ASQ Certified Reliability Engineer (CRE) — Advanced Study Guide

ASQ Certified Reliability Engineer (CRE) — Advanced Study Guide

Issuing body: American Society for Quality (ASQ) Jurisdiction: International (US-based) Level: Professional Prerequisites: 8 years on-the-job experience in CRE Body of Knowledge areas, including 3 years in a decision-making position (full-time, paid). Same education waivers as CQE (diploma 1 yr → doctorate 5 yrs, one waiver only). Prior ASQ certification experience (CQE, CQA, etc.) often counts toward the CRE experience requirement. ANAB-accredited under ISO/IEC 17024. Renewal: Every 3 years — 18 RUs or retake the exam.

1. How the credential is earned

1. Verify eligibility (8-year rule, 3 years decision-making, waivers as above).
2. Apply on asq.org; schedule the computer-based exam (Prometric). Format: 165 questions (150 scored + 15 unscored), 4.5 hours, open book, English only. (Per ASQ-published exam outlines via prep providers; confirm current format on asq.org — ASQ has been migrating exams between CBT/paper formats.)
3. Maintain: 18 RUs per 3-year cycle.

Exam strategy: the CRE is the most mathematical ASQ exam. Reliability math (distributions, system modeling, test planning) is the center of gravity — roughly half the paper rewards fluency with the equations in section 3. The exam is open book, so tab a reliability reference (the ASQ CRE Handbook's formula sections) and pre-solve every derivation below until it's reflex.

2. Knowledge domains

The CRE BoK spans reliability management, design for reliability/maintainability/safety, testing, data analysis, mathematical modeling, and life-cycle cost. Below is the theory organized the way the exam tests it.

2.1 Reliability mathematics — distributions

Reliability R(t) = P(T > t). Failure rate (hazard) h(t) = f(t)/R(t). The three distributions that carry the exam:

• Exponential: R(t) = e^(−λt), MTBF = 1/λ. Memoryless: P(T > t+s | T > s) = P(T > t). Derivation: constant hazard λ → dR/dt = −λR → R = e^(−λt). Use for: random (chance) failures, electronics in useful life, repairable-system inter-arrival times. Trap: applying it where wear-out dominates (β ≠ 1) — the exam gives Weibull data and tempts exponential.
• Weibull: R(t) = e^(−(t/η)^β), with shape β and characteristic life η (63.2% failed by t=η regardless of β). MTBF = η·Γ(1 + 1/β). β < 1: infant mortality (decreasing hazard); β = 1: exponential; β > 1: wear-out. Linearization for plotting: ln[−ln R(t)] = β ln t − β ln η — plot ln(−ln(1−F)) vs. ln t, slope = β, intercept gives η. This Weibull-plot technique is directly tested.
• Normal/lognormal: lognormal for fatigue, crack growth, repair times (right-skewed life data). Know when each applies; the exam's distribution-ID questions give failure-mode physics and expect the matching distribution.

Bathtub curve: infant mortality (β<1) → useful life (β≈1) → wear-out (β>1). Burn-in screens infant mortality; preventive replacement only pays in the wear-out region — replacing during useful life resets nothing (memoryless) and wastes money. That last sentence is an exam favorite.

2.2 System reliability modeling

• Series: R_sys = Π R_i. The chain rule — system reliability is always worse than the worst component. With n identical components: R^n.
• Parallel (active redundancy): R_sys = 1 − Π(1 − R_i). Derivation: system fails only if all fail; P(all fail) = Π(1−R_i) under independence.
• k-out-of-n: R = Σ_{i=k}^{n} C(n,i) R^i (1−R)^{n−i}.
• Standby redundancy: R = e^(−λt)(1 + λt) for one standby (perfect switching assumed — the exam tests that switching reliability must be modeled if imperfect).
• Independence assumption: all closed forms assume independent failures. Common-cause failures (shared environment, shared design flaw) break every formula — the exam's conceptual trap is the "redundant" system with a common power supply.
• Reliability apportionment/allocation: flow system requirements down to subsystems (equal apportionment, ARINC/AGREE methods); prediction (MIL-HDBK-217 parts-count/parts-stress for electronics — know it exists and its limits: it's a relative comparison tool, widely criticized as absolute prediction).

