起源故事 · 被质疑"太灵活"的天才分布
Origin Story · The Brilliant Distribution Dismissed as "Too Flexible"
1951 年，瑞典工程师 Waloddi Weibull 发表《一个广泛适用的统计分布函数》。审稿人嫌它"太灵活，什么数据都能拟合",
一度不被看好。可正是这份灵活，让它能同时描述材料疲劳、轴承寿命、风速、人类寿命 —— 改一个 β 就换一种失效模式。
如今它是可靠性工程的通用语言，你手机的保修期、电梯的维保周期，背后都是 Weibull。
In 1951, Swedish engineer Waloddi Weibull published A Statistical Distribution Function of Wide Applicability. Reviewers groaned — it was "too flexible, it can fit anything" — and the paper was initially dismissed.
But that flexibility is exactly the point: a single β describes metal fatigue, bearing life, wind speed, and human longevity all at once — change β and you change the failure mode.
Today the Weibull distribution is the lingua franca of reliability engineering. The warranty on your phone, the maintenance interval on an elevator — Weibull sits underneath all of it.
1 失效率曲线：你现在站在浴盆的哪一段？ Hazard Curve: Which Part of the Bathtub Are You On?
随机期Random2 寿命分布：产品大多在什么时候坏 Life Distribution: When Do Most Units Actually Fail?
β<1 时大量产品很早就坏（右偏严重）;β 越大，失效越集中在特征寿命 η 附近，曲线越像正态钟形。 When β<1, most units die early and the curve is heavily right-skewed. The larger β gets, the more failures cluster around characteristic life η, and the shape drifts toward a normal bell.
3 现实里的威布尔 Weibull in the Real World
保修期设计:β<1 的早夭期长 → 出厂前做老化筛选，把婴儿死亡挡在保修期外。
Warranty design: a long β<1 infant phase → run burn-in screening before shipment so the early deaths happen inside the factory, not inside the warranty window.
预防性维护:β>1 进入磨损期 → 在失效率陡升前定期更换（轴承、刀具、轮胎）。
Preventive maintenance: once β>1 wear-out kicks in, replace on a schedule before the hazard rate spikes — bearings, cutting tools, tires.
可靠性测试：用威布尔拟合寿命数据，估 B10 寿命（10% 失效点），给质保定指标。
Reliability testing: fit a Weibull to life-test data, estimate the B10 life (the 10%-failure point), and use it to set warranty targets.
风能与气象：风速分布常用威布尔，风电场选址与发电量预测离不开它。
Wind & meteorology: wind-speed distributions are routinely modeled with Weibull — wind-farm siting and energy-yield forecasts both depend on it.
一句话
In One Line
指数分布只能描述"不会老化"的随机失效（β=1 这一种）。威布尔把它推广：一个 β 就在早夭、随机、磨损之间自由切换 ——
这就是它成为可靠性工程地基的原因。看懂浴盆曲线，你就能回答最实际的问题：这东西该什么时候换？保修期定多久？
The exponential distribution only describes memoryless, never-aging failures — the single point β=1. Weibull generalizes it: one β slides freely between infant mortality, random life, and wear-out —
which is why it has become the bedrock of reliability engineering. Read the bathtub curve and you can answer the questions that actually matter on the shop floor: when should this part be replaced? how long should the warranty be?
常见误用
Common Mistakes
不看 β 直接套指数。先拟合 β：磨损件 β>1，绝不能当指数。
Slapping an exponential on everything without estimating β first. Always fit β. Wear-out parts have β>1 — modeling them as exponential is just wrong.
样本太少硬估 β。小样本 β 置信区间很宽，需足够失效数据。
Estimating β from a handful of failures. Small samples give wide confidence intervals on β — you need enough actual failures, not just censored survivors.
把 η 当成平均寿命。η 是 63.2% 失效点（特征寿命），平均寿命另算。
Calling η the "mean life." η is the 63.2%-failure point (characteristic life). The actual mean life is a separate calculation involving Γ(1 + 1/β).