Series and Parallel Systems
CSIR NET reliability: series, parallel, k-out-of-n, standby and bridge systems; system reliability, MTTF, path and cut bounds and redundancy.
- 17 min
On this page · 17 sections
- Prerequisites
- The Hook: Two Pumps Are Not Twice as Good
- Setting Up: Structure Functions
- Series Systems
- Parallel Systems
- k-out-of-n Systems
- Standby Redundancy
- Mixed Networks: Series-Parallel Reduction
- Non-Series-Parallel Systems: The Bridge
- Which Component Matters Most?
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Hazard Function, Failure Rates, Censoring and Life Testing: reliability functions, hazards, the MTTF formula \( \int_0^\infty R(t)\,dt \) and the exponential distribution.
- Sample Space, Independence and Bayes: independence of events, inclusion-exclusion and conditioning on an event, which is exactly how the bridge network is solved.
- Counting and Probability (Part A): binomial counting for k-out-of-n systems.
The Hook: Two Pumps Are Not Twice as Good
Setting Up: Structure Functions
A system has \( n \) components. Component \( i \) works with probability \( p_i \) (its reliability at a fixed mission time \( t \), so \( p_i = R_i(t) \)), independently of the others. Let \( x_i = 1 \) if component \( i \) works and 0 otherwise.
The structure function \( \phi(x_1, \dots, x_n) \in \{0, 1\} \) says whether the system works. The system reliability is
$$R_S = P(\phi(X) = 1) = E[\phi(X)].$$A system is coherent if \( \phi \) is non-decreasing in each \( x_i \) (repairing a component never hurts) and every component is relevant (each one matters in at least one state). All the systems on this page are coherent.
| System | Structure function | Works when |
|---|---|---|
| Series | \( \prod x_i = \min x_i \) | all components work |
| Parallel | \( 1 - \prod (1 - x_i) = \max x_i \) | at least one works |
| k-out-of-n | \( \mathbf{1}\{\sum x_i \ge k\} \) | at least \( k \) work |
Series is n-out-of-n and parallel is 1-out-of-n.
Series Systems
$$\boxed{R_{\text{series}} = \prod_{i=1}^n p_i \le \min_i p_i}$$A series system is weaker than its weakest component. With 0.9, 0.8 and 0.95, \( R = 0.684 \). With 50 components each of reliability 0.99, \( R = 0.99^{50} \approx 0.605 \): long chains of good parts make poor systems.
As functions of time. \( R_S(t) = \prod R_i(t) \), so the hazards add:
$$h_S(t) = \sum_{i=1}^n h_i(t).$$For exponential components with rates \( \lambda_i \), the system is exponential with rate \( \Lambda = \sum \lambda_i \) and
$$\text{MTTF}_{\text{series}} = \frac{1}{\lambda_1 + \dots + \lambda_n}.$$Example: rates 0.001, 0.002, 0.003 per hour give \( \Lambda = 0.006 \), MTTF \( \approx 166.7 \) hours and \( R(100) = e^{-0.6} \approx 0.549 \).
Parallel Systems
$$\boxed{R_{\text{parallel}} = 1 - \prod_{i=1}^n (1 - p_i) \ge \max_i p_i}$$Work with unreliabilities \( q_i = 1 - p_i \): the system fails only if every component fails. With 0.9, 0.8 and 0.95, \( R = 1 - (0.1)(0.2)(0.05) = 0.999 \).
How many in parallel? For identical components with reliability \( p \) and a target \( R^* \): \( 1 - q^n \ge R^* \iff n \ge \frac{\ln(1 - R^*)}{\ln q} \). For \( p = 0.8 \), \( R^* = 0.99 \): \( n \ge \frac{\ln 0.01}{\ln 0.2} \approx 2.86 \), so 3 components.
Parallel exponential components
For \( n \) identical exponential components with rate \( \lambda \), \( R(t) = 1 - (1 - e^{-\lambda t})^n \) and
$$\text{MTTF} = \frac{1}{\lambda}\left(1 + \frac{1}{2} + \dots + \frac{1}{n}\right).$$Why. Start with \( n \) working units. The first failure comes after the minimum of \( n \) exponentials, mean \( \frac{1}{n\lambda} \); by memorylessness the survivors are as good as new, so the next failure comes after a further mean \( \frac{1}{(n-1)\lambda} \), and so on down to the last unit.
