📚 Course ContentsChapter 4 Risk prioritization with RPN and Action Priority

4. Risk prioritization with RPN and Action Priority

Once a DFMEA team has enumerated failure modes for LS-DYNA load cases and agreed on their effects, the analysis reaches its decisive step: converting qualitative engineering judgment into a defensible ranking that tells the team where to spend limited simulation and design resources. Two complementary mechanisms serve this purpose — the traditional Risk Priority Number (RPN) and the newer Action Priority (AP) scheme introduced in AIAG-VDA. This chapter explains how to score Severity, Occurrence, and Detection specifically for explicit-solver load cases, how to compute and interpret both metrics, and how to translate the resulting ranking into concrete mitigation actions.

Scoring the three factors in an LS-DYNA context

Severity (S) describes the consequence if the failure mode reaches the field, independent of how likely or how detectable it is. For crash and impact load cases, severity should be anchored to the effect on the occupant, the structure, or regulatory compliance — not to the simulation itself. A practical 1–10 scale for this course uses: 10 for potential occupant fatality or violation of a homologation requirement; 9 for structural separation of a safety-critical joint; 7–8 for loss of load path causing intrusion beyond target; 4–6 for degraded performance without safety impact (e.g., excessive plastic deformation reducing durability margin); 2–3 for cosmetic or NVH-only effects; 1 for no discernible effect. The key discipline is that severity is a property of the effect, so it must not be lowered because a simulation is cheap to rerun.

Occurrence (O) estimates how frequently the cause will produce the failure mode across the design population and load-case envelope. In DFMEA-driven simulation, occurrence is best grounded in evidence rather than intuition. Useful anchors include: prior program history with the same joint concept, coupon or component test data, mesh-sensitivity studies, and the maturity of the CAE correlation. A team might assign O = 8 when a spot-weld model has never been correlated against physical peel tests and the same nugget layout failed on a predecessor platform; O = 4 when correlation exists but the current gauge change has not been validated; O = 2 when the design is carry-over with validated test correlation. Documenting the evidence behind each occurrence score is what makes the DFMEA auditable.

Detection (D) reflects the ability of the planned verification — here, the LS-DYNA load case and its post-processing — to catch the failure mode before release. Detection scoring is where simulation engineers most often fool themselves. A load case that runs but whose output is not interrogated for the specific failure mode provides weak detection. Anchors: D = 9–10 when no load case exercises the mode or the relevant output (e.g., no strain output requested at a critical weld); D = 6–7 when a load case exists but boundary conditions or material cards are unvalidated, so a pass is not trustworthy; D = 3–4 when the load case is correlated and the specific output (e.g., plastic strain at the heat-affected zone) is explicitly reviewed against a criterion; D = 1–2 when an automated check with a validated threshold gates the release.

Score Detection against the verification method as actually planned, not against an idealized test. If the load case will not output the quantity that reveals the failure, the detection score must reflect that gap — this is precisely the lever simulation engineers control.

From RPN to Action Priority

The RPN is the product S × O × D, ranging from 1 to 1000. It is simple and familiar, but it has well-known weaknesses: different factor combinations can yield identical products (S=9, O=2, D=5 and S=3, O=6, D=5 both give 90), and it treats a high-severity, low-occurrence event as equivalent to a low-severity, high-occurrence one. AIAG-VDA addressed this with Action Priority, a lookup logic that gives Severity a dominant role, then Occurrence, then Detection. The table below is a condensed AP decision aid for the S/O/D ranges typical of structural load cases.

Severity	Occurrence	Detection	AP	Required response
9–10	≥ 4	any	High	Action mandatory; escalate to design review
9–10	2–3	≥ 5	High	Improve detection or reduce occurrence
9–10	2–3	≤ 4	Medium	Action recommended; document rationale if deferred
7–8	≥ 5	any	High	Action mandatory
7–8	3–4	≥ 5	Medium	Action recommended
4–6	≥ 6	≥ 6	Medium	Action recommended
≤ 3	any	any	Low	Monitor; no action required

In practice, teams compute RPN for continuity with legacy DFMEAs but drive the action plan from AP. A useful convention: any item with AP = High must have an owner, a due date, and a defined verification load case before the DFMEA is signed off; AP = Medium items are reviewed at each design gate; AP = Low items are logged and monitored.

Defining mitigation actions that actually move the score

A mitigation action is only meaningful if it changes at least one factor and the team can state which. Three action classes map cleanly onto the factors:

Design actions reduce Occurrence by removing the cause — e.g., adding a gusset, increasing weld pitch, or changing material grade. These are the highest-value actions because they attack the root cause.
Detection actions reduce Detection by strengthening the load case — e.g., adding *DATABASE_ASCII strain output at the weld line, refining the mesh in the heat-affected zone, or introducing a correlated failure criterion. These are within the simulation team's direct control and should be the first response when design change is not yet feasible.
Severity actions are rare and usually involve changing the effect itself (e.g., adding a secondary load path so a joint failure no longer causes intrusion). Severity should not be "reduced" by redefining the effect to be less severe.

Consider a worked example. A front-rail-to-shotgun joint is assessed under a 64 km/h offset deformable barrier load case. The team scores S = 8 (intrusion beyond target), O = 6 (new gauge, uncorrelated spot-weld model), D = 7 (load case exists but no strain output at the joint). RPN = 336; AP = High. Two actions are defined: (1) a design action to increase weld pitch from 40 mm to 30 mm, expected to reduce O to 4; (2) a detection action to add *DATABASE_EXTENT_BINARY strain output and a shell-element erosion criterion at the joint, expected to reduce D to 4. After both actions, S = 8, O = 4, D = 4, RPN = 128, AP = Medium. The team records the revised scores and the evidence (updated load case deck, correlation memo) in the DFMEA.

Practical scoring discipline

Three habits keep the prioritization credible. First, score each factor independently and in that order — S, then O, then D — because teams that start with D tend to inflate detection confidence. Second, tie every Occurrence and Detection score to a named evidence artifact (test report, correlation memo, load case deck revision). Third, re-score after actions and record the delta; an action that does not change any factor is not a mitigation, it is a task. When RPN and AP disagree, AP governs the action plan, and the RPN is retained only for trend comparison across program phases.

Do not use RPN thresholds alone to close items. A severity-10 failure mode with O = 2 and D = 2 yields RPN = 40, which many legacy thresholds would ignore — yet AP correctly flags it as Medium or High because severity dominates. Always cross-check the AP result before deferring an action.

The output of this step is a ranked, evidence-backed action list that feeds directly into the load-case design work in later chapters: each High-AP item should map to a specific load case, a specific output request, and a specific acceptance criterion, so that the next simulation run either closes the risk or proves the action insufficient.