2.3 Maintainability and availability

• MTTR (mean time to repair), MTTF vs. MTBF (non-repairable vs. repairable — the exam enforces the distinction).
• Inherent availability: A_i = MTBF / (MTBF + MTTR). Achieved adds preventive maintenance; operational adds logistics/admin delay. Know which denominator each uses.
• Design for maintainability: modularity, diagnostics/BIT (built-in test), accessibility, standardization. The exam tests the design levers that move MTTR, not just the arithmetic.

2.4 Reliability testing

• Test types: qualification/DVT (design validation), HALT (highly accelerated life test — find margins by step-stress to destruction; it's qualitative, not a life predictor — exam trap), HASS (production screen), ALT (accelerated life test — quantitative, uses acceleration models), burn-in, ESS.
• Acceleration models: Arrhenius (temperature): AF = exp[(E_a/k)(1/T_use − 1/T_stress)] — know the form and that E_a (activation energy) is the critical, often-assumed parameter. Inverse power law (voltage/mechanical stress). Coffin-Manson (thermal cycling). The exam tests model selection and the danger of extrapolating beyond validated stress ranges.
• Test planning: success/failure (binomial) test design — demonstrate R at confidence C with n units and f allowable failures: 1 − C = Σ_{i=0}^{f} C(n,i)(1−R)^i R^{n−i}. Zero-failure test: n = ln(1−C)/ln(R). Reliability growth (Duane/AMSAA-Crow): plot cumulative MTBF vs. cumulative time on log-log; slope α is the growth rate; management strategy (test-fix-test vs. test-find-test) is examinable.
• Censoring: Type I (time-censored) vs. Type II (failure-censored); right-censoring is the norm in life data — the exam tests that censored units still contribute exposure time.

2.5 Data analysis and FMEA

• Weibull analysis of field/life data (probability plotting, rank regression vs. MLE — MLE preferred for censored data).
• FMEA/FMECA: Severity × Occurrence × Detection = RPN; FMECA adds criticality analysis. Design FMEA (before design freeze) vs. process FMEA. The exam's RPN traps mirror CQE's: severity-10 failure modes demand action regardless of RPN; RPN thresholds for action are company policy, not standard.
• Fault tree analysis: top-down Boolean logic (AND/OR gates); minimal cut sets — the smallest component-failure combinations that cause the top event. The exam tests cut-set identification on small trees and the insight that single-point cut sets are design weaknesses.
• FRACAS (failure reporting, analysis, corrective action system): closed-loop — every failure gets a report, analysis, and verified corrective action. The exam tests that FRACAS without closure verification is theater.

2.6 Reliability management, human factors, life-cycle cost

• Reliability program planning: the reliability program plan defines tasks by life-cycle phase (concept → design → test → production → field). Concurrent engineering integration.
• Human factors: human error probabilities, HRA basics, the exam's point — design for the actual user (mistake-proofing/poka-yoke reduces use error, a major field-failure driver).
• Life-cycle cost: LCC = acquisition + operating + maintenance + disposal. Reliability investments trade acquisition cost against operating/support cost — the exam's business-case questions reward total-LCC thinking over purchase-price thinking.
• Warranty analysis: warranty data is censored and biased (not all failures become claims) — Nevada charts (warranty triangles) track exposure properly.

3. Key equations

4. Worked examples

Example 1 — Weibull from data. 10 units tested; median-rank regression on Weibull paper gives β = 2.5, η = 8,000 h. R(5,000) = e^(−(5000/8000)^2.5) = e^(−(0.625)^2.5) = e^(−0.309) = 0.734. MTBF = 8000 × Γ(1.4) = 8000 × 0.887 = 7,096 h. β > 1 → wear-out; preventive replacement before ~8,000 h is justified.

Example 2 — System reliability. Three subsystems in series: R = 0.95, 0.97, 0.99 → R_sys = 0.912. Add active redundancy to the weakest (0.95 → 1 − 0.05² = 0.9975): R_sys = 0.9975 × 0.97 × 0.99 = 0.958. Lesson: redundancy on the weakest link buys the most — but check common cause first.

Example 3 — Zero-failure demonstration. Demonstrate R = 0.95 at 90% confidence: n = ln(0.1)/ln(0.95) = (−2.303)/(−0.0513) = 44.9 → 45 units, zero failures. Allowing 1 failure roughly doubles n — the exam tests the cost of the "one failure is okay" concession.