For two non-identical units, \( R(t) = e^{-\lambda_1 t} + e^{-\lambda_2 t} - e^{-(\lambda_1 + \lambda_2)t} \) and
$$\text{MTTF} = \frac{1}{\lambda_1} + \frac{1}{\lambda_2} - \frac{1}{\lambda_1 + \lambda_2}.$$The returns diminish: the \( n \)-th parallel unit adds only \( \frac{1}{n\lambda} \) to the mean life.
k-out-of-n Systems
For identical independent components with reliability \( p \):
$$R_{k/n} = \sum_{j=k}^{n} \binom{n}{j} p^j (1 - p)^{n - j}.$$The 2-out-of-3 (majority vote, “triple modular redundancy”) system has \( R = 3p^2 - 2p^3 \). At \( p = 0.9 \), \( R = 0.972 \), better than one unit. At \( p = 0.5 \), \( R = 0.5 \), and for \( p < \frac{1}{2} \) the vote is worse than a single unit: \( 3p^2 - 2p^3 - p = -p(2p - 1)(p - 1) \), which is negative for \( 0 < p < \frac{1}{2} \) and positive for \( \frac{1}{2} < p < 1 \).
For identical exponential components with rate \( \lambda \), the same “race” argument gives
$$\text{MTTF}_{k/n} = \frac{1}{\lambda}\sum_{i=k}^{n} \frac{1}{i}.$$So 2-out-of-3 has MTTF \( \frac{1}{\lambda}\left(\frac{1}{3} + \frac{1}{2}\right) = \frac{5}{6\lambda} \), less than a single unit’s \( \frac{1}{\lambda} \), even though its reliability is higher for short missions. (Check: \( \int_0^\infty (3e^{-2\lambda t} - 2e^{-3\lambda t})\,dt = \frac{3}{2\lambda} - \frac{2}{3\lambda} = \frac{5}{6\lambda} \).) The two curves cross where \( R_{2/3} = e^{-\lambda t} \), i.e. at \( e^{-\lambda t} = \frac{1}{2} \), \( t = \frac{\ln 2}{\lambda} \).
Standby Redundancy
In cold standby the spares are idle and do not fail while waiting; a switch puts a spare in service when the active unit fails. With a perfect switch, the system life is the sum of the unit lives.
| Configuration (identical exponential units, rate \( \lambda \)) | \( R(t) \) | MTTF |
|---|---|---|
| Single unit | \( e^{-\lambda t} \) | \( \frac{1}{\lambda} \) |
| 2 in active parallel | \( 2e^{-\lambda t} - e^{-2\lambda t} \) | \( \frac{3}{2\lambda} \) |
| 1 active + 1 cold spare | \( e^{-\lambda t}(1 + \lambda t) \) | \( \frac{2}{\lambda} \) |
| 1 active + \( (n-1) \) cold spares | \( e^{-\lambda t}\sum_{j=0}^{n-1} \frac{(\lambda t)^j}{j!} \) | \( \frac{n}{\lambda} \) |
The standby system life is Erlang (gamma with integer shape \( n \)), so it is IFR. For two non-identical units in cold standby, \( \text{MTTF} = \frac{1}{\lambda_1} + \frac{1}{\lambda_2} \).
Mixed Networks: Series-Parallel Reduction
Reduce a network step by step: replace each series block by the product of its reliabilities and each parallel block by one minus the product of its unreliabilities.
graph LR
S((in)) --> A["1: 0.9"]
A --> B["2: 0.8"]
A --> C["3: 0.7"]
B --> T((out))
C --> TComponent 1 in series with the parallel pair (2, 3): \( R = 0.9 \times [1 - (0.2)(0.3)] = 0.9 \times 0.94 = 0.846 \).
Redundancy at component level beats redundancy at system level
Take a two-component series system and add a second copy of each component.
- Component-level (each component duplicated in parallel, then the pairs in series): \( R_c = [1 - (1 - p)^2]^2 \).
- System-level (two complete series chains in parallel): \( R_s = 1 - (1 - p^2)^2 \).
