IEC 61123 — Reliability Testing: Compliance Test Plans for Success Ratio

💡 Standard Overview: IEC 61123, developed by IEC TC 56 (Dependability), specifies procedures for designing compliance test plans to verify that the success ratio — the probability of successfully completing a mission without failure — meets specified requirements. It complements IEC 60605 (failure rate tests) and IEC 61070 (availability compliance tests), filling a critical gap for “one-shot” devices such as missile igniters, airbag inflators, emergency shutdown systems, and pyrotechnic actuators.

1️⃣ Scope and Application Domain

IEC 61123, titled “Reliability testing — Compliance test plans for success ratio”, was first published in 1991. Its latest edition is IEC 61123:2024. The standard belongs to the IEC 60000-series dependability framework and is a key deliverable of TC 56’s reliability test methods portfolio.

The fundamental use case is straightforward: when the probability of success p for a single-shot mission must be verified against a contractual or design requirement, IEC 61123 provides the statistical sampling plans — based on the binomial distribution — needed to make accept/reject decisions with quantified risks.

⚠️ Scope Boundary: IEC 61123 applies strictly to non-repairable items or “one-shot” applications. For repairable systems, refer to IEC 61070 (availability compliance) or the IEC 60605 series (MTBF/failure rate tests). A common pitfall is applying success-ratio test plans to repairable equipment, which mischaracterizes the underlying failure process and leads to incorrect risk estimation.

Typical application domains include:

🚀 Aerospace: Stage separation mechanisms, solar array deployment actuators, parachute release systems, satellite apogee kick motors
🎯 Defense: Missile fuse initiation, pyrotechnic devices, ejection seat cartridges, warhead safing/arming mechanisms
⚡ Power Protection: Circuit breaker trip coils, protective relay output circuits, fault recorder triggers, emergency generator start signals
🏭 Industrial Safety: Emergency shutdown (ESD) pushbuttons, safety valve lift mechanisms, fire sprinkler bulb activation, gas detection alarm relays
🚗 Automotive: Airbag inflator deployment, crash sensor triggering, automatic emergency braking (AEB) system engagement, seatbelt pre-tensioner activation

2️⃣ Statistical Foundations and Key Parameters

2.1 The Binomial Model for Success Ratio Testing

Success ratio testing is built on the binomial distribution. If each independent trial has a constant success probability p, the probability of observing exactly k successes in n trials is:

P(K = k) = C(n,k) · p^k · (1-p)^(n-k)

A test plan is defined by two fundamental quality levels and two risk parameters:

p₀ (Acceptable Quality Level — AQL): The success ratio that the producer expects to be accepted with high probability. When p ≥ p₀, the probability of rejection must not exceed the producer’s risk α.
p₁ (Limiting Quality — LQ / Rejectable Quality Level — RQL): The success ratio that the consumer expects to be rejected with high probability. When p ≤ p₁, the probability of acceptance must not exceed the consumer’s risk β.
α (Producer’s Risk): The probability of rejecting a lot whose true success ratio is at least p₀.
β (Consumer’s Risk): The probability of accepting a lot whose true success ratio is at most p₁.

✅ Engineering Insight: The discrimination ratio p₀/p₁ governs the test plan’s resolving power. A ratio close to 1.0 means high discriminatory power but requires a very large sample size — often impractical. In practice, p₀/p₁ ranges from 1.5 to 5.0. For example, verifying p₀=0.99 against p₁=0.95 (ratio=1.042) demands 459 trials with at most 3 failures, whereas p₀=0.99 vs. p₁=0.90 (ratio=1.10) requires only 130 trials with at most 1 failure.

2.2 Operating Characteristic (OC) Curve

The OC curve L(p) = P(accept | p) is the primary tool for evaluating a test plan’s discriminating ability. It is a monotonically decreasing function of p: L(p₀) ≥ 1-α and L(p₁) ≤ β. The steepness of the OC curve between p₁ and p₀ reflects the plan’s efficiency — a steeper curve means better discrimination but also larger sample size.

