Converting Cpk and Process Capability Data to FMEA Occurrence Ratings: The 1-10 Mapping Practitioners Actually Use

The team has been arguing about Occurrence ratings for forty minutes. The pin-press station has six months of Cpk data showing 1.42 on the depth dimension. The quality engineer wants Occurrence 4. The process engineer thinks 6. Neither can defend their number, because neither is using the capability data the production line has already paid to generate. Process capability is the closest thing PFMEA practitioners have to objective evidence of cause frequency—and most teams leave it on the table.

This walkthrough shows how to translate Cpk and Ppk values into AIAG-VDA Occurrence ratings, why the mapping in most online templates is wrong for AIAG-VDA, and what to do when the data exists but is stale or non-normal. The math is straightforward; the judgment is in deciding which capability number describes the failure cause you are rating.

The Mapping AIAG-VDA Actually Uses

The AIAG-VDA 2019 FMEA Handbook ties the Occurrence scale to failure rate predictions in failures-per-thousand-products and supplements that with the effectiveness of current prevention controls. The handbook does not publish a Cpk-to-Occurrence equation directly, but the per-thousand-failure bands at each rating combined with the standard normal distribution give a defensible conversion. For a centered, normally distributed characteristic with two-sided specification limits, the relationship is:

Key Formula $$\text{Defect rate (DPMO)} = 2 \times 10^6 \times \Phi(-3 \times Cpk)$$

where $\Phi$ is the standard normal CDF and the factor of 2 accounts for both specification limits. For a one-sided spec, drop the factor of 2. Substituting common Cpk values gives the DPMO that anchors the Occurrence rating:

Cpk	DPMO (two-sided)	AIAG-VDA Occurrence (cause-only)	Practitioner notes
0.67	45,500	10 (very high — persistent)	Process incapable; defects routinely escape without 100% inspection
0.83	12,200	9 (very high)	Borderline; sustained production yields visible defects
1.00	2,700	8 (high)	3-sigma process; one-sided 1,350 ppm. Below capability minimum for most OEMs
1.17	466	7 (high)	Around AIAG-VDA “moderate-high” band
1.33	63	6 (moderate)	Minimum capable per AIAG; common PPAP threshold
1.50	6.8	5 (moderate)	Solidly capable; small adjustments handle drift
1.67	0.57	4 (moderate-low)	Minimum for critical characteristics under AIAG
2.00	0.002	3 (low)	Six-sigma performance; ~3.4 DPMO with the 1.5-sigma shift
2.33	~10^-6	2 (low)	Process drift is the dominant risk, not random variation
≥2.67	negligible	1 (very low — eliminated)	Reserve for designed-out causes, not statistical performance

Two practical adjustments matter when you apply this table:

Cause-level Occurrence vs. failure-mode Occurrence. The AIAG-VDA rating is for the cause of a failure mode, not the failure mode itself. If a single cause is one of several that can produce a given failure mode, Occurrence is rated on the cause-specific rate. The Cpk-derived rate is naturally cause-level when the characteristic in question is what the cause produces.
Prevention control credit is already partially baked in. AIAG-VDA Occurrence is supposed to reflect cause frequency given current prevention controls. If the Cpk data was collected with prevention controls active, the rating from the table is correct. If you imagine removing the prevention control, the Cpk degrades and the Occurrence rises—but you should rate the as-built condition, not the hypothetical.

Worked Example: Pin-Press Depth, Cpk 1.42

Return to the team in the opening scenario. The pin-press station produces a steel-pin-to-housing interference fit. The failure mode in the PFMEA is “pin under-pressed,” with the cause “press depth below LSL of 2.20 mm.” Cpk over six months on the depth characteristic is 1.42. Compute the DPMO:

Z = 3 × Cpk = 3 × 1.42 = 4.26. One-sided spec (only the lower limit matters for this cause), so DPMO = 10⁶ × $\Phi(-4.26)$ = 10⁶ × 1.02 × 10^-5 = 10.2.

Ten defects per million pressed pins, given current process state. The AIAG-VDA Occurrence band for that rate sits in the 4–5 range—closer to 4 because the rate is well inside the lower half of the band. The team should be rating Occurrence 4 if they trust the capability data, not splitting the difference between 4 and 6 because they have not done the math.

Worked Example For a one-sided specification with Cpk = 1.33: Z = 3 × 1.33 = 3.99, DPMO = 10⁶ × $\Phi(-3.99)$ = 33 ppm. That lands at AIAG-VDA Occurrence 6 (the “moderate” band, around 50–500 ppm at the bottom of the band—33 ppm is right at the boundary, defensible as 6 with a note that lower-end performance trends toward 5).

