Reliability, Availability and Maintainability Engineering

Introduction

Production systems create value only when they perform their intended functions at the required rate, quality, safety level and cost. Maintenance therefore cannot be evaluated solely by the number of jobs completed or the amount spent on spare parts. A more useful engineering perspective asks three connected questions:

1. How often does the asset fail?
2. How quickly can it be restored?
3. How much of the required time is it capable of performing its function?

Reliability, maintainability and availability provide the principal answers. Together they connect equipment design, operating conditions, maintenance policy, spare-parts strategy and production planning. They also support decisions such as whether to redesign a component, introduce condition monitoring, hold a critical spare, increase preventive-maintenance frequency or accept a specified level of operational risk.

Reliability engineering is concerned with the probability that an item performs a specified function for a specified time under stated conditions. Maintainability concerns the probability that a failed item can be restored within a specified time. Availability combines failure behaviour and restoration behaviour to describe the readiness of an asset or system for service.

Learning objectives

After studying this chapter, the reader should be able to:

• distinguish reliability, maintainability and availability;
• interpret failure rate, hazard rate and the bathtub curve;
• select and apply exponential, Weibull and normal or lognormal models appropriately;
• calculate MTBF, MTTF, MTTR and common forms of availability;
• analyse series, parallel and standby reliability structures;
• use reliability information in maintenance and production decisions;
• identify the assumptions and limitations of reliability metrics.

1. Fundamental concepts and terminology

An item is any component, subsystem, machine or system under analysis. Its intended function must be stated precisely. For example, “a pump is reliable” is incomplete; a useful statement might be “the pump delivers at least 40 m³/h at 5 bar for 1,000 operating hours under the specified fluid and temperature conditions.”

A failure is the termination of an item's ability to perform a required function. A fault is a condition that may lead to failure but does not necessarily mean that the function has already been lost. Failures may be complete or partial, sudden or gradual, independent or dependent, and evident or hidden. A safety valve that fails to open on demand has a hidden failure until the demand occurs.

Reliability is denoted by R(t), the probability of survival to time t. If T is the random variable representing time to failure, then:

The cumulative distribution function is F(t) = P(T ≤ t), so:

Reliability is dimensionless and lies between 0 and 1. It is conditional on the population, design, environment, duty cycle and definition of failure.

The probability density function f(t), when it exists, describes the distribution of failure times. The hazard or instantaneous failure rate is:

Its unit is inverse time, such as failures per hour. Hazard is not simply the proportion of items that fail in a period; it is the instantaneous conditional rate of failure among items that have survived to time t.

The cumulative hazard is:

and therefore:

These relationships assume a well-defined time scale and a valid probability model. Time may be calendar time, operating hours, cycles, starts or another exposure measure. The selected measure must match the physical failure mechanism.

Reliability, maintainability and availability

Maintainability is commonly represented by M(t), the probability that corrective maintenance is completed within time t. If Y is the repair-time random variable:

For a repairable item, mean time to repair is the expected corrective restoration time. It may include diagnosis, access, disassembly, repair, testing and return to service, depending on the declared boundary.

Availability is the probability that an item is capable of performing its function when required. The term has several forms:

• Inherent availability considers corrective maintenance and, commonly, active repair time and required corrective spares, but excludes logistics and preventive-maintenance delays.
• Achieved availability includes preventive-maintenance time as well as corrective maintenance, but generally excludes administrative and supply delays.
• Operational availability includes the full operating context: preventive and corrective maintenance, logistics, waiting, staffing and support delays.

The distinction is essential. A machine may have excellent inherent availability but poor operational availability because technicians, permits or spare parts are unavailable.

2. Failure-rate behaviour and the bathtub curve

The hazard function provides a useful engineering description of changing failure propensity. A traditional bathtub curve contains three regions:

1. Infant mortality: a decreasing failure rate caused by design defects, manufacturing variation, installation errors or commissioning problems.
2. Useful life: an approximately constant failure rate, often associated with random external stresses.
3. Wear-out: an increasing failure rate caused by fatigue, corrosion, erosion, insulation ageing or other degradation mechanisms.

