The Ishikawa diagram organizes hypotheses that may explain an effect. In maintenance it gives a multidisciplinary team a disciplined way to explore method, machine, material, manpower, measurement, and environment without treating the most popular story as a confirmed cause.
The CV-17 gearbox Ishikawa creates value only when every branch ends in a testable hypothesis. The team uses temperature, lubricant, alignment, load, assembly, and environment to decide which checks enter field execution and which assumptions still lack evidence.
When Ishikawa Diagram in Maintenance is the right tool
Use Ishikawa when the effect is bounded but competing mechanisms span Operations, lubrication, maintenance, and reliability. For CV-17, the workshop starts only after the thermal point, load band, internal limit, and event chronology are fixed, so every 6M branch addresses the same phenomenon.
The investigation fails when workshop labels become causes through voting. On CV-17, each proposal must predict a physical response, receive a test, and change state only through evidence. The number of ideas written on one branch adds no causal strength.
Investigation contract before opening the 6M branches
The CV-17 fishbone uses artifacts that differ from an action plan. The diagram holds possibilities, the test matrix defines how each possibility will be challenged, and the state register shows what the evidence allows the team to conclude.
- Effect statement: output-bearing temperature above the internal band at a defined point, during stable load above 80%, after the specified stabilization time.
- Family map: Method, Machine, Material, People, Measurement, and Environment receive relevant mechanisms without forced symmetry between branches.
- Hypothesis register: every row contains the proposed condition, expected mechanism, observable prediction, and evidence capable of refuting it.
- Test matrix: sequence, operating condition, instrument, owner, acceptance, interference risk, and reference to the field record.
- State vocabulary: confirmed when evidence matches the prediction, refuted when a valid test contradicts it, and open when the predefined exposure is still missing.
- Residual queue: overload remains at 42 of the required 100 hours above 80% load, owned by Reliability until the denominator is complete.
Missing temperature, oil, breather, alignment, or load data receives no estimated value. The matrix exposes the gap and stops state progression. This preserves the difference between an untested possibility and an explanation contradicted by valid evidence.
From the effect statement to the residual queue
| Move | Investigation artifact | Transition rule | Operating record |
|---|---|---|---|
| Fix the effect | Sheet containing variable, point, threshold, load, ambient condition, stabilization, and timeline for all eight alarms. | No hypothesis enters the fishbone while participants are describing different effects. | The original notification is cited as the event origin without assigning a premature cause. |
| Build the families | 6M map with mechanisms and the source behind each contribution. | Labels such as operator error or bad oil return to the group until they become observable conditions. | Prior orders, material records, and measurement points inform a branch without deciding it. |
| Prioritize hypotheses | Matrix considering chronology, physical plausibility, explanatory reach, safety, and refutability. | The next test is selected for the information it can add, not for the popularity of the idea. | Physical tests receive an operation, condition, instrument, and criterion that protect sequence. |
| Update states | Register containing result, evidence reference, and confirmed, refuted, or open state. | A photograph without point and load may support context but cannot change hypothesis state. | Oil level, breather, sensor, and alignment findings cite their corresponding operation or measurement. |
| Close the map | Final matrix with two confirmed hypotheses, three refuted hypotheses, and one residual open item. | Only supported mechanisms produce countermeasures; overload uncertainty remains visible. | The technical note links corrections, absence of alignment work, and the 100-hour load trigger. |
The Ishikawa output is reconstructable: versioned effect, families, hypotheses, tests, final states, and residual queue. Asset work appears only when the matrix requires physical evidence or implements a supported countermeasure.
Complete industrial case: intermittent overheating in conveyor CV-17 gearbox
CV-17 is a technical simulation built to expose Ishikawa decisions. Its alarms, pressure drops, temperatures, load hours, and hypothesis states do not describe an operating installation. Another site must derive measurement points, internal limits, predictions, and refutation criteria from the engineering authority for its gearbox.
Conveyor CV-17 transfers granular product between two process stages. The gearbox produced eight temperature alarms and three stops in 60 days. Load, lubrication, alignment, and ambient-temperature records were inconsistent.