For every \( 0 < p < 1 \), \( R_c \ge R_s \); at \( p = 0.9 \), \( R_c = 0.9801 \) and \( R_s = 0.9639 \). This is a general fact for coherent systems: redundancy at the component level is at least as good as redundancy at the system level.
Non-Series-Parallel Systems: The Bridge
graph LR
S((in)) --> C1["1"]
S --> C2["2"]
C1 --> A((a))
C2 --> B((b))
A --- C3["3 (bridge)"]
C3 --- B
A --> C4["4"]
B --> C5["5"]
C4 --> T((out))
C5 --> TThe bridge cannot be reduced by series and parallel steps. Two exact methods:
1. Pivotal decomposition (condition on the bridge). For any component \( i \),
$$R_S = p_i\,R_S(\text{given } i \text{ works}) + (1 - p_i)\,R_S(\text{given } i \text{ fails}).$$- Component 3 works: nodes \( a \) and \( b \) merge, so the system is (1 parallel 2) in series with (4 parallel 5).
- Component 3 fails: the system is (1 series 4) in parallel with (2 series 5).
With all \( p_i = p \) and \( q = 1 - p \):
$$R_{\text{bridge}} = p\,(1 - q^2)^2 + q\,[1 - (1 - p^2)^2] = 2p^2 + 2p^3 - 5p^4 + 2p^5.$$At \( p = 0.9 \): \( R = 0.97848 \). At \( p = 0.5 \): \( R = 0.5 \).
2. Minimal paths and cuts. A minimal path set is a minimal set of components whose working guarantees the system works; a minimal cut set is a minimal set whose failure guarantees system failure.
| Bridge | Sets |
|---|---|
| Minimal path sets | \( \{1, 4\}, \{2, 5\}, \{1, 3, 5\}, \{2, 3, 4\} \) |
| Minimal cut sets | \( \{1, 2\}, \{4, 5\}, \{1, 3, 5\}, \{2, 3, 4\} \) |
The system works iff some minimal path works, and fails iff some minimal cut fails completely. For independent components of a coherent system, these give the bounds
$$\prod_{\text{cuts } K}\left[1 - \prod_{i \in K} q_i\right] \;\le\; R_S \;\le\; 1 - \prod_{\text{paths } P}\left[1 - \prod_{i \in P} p_i\right].$$For the bridge at \( p = 0.9 \): \( 0.97814 \le 0.97848 \le 0.99735 \). The lower (cut) bound is sharp when the components are reliable; the upper (path) bound is sharp when they are unreliable.
Which Component Matters Most?
The Birnbaum importance of component \( i \) is
$$I_i = \frac{\partial R_S}{\partial p_i} = R_S(\text{given } i \text{ works}) - R_S(\text{given } i \text{ fails}),$$the gain in system reliability per unit gain in \( p_i \).
- Series: \( I_i = \prod_{j \ne i} p_j \), largest for the least reliable component. With 0.9, 0.8, 0.95: \( I = 0.76, 0.855, 0.72 \); improve the 0.8 component first.
- Parallel: \( I_i = \prod_{j \ne i} q_j \), largest for the most reliable component. With 0.9, 0.8, 0.95: \( I = 0.010, 0.005, 0.020 \); improve the 0.95 component first.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Active parallel: all units age together, MTTF \( \frac{3}{2\lambda} \) for two | cold standby: spares do not age, MTTF \( \frac{2}{\lambda} \) for two | |
| 2-out-of-3 has higher \( R(t) \) for small \( t \) | but lower MTTF (\( \frac{5}{6\lambda} \)) than one unit | |
| Series of IFR is IFR | parallel of IFR need not be IFR (only IFRA) | |
| Minimal path: working set that suffices | minimal cut: failed set that suffices | |
| Component-level redundancy | system-level redundancy: never better | |
| \( R = \prod p_i \) (series) | \( R = \prod p_i \) is not the parallel formula; parallel multiplies the \( q_i \) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.
Level 1: Foundation
- Write the reliability of \( n \) independent components in parallel.
- Give the MTTF of two identical exponential units (rate \( \lambda \)) in active parallel.