The bathtub curve is a conceptual model, not a universal law. Many electronic components have a long constant-rate period, whereas bearings, cutting tools and seals may show strong wear-out. Modern products may also display multiple competing failure modes, making the aggregate curve different from any individual mode.

Figure: Conceptual bathtub curve. The boundaries between regions are not fixed and depend on the item and failure mechanisms.

A key managerial implication is that preventive replacement is not automatically beneficial. If failures are random and the hazard is approximately constant, replacing an item early may not reduce failure probability and can introduce installation errors. Preventive replacement is more defensible when degradation is measurable or when the wear-out hazard rises sufficiently to offset replacement cost and risk.

3. Reliability distributions

Exponential distribution

The exponential model assumes a constant failure rate λ. It is often suitable for the useful-life region when failures are independent and memoryless:

The mean time to failure for a non-repairable item is:

For a repairable item under stable conditions, the corresponding mean time between failures is often approximated by 1/λ. The approximation requires consistent operating conditions, complete failure records and a clear definition of the interval being measured.

Example 1. A component has a constant failure rate of 0.0002 failures per operating hour. Its reliability at 1,000 hours is:

R(1,000) = e^(−0.0002 × 1,000) = e^(−0.2) ≈ 0.8187.

Thus, approximately 81.9% of comparable components are expected to survive 1,000 hours under the stated assumptions. Its MTTF is 1/0.0002 = 5,000 hours. This is a population expectation, not a guarantee that each component will operate for 5,000 hours.

Weibull distribution

The Weibull model is flexible because its hazard can decrease, remain constant or increase. With shape parameter β and scale parameter η:

The hazard is:

Here β is dimensionless and η has the same time unit as t. β < 1 indicates decreasing hazard, β = 1 gives the exponential distribution, and β > 1 indicates increasing hazard. The mean time to failure is:

where Γ is the gamma function. Weibull analysis is particularly useful for distinguishing early-life and wear-out behaviour, but parameter estimates can be unstable with small samples, censored observations or mixed failure modes.

Normal and lognormal models

A normal model may describe variation around a nominal life when negative times are not practically relevant and the spread is small relative to the mean. A lognormal model is often more plausible for repair times or degradation processes whose logarithms are approximately normally distributed. Distribution selection should be based on engineering reasoning, graphical diagnostics and statistical goodness-of-fit—not solely on the best numerical fit.

Important data issues include right-censored units that have not failed, left truncation caused by observing assets only after commissioning, changing operating conditions, and competing risks. Ignoring these issues can produce biased failure rates and poor maintenance intervals.

4. MTBF, MTTF, MTTR and availability

For a non-repairable item, MTTF is the expected time to the first failure. For a repairable item, MTBF is the average time between successive failures. MTBF must be defined carefully: some organizations measure operating time from restoration to the next failure, while others include downtime in the interval. These definitions are not interchangeable.

Mean time to repair (MTTR) is the average active repair time. A practical decomposition is:

If the terms are averaged over the same failure population, the equation is additive. Waiting time for parts or approval should be reported separately or included explicitly in operational restoration time.

Under a steady-state repairable model with constant rates and no logistics delay, inherent availability is:

Equivalently, if failure rate λ and repair rate μ are constant:

where μ = 1/MTTR. These formulas assume alternating up and down states, statistically stable behaviour and mutually exclusive operating and repair states.

Example 2. A packaging machine has MTBF = 720 operating hours and MTTR = 8 hours. Then:

Aᵢ = 720/(720 + 8) = 0.9890, or 98.90%.

If an average of 4 hours is spent waiting for a spare and 2 hours waiting for authorization per failure, operational downtime becomes 14 hours per cycle. A simple operational estimate is 720/(720 + 14) = 98.09%. The difference demonstrates why maintenance departments should not claim high availability using a narrow metric while production experiences long delays.

For a finite observation period, achieved availability can be calculated directly:

This is a descriptive measure. It does not by itself explain whether downtime resulted from frequent failures, long repairs, planned maintenance or supply delays.

5. System reliability models

System reliability depends on how component success contributes to system success. The following models assume component events are statistically independent unless stated otherwise.

