13.2 System Reliability Architecture: Series, Parallel, k-out-of-n & Redundancy

13.2 System Reliability Architecture: Series, Parallel, k-out-of-n & Redundancy

Industrial plants, automation systems, and mission-critical production facilities rarely consist of single isolated components. Instead, they comprise complex networks of mechanical drives, electrical sensors, controllers, and fluid circuits. To analyze and design reliable systems, industrial engineers construct Reliability Block Diagrams (RBDs) that map the functional dependencies of the underlying hardware.

---

1. Series System Architecture (Weakest-Link Principle)

In a Series System, all $n$ components are mutually essential for overall system function. If any single component suffers functional failure, the entire system ceases operation.

  +---------+     +---------+     +---------+     +---------+
--| Comp 1  |-----| Comp 2  |-----|  ...    |-----| Comp n  |--
  +---------+     +---------+     +---------+     +---------+

Reliability Formulation

Assuming component failures are statistically independent events:

$R_s(t) = P(T_1 > t \cap T_2 > t \cap \dots \cap T_n > t) = \prod_{i=1}^n R_i(t)$

Because $0 \le R_i(t) \le 1$, the product satisfies:

$R_s(t) \le \min\{R_1(t), R_2(t), \dots, R_n(t)\}$

Engineering Implication: The reliability of a series system is strictly lower than the reliability of its single weakest component. Adding components in series inevitably decreases system reliability.

Exponential Series Systems

If each component $i$ follows an exponential distribution with constant failure rate $\lambda_i$:

$R_s(t) = \prod_{i=1}^n e^{-\lambda_i t} = \exp\left( -\left(\sum_{i=1}^n \lambda_i\right) t \right) = e^{-\lambda_s t}$

The overall system failure rate is the direct linear sum of the individual component failure rates:

$\lambda_s = \sum_{i=1}^n \lambda_i$

$\text{MTTF}_s = \frac{1}{\lambda_s} = \frac{1}{\sum_{i=1}^n \lambda_i}$

---

2. Active Parallel System Architecture (Hot Redundancy)

In an Active Parallel System (also known as Hot Redundancy), all $n$ redundant units operate concurrently online. The system performs its intended function as long as at least one component remains operational. The system fails if and only if all $n$ components fail.

       +---------+
    +--| Comp 1  |--+
    |  +---------+  |
    |  +---------+  |
----+--| Comp 2  |--+----
    |  +---------+  |
    |  +---------+  |
    +--| Comp n  |--+
       +---------+

Reliability Formulation

The system cumulative failure probability $F_s(t)$ requires the simultaneous failure of all independent units:

$F_s(t) = \prod_{i=1}^n F_i(t) = \prod_{i=1}^n \left(1 - R_i(t)\right)$

$R_s(t) = 1 - F_s(t) = 1 - \prod_{i=1}^n \left(1 - R_i(t)\right)$

For $n$ identical components with reliability $R(t)$:

$R_s(t) = 1 - [1 - R(t)]^n$

Mean Time To Failure for Parallel Exponential Systems

Consider $n$ identical components operating in active parallel, each with exponential failure rate $\lambda$. While each individual component has a constant hazard rate, the parallel system as an entity does not have an exponential life distribution; its failure rate increases with time as redundant units fail.

$\text{MTTF}_s = \int_0^\infty R_s(t)\,dt = \int_0^\infty \left( 1 - [1 - e^{-\lambda t}]^n \right) dt$

Utilizing the standard order statistics renewal formulation:

$\text{MTTF}_s = \frac{1}{\lambda} \sum_{i=1}^n \frac{1}{i} = \frac{1}{\lambda} \left( 1 + \frac{1}{2} + \frac{1}{3} + \dots + \frac{1}{n} \right)$

Law of Diminishing Returns in Active Redundancy

• Single unit ($n = 1$): $\text{MTTF}_1 = \frac{1}{\lambda} = 1.000\,\text{MTTF}$
• Dual unit ($n = 2$): $\text{MTTF}_2 = \frac{1}{\lambda}\left(1 + \frac{1}{2}\right) = \frac{1.5}{\lambda} = 1.500\,\text{MTTF}$ (Adds 50% life)
• Triple unit ($n = 3$): $\text{MTTF}_3 = \frac{1}{\lambda}\left(1 + \frac{1}{2} + \frac{1}{3}\right) = \frac{1.833}{\lambda} = 1.833\,\text{MTTF}$ (Adds 33% life)
• Quad unit ($n = 4$): $\text{MTTF}_4 = \frac{1}{\lambda}\left(1 + \frac{1}{2} + \frac{1}{3} + \frac{1}{4}\right) = \frac{2.083}{\lambda} = 2.083\,\text{MTTF}$ (Adds 25% life)

Each successive redundant component provides progressively smaller incremental improvements in expected system operational lifetime while imposing linear increases in cost, mass, and volume.

