Reliability Engineering and Maintenance Strategy Selection

Introduction

Total Productive Maintenance (TPM) seeks to maximize equipment effectiveness across the whole life cycle. Reliability engineering provides the analytical basis for deciding how that objective should be achieved. Maintenance is not automatically beneficial: preventive work can introduce human error, consume scarce resources, and replace components that still have substantial useful life. Conversely, insufficient intervention can create safety incidents, quality losses, environmental releases, and expensive production interruptions.

Reliability engineering therefore connects three questions:

1. What can fail, and how is failure generated?
2. What are the consequences and probabilities of those failures?
3. Which maintenance policy provides an acceptable level of risk at the lowest life-cycle cost?

This chapter develops failure-mode analysis, bathtub curves, Weibull methods, FMEA/FMECA, fault-tree analysis, reliability block diagrams, Reliability-Centered Maintenance (RCM), and risk-based maintenance (RBM). The emphasis is not on applying a single tool mechanically, but on integrating evidence, engineering judgment, economics, safety, and operational context.

Learning objectives

By the end of this chapter, the reader should be able to:

• distinguish reliability, availability, maintainability, and safety;
• interpret the bathtub curve and Weibull parameters without overgeneralizing them;
• construct and prioritize FMEA/FMECA studies;
• analyze combinations of failures using fault trees and reliability block diagrams;
• apply the logic of RCM and RBM to select maintenance tasks;
• compare corrective, preventive, predictive, condition-based, redesign, and run-to-failure policies;
• evaluate maintenance decisions using risk, availability, and life-cycle cost;
• recognize data, human, ethical, and governance limitations in reliability decisions.

1. Reliability as a system and business property

Reliability is the probability that an item performs its required function, under stated conditions, for a specified time interval. If T is mission time and T_f is time to failure, reliability is:

where R(t) is dimensionless and t is elapsed operating time. The definition requires a stated function, operating environment, duty cycle, and acceptable performance level. A pump that delivers 100 m³/h at 5 bar may be reliable for one service and unreliable for another.

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

The probability density function, when it exists, is f(t) = dF(t)/dt. The hazard rate, or conditional failure intensity, is:

It represents the instantaneous rate of failure conditional on survival to time t. Its unit is inverse time, such as failures per operating hour. Hazard rate is not itself the probability of failure in an hour unless the interval is sufficiently small and the rate is approximately constant.

Maintainability concerns the speed and ease with which an item can be restored. A common measure is mean time to repair (MTTR). Availability combines reliability and maintainability. For a repairable item under stable, steady-state conditions:

where A_s is steady-state inherent or operationally defined availability, MTBF is mean time between failures, and MTTR is mean time to repair. The formula assumes that the selected definitions of MTBF and MTTR are consistent and that repair and failure processes are sufficiently stable. For a nonrepairable mission, availability is not an adequate substitute for reliability.

Operational availability may also include logistics, waiting, inspection, and administrative delays. Thus, improving technical repair time may not improve production availability if spare-part shortages or permit delays dominate.

2. Failure behavior and the bathtub curve

The classical bathtub curve divides population failure behavior into three regions:

• Early-life failures: hazard decreases with age. Causes include manufacturing defects, installation errors, commissioning mistakes, and weak components.
• Random failures: hazard is approximately constant. Failures arise from unpredictable overstress, external events, or random component weakness.
• Wear-out failures: hazard increases with age because of fatigue, corrosion, erosion, insulation deterioration, or accumulated damage.

The bathtub curve is a useful teaching model, not a universal empirical law. Many modern electronic components exhibit approximately constant or decreasing hazard, while mechanical assemblies may contain several overlapping failure modes. A population curve may also conceal different subpopulations, operating regimes, or maintenance effects.

Figure: Conceptual bathtub curve. Actual equipment may not display all three regions.

A key managerial implication is that age-based replacement is rational only when failure risk increases with age and the replacement interval is economically justified. Replacing an item during a constant-hazard period may provide little reliability benefit and may create infant mortality through installation errors.

