
- Evidence for and against each:
- Hypotheses rejected and why:
7. Findings
| Finding | Type (root cause / contributing / trigger / detection gap) | Rung | Evidence (E#) | Premises if inferred |
|---|---|---|---|---|
8. Confidence and open questions
| Open question | Why it matters | Test that would settle it | Owner | Date |
|---|---|---|---|---|
9. Corrective and preventive actions
| Action | Addresses (finding) | Owner | Due | How we will know it worked |
|---|---|---|---|---|
10. Appendices: (list)
Where RECD evidence usually lives
Experienced investigators gather evidence from sources like these. Use the list to check what is missing from your evidence register.
- Controller history: ΔP, temperatures, heater events, alarms and regeneration logs. Export before anything is reset. See why an alarm returns after a reset and RECD regeneration history report contents.
- Independent measurements: portable pressure and temperature readings, with instrument IDs and calibration dates. See separating a sensor fault from an RECD performance problem.
- Duty records: load logs, run hours and starts. See using DG load logs.
- Fuel and oil records: delivery notes, supplier, tank dips, analyses and oil consumption.
- Installation records: drawings, commissioning data, changes since. See updating drawings after exhaust modifications.
- Service records: what was done, by whom, with which parts. See what an RECD service report should contain.
- Physical evidence: the filter or core, deposits, gaskets and tubes, kept in labelled bags with photographs. Mark samples with the date and position.
- Environmental records: weather, ambient temperatures, flooding or construction nearby.
- Interviews: operators and technicians, taken early and separately, with their words written down as rung 1 statements (what they say they saw) and not as conclusions.
Language that keeps the report honest
Small wording choices keep the ladder visible.
| Instead of | Write |
|---|---|
| “The filter was blocked by poor fuel.” | “ΔP rose from 31 to 52 mbar (E3) after the delivery of 3 Oct (E5). The delivery’s fuel quality has not been tested (assumed). The timing is consistent with a fuel contribution.” |
| “The sensor is faulty.” | “The controller read 9 mbar above a portable gauge at the same taps (E7, E8). A sensing-tube sag was observed (photo 11).” |
| “Operator error.” | “The reset history was cleared on 5 Oct (E2). No procedure required it to be exported first (E12).” |
| “The core is cracked.” | “Soot was observed on the outlet face (photo 14). This is consistent with a breach. The core has not been opened.” |
| “It will not happen again.” | “Action 4 adds a pre-reset export step. We will check compliance at the next three events.” |
Common mistakes
- Writing the conclusion first and arranging the evidence to fit it.
- Naming a person or a part and stopping there.
- Mixing the summary and the evidence, so the reader cannot check anything.
- Hiding the rejected alternatives, which makes the analysis look thin.
- Presenting assumptions as facts because they were repeated in meetings.
- Listing actions without owners, dates or checks.
- Skipping the open questions, which signals false certainty.
- Leaving out the timeline, the single most useful tool for spotting a missing cause.
- Treating one failure as proof of a general rule. A single event shows what can happen and not how often.
- Never going back. Revisit the report after the corrective actions to see whether they worked.
A short worked example
The following is an illustrative scenario, not a real customer. A logistics warehouse reports that its 250 kVA set’s RECD showed a high-pressure alarm in October, was regenerated twice with no lasting effect and had its core replaced. The first draft of the report read: “The RECD core failed because of soot overload caused by poor-quality fuel. Replace fuel supplier.”
Rebuilding the report.
Summary (as rewritten). ΔP at 60% load rose from 31 to 52 mbar between 28 September and 10 October. Two regenerations gave only temporary relief. The core was replaced on 18 October. The most probable root cause is a sustained pattern of light-load running after a duty change in August, which the RECD was not specified for (confidence: medium). Contributing factors are a long idle-run policy and a possible fuel-quality effect, which has not been tested. The key action is to review duty, specification and operating policy.