Case baseline
| Data | Didactic value | Interpretation |
|---|---|---|
| Temperature alarms in 60 days | 8 | Events occurred across two shifts |
| Associated stops | 3 | Didactic total of 31 downtime hours |
| Orders with measured oil volume | 2 of 8 | Other records said only that level was checked |
| Maximum load difference between shifts | 22% | Didactic operating-history value |
Case target: test the hypotheses that could explain overheating, control the confirmed mechanism, and run for 60 comparable days with no alarm attributed to the same cause.
The closure rule crosses thermal response by load band with the documented state of every hypothesis. Correcting oil level and breather does not erase the overload question. If D+60 does not supply the planned 100 hours above 80% load, that row remains open even though the other tests are complete.
Decision produced by the method
The team confirmed unmeasured top-ups and an obstructed breather as conditions supporting the overheating mechanism. Alignment, sensor error, and ambient temperature were refuted by their specified checks. The decision corrected the breather, drained and refilled a measured oil volume, revised the task list, and inspected equivalent gearboxes. The overload hypothesis remained open until sufficient exposure could be accumulated.
| Hypothesis | Final state | Didactic evidence and decision |
|---|---|---|
| Overfill after unmeasured top-ups | Confirmed | Drained volume was 12% above the internal target; correct the level and require volume records. |
| Obstructed breather | Confirmed | Inspection found a saturated element and 6.2 kPa pressure drop against a 3.0 kPa internal limit; replace it and inspect equivalent assets. |
| Misalignment | Refuted | Two laser measurements met the internal criterion before oil and breather correction; no alignment action entered the order. |
| Sensor error | Refuted | Comparison with a calibrated reference showed a 0.6 °C difference against ±1.5 °C acceptance. |
| Ambient temperature | Refuted | Alarms occurred from 21 °C to 29 °C and remained elevated after normalization for ambient condition. |
| Sustained overload | Open | Only 42 h above 80% load were available, below the 100 h defined to close the correlation. |
The final matrix contains two confirmed hypotheses, three refuted hypotheses, and one open item. Oil level and breather generate countermeasures; misalignment produces no work and leaves the queue; overload remains in the residual queue until 100 hours above 80% load are accumulated, with the existing 42 hours preserved in its denominator.
Indicator and verification window
Primary indicator: temperature alarms per 100 hours above 80% load, supported by compliance with recorded oil volume.
Case calculation: Alarm rate = valid alarms ÷ hours at comparable load × 100.
Window: D+3 for implementation, D+15 for equivalent assets, and D+60 for effectiveness at representative load.
The completed order records oil-level and breather correction, not automatic closure of the investigation. The state matrix still requires thermal response under representative load and keeps overload in the residual queue with 42 valid hours. Its state changes only after the series reaches the exposure rule and the correlation can be judged.
Limits and neighboring methods
Limits: Ishikawa expands the investigation field but does not demonstrate causality. The 6M categories are prompts, not boxes that must be filled. Complex events may require FTA, barrier analysis, or specialist tests.
Neighboring methods: Five Whys deepens a short chain, FTA represents logical combinations, and FMEA anticipates failure modes. Ishikawa owns multidisciplinary hypothesis organization.
Method selection follows the question raised by the CV-17 gearbox. Five Whys deepens one chain, FTA tests combinations, and RCA governs a formal investigation. Ishikawa completes its job when every relevant branch is confirmed, eliminated, or retained as an open hypothesis; physical action proceeds only from a test that supports the decision.
From a precise effect to falsifiable branches
The CV-17 effect is temperature above the internal range during stable load, at a defined measurement point. Writing only “hot gearbox” would let each participant discuss a different phenomenon. The record therefore includes load band, ambient condition, measurement location, and the intervention timeline.
Method examines filling and level-reading practice. Machine covers the breather and heat dissipation. Material addresses lubricant and contamination. People examines the assembly step without turning a person into a cause. Measurement challenges instrument and position. Environment covers dust, ventilation, and ambient temperature.
Every branch becomes a causal statement with an expected physical mechanism and evidence able to disprove it. “Lack of training” is rejected until the team identifies the affected step, the resulting condition, and the record that would reveal the difference.
The first test order checks oil level and breather condition. Alignment receives a separate measured operation; overload is tested through correlation between load and temperature. Results close, retain, or reformulate branches. The team avoids implementing four corrections at once because that would destroy causal attribution.
