A component can perform perfectly in one test and still fail the production line when a shared utility goes down. Industrial reliability needs a defined mission, realistic operating evidence and a valid model of dependencies.
The Gold Standard of Industrial Reliability Engineering separates uninterrupted success from availability, compares failure modes and repair capacity, and examines what redundant design really protects. The approach favours reproducible evidence over promises.
This guide contains fictional exercises for observed failure rates, MTBF, series and parallel reliability and downtime costs. All examples declare their assumptions and should not be applied to certify a real plant.
A practical reliability question often involves a costly disagreement: procurement reports a predicted component MTBF, operations reports outages, maintenance reports short repair durations and a customer reports incomplete service. These accounts may all be individually accurate because they use different boundaries. A useful investigation begins by writing down the service being promised, its required demand, the clock used to measure it and the event that constitutes a failure. Only then should the engineer choose between a failure-rate summary, mission reliability, availability or a restoration-time distribution.
How to read the worked examples: all equipment, loads, failure counts, probabilities and currency values are invented. The series and parallel calculations assume particular forms of independence and adequate surviving capacity, while field-rate calculations describe a past selected population. Every example is an invitation to test what happens when the assumptions change, not a recommended safety target. In a real industrial setting, qualified personnel must use the applicable design standards, test methods, plant records and approval processes before selecting equipment or changing a safety-critical configuration.
Chapter index
Requirements and probability models · Chapters 1–17
- Reliability starts with a declared mission
- Reliability and availability answer different questions
- Maintainability affects lost service
- Asset management provides the organisational context
- Define failure in functional language
- Choose the right system boundary
- Failure mode, symptom and cause must be distinguished
- Load and environment belong to the data
- Failure counts require exposure
- Worked example: two exposure-adjusted rates
- MTBF is not a guaranteed lifetime
- Worked example: MTBF and corrective downtime
- Availability has more than one valid denominator
- Reliability block diagrams model dependencies
- Worked example: two components in series
- Redundancy needs an alternative capable of working
- Worked example: ideal independent parallel units
Failures, stress, software and people · Chapters 18–30
- Common-cause failure can defeat separate equipment
- FMEA links failure modes to their effects
- Avoid treating a risk-priority product as a probability
- Fault trees start with the unwanted outcome
- Constant failure rate is a model assumption
- Weibull models allow different failure patterns
- Survivors in tests carry useful information
- Accelerated stress can change the mechanism
- Environmental and supplier variations belong in analysis
- Software defects are often correlated
- Human performance belongs in system design
- Reliability growth should follow identifiable changes
- Preventive maintenance can both help and harm
Maintenance, economics and proof · Chapters 31–39
- Restoration delay is an engineering outcome
- Reliability investment requires whole-life comparison
- Worked example: direct cost of disruption
- Procurement needs measurable acceptance criteria
- Change control preserves the scope of previous tests
- Averages can hide high-consequence events
- Audit one complete mission under degraded conditions
- Frequently asked questions on industrial reliability
- Continue through connected Gold Standard articles
Quick comparison: dependability properties
| Property | Question |
|---|---|
| Reliability | Can the function survive its mission? |
| Availability | Can it operate when required? |
| Maintainability | Can it be restored? |
| Redundancy | Can an alternative meet demand after failure? |
1. Reliability starts with a declared mission
Industrial reliability measures whether an asset performs a required function under specified conditions for a defined period. A motor can run while its conveyor fails to move material at the required rate. The gold standard is a testable claim about the real service, not a generic product adjective.
State the mission duration, demand, operating conditions, acceptable degradation and the precise event that counts as failure. A supplier’s component prediction is not automatically a prediction for the installed system.
2. Reliability and availability answer different questions
Reliability concerns continued mission success without the specified failure. Availability concerns whether the service is usable when required, including any restoration after failure. A machine that fails frequently but recovers rapidly may show good availability but poor uninterrupted reliability.
A dashboard reporting uptime cannot replace failure-count and user-outcome measures. Define whether planned maintenance, partial service and legitimate protective shutdowns are included.
