System reliability.
Combine component reliabilities into a system number. Put parts in series and every one drags the total down. Add redundancy in parallel and the system beats its best part. Set a k-out-of-n rule for identical units. Watch the block diagram build as you type.
One per line or comma separated. Enter each as a fraction (0.99) or a percent (99%). All components must work, so Rsys is the product of them all.
One per line or comma separated. Fraction (0.99) or percent (99%). The system works if any one part works, so Rsys is one minus the product of all the failure chances.
n identical parts, each with the same reliability, and at least k of them must work. Fraction (0.95) or percent (95%). k of n covers the whole range from series (k = n) to full redundancy (k = 1).
Reliability block diagrams are the easy part. Crucible adds Weibull life-data fitting, fault trees, MIL-HDBK-217 MTBF, and AIAG-VDA FMEA action priority, deterministic and in the open.
The method.
Every component has a reliability, the chance it works over the period you care about, a number between zero and one. The system reliability is how those combine, and the wiring decides it.
In series, all parts must work, so the chances multiply. Each extra block can only pull the product down, which is why a series chain is always weaker than its weakest link. In parallel, the system survives if any one part survives, so instead you multiply the failure chances, one minus each reliability, and subtract from one. Redundancy pushes the system above its best single part.
k-out-of-n generalises both. With n identical units and at least k needed, you sum the binomial chances of exactly k, k plus one, up to n working. Series is k equal to n, simple parallel is k equal to one.
The maths, in the open.
series: R = ∏ Rᵢ · parallel: R = 1 − ∏ (1 − Rᵢ) · k-of-n: R = ∑ᵢ₋ₖⁿ C(n,i) pⁱ (1−p)ⁿ₋ⁱ
Worked example. Three parts at 0.99, 0.98, 0.95. In series: 0.99 × 0.98 × 0.95 = 0.92169, below the weakest part 0.95. The same three in parallel: 1 − (0.01)(0.02)(0.05) = 0.99999, above the best part 0.99. For k-of-n with four parts at 0.95 needing three, sum the chances of exactly three and exactly four working, about 0.9860.
Common pitfalls.
The big assumption is independence. Real redundant parts often share a power supply, a cooling loop, a firmware image, or a design flaw, and a single common cause can fail them together, so the clean parallel number overstates the truth. Redundancy is not free either: the switch that picks the live unit, or the voting logic that decides who is right, becomes a new part that can fail. And these are period reliabilities, not rates, so make sure every component number covers the same mission time before you multiply.
System reliability, answered.
How do you calculate system reliability in series?
For components that must all work, multiply their reliabilities together. Rsys equals R1 times R2 times R3 and so on. Every extra part can only lower the total, so a series system is always less reliable than its weakest part.
How does redundancy change reliability?
In a parallel or redundant system the whole works if any one part works. You multiply the failure probabilities, one minus each reliability, and subtract from one. A parallel system is always more reliable than its best single part.
What is k-out-of-n reliability?
With n identical components each of reliability p, and at least k needed to succeed, the system reliability is the sum over i from k to n of the binomial probability of exactly i working. Series is the n-out-of-n case and simple parallel is the 1-out-of-n case.
Why do these formulas overstate real reliability?
They assume components fail independently. In real systems a common cause such as shared power, heat, or a design flaw can take out redundant parts at once, so true parallel reliability is usually lower than the ideal formula suggests.
Is this calculator free?
Yes, free and no signup, running entirely in your browser.