3. Standby Redundancy (Cold, Warm, Hot)

In standby redundancy, secondary components remain offline or at reduced operational load until an active primary component fails, at which point a sensing and switching mechanism transfers operation to the standby unit.

Comparison of Standby Operational Modes

Mathematical Formulation: Cold Standby with Perfect Switching

Consider two identical components with failure rate $\lambda$. Component 1 operates initially while Component 2 resides in cold standby (dormant failure rate $\lambda_2 = 0$). When Component 1 fails at time $T_1$, an ideal automatic transfer switch ($P_{sw} = 1.0$) instantly energizes Component 2, which then operates until failure at time $T_2$.

Because the two operational intervals are independent and identically distributed exponential random variables, the total time to system failure is $T = T_1 + T_2$, which follows a 2-stage Erlang distribution (Gamma distribution with $\alpha = 2, \beta = \lambda$):

$R_s(t) = P(T_1 + T_2 > t) = P(T_1 > t) + P(T_1 \le t \cap T_1 + T_2 > t)$

$R_s(t) = e^{-\lambda t} + \int_0^t (\lambda e^{-\lambda u}) e^{-\lambda(t - u)}\,du = e^{-\lambda t} + \lambda t e^{-\lambda t} = e^{-\lambda t}(1 + \lambda t)$

The system MTTF is the direct sum of the expected lives:

$\text{MTTF}_{\text{cold}} = E[T_1] + E[T_2] = \frac{1}{\lambda} + \frac{1}{\lambda} = \frac{2}{\lambda}$

Decisive Engineering Comparison:

• Active parallel dual-unit system: $\text{MTTF}_{\text{active}} = \frac{1.5}{\lambda}$
• Cold standby dual-unit system: $\text{MTTF}_{\text{cold}} = \frac{2.0}{\lambda}$

Cold standby provides a 33.3% higher MTTF than active parallel redundancy because the standby unit preserves its full operational lifespan while the primary unit operates.

Impact of Imperfect Switching

If the transfer switch has a static operational probability of successful transfer $P_{sw} < 1.0$, the standby reliability equation modifies to:

$R_s(t) = e^{-\lambda t} + P_{sw} \lambda t e^{-\lambda t} = e^{-\lambda t}(1 + P_{sw} \lambda t)$

$\text{MTTF}_{\text{imperfect}} = \frac{1}{\lambda} + P_{sw} \frac{1}{\lambda} = \frac{1 + P_{sw}}{\lambda}$

If switch reliability drops to $P_{sw} < 0.50$, the cold standby architecture yields lower MTTF than active parallel redundancy.

4. k-out-of-n System Architecture ($k/n:G$)

A $k$-out-of-$n:G$ System (Good) consists of $n$ total components operating simultaneously, of which at least $k$ must remain functional for the system to operate successfully. If $(n - k + 1)$ components fail, the system suffers functional breakdown.

Binomial Formulation for Identical Components

When all $n$ components are mutually independent and possess identical operational reliability $R(t)$:

$R_{k/n}(t) = \sum_{j=k}^n \binom{n}{j} [R(t)]^j [1 - R(t)]^{n - j}$

Where $\binom{n}{j} = \frac{n!}{j!(n - j)!}$ represents the binomial combination coefficient.

Special Boundary Cases

1. $k = n$ (Series System): $R_{n/n}(t) = \binom{n}{n} [R(t)]^n = [R(t)]^n$
2. $k = 1$ (Parallel System): $R_{1/n}(t) = \sum_{j=1}^n \binom{n}{j} R^j (1 - R)^{n-j} = 1 - [1 - R(t)]^n$

Triple Modular Redundancy (TMR / 2-out-of-3 Voting)

Triple Modular Redundancy is the standard fault-tolerant architecture utilized in industrial safety instrumented systems (SIS), nuclear power plant trip computers, and fly-by-wire aerospace avionics. Three identical channels process data in parallel, feeding into a majority voting gate where the system output matches the majority consensus ($k = 2, n = 3$):

$R_{2/3}(t) = \binom{3}{2} R^2(1 - R) + \binom{3}{3} R^3 = 3R^2(1 - R) + R^3 = 3R^2 - 2R^3$

Crossover Analysis of Voting Systems

To determine when TMR provides a reliability benefit over a non-redundant single component, set $R_{2/3} = R$:

$3R^2 - 2R^3 = R \implies 2R^3 - 3R^2 + R = 0$

Factoring out $R(2R - 1)(R - 1) = 0$ yields critical threshold points at $R = 0$, $R = 0.5$, and $R = 1.0$.

