A conveyor motor trips on overload. Maintenance resets the system, and production resumes. Two days later the same alarm returns. The motor is replaced. For a short period everything appears normal, and then the fault happens again.
The replacement has solved a visible symptom only temporarily. It has not demonstrated why the machine developed excessive load.
Repeated failures are especially expensive because a factory pays for the same disturbance multiple times: lost production, repeated troubleshooting, spare parts, planning changes and reduced trust in the equipment.
The central engineering question is not simply "Which component stopped?" It is "Which physical and organizational conditions made the failure possible, and how can we prove that those conditions have been corrected?"
This article develops a practical Root Cause Analysis (RCA) method using a reconstructed conveyor-drive failure. All case-specific events, measurements and final findings below are illustrative, not records from an identifiable factory.
1. Symptom, failure mechanism and root cause are not the same
A machine alarm describes an observed condition. The mechanism explains how that condition developed. The root cause investigation goes further and asks what underlying conditions allowed the mechanism to occur and recur.
| Level | Illustrative conveyor example | What it tells us |
|---|---|---|
| Visible symptom | Production stops with an overload event | What the machine reported |
| Immediate mechanism | Motor load and current increase | How the protective event became possible |
| Physical contributing condition | Conveyor drag increases at one guide position | Why the driven load grew |
| System-level cause | Changeover setup has an ambiguous guide-location reference | Why an unsuitable guide position can recur |
| Weak detection control | Changeover acceptance does not evaluate the relevant mechanical margin | Why the condition escaped before production |
A chain can have more than one contributing branch. A weak guide design, changing product loading and inadequate setup verification may all be relevant.
A root cause is not necessarily a single failed part or a single person who touched the machine.
2. What makes a fault a recurring reliability problem?
A machine can fail repeatedly because the original defect was never removed, because a component is exposed to the same damaging conditions, or because different failures produce the same general alarm.
These possibilities must be distinguished.
A repeated motor-overload indication does not automatically prove that every occurrence had the same cause. A drive may report the same alarm for different load conditions.
Likewise, two separate faults can appear identical on the HMI if the control software groups many conditions under one message.
Useful records therefore include:
- Equipment identity and relevant operating state.
- Exact alarm or diagnostic code, including additional status information when available.
- The earliest recorded abnormal event.
- Product type, load and production sequence.
- Changes made after previous occurrences.
- Evidence supporting or contradicting each attempted correction.
A useful reliability record links the alarm to the specific operating circumstances, not merely to the date of the latest shutdown.
3. Start with a measurable problem statement
A vague statement such as "the conveyor keeps failing" cannot support an effective investigation.
A more useful illustrative problem statement is:
"During the last three production weeks, conveyor CV-02 experienced four overload stops while transporting the wide product format. No corresponding stops were recorded in the same period for the narrow format. Two previous component replacements did not prevent recurrence."
The counts here are hypothetical. They demonstrate the kind of specificity required, not a historical incident.
A good problem statement describes what happened, where, when and under what conditions. It should not already contain an assumed explanation such as "because the operator positioned the guide incorrectly."
The problem statement defines what evidence the investigation must explain.
4. Separate containment from permanent corrective action
Production often needs a temporary response before the complete cause has been established. This is containment.
Containment limits the immediate consequence and should be approved under the plant's safety, quality and production processes. It is not proof that the underlying defect has been eliminated.
| Type of action | Primary purpose | What it does not prove |
|---|---|---|
| Containment | Limit immediate consequences while investigation continues | Root cause eliminated |
| Correction | Restore a specific item or process condition | The same defect cannot recur |
| Root cause corrective action | Remove or control a verified causal condition | Similar weaknesses elsewhere have been addressed |
| Preventive system improvement | Reduce exposure to related failure mechanisms | Every possible fault has been eliminated |
| Effectiveness check | Confirm the intended reduction in failure under suitable exposure | Future reliability under all conditions |
A motor replacement might be a valid correction if the motor is damaged. However, replacing it is not a permanent corrective action for a recurring mechanical overload unless the motor itself is the verified initiating cause.