3. Maintainability affects lost service
Maintainability is the ability to restore function under specified repair conditions with available people, spares and procedures. A component that seldom fails but needs a month for replacement can threaten essential operations.
Design access, diagnostics, controlled configuration backups and suitable recovery arrangements from the beginning. Mean repair time and full calendar downtime are not always the same quantity.
4. Asset management provides the organisational context
ISO 55001:2024 specifies requirements for an asset-management system linking objectives, lifecycle decisions, value and risk. It is not a universal component test or a certificate that one machine can never fail.
Use the framework to decide which functions merit investment and evidence. A written asset policy without corresponding inspection, operation and decision records is weak assurance.
5. Define failure in functional language
Failure can include complete stoppage, insufficient flow, inaccurate output or loss of a necessary protective function. The same physical equipment can meet one use case while failing a more demanding one.
Write acceptance criteria before testing. A healthy motor indicator cannot prove the manufactured product met specification, and a correctly actuated safety shutdown is not necessarily evidence of failed safety.
6. Choose the right system boundary
Draw dependencies from equipment through power, controls, network, cooling, operators and downstream steps. Components can meet individual claims while the complete service fails because of a single missing dependency.
The boundary for a machine vendor differs from that of a factory owner. A reliable report tells readers exactly which services and subsystems were counted.
7. Failure mode, symptom and cause must be distinguished
A vibration alert is an observation; loss of bearing function is one possible failure mode; contamination may be one cause. Several causes can produce similar symptoms, and one cause can create several symptoms.
Record evidence first and assign causes only when supported. This reduces circular claims in which an operator’s initial guess is later treated as confirmed truth.
8. Load and environment belong to the data
Machine reliability can change with starts, speed, operating temperature, contamination, moisture and variation in load. Two identical model numbers are not necessarily comparable if duty differs.
Store operating context and date with incident counts. Failure statistics from one laboratory or site do not automatically transfer to a more severe environment.
9. Failure counts require exposure
An asset with five failures in 500 hours has a different observed frequency from an asset with five failures in 5,000 hours. Appropriate denominators may include operating hours, cycles, starts or units produced.
Use the measure linked to the mechanism and define it before comparing populations. Calendar months alone may be misleading when production schedules differ.
10. Worked example: two exposure-adjusted rates
A fictional line A records twelve failures over 24,000 operating hours: 12 ÷ 24,000 × 1,000 = 0.5 failures per 1,000 hours. Line B has nine failures over 9,000 hours, giving 1.0 failure per 1,000 hours.
Raw count would rank A worse, while exposure-adjusted observation ranks it better. Neither proves a causal difference or forecasts the next failure without studying duty, age, coding and uncertainty.
11. MTBF is not a guaranteed lifetime
Mean time between failures for a repairable population is often calculated as total operating time divided by defined failures. It is a summary of the observed or modelled population under stated rules, not the minimum time until breakdown.
Check whether records refer to failures, repairs or restarts and whether the machine returned to an as-good-as-new state. These distinctions affect interpretation.
12. Worked example: MTBF and corrective downtime
Suppose a fictional repairable unit accumulates 1,980 running hours, six breakdowns and 54 hours of corrective downtime. Observed MTBF is 1,980 ÷ 6 = 330 hours, and mean corrective downtime is 54 ÷ 6 = 9 hours.
Under the selected alternating up/down convention, observed availability is 1,980 ÷ (1,980 + 54) ≈ 97.35%. This is not future reliability or a claim that planned downtime was counted.
13. Availability has more than one valid denominator
Availability may be measured over scheduled production, all calendar time, service demand or successful transactions. An attractive percentage can conceal an excluded maintenance period or users who could not complete a real task.
State numerator, denominator, time window, operational boundary and treatment of degraded output. Never compare suppliers using different conventions without adjusting the basis.
14. Reliability block diagrams model dependencies
Functional block diagrams show which combinations of successful components permit the service outcome. Series configurations require all essential blocks; parallel paths may succeed when one alternative remains.
IEC 61078:2016 documents reliability-block modelling and its limitations. A diagram is not evidence that failure probabilities are measured or independent.
