起源故事 · 田口的"规格门是断崖"之问
Origin Story · Taguchi's "The Spec Gate Is a Cliff" Challenge
日本工程师 田口玄一(Genichi Taguchi)对传统的"规格门思维"很不满:在他看来,把质量看成"门内全合格、门外全不合格",
就像把世界画成了一道断崖 —— 可现实哪有这么干脆?他提出质量损失函数 L(x)=k(x−T)²:
只要偏离目标值 T,无论是否还在规格内,社会都在承担损失,且损失随偏离呈二次方增长。
后来 Chan、Cheng、Spiring(1988)据此提出 Cpm,把"偏离目标"正式写进能力指数。
一句话:合格不是目的,逼近靶心才是。
Japanese engineer Genichi Taguchi was deeply unhappy with the classical "spec-gate" view of quality. To him, calling a part "perfect inside the gate, garbage outside" turned the world into a cliff — and reality is never that abrupt.
He proposed the quality loss function L(x) = k(x − T)²: any drift from target T, in spec or not, makes society pay, and the bill grows quadratically with the offset.
Chan, Cheng, & Spiring (1988) later turned this insight into Cpm, baking "off-target" directly into the capability index.
In one line: passing spec is not the goal — closing on the bullseye is.
1 分布越偏离靶心,抛物线损失越往上爬 The Further You Drift, the Higher the Parabolic Loss Climbs
瞄准靶心试试aim for center2 Cpm 比 Cpk 更"严厉":偏移罚得更狠 Cpm Is Stricter Than Cpk — It Punishes Off-Center Harder
居中时 Cpm 与 Cpk 接近;一旦偏移,Cpm 掉得比 Cpk 更快 —— 因为 Cpm 把偏移当成"等效的额外波动"塞进分母。它在乎的不是"别出门",而是"别离靶心"。 When centered, Cpm and Cpk are close. Drift away and Cpm falls faster than Cpk — because Cpm folds the offset into the denominator as "effective extra spread". It does not care about "stay inside the gate" — it cares about "stay near the bullseye".
3 现实里的 Cpm 与田口思想 Cpm & Taguchi Thinking in the Real World
精密制造:齿轮中心距、镜片曲率等,差一点就影响配合与成像,必须紧贴标称值,Cpm 比 Cpk 更合适。
Precision manufacturing: gear center distance, lens curvature — even a sliver of drift wrecks fit or focus. Stay tight to nominal — Cpm fits better than Cpk here.
田口稳健设计:在产品设计阶段就让性能对噪声不敏感,把"瞄准靶心"提前到源头。
Taguchi robust design: make performance insensitive to noise at the design stage, so "aiming for the bullseye" begins upstream rather than on the line.
电池/电源标称:输出电压、容量必须贴近标称值,偏高偏低都会缩短寿命或触发保护。
Battery / power nominal: output voltage and capacity must hug the nominal value — drift either way trims lifetime or trips protection circuits.
含量目标控制:药品有效成分追求精准目标含量,多了有副作用、少了无疗效,典型目标值导向。
Active-ingredient targeting: pharma APIs chase a precise nominal dose — too much brings side effects, too little kills efficacy. A textbook target-driven characteristic.
一句话In One Line
Cpk 的分母是 3σ,只衡量散布与"到最近规格门"的距离;Cpm 的分母换成 3√(σ²+(μ−T)²) ——
它把"离靶心的偏移"当作等效的额外波动加进去。于是即便你没出规格门,只要 μ 偏离 T,分母就变大、Cpm 就下跌。
这背后是田口的世界观:质量损失是连续的二次曲线,不是非黑即白的断崖。
想提升 Cpm 只有一条心法 —— 既要瘦(降 σ),更要正(μ 贴 T),缺一不可。
Cpk's denominator is 3σ — it weighs spread and the distance to the nearest spec limit. Cpm swaps that out for 3√(σ² + (μ − T)²) —
it folds "offset from the bullseye" in as effective extra spread. So even while you stay inside the spec gate, any drift of μ away from T widens the denominator and Cpm falls.
That is Taguchi's worldview: quality loss is a smooth quadratic curve, not a black-and-white cliff at the spec limit.
Lifting Cpm therefore takes two moves, not one — be lean (cut σ) AND be on target (drag μ onto T). Neither alone is enough.
常见误用Common Mistakes
目标值 T 不在规格中点也照搬公式。Cpm 以 T 为基准,T 可偏离规格中点,须按真实目标设定。
Plugging in the formula when T isn't at the spec midpoint. Cpm anchors on T, and T can sit off-center — always set it from the real engineering target, not the spec midpoint.
把 Cpm 当成更高的 Cpk 来吹。偏移时 Cpm 通常更低,它是更严格而非更宽松的指标。
Marketing Cpm as a "higher Cpk". Once you're off-center Cpm is usually lower — it is a stricter index, not a looser one.
有自然单边目标也强行用二次损失。望小/望大特性(如杂质、强度)用单边损失函数,别套对称抛物线。
Forcing a symmetric quadratic loss on a one-sided target. Smaller-is-better and larger-is-better characteristics (impurity, strength) deserve a one-sided loss function — don't bolt on a symmetric parabola.