3. Weibull analysis

The two-parameter Weibull distribution is widely used because it can represent decreasing, constant, or increasing hazard. Its reliability function is:

where β is the shape parameter and η is the scale parameter. The scale parameter has the same unit as time; η is the time at which cumulative unreliability reaches approximately 63.2%, since R(η) = e⁻¹ ≈ 0.368. The shape parameter is dimensionless.

The corresponding hazard function is:

Interpretation is central:

These interpretations are hypotheses about the observed failure mechanism, not proof of causation. A fitted value of β > 1 may result from mixed failure modes, changing loads, or incomplete censoring treatment.

For n independent identical components in series, system reliability is:

If all components have equal reliability R(t), then R_s(t) = R(t)ⁿ. Series reliability decreases as the number of required components increases.

Worked example: Weibull reliability

Suppose bearing failures are modeled with β = 2 and η = 10,000 operating hours. Reliability at 6,000 hours is:

R(6000) = exp[−(6000/10000)²] = exp(−0.36) ≈ 0.698.

Thus, the estimated probability of surviving 6,000 hours is approximately 69.8%, under the model assumptions. This is not a guarantee for an individual bearing and should not be used without checking whether load, lubrication, alignment, and censoring conditions match the data.

For field data, censored observations—items that have not failed when observation ends—must not be treated as failures. Weibull plots and maximum-likelihood estimation can be useful, but parameter uncertainty should be reported through confidence intervals or sensitivity analysis.

4. Failure modes and mechanisms

A failure mode is the manner in which an item fails to perform its function: seal leakage, shaft fracture, valve stuck open, sensor drift, or loss of insulation resistance. A failure cause is the physical or organizational reason for the failure, such as abrasive contamination, incorrect torque, overload, or inadequate training. A failure effect is the local, system, or plant-level consequence.

The distinction matters because maintenance should address causes and functional consequences rather than merely replace failed parts. Failure mechanisms may include fatigue, wear, corrosion, erosion, electrical breakdown, thermal degradation, software error, human error, and process-induced overstress.

5. FMEA and FMECA

Failure Modes and Effects Analysis (FMEA) is a structured, bottom-up method. For each function, the team identifies failure modes, causes, effects, existing controls, and recommended actions. FMECA adds explicit criticality analysis, often emphasizing severity and probability.

A robust FMEA normally includes:

1. system boundary and operating context;
2. functions and performance standards;
3. functional failures;
4. failure modes;
5. causes and mechanisms;
6. local, system, and end effects;
7. current prevention and detection controls;
8. safety, environmental, quality, production, and financial consequences;
9. recommended treatment and accountable owner.

Many organizations use severity, occurrence, and detection ratings. A traditional Risk Priority Number is:

where S, O, and D are ordinal ratings. RPN is a prioritization aid, not a probability of loss. Multiplication can rank a high-severity, low-occurrence event below several moderate events, and equal RPNs can represent very different risks. Modern practice therefore supplements or replaces RPN with explicit risk matrices, criticality categories, probability distributions, and management review.

FMEA is strongest when failure modes are relatively discrete and the team has process knowledge. It is weaker when complex interactions, common-cause failures, software dependencies, or organizational controls dominate.

6. Fault-tree analysis and reliability block diagrams

Fault-tree analysis (FTA) is a top-down, deductive technique. It begins with a top event, such as “loss of cooling,” and logically decomposes it into combinations of causes using AND and OR gates. An OR gate occurs when any input can cause the output. An AND gate occurs only when all required inputs occur.

For independent basic events A and B:

and:

The AND expression assumes statistical independence. Common-cause failures violate this assumption and must be modeled explicitly or treated conservatively.

A minimal cut set is the smallest combination of basic events sufficient to produce the top event. FTA is valuable for safety, protection systems, and complex dependencies, but Boolean simplification can hide time sequence, repair, standby behavior, and imperfect switching unless the model is extended.

A reliability block diagram (RBD) represents success logic. For independent components in series:

For independent components in parallel, where the system succeeds if at least one succeeds:

For two identical parallel units with reliability 0.90, system reliability is 1 − (1 − 0.90)² = 0.99. However, physical redundancy may not produce this result if units share a power supply, cooling system, control logic, or environmental exposure.

