4. Root cause analysis techniques for crash test anomalies

When a crash test produces an anomalous result — a dummy reading that exceeds the injury assessment reference value (IARV), a sensor that drops out mid-event, a barrier that fails to break away, or a post-test deformation pattern that contradicts the CAE prediction — the anomaly report must move beyond description into causation. Three techniques form the analytical backbone of this transition: the fishbone (Ishikawa) diagram for breadth-first cause enumeration, the 5 Whys for depth-first causal chaining, and fault tree analysis (FTA) for quantitative, logic-based top-down decomposition. Each answers a different question, and mature test organizations use them in combination rather than choosing one.

Fishbone diagram: structuring the search space

The fishbone diagram organizes potential causes into categories so that the investigation team does not fixate on the first plausible explanation. In crash testing, the conventional 6M categories map naturally onto the test system: Man (technician setup, dummy seating procedure, tightening torque application), Machine (sled or barrier hardware, tow system, onboard data acquisition), Material (dummy components, foam, restraint hardware lot variation), Method (test procedure, sensor calibration chain, filtering per SAE J211), Measurement (transducer range, sampling rate, channel zeroing), and Environment (ambient temperature and humidity affecting foam and pyrotechnic behavior).

The practical discipline is to populate the bones with specific, testable candidate causes rather than generic labels. "Measurement" should carry entries such as "chest potentiometer zero offset exceeded ±0.5 mm at pre-test check" or "filter class CFC 180 applied where CFC 600 was specified," not simply "sensor problem." A useful convention is to mark each candidate with an evidence status — confirmed, plausible but untested, or eliminated — so the diagram becomes a living record rather than a one-off brainstorm. The fishbone's weakness is that it does not express interactions between causes; a chest deflection anomaly caused jointly by a mis-seated dummy and a low ambient temperature will appear on two separate bones with no connecting logic. That gap is what FTA addresses.

5 Whys: driving to the actionable root

The 5 Whys is a linear causal chain: state the anomaly, ask why it occurred, then ask why that occurred, and continue until the answer identifies a cause that can be acted upon. The "five" is a heuristic, not a rule; chains may terminate in three or extend to seven.

Consider a worked example. A 64 km/h offset deformable barrier (ODB) test returns a driver thorax deflection of 41 mm against a 35 mm IARV, while the CAE model predicted 31 mm.

1. Why did thorax deflection exceed the IARV? The restraint system allowed more forward torso excursion than predicted.
2. Why was excursion greater? The seatbelt load limiter reached its force plateau earlier and at a lower level than the build sheet specified.
3. Why was the plateau lower? The torsion-bar load limiter installed was a different part number from the one released for this build.
4. Why was the wrong part installed? Two load limiter variants share the same external housing and are distinguished only by a laser-etched suffix on the retractor flange.
5. Why was the variant not verified? The pre-test build verification checklist records the retractor part number but not the load limiter variant suffix.

The chain terminates at a process control gap — the checklist field — which is actionable. Note the discipline required: at step 3, a weaker investigator might stop at "supplier nonconformity" and issue a corrective action to the supplier, leaving the internal verification gap intact. The 5 Whys is only as good as the honesty of each link, and each "why" should be supported by physical evidence (part markings, build records, torque logs) rather than inference. Where evidence is absent, the chain should branch rather than proceed on assumption.

Fault tree analysis: logic, combinations, and probability

FTA starts from a defined top event — for example, "driver thorax deflection exceeds 35 mm" — and decomposes it through Boolean gates into basic events. AND gates represent combinations that must co-occur; OR gates represent alternatives. The structure exposes exactly what the fishbone cannot: interaction. A minimal cut set analysis then identifies the smallest combinations of basic events sufficient to produce the top event, which is where corrective actions have the greatest leverage.

For the thorax case, a simplified tree might place an OR gate beneath the top event with two branches: "excessive torso excursion" and "measurement over-reporting." The excursion branch could carry an AND gate combining "load limiter force below specification" with "pretensioner stroke below specification" — either alone might be tolerable, but together they exceed the threshold. The measurement branch would carry basic events such as "potentiometer zero offset" and "incorrect filter class." Assigning probabilities to basic events from historical test data, supplier records, and calibration histories allows the tree to be evaluated for the top-event probability, and more usefully, to rank cut sets by contribution.

The tree above illustrates a key insight: the AND gate means neither restraint defect alone would have produced the anomaly, so a corrective action addressing only the load limiter would leave the system vulnerable to the same combination recurring. The minimal cut sets also show that the measurement branch, though it looks like a secondary concern, carries a higher individual probability than the restraint combination — a finding that should redirect pre-test verification effort toward the calibration chain.

Choosing and combining the techniques

Use the fishbone early, when the cause space is still wide and the team risks anchoring on a single hypothesis. Use the 5 Whys once a specific causal path has physical evidence behind it, to drive from symptom to the process control that failed. Use FTA when the anomaly involves plausible interactions between causes, when the same top event has recurred across tests, or when the retest decision needs a defensible probability argument. In practice the three chain together: fishbone to enumerate, 5 Whys to deepen the most credible bone, FTA to formalize the logic and test whether the identified cause is sufficient or merely contributory.

Deliverable: anomaly causal analysis record

The following record structure captures the output of all three techniques in a form that supports the retest decision. It is filled with the worked example above; values marked as illustrative are teaching assumptions, not measured results.

Two disciplines make this record trustworthy. First, every "confirmed" entry must cite a physical artifact — a part marking, a calibration certificate, a DAQ configuration export — not a recollection. Second, every "eliminated" entry must state the check performed and its result, because the retest decision rests as much on what was excluded as on what was found. When the record is complete, the retest decision becomes a matter of reading the cut sets: if the confirmed cause accounts for the anomaly and the correction is verifiable before the next event, a retest is justified; if the confirmed cause is only one basic event in an AND combination whose partner remains unverified, the retest should be deferred until that partner is resolved.