Traceability for the CV-17 tests
The versioned fishbone remains a technical attachment, while every test-matrix row points to the operation or measurement point that produced its result. The inspection order preserves the condition used to observe oil level, breather, and alignment; the load and temperature series supports the residual queue without blending it into completed tests.
PM Run Mobility can return measurements, notes, and images to the SAP PM workflow, including after offline execution, while PM Run Planning supports capacity for approved tests. The platform does not measure temperature, draw the fishbone, or classify hypotheses; instruments, interpretation, and final judgment remain with the technical team.
The diagram closes when relevant branches are tested, justified as pending, or eliminated. The accepted mechanism must explain the effect and survive comparable load. A photograph without position and context is supporting material, not causal proof.
- Is the effect observable and bounded?
- Does each branch contain a test rather than an opinion?
- Are open hypotheses still visible?
- Did the countermeasure receive an order and acceptance rule?
Investigation quality after the workshop
The facilitator rejects labels that cannot be tested. One causal chain may continue through Five Whys; a higher-consequence event may require the governance of a full root cause analysis.
Approved tests receive separate maintenance orders when their conditions interfere. Correcting the breather before the planned alignment measurement could erase evidence, so sequence protects learning.
Every result updates the branch explicitly: confirmed, rejected, reformulated, or pending. An attachment without measurement context cannot close a hypothesis.
Effectiveness is assessed after the countermeasure under the same load band. Lower temperature supports the result, while causal acceptance also depends on the predicted mechanism and repeatable evidence.
A new alarm pattern starts a new effect statement rather than being forced into the old fishbone. This boundary keeps the diagram useful and prevents a permanent catalogue of every possible gearbox problem.
Additional control questions
Technical review reads the fishbone backward. It starts from the accepted conclusion and asks which test supports it, which order executed that test, and which competing branch was eliminated. A broken chain returns the cause to hypothesis status.
A branch without an immediate test receives a reopening condition rather than a generic action. Overload stays open until a stable period includes recorded load. Uncertainty remains visible without blocking the correction supported by oil level and breather evidence.
Facilitation quality is also audited. Participation across Operations, maintenance, reliability, and lubrication matters only when contributions become precise mechanisms. A longer list is not a stronger investigation.
The final branch map is versioned with date, participants, effect statement, and evidence references. If the temperature pattern changes, the team creates a new effect definition rather than stretching the old explanation.
The test plan records expected observations before execution. If a blocked breather is causal, pressure behavior and temperature response should move in a defined direction. Writing the prediction in advance reduces hindsight bias. A result that contradicts prediction changes the branch, even when the original idea came from a senior participant.
Sampling conditions are preserved with each measurement: load, ambient temperature, instrument, point, and stabilization time. This allows later engineers to compare evidence without recreating context from memory. The branch conclusion cites those records rather than a meeting slide.
The branch register distinguishes absence of evidence from evidence of absence. A hypothesis with no scheduled test remains open; a hypothesis contradicted by a valid test is rejected. This language prevents pending work from being presented as causal elimination. The facilitator also checks interactions between branches: an incorrect oil level may alter temperature response to load, while ventilation can amplify but not originate internal friction. When interaction matters, the test sequence isolates conditions or records them explicitly. Countermeasure review asks whether the accepted mechanism predicts the observed response, whether competing explanations remain plausible, and whether the same effect returns after comparable service. The final technical note references measurement records and order operations, allowing another engineer to audit the logic without reopening the workshop discussion.
A branch review sheet names the proposed condition, predicted effect, test, acceptance, result, and next state. Facilitators use it to prevent a plausible statement from surviving simply because nobody challenged it. When oil level and breather are corrected, the team records which response should occur first and how long temperature needs to stabilize. The alignment hypothesis was refuted by two laser measurements and removed from further testing. Only the overload hypothesis remains open until the defined exposure is reached. The conclusion distinguishes contributing condition from primary mechanism; more than one branch may matter without every branch becoming a root cause. Engineering also checks whether the countermeasure controls the accepted mechanism and whether a later recurrence follows the same effect definition. A different thermal pattern begins a new investigation instead of being forced into the former diagram.
Technical sources
- ASQ, Fishbone Diagram, consulted for organizing potential causes into testable branches.
- SAP Help Portal, Maintenance Order, used to define the operating record for hypothesis tests.