• If $R > 0.50$: $R_{2/3} > R$ (TMR improves system reliability). Example: For $R = 0.90$, $R_{2/3} = 3(0.90)^2 - 2(0.90)^3 = 2.43 - 1.458 = 0.9720$ (Reliability gains 7.2%).
• If $R < 0.50$: $R_{2/3} < R$ (TMR degrades system reliability because the probability of two or more simultaneous faulty channels exceeds the probability of a single channel failing). Example: For $R = 0.40$, $R_{2/3} = 3(0.40)^2 - 2(0.40)^3 = 0.48 - 0.128 = 0.3520$.

5. Component-Level vs. System-Level Redundancy

A critical optimization problem in systems engineering addresses where to implement redundancy within multi-stage architectures: should entire systems be duplicated in parallel, or should parallel redundancy be applied to individual components at each sub-stage?

Mathematical Formulation

Consider a two-stage functional chain requiring Subsystem $A$ (reliability $R_A$) in series with Subsystem $B$ (reliability $R_B$). Suppose an engineering budget permits doubling hardware resources (2 units of $A$ and 2 units of $B$).

Configuration 1: System-Level Redundancy

Two complete series channels are connected in active parallel. The system functions if Channel 1 OR Channel 2 functions.

       +---------+     +---------+
    +--| Comp A1 |-----| Comp B1 |--+
    |  +---------+     +---------+  |
----+                               +----
    |  +---------+     +---------+  |
    +--| Comp A2 |-----| Comp B2 |--+
       +---------+     +---------+

Channel reliability: $R_{\text{chan}} = R_A R_B$

$R_{\text{sys}} = 1 - (1 - R_{\text{chan}})^2 = 1 - (1 - R_A R_B)^2 = 2 R_A R_B - R_A^2 R_B^2$

Configuration 2: Component-Level Redundancy

Parallel redundancy is applied to Subsystem $A$, which is connected in series with a parallel redundant bank of Subsystem $B$.

       +---------+          +---------+
    +--| Comp A1 |--+    +--| Comp B1 |--+
    |  +---------+  |    |  +---------+  |
----+               +----+               +----
    |  +---------+  |    |  +---------+  |
    +--| Comp A2 |--+    +--| Comp B2 |--+
       +---------+          +---------+

$R_{A,\text{par}} = 1 - (1 - R_A)^2 = 2R_A - R_A^2$

$R_{B,\text{par}} = 1 - (1 - R_B)^2 = 2R_B - R_B^2$

$R_{\text{comp}} = R_{A,\text{par}} \times R_{B,\text{par}} = (2R_A - R_A^2)(2R_B - R_B^2) = 4 R_A R_B - 2 R_A R_B^2 - 2 R_A^2 R_B + R_A^2 R_B^2$

Formal Mathematical Proof ($R_{\text{comp}} \ge R_{\text{sys}}$)

Subtracting $R_{\text{sys}}$ from $R_{\text{comp}}$:

$R_{\text{comp}} - R_{\text{sys}} = \left(4 R_A R_B - 2 R_A R_B^2 - 2 R_A^2 R_B + R_A^2 R_B^2\right) - \left(2 R_A R_B - R_A^2 R_B^2\right)$

$R_{\text{comp}} - R_{\text{sys}} = 2 R_A R_B - 2 R_A R_B^2 - 2 R_A^2 R_B + 2 R_A^2 R_B^2$

Factoring $2 R_A R_B$:

$R_{\text{comp}} - R_{\text{sys}} = 2 R_A R_B \left( 1 - R_B - R_A + R_A R_B \right) = 2 R_A R_B (1 - R_A)(1 - R_B)$

Because $0 \le R_A \le 1$ and $0 \le R_B \le 1$, every term in the product $2 R_A R_B (1 - R_A)(1 - R_B)$ is non-negative:

$R_{\text{comp}} - R_{\text{sys}} \ge 0 \implies R_{\text{comp}} \ge R_{\text{sys}}$

Physical Reliability Insight

Why does component-level redundancy guarantee equal or superior reliability? In System-Level Redundancy, if Component $A_1$ fails in Channel 1 and Component $B_2$ fails in Channel 2, both channels are disabled, precipitating complete system failure. In Component-Level Redundancy, internal cross-strapping between stages enables the surviving units to cross-connect: operational unit $A_2$ powers operational unit $B_1$, preserving system continuity.

---

6. Complex Networks: Minimal Tie Sets and Minimal Cut Sets

When systems cannot be resolved by simple series and parallel reductions (e.g., bridge circuits, cross-strapped distributed networks), reliability engineers employ Minimal Tie Sets and Minimal Cut Sets.