2.3 Producer’s and Consumer’s Risk in Practice

The standard typically recommends α = 5%~10% and β = 5%~20%. A critical observation from field experience: many engineering teams focus exclusively on α (producer’s risk) while neglecting β (consumer’s risk). This imbalance is dangerous in safety-critical applications — it implies a non-negligible probability of accepting a batch whose actual success ratio falls below the minimum acceptable level.

Parameter	Symbol	Typical Range	Description
Acceptable Quality Level	p₀	0.90 ~ 0.999	High probability of acceptance when p ≥ p₀
Limiting Quality Level	p₁	0.50 ~ 0.95	High probability of rejection when p ≤ p₁
Producer’s Risk	α	0.05 ~ 0.10	Probability of rejecting a conforming lot
Consumer’s Risk	β	0.05 ~ 0.20	Probability of accepting a non-conforming lot
Discrimination Ratio	p₀ / p₁	1.5 ~ 5.0	Smaller ratio → larger required sample size

3️⃣ Test Plan Types and Design Methodology

3.1 Fixed Sample Size Plan

The fixed sample size plan is the most straightforward approach. The engineer predetermines the number of trials n and the acceptance number c (maximum allowable failures). The lot is accepted if the observed number of failures d ≤ c, and rejected otherwise. The plan (n, c) is found by solving the following simultaneous inequalities:

L(p₀) = P(accept | p₀) ≥ 1 – α —— Producer’s risk constraint
L(p₁) = P(accept | p₁) ≤ β —— Consumer’s risk constraint

The standard’s annexes provide pre-computed tables for various (p₀, p₁, α, β) combinations. For the commonly used α=0.05 and β=0.10:

p₀	p₁	Required Sample Size n	Accept Number c
0.99	0.95	459	3
0.99	0.90	130	1
0.95	0.80	77	3
0.90	0.70	43	2
0.85	0.60	28	2

⚠️ Engineering Trap: A disturbingly common mistake is running far fewer trials than the plan requires. For p₀=0.99 vs. p₁=0.95, running only 50 trials instead of the required 459 yields an OC curve with almost no discriminating power — a lot with true p=0.96 (below AQL) would still have a ~85% acceptance probability. Always verify that the planned sample size is achievable before committing to a specific (p₀, p₁) pair.

3.2 Sequential Test Plan (SPRT)

Sequential test plans — also called truncated sequential plans or Sequential Probability Ratio Tests (SPRT) — are significantly more efficient than fixed-sample-size plans. After each trial, one of three decisions is made: accept, reject, or continue testing. The Average Sample Number (ASN) is typically 30%~50% lower than a comparable fixed plan, especially when the product quality is clearly good (early acceptance) or clearly bad (early rejection).

The decision boundaries are derived from Wald’s SPRT theory:

Accept boundary: ln(A) = ln(β / (1 – α))
Reject boundary: ln(B) = ln((1 – β) / α)
Log-likelihood ratio: LR(n) = d·ln(p₁/p₀) + (n-d)·ln((1-p₁)/(1-p₀))

Decision rule:
– LR(n) ≤ ln(A) → Accept the lot
– LR(n) ≥ ln(B) → Reject the lot
– ln(A) < LR(n) < ln(B) → Continue testing

where d = cumulative number of failures after n trials

✅ Engineering Recommendation: Sequential plans are most advantageous when: ① each trial is expensive or time-consuming (e.g., full-scale rocket launch tests); ② there is prior confidence in product quality (expecting quick pass); or ③ flexible test sizing is needed. However, the maximum number of trials (truncation point) must be pre-established — otherwise the test could theoretically continue indefinitely. The standard provides truncated sequential tables with maximum sample sizes for common parameter combinations.

3.3 Zero-Failure Plan

When safety requirements are extremely stringent — nuclear safety systems, manned spaceflight critical functions, flight-critical avionics — a zero-failure plan (c=0) is often mandated. In this plan, any single failure triggers rejection. The required sample size for a zero-failure plan with consumer’s risk β is:

n ≥ ln(β) / ln(1 – p₀) (to verify p ≥ p₀ with consumer risk β)

Example: Verify success ratio ≥ 0.99 with β = 0.10
n ≥ ln(0.10) / ln(0.01) = 229.1 → n = 230 trials with zero failures

🛑 Critical Caution: Zero-failure plans are extremely demanding. A single failure out of 230 trials means rejection — even though a failure rate of 1/230 ≈ 0.43% might be considered acceptable in many non-safety applications. Use zero-failure plans only when the consequence of a single failure in the field is truly catastrophic. For less critical systems, consider plans with c ≥ 1 to avoid excessive sensitivity to isolated defects.