Series systems

In a series system, every component must function. If component reliabilities are R₁, R₂, …, Rₙ:

For two components with reliabilities 0.98 and 0.97, system reliability is 0.98 × 0.97 = 0.9506. The system is less reliable than either component. Series analysis identifies critical components and encourages design simplification, higher-quality parts or condition monitoring.

Parallel systems

In an active parallel system, the system functions if at least one component functions. For independent components:

Two identical components each with reliability 0.90 give Rₚ = 1 − (0.10 × 0.10) = 0.99. This benefit is achieved only if the standby or redundant path is genuinely available and common-cause failures are controlled.

k-out-of-n systems

A k-out-of-n system succeeds when at least k of n components operate. For identical independent components with reliability R:

where C(n,j) is the number of combinations of j functioning components. The formula does not apply directly when components have unequal reliabilities, dependencies or load-sharing effects.

Standby and common-cause limitations

In a standby system, a reserve component operates only after the primary fails. Reliability depends on switching success, standby state, detection and the reserve's exposure to ageing. A simple parallel formula may overstate benefit because a common power supply, control system, environment or maintenance error can defeat both channels. Fault-tree analysis, reliability block diagrams and failure-mode, effects and criticality analysis (FMECA) are useful tools for exposing such dependencies.

6. Maintainability engineering and maintenance decisions

Maintainability should be designed, not merely measured after failure. Design features include accessible components, modular replacement, standardized fasteners, diagnostic interfaces, error-proof connectors, lifting provisions, isolation points and safe work instructions. Maintainability is affected by human factors, lighting, ergonomics, contamination, documentation and permit systems as much as by the physical machine.

If repair time follows an exponential distribution with repair rate μ, the probability of completing repair by time t is:

Since μ = 1/MTTR, this model is appropriate only when repair times are approximately memoryless and conditions are stable. For non-exponential repair times, empirical distributions or lognormal, gamma or Weibull models may be more suitable.

A useful decision framework is:

1. Define the required function and consequence of failure.
2. Separate failure modes and identify their physical causes.
3. Estimate failure and repair distributions from credible data.
4. Determine whether a failure is age-related, condition-related or random.
5. Compare corrective, time-based, condition-based and redesign options.
6. Include production loss, safety, environmental effects, spare parts and labor.
7. Validate the policy through performance feedback and updated data.

Preventive replacement is justified when failure consequences are high and age or condition predicts failure. Condition-based maintenance is preferred when a measurable indicator—such as vibration, oil debris, temperature or electrical signature—provides sufficient warning. Run-to-failure can be rational for low-consequence, inexpensive and readily replaceable items, but it is not a substitute for risk assessment.

7. Applied case: illustrative centrifugal-pump fleet

Background

A process plant operates ten identical cooling-water pumps. Each pump is required for production, but one installed standby pump can support the process during a single pump outage. Historical records show 18 pump failures in 36,000 operating pump-hours. Average active repair time is 6 hours, while logistics and authorization add an average of 5 hours.

Problem

Management must decide whether to purchase additional spares and introduce vibration monitoring. Production has treated all downtime as MTTR, making the reported maintenance performance difficult to interpret.

Analysis

The observed failure-rate estimate is:

λ̂ = 18/36,000 = 0.0005 failures per operating hour.

The corresponding observed MTBF is 36,000/18 = 2,000 operating hours. Active inherent availability is approximately:

Aᵢ = 2,000/(2,000 + 6) = 0.9970, or 99.70%.

Including logistics and authorization, the restoration interval is 11 hours and the operational estimate becomes:

Aₒ = 2,000/(2,000 + 11) = 0.9945, or 99.45%.

The numerical difference appears small, but across ten pumps and a high-throughput process, the lost production hours can be economically significant. Moreover, the standby arrangement may not provide full protection if the standby pump is unavailable or if both pumps share a suction line, power source or control system.

The plant examines failure records and finds that 12 of the 18 failures involved bearing vibration and rising temperature. The remaining six were electrical or seal-related. A Weibull review of bearing failures indicates an increasing hazard after approximately 18,000 shaft-hours, although the sample is small and confidence limits are wide.