FTA asks, “How can failure occur?” RBD asks, “What combination must operate for success?” They are complementary, not interchangeable in every system.

7. RCM logic and maintenance task selection

Reliability-Centered Maintenance begins with functions and performance standards rather than equipment names. A typical RCM analysis asks:

• What should the asset do?
• In what ways can it fail to perform its function?
• What causes each functional failure?
• What happens when the failure occurs?
• Why does the failure matter?
• What proactive task is technically feasible and worth doing?
• What should be done if no proactive task is suitable?

The principal task categories are:

The concept of a potential failure is important in condition-based maintenance. A potential failure is an identifiable condition showing that functional failure is developing. The P–F interval is the time between detection of that condition and functional failure. Inspection interval must be shorter than the P–F interval with sufficient margin for analysis, planning, and intervention.

Age-based maintenance is effective only for failure modes with a meaningful age relationship. Many random failures do not become less likely after replacement; scheduled replacement can therefore be wasteful or harmful.

8. Risk-based maintenance

Risk is commonly expressed conceptually as a combination of likelihood and consequence:

This expression is useful for ranking but should not be mistaken for a universally valid probabilistic identity. Likelihood may be a frequency, probability over a defined period, or ordinal category; consequence may be financial loss, injury severity, environmental damage, or production impact. Units and definitions must be explicit.

RBM prioritizes inspection and maintenance according to risk. A practical process is:

1. define assets, boundaries, and hazardous scenarios;
2. identify initiating failures and escalation pathways;
3. estimate likelihood using failure data, models, or expert judgment;
4. estimate consequences across safety, environment, quality, delivery, and cost;
5. evaluate existing barriers and uncertainty;
6. rank risk and define tolerability criteria;
7. select inspection, monitoring, testing, redesign, or operational controls;
8. document residual risk and review assumptions.

Risk matrices are easy to communicate but can distort risk through coarse categories. High-consequence, low-probability events require scenario analysis, safeguards review, and uncertainty treatment rather than simple multiplication.

9. Economic selection of maintenance policy

Technical feasibility is necessary but insufficient. Policy selection should compare life-cycle consequences. A basic expected annual loss model is:

where EAC is expected annual cost, C_pm is planned-maintenance cost, λ_f is failure frequency per year, C_f is expected cost per failure, C_ins is inspection cost, and C_pred is predictive-monitoring cost. The model assumes stable rates and costs and should be expanded when downtime, discounting, changing hazard, or multiple failure modes matter.

For alternatives over different lives, discounted life-cycle cost is more appropriate. The present value of a future cost C_t at time t is:

where i is the discount rate per period. Safety and legal obligations cannot be traded away merely because a cheaper policy has lower expected cost.

A sound decision compares total consequences: labor, spares, downtime, quality losses, energy, environmental exposure, safety controls, contractor risk, and opportunity cost. It also considers maintainability improvements, standardization, and the value of better information.

10. Integrated selection workflow

A practical strategy-selection workflow is:

Define the operating context. Document duty cycle, loads, environment, production role, redundancy, controls, legal requirements, and consequence categories.

Identify functions and failure modes. Use functional analysis, FMEA, historical data, operator knowledge, and engineering standards.

Model critical dependencies. Apply FTA for hazardous top events and RBD or availability models for system success logic.

Characterize failure behavior. Use Weibull or other statistical methods only when data quality and mechanism homogeneity support them.

Select technically valid tasks. Test whether a task detects deterioration, has an adequate P–F interval, and can be executed safely.

Evaluate economics and resources. Include labor, spares, access, outage windows, monitoring systems, and production constraints.

Assign ownership and feedback. Integrate work orders, inspection findings, defect elimination, and post-maintenance verification into the computerized maintenance management system (CMMS).

Review performance. Monitor availability, mean time between failures, mean time to repair, schedule compliance, repeat failures, maintenance-induced failures, and risk indicators. Metrics should support learning, not encourage concealment or premature closure.