- What is the hazard of a series system in terms of the component hazards?
Answers
- \( 1 - \prod_{i=1}^n (1 - p_i) \).
- \( \frac{1}{\lambda} + \frac{1}{2\lambda} = \frac{3}{2\lambda} \).
- The sum: \( h_S(t) = \sum_i h_i(t) \) (independent components).
Level 2: Part B Style
Practice Q1. Component 1 (reliability 0.9) is connected in series with a parallel pair of components 2 and 3 (reliabilities 0.8 and 0.7). All are independent. The system reliability is
(a) 0.504 (b) 0.846 (c) 0.956 (d) 0.994
Answer: (b)
Parallel pair: \( 1 - (0.2)(0.3) = 0.94 \). In series with 0.9: \( 0.9 \times 0.94 = 0.846 \). (a) is all three in series, (d) all three in parallel, (c) is 1 in parallel with the series pair (2, 3).
Practice Q2. Identical independent units of reliability 0.7 are placed in parallel. The least number of units for system reliability at least 0.99 is
(a) 3 (b) 4 (c) 5 (d) 7
Answer: (b)
Need \( 0.3^n \le 0.01 \): \( 0.3^3 = 0.027 \), \( 0.3^4 = 0.0081 \). So \( n = 4 \) (\( \frac{\ln 0.01}{\ln 0.3} \approx 3.82 \)).
Practice Q3. A 2-out-of-3 system has identical independent exponential components with rate \( \lambda \). Its mean time to failure is
(a) \( \frac{1}{3\lambda} \) (b) \( \frac{5}{6\lambda} \) (c) \( \frac{3}{2\lambda} \) (d) \( \frac{11}{6\lambda} \)
Answer: (b)
\( \frac{1}{\lambda}\left(\frac{1}{3} + \frac{1}{2}\right) = \frac{5}{6\lambda} \): the first failure (mean \( \frac{1}{3\lambda} \)) leaves 2 working, the second (a further \( \frac{1}{2\lambda} \)) kills the system. (a) is the series value, (d) the parallel value.
Practice Q4. Two identical units with exponential lives (rate 0.01 per hour) form a cold standby system with a perfect switch. The reliability for a 100-hour mission is
(a) \( e^{-1} \) (b) \( 2e^{-1} \) (c) \( 1 - (1 - e^{-1})^2 \) (d) \( e^{-2} \)
Answer: (b)
\( R(t) = e^{-\lambda t}(1 + \lambda t) \) with \( \lambda t = 1 \): \( 2e^{-1} \approx 0.736 \). (c) \( \approx 0.600 \) is active parallel; (a) is one unit; (d) is the series pair.
Level 3: Part C Style
Practice Q5. In the bridge network drawn above, all five components are independent with the same reliability \( p \). Which statements are true?
(a) The system reliability is \( 2p^2 + 2p^3 - 5p^4 + 2p^5 \). (b) For \( p = 0.9 \) the system reliability is less than \( 1 - (1 - p^2)^2 = 0.9639 \). (c) \( \{1, 3, 5\} \) is a minimal path set and also a minimal cut set. (d) For \( p = 0.5 \) the system reliability is 0.5.
Answer: (a), (c) and (d)
- (a) True. Pivot on 3: \( p(1 - q^2)^2 + q[1 - (1 - p^2)^2] \); expanding gives \( 2p^2 + 2p^3 - 5p^4 + 2p^5 \).
- (b) False. At \( p = 0.9 \) the bridge gives 0.97848, more than 0.9639 (the reliability without the bridge component, i.e. two parallel series paths). Adding component 3 can only help a coherent system.
- (c) True. 1, 3, 5 working gives the path in → a → b → out, and no proper subset is a path. If 1, 3, 5 all fail, the only remaining route from the input is through 2 to \( b \), and from \( b \) the edges 3 and 5 are dead, so the system fails; no proper subset is a cut.
- (d) True. \( 2(0.25) + 2(0.125) - 5(0.0625) + 2(0.03125) = 0.5 + 0.25 - 0.3125 + 0.0625 = 0.5 \).
Practice Q6. Two independent components with exponential lives, rates \( \lambda_1 = 1 \) and \( \lambda_2 = 2 \) (per 1000 hours), are connected in active parallel. Which statements are true?