5. The illustrative event history
Consider the following reconstructed record from an industrial conveyor driven by a variable-frequency drive.
| Event | Production condition | Observation |
|---|---|---|
| First stop | Wide cartons | Motor overload indication |
| Initial response | Machine returned to service | Event disappears temporarily |
| Second stop | Wide cartons | Same overload indication |
| Component replacement | Planned maintenance | Motor replaced, but no load cause established |
| Third stop | Wide cartons | Overload indication returns |
| Engineering review | Changeover and process records | Higher load coincides with a particular guide setup |
This history does not by itself establish a mechanical guide problem. It does show that replacing the motor was not sufficient to prevent the observed recurrence.
The investigation now needs to compare alternative explanations against independent evidence.
6. Build hypotheses before using Five Whys
Jumping directly from "motor overload" to a five-step narrative can make the result sound convincing even when the first assumption is wrong.
A better approach first identifies competing hypotheses.
| Hypothesis | Evidence that could support it | Evidence that would weaken it |
|---|---|---|
| Motor electrical defect | Specific electrical diagnostic evidence or measured condition outside approved limits | Same load-related pattern returns after a documented compatible replacement |
| Drive configuration mismatch | Relevant project/change history and a corresponding diagnostic inconsistency | Configuration is documented unchanged and appropriate |
| Mechanical resistance | Load trend correlates with a repeatable physical condition | No correlation between the condition and available load evidence |
| Product mass or process demand | Fault follows a specific product and operating profile | Same demand remains within validated design conditions |
| Supply or power-quality event | Matching electrical supply events | No supporting supply disturbance in available records |
| Control sequencing issue | Command chronology produces unexpected acceleration, overlap or process loading | Control sequence matches the documented process |
Every row is a testable explanation, not a conclusion. The actual evidence should drive which branch receives attention.
7. Distinguish an overload indication from its initiating cause
A protective device or drive can report that an electrical or thermal limit has been reached. That does not automatically identify the physical cause of the load.
In the reconstructed case, event records show the commanded conveyor operation continues, while the motor's recorded current rises during a particular product-transfer condition before the overload event.
For illustration only, the drive trend may show:
| Operating condition | Illustrative motor current |
|---|---|
| Normal transfer with narrow format | 5.4 A |
| Wide-format transfer before the fault | 7.9 A |
| Repeated high-load interval | 9.1 A |
These numbers are not motor ratings, trip settings or acceptable operating limits. Their purpose is to illustrate a trend of increased load. Actual limits depend on motor nameplate data, drive configuration and manufacturer specifications.
The observed current increase supports a load-related hypothesis but does not on its own prove whether the cause is mechanical friction, process mass, electrical supply conditions or another factor.
8. Apply Five Whys only where evidence supports each answer
The Five Whys technique asks successive questions to move from a symptom toward underlying causes. The number five is a guide, not a requirement to stop after exactly five answers.
A well-constructed chain for this reconstructed case could be:
| Question | Evidence-based answer in the scenario |
|---|---|
| Why did the conveyor stop? | The drive recorded an overload event and stopped the commanded motion |
| Why did the overload condition develop? | The recorded motor load increased during one repeatable product-transfer condition |
| Why did the mechanical load increase? | The conveyor guide contacted the moving product/belt path under that format's configuration |
| Why could the guide reach that unsuitable position? | The changeover arrangement lacked a sufficiently repeatable positive locating reference |
| Why was the unsuitable arrangement accepted? | The documented setup check covered product size but did not verify the mechanical clearance and loading margin |
A weak Five Whys analysis would fill each cell with a guess and then declare the final guess a root cause.
A stronger analysis documents which observation, drawing, event history or approved engineering finding supports each answer.
9. When Five Whys becomes too simplistic
Industrial machines rarely have only one possible causal path.
A mechanical guide could be positioned correctly but move under vibration. A worn bearing could cause drag independently. A process change could create an unexpected load even when the guide is stable.
For these situations, a cause-and-effect (Ishikawa) diagram can organize hypotheses by categories such as machine, method, material, measurement, environment and people.
The categories should be used to broaden the search, not to assign blame.
| Causal branch | Example question |
|---|---|
| Machine | Is the physical mechanism behaving within its documented design? |
| Method | Does the changeover method establish a repeatable safe operating arrangement? |
| Material | Has the product format or load changed? |
| Measurement | Can the available records distinguish demand, resistance and protection events? |
| Environment | Are temperature, contamination or vibration contributing factors? |
| Human-system interface | Does the setup procedure provide clear references and acceptance criteria? |
The investigation can then apply deeper questioning to each credible branch. A single linear story should never replace contradictory evidence.