15. Worked example: two components in series
Assume fictional blocks A and B must both survive the same mission, with independent success probabilities of 0.98 and 0.97. Simple series reliability is 0.98 × 0.97 = 0.9506, or 95.06%.
If the two functions share a power dependency or their claims cover different time intervals, the multiplication may not represent the actual system. Independence must be justified.
16. Redundancy needs an alternative capable of working
Installing two pumps is not enough if one alone cannot handle the required demand, both share an unprotected utility or both rely on the same defective controls. Functional redundancy must be demonstrated under credible failure.
Test the surviving capacity and changeover conditions safely. A green backup signal is not proof that the user still received correct output.
17. Worked example: ideal independent parallel units
Suppose two fully capable alternative units each independently fail a defined mission with probability 0.02. If either can meet demand, the ideal combined failure probability is 0.02 × 0.02 = 0.0004, implying 99.96% success.
Shared power, software or maintenance conditions can invalidate independence. This teaching calculation is not a functional-safety certification or approved reliability requirement.
18. Common-cause failure can defeat separate equipment
Flooding, unstable utilities, shared credentials, faulty software or one mistaken maintenance procedure can disable several nominally independent units. These common causes should be explicitly investigated instead of ignored in a multiplication model.
Identify where power, networking, physical environment and personnel decisions overlap. A redundancy claim is credible only when its major shared dependencies are known.
19. FMEA links failure modes to their effects
Failure mode and effects analysis examines how items or processes can fail, what follows and how risks can be addressed. It may cover hardware, software, tasks and interfaces.
IEC 60812:2018 explains FMEA and FMECA. It is a general dependability method, not proof that every safety or regulatory obligation has been met.
20. Avoid treating a risk-priority product as a probability
Some FMEA methods multiply ordinal severity, occurrence and detection ratings. The output helps structure discussion but is not necessarily a statistically meaningful risk probability.
Preserve high-severity cases as separately visible decisions. An arbitrary score should not permit a serious unresolved safety hazard to disappear inside averages.
21. Fault trees start with the unwanted outcome
Fault-tree analysis explores logical combinations leading to a specified event. It can expose failure pathways omitted in a component-by-component review.
Check whether the underlying causal model is reasonable before estimating probabilities. The most detailed diagram still requires real operating evidence and review of dependencies.
22. Constant failure rate is a model assumption
An exponential reliability model assumes a constant hazard rate within the relevant conditions. Real equipment can experience infant failures, wear-related deterioration or changing risks after maintenance.
Do not turn an observed MTBF into an unqualified future failure probability. Model selection and parameter uncertainty should be reported alongside calculations.
23. Weibull models allow different failure patterns
Weibull distributions can represent a variety of lifetime hazard shapes under their assumptions. They are useful when there is enough representative life data and an identified population.
Censoring, changing duty and mixed failure mechanisms complicate fitting. A smooth survival curve cannot predict the exact failure day of one named machine.
24. Survivors in tests carry useful information
An item still working at the observation cutoff provides survival evidence even though its failure time is unknown. Ignoring these right-censored cases and averaging only broken units can bias conclusions.
Report test entry, exit, repairs and removals. Suitable survival-analysis methods are preferable to discarding data that do not contain a convenient failure date.
25. Accelerated stress can change the mechanism
Testing under more heat, load or cycling may shorten the time to observe a degradation process. But excessive stress can produce damage unlike the intended service condition.
Validate that the observed physical failure mode transfers before extrapolating lifetime. A high-temperature chamber hour is not universally equal to a prescribed number of years in use.
26. Environmental and supplier variations belong in analysis
Components from different batches, suppliers or installations may differ. Dust, humidity, vibration and duty can alter wear and electrical performance, while repair quality can affect repeat failures.
Use configuration and lot identifiers with test results. An aggregate failure rate that hides the affected batch may misdirect the corrective action.
27. Software defects are often correlated
Two controllers running the same incorrect program can fail on the same input even when their electronics are separate. Duplicating equipment does not automatically protect against systematic requirements or coding errors.
