故事 · 摩托罗拉为什么取名"六西格玛"
Origin Story · Why Motorola Called It "Six Sigma"
1980 年代,摩托罗拉的工程师 Bill Smith 发现一个规律:出厂时检验合格的产品,到了客户手里仍会失效。
根源是过程的波动太大 —— 规格再宽,只要 σ 不够小,长期运行总会蹭到规格边缘出废品。
他们于是定下一个雄心勃勃的目标:让规格限离过程均值足足 6 个 σ,并预留 1.5σ 给现实中无法避免的长期漂移。
算下来,这意味着每百万次只有 3.4 个缺陷。这个数字太有冲击力,公司干脆用它命名整套方法 —— Six Sigma。
所以"六西格玛"从一开始就是个统计承诺:把波动压到规格的六分之一以内。
In the 1980s, Motorola engineer Bill Smith noticed a pattern: products that passed final inspection still failed once they reached the customer.
The root cause was process variation — no matter how generous the spec, if σ was not small enough, long runs would keep nicking the edges of the spec and producing defects.
So Motorola set an audacious target: put the spec limit a full 6σ from the process mean, and reserve 1.5σ for the long-term drift that no real process can avoid.
The arithmetic landed at only 3.4 defects per million opportunities. The number was so striking that the company simply named the whole methodology after it — Six Sigma.
From day one, "Six Sigma" was a statistical promise: keep variation inside one-sixth of the spec.
1 σ 越多,规格离均值越远,尾巴越细 More σ Between Mean and Spec → Thinner Tails
6σ2 质量阶梯:从 3σ 到 6σ 的断崖式下降 The Quality Ladder: From 3σ to 6σ, a Cliff-Drop in Defects
每升一级 σ,DPMO 不是线性下降而是断崖式暴跌(含 1.5σ 漂移): 3σ ≈ 6.7 万 → 4σ ≈ 6210 → 5σ ≈ 233 → 6σ ≈ 3.4。这就是为什么从 4σ 到 6σ 价值连城。 Each extra σ does not lower DPMO linearly — it collapses it (with the 1.5σ drift baked in): 3σ ≈ 66,800 → 4σ ≈ 6,210 → 5σ ≈ 233 → 6σ ≈ 3.4. That is why every step from 4σ toward 6σ is worth a fortune.
3 现实里的六西格玛标尺 The Six Sigma Ruler in the Real World
命名由来:6σ = 规格半宽是过程标准差的 6 倍,波动只占规格的六分之一,是摩托罗拉立下的统计目标。
Where the name comes from: 6σ means the half-spec-width equals six times the process standard deviation — variation is only one-sixth of the spec. This is the statistical target Motorola committed to.
规格-波动比:同样的规格,σ 越小水平越高;同样的 σ,规格越宽水平越高。两者之比就是 σ 水平。
Spec-to-spread ratio: for a fixed spec, a smaller σ means a higher sigma level; for a fixed σ, a wider spec means a higher sigma level. The ratio between the two is exactly the sigma level.
DPMO:把超规格概率乘以一百万,得到"每百万缺陷数",便于跨产品、跨行业横向比质量。
DPMO: multiply the out-of-spec probability by one million and you get "defects per million opportunities" — a single number that compares quality across products and industries.
质量标尺:航空、医疗常要 6σ 甚至更高;普通制造业多在 3~4σ。σ 水平给所有过程一把统一的尺。
A universal quality ruler: aerospace and medical devices commonly demand 6σ or higher; everyday manufacturing usually sits at 3–4σ. The sigma level puts every process on the same scale.
一句话In One Line
六西格玛的内核,是把质量翻译成一个可比较的数:规格离均值有几个 σ。
σ 数越大,意味着过程波动相对规格越小,规格那一侧留的余量越足,超规格的概率越低。
关键的 1.5σ 漂移是个务实的让步:没有过程能永远站在中心,给它预留 1.5σ 的晃动空间,
6σ 的短期 2.0 ppb 才落地成长期的 3.4 DPMO。从此"质量好不好"不再靠感觉,而是一个 σ 水平、一个 DPMO 数字,
航空和螺丝厂可以放在同一把尺上比。这就是为什么六西格玛把一切偏离都换算成 σ。
The core of Six Sigma is translating "quality" into one comparable number: how many σ sit between the mean and the spec.
A bigger σ-count means process variation is smaller relative to the spec, the headroom on the spec side is larger, and the chance of falling out of spec is lower.
The crucial 1.5σ drift is a pragmatic concession: no real process holds dead-center forever, so you give it 1.5σ of room to wander.
That is why the ideal 2.0 ppb at short-term 6σ lands at the famous long-term 3.4 DPMO. From then on, "is the quality good?" stops being a gut feel — it becomes a sigma level plus a DPMO number,
so an aerospace plant and a screw factory can be compared on the same ruler. This is why Six Sigma converts every deviation back into σ.
常见误用Common Mistakes
忘了 1.5σ 漂移,直接用 6σ 的理想良率(约 2 ppb)报数。行业惯例计入 1.5σ 漂移,6σ 对应 3.4 DPMO。
Forgetting the 1.5σ drift and reporting the ideal 6σ yield (~2 ppb). Industry convention bakes in the 1.5σ drift, so 6σ corresponds to 3.4 DPMO.
数据非正态还套 σ 水平表。DPMO 换算建立在正态上,偏态数据要先变换或用其他方法。
Applying the sigma-level table to non-normal data. The DPMO conversion assumes normality — skewed data must be transformed first or evaluated with a different method.
把"六西格玛"只当成管理口号。它有精确的统计定义:规格半宽 = 6 倍过程 σ。
Treating "Six Sigma" as a management slogan. It has a precise statistical definition: the half-spec-width equals six times the process σ.