Example 4 — Arrhenius acceleration. E_a = 0.7 eV, T_use = 313 K (40°C), T_stress = 358 K (85°C). AF = exp[(0.7/8.617×10⁻⁵)(1/313 − 1/358)] = exp[8123 × (0.003195 − 0.002793)] = exp(3.266) = 26.2. 1,000 h at 85°C ≈ 26,200 h at 40°C. Trap: this assumes temperature is the only accelerating stress and the mechanism doesn't change — state the assumption.

Example 5 — Availability. MTBF = 2,000 h, MTTR = 4 h → A_i = 2000/2004 = 0.998. Adding a second technician cuts MTTR to 2 h → 0.999. Diminishing returns are visible immediately — the exam wants you to price the improvement against the availability gain.

5. Standards/regulations cross-reference

• MIL-HDBK-217 / 217Plus, Telcordia SR-332 — reliability prediction methods; know their role (relative comparison) and limits.
• MIL-STD-1629 — FMECA procedure; MIL-STD-882 — system safety (the safety-engineering sibling).
• IEC 61025 — fault tree analysis; IEC 60812 — FMEA procedure.
• ISO 9001 / AS9100 — reliability program plans live inside the QMS; AS9100's risk and configuration emphasis overlaps CRE practice.
• FDA QMSR / EU MDR — medical-device reliability: design validation, risk management file (ISO 14971), post-market surveillance.

6. Pitfalls and trap patterns

• Exponential by default. The exam hands you wear-out data hoping you'll use e^(−λt). Check β first.
• MTBF ≠ life. MTBF = 10,000 h does not mean the unit lasts 10,000 h — for exponential, only 37% survive to one MTBF. The exam exploits this mercilessly.
• HALT is not ALT. HALT finds design margins (qualitative); ALT predicts life (quantitative). Swapping them fails the question.
• Common cause. Every redundancy formula dies on shared failure modes — always scan for them before calculating.
• Preventive replacement in useful life. Memoryless means PM doesn't improve reliability there — only in wear-out (β>1).
• RPN thresholds. No standard sets them; "RPN > 100 requires action" is company policy, and the exam will present it as if it were universal.

7. Practice problems (answers at end)

1. Weibull β=1.8, η=5,000 h. Compute R(3,000) and state the failure regime.
2. Two components in series, R=0.98 each. What parallel redundancy on one component raises system R above 0.99?
3. Demonstrate R=0.90 at 80% confidence, zero failures. How many units?
4. MTBF=1,500 h, MTTR=6 h. Inherent availability? If MTTR halves, new A?
5. A test runs 2,000 h at 70°C to simulate 25°C operation (298 K → 343 K), E_a=0.8 eV. Compute AF and equivalent field hours.
6. Draw the minimal cut sets (list them) for: TOP = (A AND B) OR C.
7. Why is burn-in ineffective against wear-out failures?

Answers: 1. R = e^(−(0.6)^1.8) = e^(−0.399) = 0.671; β>1 → wear-out. 2. Current R_sys = 0.9604. Redundant component: 1−0.02² = 0.9996 → R_sys = 0.9996×0.98 = 0.9796. Still <0.99 — need redundancy on both or better components (shows the limit: 0.9996×0.9996 = 0.9992 ✓ with both redundant). 3. n = ln(0.2)/ln(0.9) = 15.3 → 16 units. 4. A = 1500/1506 = 0.99602; MTTR=3 → 1500/1503 = 0.99800. 5. AF = exp[(0.8/8.617e−5)(1/298 − 1/343)] = exp[9284×0.0004403] = exp(4.088) = 59.6 → ≈119,200 field hours. 6. Cut sets: {C}, {A,B} — C alone is a single-point failure. 7. Burn-in screens infant mortality (β<1 region); wear-out (β>1) hasn't started yet — burn-in consumes useful life without removing the wear-out mechanism.

Related credentials in this discipline

• CQE → asq-certified-quality-engineer-cqe-study-guide.md
• CQA → asq-certified-quality-auditor-cqa-study-guide.md
• Six Sigma Green/Black Belt → shared: ../shared/six-sigma-green-black-belt-study-guide.md

Original ETS reference writing, verified against public sources on 2026-09-30. Thin sourcing is flagged inline. Not legal advice — confirm current editions with the issuing body.