Illustrative case: cooling-water pumps in a process plant

Background

A process plant uses two electrically driven cooling-water pumps. One pump is normally operating and the other is standby. Loss of cooling can force a controlled shutdown and, under unfavorable conditions, damage temperature-sensitive equipment.

Problem

The plant historically replaced pump bearings every 6,000 operating hours. Failures continued, and maintenance records showed that many bearing failures were associated with misalignment, contamination, and lubrication errors rather than age alone.

Analysis

The team defined the function as delivering the required cooling flow and pressure under specified process conditions. FMEA identified bearing seizure, seal leakage, motor insulation failure, blocked suction, and standby-start failure. FTA for “loss of cooling” showed that either operating-pump failure combined with standby unavailability, or common loss of both pumps, could produce the top event. The RBD indicated nominal parallel redundancy, but the pumps shared a suction header and electrical bus; therefore, the redundancy was not complete.

Inspection data were censored because several bearings were replaced preventively. A preliminary Weibull analysis gave an estimated shape parameter near 1.0 for bearing failures, with wide uncertainty. The evidence did not justify a strict age-based replacement policy.

Approach and solution

The revised strategy was:

• replace routine bearing replacement with vibration and temperature monitoring;
• establish alignment and lubrication standards;
• sample lubricant for contamination and wear debris;
• test the standby pump automatically each week;
• perform a failure-finding test of the standby start and associated valves;
• inspect the shared suction header and electrical protection;
• retain corrective replacement for confirmed defects;
• review condition-monitoring alarms using defined response times.

A criticality review classified loss of cooling as high consequence, requiring documented escalation and management of residual risk. The CMMS linked alarm limits, inspection routes, work instructions, and verification records.

Outcome

In this illustrative case, bearing replacements decreased, repeat failures declined, and standby availability improved. The greatest benefit did not come from a more sophisticated sensor alone; it came from correcting common causes, testing hidden functions, and recognizing that nominal redundancy was weakened by shared dependencies.

Learning points

• Do not infer wear-out from repeated failures without considering causes and censoring.
• A standby asset requires failure-finding tasks, not merely installation.
• Redundancy is only as strong as its common dependencies.
• Condition monitoring is a management process, not a hardware purchase.
• Criticality should influence response time, governance, and evidence quality.

Challenges, limitations, and governance

Reliability analysis is vulnerable to incomplete failure coding, survivorship bias, changing operating conditions, and inconsistent definitions of failure. A CMMS record may say “pump failed” without distinguishing seal leakage, process blockage, motor trip, or operator intervention. Data cleansing and failure-mode coding are therefore engineering activities, not clerical tasks.

Human factors are equally important. Poor access, confusing alarms, inadequate procedures, fatigue, and production pressure can create or amplify failures. Maintenance strategies should be designed with operators and technicians, while preserving clear accountability for safety-critical decisions.

Artificial intelligence can classify work-order text, detect anomalies, estimate remaining useful life, and prioritize inspections. However, models may fail under distribution shift, sparse failure data, sensor drift, and changing operating regimes. AI recommendations require traceability, validation, cybersecurity controls, human review, and a defined response process. Predictive accuracy is not the same as maintenance value: a forecast is useful only when it enables a timely, safe, and economically justified action.

Safety, environmental, and regulatory requirements establish minimum controls. They should not be overridden by an economic optimization. Risk acceptance must be transparent, documented, and authorized at the appropriate level. Contractors and suppliers should provide evidence of design assumptions, maintenance limits, and safety-critical inspection requirements.

Best-practice checklist

• Define functions and operating context before selecting tasks.
• Separate failure mode, cause, effect, and consequence.
• Treat FMEA ratings as structured judgment, not objective truth.
• Use Weibull parameters with confidence bounds and mechanism checks.
• Model common-cause and dependency failures explicitly.
• Match inspection intervals to the P–F interval and decision lead time.
• Test hidden protective functions.
• Use run-to-failure only where consequences and recovery are acceptable.
• Include maintenance-induced risk and life-cycle cost.
• Close the learning loop through verified defect elimination.

