审厂时客户问"你报的是 Cpk 还是 Ppk"，很多人脱口而出："Cpk 看组内变差，Ppk 看总体变差，Ppk 永远比 Cpk 小。"这套说法在 2026 版 AIAG-VDA SPC 手册里已经过时。新手册按 ISO 22514-2 和 ISO 3534 统一了算法：Pm/Pmk、Pp/Ppk、Cp/Cpk 六种指数的计算方法完全相同，用的是同一个公式、同一个变差；区别只在于数据怎么采的、过程是否证实稳定。
▌ 一个公式，六种指数
手册写得明白：指数计算基于分布 0.135% 和 99.865% 的分位数，正态分布下就是均值 ±3σ。同一批数据不管报 Pp 还是 Cp，数值算出来都一样——名称前的字母不是算法，是"身份证"：
- Pm/Pmk：机器性能，只反映机器影响，其余 4M（人、机、料、法、环）固定，用于设备放行；
- Pp/Ppk：过程性能，考虑全部 5M 影响，但稳定性未证实、未检验或不能评定为受控，用于过程放行（试产 / PPAP 初始研究）；
- Cp/Cpk：过程能力，考虑全部 5M 且过程受控，用于量产监视。
Cpk 这个名字也解释了含义：C = 能力，p = 过程，k = Katayori（偏移）。
▌ 稳定性是分水岭
手册原话："只有在过程的稳定性得到证实的情况下，才使用'能力'（Capability）一词。"因此手册明确区分初始性能 Ppk 和最终能力 Cpk。各阶段该报哪个指数？
- 设备放行看 Pm/Pmk（n=50 件连续生产）；
- 过程放行看 Pp/Ppk（至少 n=125 件、k=25 个子组，典型策略 25 组 × 5 件）；
- 量产阶段，用控制图证实稳定后才能报 Cp/Cpk，没证实就只能报 Pp/Ppk。
过程不稳定还报 Cpk，等于把"性能"硬说成"能力"，审核时是硬伤。
▌ 目标值没有固定值
手册表 8-3 按特性等级给出目标值示例（N≥125）：关键特性 Pp≥2.00、Ppk/Cpk≥1.67；主要特性 Pp≥1.67、Ppk/Cpk≥1.33；次要、其他逐级下调。具体目标由组织与顾客协议确定，不是全国统一价。
样本量不足时目标值还要上调：按 99.99% 置信水平，样本从 125 件降到 75 件，原本 1.67 的目标要抬到 1.82、1.33 要抬到 1.45。拿 20 件数据报 1.67，等于自欺欺人。
▌ 先稳定，再能力
手册图 9-26 的四象限里，只有"有能力且稳定"这一格才合格，其余三格都要采取措施。拿到指数报告先问三件事：
先画控制图证实稳定，再选对指数，最后按特性等级和样本量核对门槛。用对了，数据才说话。
想学透 SPC 能力指数与 2026 版新规则？欢迎联系续创信息技术咨询。
When a customer asks during an audit, "Is that Cpk or Ppk you're reporting?", many people answer without hesitation: "Cpk is within-subgroup variation, Ppk is overall variation, and Ppk is always smaller than Cpk." In the 2026 edition of the AIAG-VDA SPC manual, that claim is outdated. The new manual aligns with ISO 22514-2 and ISO 3534: Pm/Pmk, Pp/Ppk and Cp/Cpk are calculated in exactly the same way — the same formula, the same variation. What differs is only how the data was collected and whether the process has been demonstrated to be stable.
▌ One Formula, Six Indices
The manual is explicit: indices are computed from the 0.135% and 99.865% quantiles of the distribution — under normality, mean ±3σ. For the same data set, whether you label it Pp or Cp, the number comes out identical. The letters in front are not the algorithm; they are an ID card:
- Pm/Pmk — machine performance: reflects the influence of the machine only, with the other 4Ms (man, material, method, environment) held fixed; used for equipment release;
- Pp/Ppk — process performance: covers all 5M influences, but stability is unproven, unverified or cannot be judged as in control; used for process release (trial runs / initial PPAP studies);
- Cp/Cpk — process capability: covers all 5M with the process in a state of control; used for series production monitoring.
The name Cpk also spells out its meaning: C = capability, p = process, k = Katayori (offset).
▌ Stability Is the Dividing Line
In the manual's own words: "The term capability is only used when the stability of the process has been demonstrated." The manual therefore distinguishes explicitly between initial performance Ppk and final capability Cpk. Which index to report at each stage?
- Equipment release → Pm/Pmk (n = 50 consecutive parts);
- Process release → Pp/Ppk (at least n = 125 parts, k = 25 subgroups; the typical strategy is 25 groups × 5 parts);
- Series production → report Cp/Cpk only after stability has been demonstrated with a control chart; without that evidence, only Pp/Ppk may be reported.
Reporting Cpk for an unstable process is dressing performance up as capability — a serious finding in any audit.
▌ There Is No Fixed Target Value
Table 8-3 of the manual gives example targets by characteristic class (N ≥ 125): for critical characteristics, Pp ≥ 2.00 and Ppk/Cpk ≥ 1.67; for major characteristics, Pp ≥ 1.67 and Ppk/Cpk ≥ 1.33; minor and other classes step down accordingly. The specific targets are agreed between the organization and the customer — there is no one-size-fits-all number.
When the sample size is insufficient, the target must be raised: at a 99.99% confidence level, moving from 125 down to 75 parts lifts a 1.67 target to 1.82 and a 1.33 target to 1.45. Reporting 1.67 from 20 data points is fooling no one but yourself.
▌ Stability First, Capability Second
In the four quadrants of Figure 9-26, only "capable and stable" passes; the other three all require action. When an index report lands on your desk, ask three questions first:
- Has stability been demonstrated?
- Is the right index being used?
- Is the sample size sufficient?
Plot the control chart to prove stability, then pick the correct index, and finally check the threshold against characteristic class and sample size. Get it right, and the data will speak for itself.
Want to master SPC capability indices and the new 2026 rules? Contact CINO Information Technology Consulting.