故事 · 为什么飞机要装两台发动机 Story · Why Aircraft Carry Two Engines
一架双引擎飞机,单台引擎在一次飞行中失效的概率约 1/100000。如果两台是串联(任一停就坠机),那比单引擎更危险 —— 坏的机会翻倍。可工程师把它设计成并联:单引擎也能安全降落,于是「两台同时失效」才致命,概率掉到约 1/100亿。 同一批零件,怎么连决定了系统是更脆还是更稳。串联是「短板效应」——一环断、全链断; 并联是「冗余」——多一条命,可靠度指数级提升。这就是系统可靠性工程最核心的两个公式。 On a twin-engine aircraft, a single engine has roughly a 1-in-100,000 chance of failing during a flight. If the two engines were in series (any one failure crashes the plane), the aircraft would be more dangerous than a single-engine plane — you would have doubled the chance of failure. So engineers wire them in parallel: either engine alone can land the aircraft safely, so only "both engines fail simultaneously" is fatal — and that probability collapses to about 1 in 10 billion. Same parts, different topology — and that one choice flips the system between fragile and robust. Series is the weakest-link effect: one bad link breaks the chain; parallel is redundancy: each extra leg multiplies the system's life. These two formulas are the bedrock of system reliability engineering.

1 结构框图:信号从左流到右 Block Diagram: Signal Flows Left to Right

串联Series

2 同样三个部件:串起来 vs 并起来 Same Three Parts: Series vs Parallel

串联永远低于最弱部件(乘法越乘越小);并联永远高于最强部件(失效概率连乘越乘越小)。这就是「短板 vs 冗余」的数学。 Series always lands below the weakest part (multiplying numbers under 1 only shrinks them); parallel always lands above the strongest part (multiplying failure probabilities shrinks them). That is the math of "weakest link vs redundancy".

3 现实里的串联与并联 Series and Parallel in Practice

串联短板:一条产线上 20 道工序串联,每道 99% 良率,整线良率只剩 0.99²⁰≈82% —— 工序越多越要保单工序质量。 Series weakest-link: a line with 20 sequential operations at 99% yield each ends at 0.99²⁰ ≈ 82% overall — the more steps you add, the more single-step yield matters.
冗余设计:数据中心双电源、RAID 磁盘阵列、双网卡,都是并联冗余 —— 单点失效不致命。 Redundant design: dual data-center power feeds, RAID disk arrays, dual NICs — all parallel redundancy, so a single point of failure never takes the service down.
找瓶颈:系统可靠性由最弱环节主导。先算各环节贡献,把冗余加在瓶颈上,投入产出比最高。 Spot the bottleneck: system reliability is dominated by the weakest segment. Score each segment's contribution first, then put your redundancy on the bottleneck — that is where the ROI lives.
高安全行业:航空、核电、医疗设备用「N+1 / 2oo3」表决冗余,把致命失效压到极低概率。 Safety-critical industries: aviation, nuclear, and medical devices lean on "N+1" and "2-out-of-3" voting redundancy to drive catastrophic failure probability into the deep tail.
一句话In One Line
两个公式其实是同一个逻辑的两面。串联要「全都好」,可靠度是好的概率连乘(R = ∏Rᵢ)—— 乘起来只会越来越小,被最弱的拖下去。并联要「至少一个好」,反过来算「全都坏」的概率连乘 (1 − ∏(1−Rᵢ))—— 失效概率越乘越小,于是系统越加越稳。看懂这一点,你就明白: 提升系统可靠性,要么消灭最弱的串联环节,要么在关键处并上冗余,没有第三条路。 The two formulas are two sides of the same logic. Series demands "all must work", so reliability is the product of the good probabilities (R = ∏ Rᵢ) — and a product of numbers below 1 only shrinks, dragged down by the weakest. Parallel demands "at least one works", so you flip it around and take the product of the failure probabilities (1 − ∏ (1 − Rᵢ)) — that product shrinks toward zero, which is exactly why each extra leg makes the whole stronger. Once that lands, the design rule is clear: to raise system reliability, either kill the weakest series link or add redundancy where it matters. There is no third path.
常见误用Common Mistakes
以为部件越多系统越可靠。串联恰恰相反,部件越多越脆;只有并联冗余才越多越稳。 Assuming "more components = more reliable". Series is the opposite — more parts make it more fragile. Only parallel redundancy gets stronger with more legs.
把可靠度直接相加。是连乘不是相加;0.9+0.9 不等于 1.8,串联是 0.81、并联是 0.99。 Adding reliabilities together. Use products, not sums: 0.9 and 0.9 give 0.81 in series and 0.99 in parallel — never 1.8.
假设部件失效相互独立。共因失效(同一断电、同批次缺陷)会让冗余形同虚设,公式会高估真实可靠度。 Assuming component failures are independent. Common-cause failures — a shared power outage, a single bad lot — collapse redundancy and make the formula overstate real reliability.

系统可靠性