10. Do not end the analysis with "operator error"
"Operator set the guide incorrectly" is an incomplete root cause statement.
A robust engineering investigation asks why the setup permitted that condition, whether the expected position was clearly defined, whether the machine's locating features supported repeatability, and how the condition could be detected before it affected production.
This does not mean actions are never relevant. It means that blaming the last person involved does not create a dependable reliability control.
The useful question is: What design, process or verification condition can be improved so that the same error is less likely to cause a repeat failure?
That leads to changes with a clearer engineering basis than an instruction to "be more careful."
11. Verify the suspected cause using independent evidence
A proposed root cause remains a hypothesis until it is sufficiently supported.
In the reconstructed case, a credible verification package would relate several findings:
- Product-changeover records show that events follow one configuration.
- Existing motor-load trends correlate with the problem condition.
- Approved mechanical inspection documents the guide's actual position and interference.
- The original design and changeover documentation lack an unambiguous locating reference.
- Subsequent authorized acceptance evidence shows the issue is no longer present under representative conditions.
Not all observations have the same evidential strength. A single successful restart is much weaker than a repeatable relationship across the relevant operating conditions.
The final report must distinguish direct observations, engineering interpretation and conclusions supported by the combined evidence.
12. The difference between physical and system-level causes
Even where a mechanical interference condition is verified, the engineering work is not necessarily complete.
There may be a direct physical cause: excessive drag from contact at the guide. There may also be a system-level cause: a setup method and locating arrangement that allow the same contact to return after changeover.
Both are relevant because they answer different questions:
What physically made the machine stop? Increased mechanical load led to the protective event.
What allowed the condition to recur? The production changeover did not reliably establish or verify the required guide position.
Only addressing the first may restore operation once. Addressing the system-level condition provides a stronger basis for preventing recurrence.
13. Define corrective action around the verified mechanism
In the reconstructed scenario, a permanent solution would be evaluated against the machine's design constraints and approved engineering process.
The solution might involve a more repeatable guide-location feature and a clearly documented setup acceptance requirement for the relevant product formats.
This is an engineering design objective, not a hands-on adjustment procedure. Any actual modification must be designed, risk-assessed, implemented and validated by authorized personnel.
The corrective-action record should establish:
| Required element | What should be documented |
|---|---|
| Verified causal condition | How the cause was supported by evidence |
| Proposed engineering control | Which failure mechanism the change addresses |
| Owner and implementation plan | Who is accountable for approved work |
| Design and safety review | How relevant machine requirements are preserved |
| Acceptance evidence | Whether the corrected arrangement meets its documented function |
| Effectiveness period | When recurrence and related indicators will be reviewed |
A corrective action is stronger when it directly interrupts the verified causal chain.
14. Reliability improvement requires an effectiveness check
A repair can appear successful because no failure has occurred yet.
Suppose the original fault appeared only when the wide product format was used. Running the narrow format for three weeks without an overload does not establish that the correction worked.
A valid effectiveness check must consider exposure to the conditions that previously caused the fault.
Useful outcomes may include:
- Number of relevant production cycles with the affected format.
- Frequency of the original fault per comparable operating exposure.
- Trends in motor load or other appropriate condition indicators.
- Completion and validity of approved changeover acceptance checks.
- Any new problems introduced by the corrective work.
Zero failures over a very small exposure period is weaker evidence than zero failures over a sufficiently representative and documented period.
15. A simple recurrence-rate illustration
Consider hypothetical pre- and post-correction operating records:
| Period | Relevant wide-format cycles | Matching overload stops | Stops per 10,000 cycles |
|---|---|---|---|
| Before corrective action | 20,000 | 4 | 2.0 |
| After corrective action | 30,000 | 0 | 0.0 observed |
These values are illustrative, not real factory data.
The observed rate has improved, but a zero count does not mathematically prove that future failure probability is zero. The result also assumes consistent fault definitions, comparable operating conditions and trustworthy cycle records.
A reliability review should describe the observed improvement and the remaining uncertainty rather than claiming perfect reliability.
16. How 8D differs from a quick Five Whys discussion
The Eight Disciplines (8D) approach provides a more complete structure for problem solving.
It typically includes establishing a team, defining the problem, developing interim containment, verifying root causes, selecting corrective actions, implementing them, preventing recurrence and recognizing the team's work.