When the Cpk Data Does Not Apply

The conversion is only honest when the capability data describes the cause you are rating. Four situations break the mapping:

Non-normal distributions. Skewed processes (one-sided physical limits like contamination, wear, or scrap-rate phenomena) produce defect rates that the standard normal formula understates or overstates. Use the actual observed defect rate from the capability study, not the Cpk-derived prediction, when the distribution clearly is not normal. The Anderson-Darling test on the capability dataset is the standard sanity check.
Insufficient sample. Cpk computed from a one-shot 30-piece run is not capable enough to support the 6-sigma tail estimates the conversion relies on. Practitioners use Ppk (long-term) over Cpk (short-term) when sample size is below 100–150 pieces, and they hedge ratings up by 1 if the sample is below 30. The AIAG-VDA spirit is to derate when data is thin, not to claim precision the data cannot support.
The failure cause is not characteristic-driven. If the cause is “operator loads the wrong part into the fixture,” Cpk on the pressed dimension is irrelevant—the cause is upstream of the measured characteristic. Capability data does not bound human-error rate; estimate Occurrence from incident history or use the AIAG-VDA prevention-control criteria directly.
Process drift between data collection and rating. A Cpk from twelve months ago is not the current state if equipment, materials, or operators have changed. Recompute capability from current data before anchoring an Occurrence rating to it; otherwise the rating reflects historical performance, not current risk.

For the broader question of how to interpret a misleading risk score once Occurrence is set, our walkthrough on why high RPN is not always the biggest risk covers the related trap on the AP-vs-RPN side.

How to Defend the Rating in an FMEA Review

The reason the pin-press argument lasted forty minutes was that nobody had the data ready. The team is not arguing about Cpk; they are arguing about who has authority to assign the number. Capability data settles that by transferring authority from the loudest voice to the production record. To make the data hold up in a review:

Cite the specific characteristic, the data collection window, the sample size, and the Cpk or Ppk value. “Depth, last 90 days, n=2,400, Ppk 1.42” closes the discussion.
Compute the implied DPMO at the meeting using the formula above (or have it precomputed in a tool you bring in—the RPN and Action Priority calculator can chain a capability-to-Occurrence pre-step). Show the team the AIAG-VDA Occurrence band that DPMO falls in.
Note any prevention controls active during the capability study and confirm the rating reflects current state. If the control will be removed (cost-down project, supplier change), re-rate at the projected post-change Cpk.
For causes without capability data, mark Occurrence as “engineering judgment, no capability data available” in the FMEA notes. Auditors prefer honest hedging over false precision.

Tip The fastest way to clean up an old PFMEA’s Occurrence ratings is to pull current capability data on every measured characteristic, compute the DPMO, and re-rate every cause that is characteristic-driven. The non-characteristic causes (human error, equipment malfunction, supplier variation) take more judgment, but the measured ones become deterministic.

Edge Cases: Cpk Above 1.67, One-Sided Specs, and Stable vs Unstable Processes

A few edge cases that show up in audit-prep sessions:

Cpk > 1.67 (often around 2.0): the implied DPMO is below 1 ppm. That rates Occurrence 2 or 3 on most AIAG-VDA tables, but the FMEA should reserve Occurrence 1 (“eliminated”) for causes that have been physically designed out, not merely statistically suppressed. Statistical suppression can degrade. Designed-out cannot.
One-sided specifications: use Cpk computed against the relevant limit only (sometimes called Cpu or Cpl). The DPMO formula drops the factor of 2. A torque cause where over-torque damages the part but under-torque is benign has only the upper limit’s DPMO contribution.
Unstable processes: Cpk is meaningless on a non-stable process. If the control chart shows out-of-control signals, the long-term variation is dominated by special-cause excursions, not the within-group spread the Cpk formula measures. Rate Occurrence based on the actual scrap rate during the unstable period, then put process stabilization on the FMEA action list.
Cpk vs Ppk: Cpk uses within-subgroup variation (a short-term estimate). Ppk uses total variation including between-subgroup drift (a long-term estimate). For FMEA Occurrence ratings, Ppk is the more honest input when you have it—customer experience reflects long-term performance, not best-case within-shift consistency.

When Occurrence Ratings Disagree With Field Data

Sometimes the capability data says Occurrence 3 but the warranty database says Occurrence 7. That divergence is informative. Either:

The measured characteristic is not the only path to the failure mode (other causes are driving warranty), or
The capability data is stale or non-representative (newer parts, different operators, supplier change), or
The failure mechanism is environmental (corrosion, thermal cycling, vibration) and does not appear in production capability data at all.

Use the higher-rated number when in doubt. The cost of over-rating Occurrence is an unnecessary action item; the cost of under-rating is a missed failure mode and a warranty surprise. For the broader workflow of how field failures should flow back into FMEA updates, our walkthrough on action closure and re-rating covers the loopback structure on the post-action side.

The AIAG-VDA Occurrence guidance lives in Section V of the 2019 FMEA Handbook; the underlying process capability conventions trace back to ASQ’s process capability guidance. Once Occurrence is anchored to data, severity and detection arguments tend to compress—the team has less surface area to debate.