起源故事 · 客户视角的反击
Origin Story · The Customer Strikes Back
和 Pp 同根,1989 年 QS-9000 三巨头同时引入了 Ppk。背景一模一样:供应商爱报 Cpk —— 用瞬时最稳的组内 σ + 最近的 30 个样本,数字漂亮。
可客户用几个月的累积数据 + 含季节漂移再算,Ppk 直接腰斩。两个数字差距 = 你工厂"自我感觉"和客户"真实体验"的鸿沟。
今天 IATF 16949 / PPAP / VDA 6.3 全在要求双报 Cpk 与 Ppk,并把 Ppk ≥ 1.67 作为关键件量产放行门槛 —— 它是 Cpk 的真相版本。
Ppk shares Pp's birthday: 1989, when the QS-9000 Big Three rolled them out together. The motive was the same — suppliers reported Cpk using the cleanest σ_within and the latest 30 samples,
and the numbers looked great. The customer would then crunch a few months of accumulated data with seasonal drift mixed in, and Ppk would be cut in half.
That gap is the chasm between the factory's self-image and the customer's real experience. Today IATF 16949, PPAP, and VDA 6.3 all require both Cpk and Ppk,
and Ppk ≥ 1.67 is the mass-production gate for critical characteristics — Ppk is simply the truth-telling version of Cpk.
1 同规格、双钟形:工厂瘦 / 客户胖 Same Spec, Two Bells: Supplier Lean / Customer Fat
差距正常2 一周趋势:Cpk 始终 ≥ Ppk A Week of Trend: Cpk Always Sits Above Ppk
蓝线 = 每天的 Cpk(组内瞬时),红线 = 累计 Ppk(含组间漂移)。红线永远在蓝线下方,差距是过程偷漂的累计代价。 Blue line = daily Cpk (within-subgroup snapshot); red line = cumulative Ppk (with between-subgroup drift). The red line always rides below the blue — the gap is the compounding price of drift.
3 现实里的 Ppk Ppk in the Real World
IATF 16949 关键件:要求 Ppk ≥ 1.67。Cpk 1.67 但 Ppk 1.2 = PPAP 不通过,必须先稳定化再来。
IATF 16949 critical characteristic: required Ppk ≥ 1.67. Cpk 1.67 but Ppk 1.2 → PPAP fails; stabilize the process before resubmitting.
新设备 PPAP:跑 30 天,Cpk=1.5 但 Ppk=1.2 → 设备本身行,但安装后第一周飘了 → 需要稳定化期才能批量放行。
New equipment PPAP: 30-day run, Cpk = 1.5 but Ppk = 1.2 → the equipment is fine, but it drifted during the first week of install. A burn-in window is required before mass-production release.
注塑机换班次:白班 Ppk 1.4,夜班降到 1.0 → 操作员培训不一致 → SOP 加 OJT 是根因,不是设备问题。
Injection-molding shift handover: day shift Ppk 1.4, night shift drops to 1.0 → inconsistent operator training. The root cause is SOP plus on-the-job training, not the equipment.
季节温度影响:夏天 Ppk 1.4,冬天 Ppk 1.0 → 增加 HVAC 或季节参数补偿,比换设备便宜十倍。
Seasonal temperature effect: summer Ppk 1.4, winter Ppk 1.0 → HVAC upgrade or seasonal parameter compensation runs ten times cheaper than replacing the machine.
一句话In One Line
Cpk 是供应商日报,Ppk 是客户月报。差距大 = 你的 SPC 没起作用,过程在偷漂。
改善路径很标准:① 用 X̄-R 图 / 多元方差找出 σ_between 的来源(班次?温度?批次?刀具?)② 用控制规则把异常找出来锁住
③ 重测 Cpk 和 Ppk 直到两者收敛到 5% 以内。这一刻你的过程才真正"在控"。
记住:客户不看你的日报,客户看月报。
Cpk is the supplier's daily report; Ppk is the customer's monthly truth. A wide gap means SPC is not earning its keep and the process is drifting.
The improvement loop is well known: (1) use X̄-R charts or variance-component analysis to identify σ_between's source — shift? temperature? batch? tooling?
(2) apply control-chart rules to flag and lock down the anomalies; (3) re-measure Cpk and Ppk until they converge to within 5 %. Only then is the process truly "in control".
Remember: the customer doesn't read your daily — the customer reads your monthly.
常见误用Common Mistakes
Ppk 用瞬时组内 σ 算。必须 σ_overall(全部数据合并 STDEV),不是 R̄/d₂。两个尺子用错就成 Cpk 了。
Computing Ppk with the short-term σ_within. Use σ_overall (pooled STDEV of all data), not R̄/d₂. Swap the rulers and you're just reporting Cpk under a different name.
30 个数据就算 Ppk 报客户。至少 100+ 数据,且必须跨多个班次 / 批次 / 天,才能让 σ_between 显形。短窗口算的 Ppk ≈ Cpk,是假的。
Quoting Ppk from a 30-sample slice. Need 100+ data points spanning multiple shifts, batches, and days before σ_between can show its face. A short-window Ppk ≈ Cpk — it's fake.
Cpk 高就交付,Ppk 不管。客户审 Ppk;Cpk 高 Ppk 低 = PPAP 不通过,强制返工稳定化。IATF 关键件 Ppk ≥ 1.67 不容商量。
Shipping on a strong Cpk and ignoring Ppk. The customer audits Ppk. High Cpk + low Ppk fails PPAP and forces a stabilization loop. IATF critical-characteristic Ppk ≥ 1.67 is non-negotiable.