Five Whys is a questioning tool, not an entire corrective-action management system. It can support 8D or another suitable investigation framework.
For a high-consequence or repeated factory failure, a team often needs more than one worksheet: reliable evidence, technical verification, design responsibility, documented decisions and follow-up results.
The American Society for Quality emphasizes the verification of root causes and the effectiveness of permanent corrections within a structured process.
17. Root Cause Analysis and FMEA are related but different
Root Cause Analysis investigates why a known event occurred.
Failure Mode and Effects Analysis (FMEA) is commonly used to evaluate potential failure modes and their consequences before or during design and process planning.
An RCA finding can inform an FMEA review. For example, a verified guide-position vulnerability might justify revisiting the relevant setup or mechanical-interface failure mode in the equipment's risk and quality documentation.
But an FMEA ranking cannot by itself prove what caused a past event. Likewise, a single historical root cause does not describe every future way a machine could fail.
These methods complement one another when used with clear purposes.
18. The root cause report should survive a change of personnel
A useful investigation record should allow another engineer to understand the conclusion without relying on the investigator's memory.
| Report section | Essential content |
|---|---|
| Problem definition | Machine, symptom, scope and relevant operating exposure |
| Chronology | Ordered fault events, earlier repairs and production changes |
| Available evidence | Controller history, condition data, drawings and approved inspection records |
| Considered hypotheses | What was supported, contradicted or left unresolved |
| Causal explanation | Immediate failure mechanism and underlying conditions |
| Corrective action | Engineering response connected to the verified mechanism |
| Acceptance | Evidence that approved work achieved the intended function |
| Effectiveness review | Relevant exposure, recurrence results and limitations |
| Lessons learned | Where similar vulnerabilities may exist |
The report should not erase uncertainty or turn a plausible engineering story into a claimed fact.
19. Common Root Cause Analysis mistakes
- Stopping at the alarm: An overload report is not an explanation of the physical load.
- Replacing the most visible component: A new motor cannot remove an unrelated mechanical interference condition.
- Beginning Five Whys with a guess: A logical sequence of assumptions is still not evidence.
- Demanding exactly five answers: A causal explanation may need fewer or more questions and multiple branches.
- Confusing correction with prevention: Restarting the machine does not prove recurrence has been prevented.
- Blaming an operator instead of examining the system: Ambiguous setup references may make a repeated error predictable.
- Ignoring competing hypotheses: An attractive explanation may conceal a different or additional cause.
- Checking effectiveness without relevant exposure: A long period using the wrong product format proves little.
- Declaring zero risk from zero observed events: A finite test period cannot guarantee that all future failures are eliminated.
Technical conclusion
Recurring machine failures require a different mindset from one-off fault restoration.
The alarm describes what the machine observed. The immediate mechanism explains how the stop occurred. The root cause investigation explains why the conditions that produced the failure existed and why they could recur.
In this reconstructed conveyor example, a motor overload was the visible event; increased mechanical resistance was the immediate mechanism; and a changeover arrangement without a sufficiently repeatable position reference was the underlying vulnerability.
An effective RCA combines a clear problem statement, trustworthy chronology, competing hypotheses, evidence-based Five Whys or cause-and-effect analysis, and a measurable effectiveness check.
The investigation is complete only when the causal explanation and corrective action have been validated under the conditions relevant to the original failure.
For additional reading, see Industrial Preventive Maintenance: How to Build a Checklist That Actually Prevents Failures, Intermittent Conveyor Stops After Changeover: A Reconstructed Sensor Fault Case Study and Why Does a Machine Stop Without an Alarm?.
Authoritative references
- American Society for Quality — Root Cause Analysis: principles and methods of investigating underlying causes.
- American Society for Quality — Five Whys and Five Hows: repeated questioning as a method for developing a causal explanation.
- American Society for Quality — Eight Disciplines (8D): verification of root causes, permanent corrective actions and prevention of recurrence.
- American Society for Quality — Problem Solving: combining problem definition, evidence and evaluation of alternative solutions.
Cover photograph: industrial electric motor and mechanical drive mechanism by LJamesCPH, Wikimedia Commons. Licensed under Creative Commons Attribution–ShareAlike 4.0. The photograph was resized, repositioned on a blurred background and converted to WebP; the adapted image is shared under the same license. It is an illustrative photograph, not the machine from the reconstructed case.