(a) The system MTTF is \( \frac{7}{6} \) thousand hours. (b) The system reliability is \( e^{-t} + e^{-2t} - e^{-3t} \). (c) The system hazard is non-decreasing in \( t \). (d) The system hazard tends to 1 as \( t \to \infty \).
Answer: (a), (b) and (d)
- (a) True. \( 1 + \frac{1}{2} - \frac{1}{3} = \frac{7}{6} \).
- (b) True. \( 1 - (1 - e^{-t})(1 - e^{-2t}) \).
- (c) False. The hazard is 0 at \( t = 0 \), about 1.105 at \( t = 1.5 \), and about 1.043 at \( t = 3 \): it rises above 1 and then falls.
- (d) True. For large \( t \) the weaker component is almost surely dead, and the system behaves like the survivor, rate \( \lambda_1 = 1 \). Formally \( f(t)/R(t) \to 1 \) because both are dominated by the \( e^{-t} \) terms.
Practice Q7. A system needs components A and B (series). Each has reliability \( p \in (0, 1) \), independently. One extra copy of each is available. Let \( R_c \) be the reliability when each component is duplicated in parallel (component-level redundancy) and \( R_s \) when two complete A-B chains are put in parallel (system-level). Which statements are true?
(a) \( R_c = [1 - (1 - p)^2]^2 \). (b) \( R_s = 1 - (1 - p^2)^2 \). (c) At \( p = 0.9 \), \( R_c - R_s = 0.0162 \). (d) There is a \( p \in (0, 1) \) with \( R_s > R_c \).
Answer: (a), (b) and (c)
- (a), (b) True by series-parallel reduction.
- (c) True. \( R_c = 0.99^2 = 0.9801 \); \( R_s = 1 - 0.19^2 = 0.9639 \); difference 0.0162.
- (d) False. \( R_c - R_s = 2p^2(1 - p)^2 \ge 0 \) (expand both: \( R_c = 4p^2 - 4p^3 + p^4 \), \( R_s = 2p^2 - p^4 \), so \( R_c - R_s = 2p^2 - 4p^3 + 2p^4 = 2p^2(1 - p)^2 \)), with equality only at \( p = 0, 1 \).
Quick Revision Table
| System | Reliability | MTTF (identical exponential, rate \( \lambda \)) |
|---|---|---|
| Series of \( n \) | \( \prod p_i \) | \( \frac{1}{n\lambda} \) (in general \( \frac{1}{\sum \lambda_i} \)) |
| Parallel of \( n \) | \( 1 - \prod q_i \) | \( \frac{1}{\lambda}\sum_{i=1}^n \frac{1}{i} \) |
| Two parallel, different rates | \( e^{-\lambda_1 t} + e^{-\lambda_2 t} - e^{-(\lambda_1 + \lambda_2)t} \) | \( \frac{1}{\lambda_1} + \frac{1}{\lambda_2} - \frac{1}{\lambda_1 + \lambda_2} \) |
| k-out-of-n | \( \sum_{j \ge k}\binom{n}{j}p^j q^{n-j} \) | \( \frac{1}{\lambda}\sum_{i=k}^n \frac{1}{i} \) |
| 2-out-of-3 | \( 3p^2 - 2p^3 \) | \( \frac{5}{6\lambda} \) |
| Cold standby, \( n \) units | \( e^{-\lambda t}\sum_{j| \( \frac{n}{\lambda} \) | |
| Bridge (all \( p \)) | \( 2p^2 + 2p^3 - 5p^4 + 2p^5 \) | |
| Bounds | \( \prod_K (1 - \prod_K q_i) \le R \le 1 - \prod_P (1 - \prod_P p_i) \) | |
| Birnbaum importance | \( \frac{\partial R_S}{\partial p_i} \) | series: weakest; parallel: strongest |
Key Takeaways
What’s Next
- Linear Programming: Simplex and Duality: from systems that fail to resources that must be allocated optimally.
- Hazard Function, Failure Rates, Censoring and Life Testing: revisit the ageing classes that system structure preserves or destroys.
- Reliability and Operations Research: Quick Revision.