Approach and solution

The plant adopts a staged policy:

• install online vibration and temperature monitoring on the most critical pumps;
• inspect the standby pump weekly and run a scheduled functional test;
• hold a critical bearing and seal kit locally;
• separate active repair, waiting and production-loss time in the computerized maintenance management system;
• review common-cause exposure in power, suction and control systems;
• replace bearings based on condition and verified degradation rather than an arbitrary calendar interval;
• conduct a quarterly reliability review using censored data and failure-mode coding.

Outcome

After six months, the plant reports active MTTR separately from logistics delay, enabling targeted action. Local kits reduce average waiting time, while condition monitoring identifies two deteriorating bearings before functional failure. The organization does not claim that the observed improvement proves the monitoring system caused every gain; it treats the result as preliminary evidence and continues the evaluation with exposure-adjusted measures.

Learning points

The case demonstrates that availability improvement may come from reducing repair duration, reducing waiting time, preventing failure, or improving redundancy. These interventions are not substitutes for one another. It also shows why a single MTBF number cannot determine a maintenance policy without failure-mode, consequence and data-quality analysis.

8. Limitations, governance and contemporary practice

Reliability metrics can be misleading when failure definitions change, minor stoppages are omitted, operating hours are estimated, or repaired items are treated as new without accounting for age. MTBF can increase because failures are underreported, not because equipment improved. Availability can rise while safety risk worsens if protective devices are bypassed or inspections are deferred.

Reliability-centred maintenance (RCM) links functions, functional failures, failure modes, consequences and technically feasible tasks. FMECA ranks failure modes using severity, occurrence and detectability, although numerical risk-priority numbers should not be treated as objective truth. Fault-tree analysis reasons deductively from a top event, while event-tree analysis explores consequences following an initiating event.

Digital condition monitoring, industrial internet-of-things platforms and machine-learning models can improve early-warning capability. However, predictive models require representative data, stable sensor quality, traceable labels and control of model drift. A model that predicts “normal” because failures are rare may create false confidence. Human review, alarm governance, cybersecurity, access control and audit trails are therefore essential. Maintenance decisions affecting pressure equipment, lifting systems, safety functions or environmental controls must comply with applicable laws, standards and organizational authorization requirements.

Best practice is to:

• define metrics and boundaries before collecting data;
• use operating exposure rather than calendar time where appropriate;
• distinguish failure, fault, inspection finding and planned replacement;
• report uncertainty, sample size and censoring;
• investigate common-cause and human-factor contributors;
• connect reliability metrics to production, safety and lifecycle cost;
• review models when design, duty cycle or environment changes.

Chapter Summary

Reliability is the probability that an item performs its required function for a specified time; maintainability is the probability of restoring it within a specified time; and availability is its readiness for service. Reliability is represented through survival, failure distributions and hazard functions. The bathtub curve provides a useful but simplified view of infant mortality, useful life and wear-out.

The exponential model describes constant hazard, while the Weibull model accommodates decreasing, constant and increasing hazard through its shape parameter. MTTF applies primarily to non-repairable items, MTBF to repairable failure intervals, and MTTR to restoration time. Under stated steady-state assumptions, inherent availability is MTBF divided by MTBF plus MTTR; operational availability must also consider logistics and organizational delays.

Series systems multiply component reliabilities, parallel systems gain redundancy, and k-out-of-n models represent voting or capacity structures. Independence assumptions, switching reliability and common-cause failures must be examined before claiming a redundancy benefit. Maintenance policy should be selected from failure mechanisms, consequences, data quality and lifecycle cost—not from MTBF alone. Digital monitoring can support predictive maintenance, but governance, cybersecurity, human judgment and regulatory compliance remain essential.

Key Terms

Reliability: Probability that an item performs its required function for a specified time under stated conditions.

Failure: Termination of the ability to perform a required function.

Fault: An abnormal condition that may lead to failure.

Failure rate: Frequency of failure relative to exposure time.

Hazard rate: Instantaneous conditional rate of failure among surviving items.

Bathtub curve: Conceptual curve showing decreasing, constant and increasing failure-rate regions.

MTTF: Mean time to failure of a non-repairable item.

MTBF: Mean time between failures of a repairable item.

MTTR: Mean active time required to restore an item.

Maintainability: Probability of completing restoration within a specified time.

Availability: Probability that an item is capable of performing its function when required.

Inherent availability: Availability based mainly on failure and active corrective repair time.