Evidence and rungs.
- Rung 2: ΔP trend from the controller (E1), confirmed by a portable gauge on 10 October (E2).
- Rung 2: load logs from the building system show that from 1 August the set carried 8 to 15% of rating on most test runs (E3).
- Rung 1: a service report from September records “ran for 10 minutes on no load” as the standard monthly test (E4).
- Rung 4: “fuel from the same supplier as always” (verbal). No delivery notes or sample available.
- Rung 1: the removed core, photographed, shows heavy dry soot and no cracking (E6).
Analysis. Four hypotheses were compared: heavy soot from light-load duty (predicts dry soot, ΔP rising with run history, partial recovery after regeneration, matches E3 and E6); wet or contaminated fuel (predicts wetter deposits and a step after a delivery, but no delivery data exist, so it cannot be tested); a sensing fault (rejected, since the portable gauge agreed within 2 mbar); and a damaged core (rejected, since the pressure rose, the core showed no crack and the outlet face was clean).
Findings.
- Root cause (inferred, rung 3): duty had changed to extended light-load running, outside the conditions the RECD was specified for. Premises: E3, E4 and E6.
- Contributing: the monthly test procedure did not run the set at a load sufficient for regeneration (observed).
- Contributing, not excluded: fuel quality (assumed, untested).
- Detection gap: ΔP trend was not reviewed between August and October (observed: no review records).
Actions. Review the duty with the supplier, with an owner and a date. Change the monthly test to a loaded run with the load bank. Keep delivery notes and test a retained fuel sample at the next delivery. Add a monthly ΔP trend review with a floor and alarm line. Check at the next three months that ΔP at a reference load stays within range.
The rewritten report is longer than the first draft, but a reader can see what was known, what was inferred and what remains open. The first draft’s confident fuel conclusion, which would have changed the supplier and left the duty problem in place, has been replaced by a plan that addresses the likely cause and still tests the fuel.
After the report is written
- Review before release. Have someone who did not do the investigation read it and challenge every rung-3 inference.
- Share the lessons, not just the incident. Add the key points to your service plan, asset record and operator training.
- Trace the actions. A report with no follow-up is paperwork. Put actions into the same system as work orders and check them at an agreed date.
- Use it for claims with care. A well-evidenced report helps a warranty claim, but it should be accurate. See what to record in an RECD warranty claim.
- Feed the cost record. See tracking RECD service cost per operating hour.
Questions people ask
How long should a root-cause report be?
As long as the evidence needs. A simple event may need two pages and a complex one twenty. The test is whether a reader can check every claim.
What if the root cause cannot be determined?
Say so. Report the most probable cause, the alternatives, the open questions and the test that would settle them. An honest “not yet known” is better than a guessed answer.
Should the supplier write the report or should we?
Ideally both contribute. The supplier brings device knowledge, and the site brings duty, fuel, installation and operating facts. A joint report with the ladder makes it clear who is responsible for which statements.
Is 5 Whys enough?
For simple faults, yes. For anything with several contributing conditions, add a hypothesis table or a fault tree so that branches are not lost.
Do we need a root-cause report for every alarm?
No. Use a short event note for routine alarms and keep the full report for repeat failures, expensive failures, compliance-affecting events and any event where the cause is disputed.
Closing thought
A root-cause report is a tool for preventing the next failure, and it only works if the reader can trust what it says. Label every statement with the claims ladder, follow the ten-part anatomy, keep root cause and contributing factors apart, list the alternatives you rejected and be open about what remains unknown. If you would like help structuring an RECD investigation or reviewing a draft report, see our retrofit emission control devices, the FAQ or contact the Aceget team.
Related Reading
- What to Record in an RECD Event Log
- Auditing an RECD Service Report for Missing Information
- RECD Warranty Claim: What to Record
- RECD Fault Business Continuity Plan
- RECD Regeneration History Report: What It Should Contain