Minimal Tie Sets (Path Sets)

A Tie Set is a group of components whose operation ensures system operation. A Minimal Tie Set is a path set that contains no redundant components; if any component in the minimal tie set fails, that specific operational path is broken.

Let $T_1, T_2, \dots, T_m$ be the $m$ minimal tie sets of a system. The system functions if at least one minimal tie set functions:

$R_s = P\left( \bigcup_{j=1}^m T_j \right)$

Minimal Cut Sets

A Cut Set is a group of components whose simultaneous failure causes total system failure. A Minimal Cut Set contains no subset of components whose failure would independently fail the system.

Let $C_1, C_2, \dots, C_k$ be the $k$ minimal cut sets. The system fails if any minimal cut set fails:

$F_s = P\left( \bigcup_{i=1}^k C_i \right) \implies R_s = 1 - P\left( \bigcup_{i=1}^k C_i \right)$

Esary-Proschan Reliability Bounds

For complex positive-dependent networks:

$\prod_{i=1}^k \left(1 - P(C_i)\right) \le R_s \le 1 - \prod_{j=1}^m \left(1 - P(T_j)\right)$

7. Worked Numerical Example: Architectural Comparison

Problem Formulation

A safety instrumented shutdown system in a chemical synthesis reactor consists of two functional stages in series:

• Stage A: Pressure Transmitter Subsystem, with single-unit reliability $R_A = 0.90$.
• Stage B: Automated Depressurization Valve, with single-unit reliability $R_B = 0.85$.

An engineering reliability audit mandates evaluating four architectural options:

1. Architecture 1: Baseline non-redundant series configuration ($A - B$).
2. Architecture 2: System-level active redundancy (two identical series channels in parallel).
3. Architecture 3: Component-level active redundancy (parallel pair of $A$ in series with parallel pair of $B$).
4. Architecture 4: Cold standby configuration for Stage A (primary $A_1$ with $\lambda_A = 1.0 \times 10^{-3}\text{ hr}^{-1}$, identical cold backup $A_2$ with perfect switch $P_{sw} = 1.0$, mission time $t = 200$ hours) driving a single Stage B valve with $R_B = 0.85$.

---

Step-by-Step Numerical Calculations

Architecture 1: Baseline Series System

$R_{\text{arch1}} = R_A \times R_B = (0.9000)(0.8500) = 0.765000 \implies 76.50\%$

Architecture 2: System-Level Redundancy

Each channel has reliability $R_{\text{chan}} = 0.7650$. $R_{\text{arch2}} = 1 - (1 - R_{\text{chan}})^2 = 1 - (1 - 0.7650)^2 = 1 - (0.2350)^2$ $R_{\text{arch2}} = 1 - 0.055225 = 0.944775 \implies 94.48\%$

Architecture 3: Component-Level Redundancy

• Stage A parallel reliability: $R_{A,\text{par}} = 1 - (1 - R_A)^2 = 1 - (1 - 0.90)^2 = 1 - (0.10)^2 = 0.990000$
• Stage B parallel reliability: $R_{B,\text{par}} = 1 - (1 - R_B)^2 = 1 - (1 - 0.85)^2 = 1 - (0.15)^2 = 1 - 0.0225 = 0.977500$
• Overall system reliability: $R_{\text{arch3}} = R_{A,\text{par}} \times R_{B,\text{par}} = (0.990000)(0.977500) = 0.967725 \implies 96.77\%$

Reliability Differential ($R_{\text{comp}} - R_{\text{sys}}$): $\Delta R = 0.967725 - 0.944775 = 0.022950 \implies +2.30\%$ Using our derived algebraic formula: $2 R_A R_B (1 - R_A)(1 - R_B) = 2(0.90)(0.85)(0.10)(0.15) = 2(0.765)(0.015) = 0.022950$ The calculated delta matches the analytical identity with 100% precision. Unreliability drops from $5.52\%$ to $3.23\%$, representing a $41.5\%$ reduction in probability of system failure.

Architecture 4: Cold Standby Subsystem

• Operating product for Stage A: $\lambda_A t = (1.0 \times 10^{-3}\text{ hr}^{-1})(200\text{ hr}) = 0.200$
• Stage A cold standby reliability: $R_{A,\text{cold}}(200) = e^{-\lambda_A t}(1 + \lambda_A t) = e^{-0.200}(1 + 0.200) = (0.818731)(1.200) = 0.982477$ (Note: An active parallel pair of identical units would achieve only $1 - (1 - 0.818731)^2 = 0.967140$).
• Overall Architecture 4 reliability (Stage A cold standby in series with single Stage B valve): $R_{\text{arch4}} = R_{A,\text{cold}} \times R_B = (0.982477)(0.850000) = 0.835105 \implies 83.51\%$

---

Architectural Summary Table