Operational availability: Availability including logistics and organizational delays.

Exponential distribution: Reliability distribution with constant hazard rate.

Weibull distribution: Flexible distribution whose hazard may decrease, remain constant or increase.

Series system: System that fails when any required component fails.

Parallel system: System that operates when at least one redundant component operates.

Common-cause failure: Failure of multiple elements due to a shared cause.

Censoring: Observation in which the exact failure time is not known, often because the unit survives observation.

Reliability-centred maintenance: Structured method linking functions, failure modes, consequences and maintenance tasks.

Review Questions

Short Answer Questions

1. Define reliability, maintainability and availability, and explain how they differ.
2. What are the three regions of the bathtub curve?
3. State the assumptions underlying the exponential reliability model.
4. Distinguish MTTF from MTBF.
5. Why can operational availability be lower than inherent availability?

Long Answer Questions

1. Explain the reliability function, failure distribution, hazard rate and cumulative hazard, showing how they are mathematically related.
2. Compare exponential and Weibull models, including their parameters, interpretations and maintenance applications.
3. Derive and discuss the reliability expressions for independent series and parallel systems.
4. Explain how maintainability engineering influences equipment design and maintenance strategy.
5. Critically evaluate the use of MTBF as a performance and decision-making metric.

Analytical / Application Questions

1. A repairable machine has an MTBF of 1,200 hours and an active MTTR of 10 hours. Calculate inherent availability. If average logistics delay is 8 hours per failure, calculate a simple operational availability estimate and interpret the difference.
2. Three independent components in series have reliabilities 0.99, 0.98 and 0.96 for a specified mission. Calculate system reliability and identify one engineering action that could improve it.
3. A plant observes decreasing failures during commissioning, followed by a constant rate and then increasing bearing failures. Propose a maintenance and data-collection policy for each region, and state the risks of applying one policy to all three regions.

Multiple Choice Questions with Answers

Question 1

Which expression defines reliability at time t?

A. P(T ≤ t)

B. P(T > t)

C. P(Y > t)

D. 1/P(T > t)

Correct Answer: B. P(T > t)

Question 2

A constant failure rate of 0.0002 failures per hour implies an exponential-model MTTF of:

A. 200 hours

B. 2,000 hours

C. 5,000 hours

D. 20,000 hours

Correct Answer: C. 5,000 hours

Question 3

For two independent components in series, with reliabilities 0.98 and 0.97, system reliability is:

A. 0.9506

B. 0.9700

C. 0.9800

D. 0.9994

Correct Answer: A. 0.9506

Question 4

For two independent parallel components, each with reliability 0.90, system reliability is:

A. 0.81

B. 0.90

C. 0.91

D. 0.99

Correct Answer: D. 0.99

Question 5

A machine has MTBF 720 hours and MTTR 8 hours. Its inherent availability is approximately:

A. 88.9%

B. 98.90%

C. 99.89%

D. 90.00%

Correct Answer: B. 98.90%

Question 6

In a Weibull model, β > 1 generally indicates:

A. Decreasing hazard

B. Constant hazard

C. Increasing hazard

D. Zero hazard

Correct Answer: C. Increasing hazard

Question 7

Which activity is normally included in operational availability but excluded from a narrow inherent-availability calculation?

A. Equipment design

B. Logistics waiting for a spare

C. Component reliability

D. Required operating function

Correct Answer: B. Logistics waiting for a spare

Question 8

A system that operates when at least k of n components function is called a:

A. Series system

B. Single-point system

C. k-out-of-n system

D. Non-repairable system

Correct Answer: C. k-out-of-n system

Question 9

Which condition most strongly supports condition-based maintenance?

A. A measurable indicator provides warning of degradation

B. Failure is completely random and unobservable

C. The asset has no consequence of failure

D. Repair time is always zero

Correct Answer: A. A measurable indicator provides warning of degradation

Question 10

Why can a simple parallel reliability formula overstate redundancy benefits?

A. Parallel systems can never improve reliability

B. Component reliability is always zero in parallel systems

C. It ignores possible common-cause failures and switching failures

D. It applies only to non-engineering systems

Correct Answer: C. It ignores possible common-cause failures and switching failures