4️⃣ Engineering Practice and Common Pitfalls

Beyond the mathematics, successful application of IEC 61123 requires careful attention to several practical aspects:

Define success/failure criteria unambiguously: Before any testing begins, the criteria must be documented and agreed upon by all stakeholders. Ambiguity here is the single most common source of disputes. For instance, does a “delayed but successful” deployment count as success? What is the allowable time window? What environmental conditions apply?
Sample representativeness: Test samples must come from the same production lot under controlled manufacturing conditions. Mixing samples from different lots introduces batch-to-batch variability that confounds the true success ratio estimate.
Environmental fidelity: Test conditions (temperature, vibration, EMI, humidity) should realistically simulate the operational environment. Overly benign conditions produce inflated in-lab success ratios that do not transfer to the field; overly severe conditions cause unnecessary rejections.
Bayesian augmentation: When prior data exist (historical lots, similar products, engineering judgment), a Bayesian binomial model can incorporate prior information to reduce the required sample size while maintaining equivalent risk protection. The standard’s deterministic approach can be complemented with Bayesian methods, though contractual acceptance criteria must still be agreed upon.
Plan selection strategy: Use sequential plans for high-cost, long-cycle tests; fixed plans for standardized mass-production acceptance; zero-failure plans only for safety-critical applications with catastrophic failure consequences.

💡 Field Insight: In defense and aerospace programs, success-ratio test plans are commonly used in two distinct phases: qualification testing (design verification, typically with tighter p₁ and lower α/β to provide high-confidence design validation) and acceptance testing (lot-by-lot production verification, often with relaxed sample sizes to balance cost and quality assurance). Never use acceptance-level plans for qualification purposes — the risks are not equivalent.

❓ Frequently Asked Questions

❓ Q1: What is the difference between IEC 61123 and the IEC 60605 series (failure rate tests)?

IEC 60605 focuses on failure rate λ(t) or Mean Time Between Failures (MTBF) — parameters that describe repairable equipment operating over time, modeled with exponential or Weibull distributions. IEC 61123 focuses on the single-shot success probability p for non-repairable items, modeled with the binomial distribution. A simple mnemonic: 60605 asks “how long does it run?” while 61123 asks “does it work when called upon?”

❓ Q2: What can I do when the required sample size is unattainable?

When the sample size dictated by the standard is impractical (e.g., only 5 flight-qualified units exist for a satellite program), several strategies are available: ① Switch to a sequential plan which reduces average sample size by 30-50%; ② Use Bayesian methods to incorporate prior data from component-level tests, similar programs, or engineering analysis; ③ Adopt a “pyramid testing” approach — extensive testing at lower assembly levels combined with limited system-level tests. Regardless of the approach, the alternative plan’s OC curve and resulting α/β risks must be clearly communicated and contractually agreed upon with the customer.

❓ Q3: How should p₀ and p₁ be determined in practice?

p₀ is typically the design specification or contract requirement value — the success ratio the producer commits to achieving. p₁ is the minimum acceptable success ratio from the consumer’s perspective, often derived from system safety analysis (FTA, FMEA, PSA), field data from similar systems, or regulatory requirements. The discrimination ratio p₀/p₁ should be selected by balancing technical feasibility against test cost. A good starting point for many industrial applications is p₀/p₁ ≈ 1.1-1.3, adjusted based on risk tolerance and budget constraints.

❓ Q4: How should “partial success” outcomes be handled?

The binomial model underlying IEC 61123 supports only binary outcomes (success/failure). Partial successes must be mapped to one of these two categories before testing begins. If multiple performance parameters must be simultaneously satisfied, use a logical AND — failure in any single parameter triggers classification as “fail.” For applications where graded outcomes are essential, consider supplementing the IEC 61123 approach with MIL-STD-1916 or Bayesian ordinal models. The treatment of ambiguous outcomes is the most common source of contractual disputes in success-ratio testing — document everything in advance.