Version and test software, interfaces and error recovery with representative cases. Keep controlled rollback and independent review where the consequences warrant it.
28. Human performance belongs in system design
Operators, technicians and managers work under workload, time pressure and information constraints. Inadequate displays, misleading alarms and poorly timed procedures can create predictable errors.
Analyse the task and environment rather than blaming the last person who touched the asset. Good design supports detection and safe correction.
29. Reliability growth should follow identifiable changes
Prototype improvements may reduce failures through better design, materials or controls. Evidence is stronger when every configuration version and test is recorded and failure modes are compared consistently.
Do not combine early and late builds indiscriminately into one improving trend. A claimed fix needs representative retest and a way to detect recurrence.
30. Preventive maintenance can both help and harm
Appropriate inspection, lubrication and replacement may control known failure mechanisms. Unnecessary disassembly, contamination or incompatible spares can create new problems.
Connect interventions to proven mechanisms and observe the subsequent duty. Industrial Maintenance explores the practical work management.
31. Restoration delay is an engineering outcome
Full downtime can involve diagnosis, authorisation, procurement, repair, verification and return to service. The time spent actively replacing a part is often only one portion.
Capture the stages to identify the limiting factor. A redesigned spare-parts process may improve service more than increasing bench reliability by a small theoretical amount.
32. Reliability investment requires whole-life comparison
Redundancy, monitoring and redesign can impose new costs while reducing lost output and repairs. Economic evaluation should include the relevant time horizon, service demand and uncertainty.
Avoid claiming that modelled savings are guaranteed; expected financial outcomes do not override safety obligations. A decision should retain the assumptions and evidence behind the forecast.
33. Worked example: direct cost of disruption
Suppose a fictional line has five breakdowns annually, each causing six hours of lost contribution at S$1,500 per hour. The simplified annual consequence is 5 × 6 × 1,500 = S$45,000.
If a proposed change hypothetically reduces breakdowns to two at unchanged duration, modelled loss becomes S$18,000, a S$27,000 difference before intervention costs. Those frequencies are illustrative, not actual forecasts.
34. Procurement needs measurable acceptance criteria
Specify required output, environment, mission period, fault handling, diagnostics, maintenance support and system-level acceptance tests. Ask suppliers what their reliability figures include.
Marketing descriptions such as heavy-duty or industrial grade do not independently establish performance. Compare verified evidence against the actual installation and its users.
35. Change control preserves the scope of previous tests
A new load, sensor, software version, supply arrangement or material can invalidate an earlier reliability demonstration. Document the change and decide which acceptance evidence needs renewal.
Retest under the changed configuration rather than assuming the previous percentage remains true. Configuration traceability enables a meaningful comparison of new and old behaviour.
36. Averages can hide high-consequence events
A good mean failure rate may coexist with a rare fault that causes severe safety or production effects. Segmentation by mode, function and consequence reveals risks that a single target obscures.
Track critical unresolved problems separately. One serious weakness cannot be offset merely by many successful routine cycles.
37. Audit one complete mission under degraded conditions
Select an essential outcome and trace functional dependencies, field failure records, exposure, recovery and changes. Rebuild the key calculations from original evidence and check assumptions.
Then test conceptually whether one utility or control failure defeats multiple blocks. A gold-standard argument is reproducible and remains candid when conditions depart from the ideal model.
38. Frequently asked questions on industrial reliability
Does MTBF guarantee a minimum operating period? No. Is availability the same as reliability? No. Does a second machine necessarily provide resilience? No. Can common causes defeat independent-looking equipment? Yes.
Does an FMEA score equal a failure probability? No. Do laboratory accelerated tests guarantee a specific lifetime? No. Is successful repair proof the original failure mode has been removed? Not without verification.
39. Continue through connected Gold Standard articles
Study Industrial Maintenance, Industrial Tribology, Industrial Automation, Industrial Materials Testing, Industrial Process Safety and Instrumentation Engineering.
Primary references: ISO 55001:2024, IEC 61078:2016 and IEC 60812:2018. The calculations are fictional education, not acceptable machine failure rates or certification.