Chapter Summary

Reliability engineering supplies the logic for selecting maintenance strategies in TPM. Reliability describes the probability of performing a required function for a stated time, while maintainability concerns restoration and availability combines successful operation with repair performance. The bathtub curve and Weibull distribution help interpret changing hazard, but neither should be applied without attention to failure mechanisms, censoring, and uncertainty.

FMEA/FMECA provides a bottom-up examination of functions, failure modes, causes, effects, and controls. FTA analyzes how combinations of basic events produce a top event, while RBD models the combinations required for system success. RCM selects technically feasible tasks according to failure consequences and detectable deterioration. RBM prioritizes action according to likelihood, consequence, safeguards, and uncertainty. Economic evaluation then compares life-cycle cost without permitting cost optimization to override safety or legal obligations.

The strongest maintenance strategy integrates engineering models, field knowledge, data quality, human factors, governance, and continuous feedback. The aim is not maximum maintenance, but the right intervention for each failure mode.

Key Terms

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

Maintainability: Ability and speed with which an item can be restored after failure.

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

Hazard rate: Conditional instantaneous rate of failure among surviving items.

Bathtub curve: Conceptual pattern of decreasing, constant, and increasing hazard regions.

Weibull distribution: Flexible lifetime distribution characterized by shape and scale parameters.

Failure mode: Specific manner in which a functional failure occurs.

Failure mechanism: Physical or chemical process producing degradation or failure.

FMEA: Bottom-up analysis of functions, failure modes, causes, effects, and controls.

FMECA: FMEA expanded with criticality assessment.

Fault-tree analysis: Top-down deductive analysis of combinations causing a top event.

Minimal cut set: Smallest combination of basic events sufficient to cause a top event.

Reliability block diagram: Representation of component success relationships.

RCM: Structured method for selecting maintenance based on functions and consequences.

Risk-based maintenance: Maintenance prioritization using likelihood, consequence, and safeguards.

Potential failure: Detectable condition indicating that functional failure is developing.

P–F interval: Time between potential failure detection and functional failure.

Common-cause failure: Failure affecting multiple redundant elements through a shared cause.

Life-cycle cost: Total cost of acquisition, operation, maintenance, risk, and disposal over the asset life.

Review Questions

Short Answer Questions

1. Define reliability and explain why operating context is necessary.
2. What do the three regions of the bathtub curve represent?
3. Interpret Weibull shape parameters β < 1, β = 1, and β > 1.
4. Distinguish a failure mode from a failure cause and a failure effect.
5. What is the purpose of a failure-finding task?

Long Answer Questions

1. Explain how FMEA, FTA, and RBD complement one another in maintenance strategy development.
2. Discuss the technical and economic conditions under which condition-based maintenance is preferable to scheduled replacement.
3. Explain the RCM decision logic and the limitations of using RPN as the sole basis for prioritization.
4. Analyze how common-cause failures affect claims of redundancy and system availability.
5. Discuss the governance requirements for using AI-based predictive maintenance in safety-critical assets.

Analytical / Application Questions

1. A component has Weibull parameters β = 2.5 and η = 8,000 hours. Calculate its reliability at 5,000 hours and explain whether the estimated hazard is increasing or decreasing.
2. Two independent parallel pumps each have reliability 0.92 over a mission. Calculate the nominal parallel reliability and identify two real-world dependencies that could make the actual reliability lower.
3. Construct a mini-FMEA for a conveyor drive in a packaging line. Identify three failure modes, their causes and effects, controls, consequences, and recommended maintenance policies.

Multiple Choice Questions with Answers

Question 1

Which Weibull shape parameter indicates an increasing hazard rate?

A. β < 1

B. β = 0

C. β = 1

D. β > 1

Correct Answer: D. β > 1

Question 2

For two independent components in a series system, system reliability is:

A. The sum of component reliabilities

B. The product of component reliabilities

C. One minus the product of component reliabilities

D. The average of component reliabilities

Correct Answer: B. The product of component reliabilities

Question 3

Which method begins with a top event and deductively identifies combinations of causes?

A. FMEA

B. FTA

C. RBD

D. Weibull analysis

Correct Answer: B. FTA

Question 4

A failure-finding task is primarily intended to detect failure of:

A. Visible wear mechanisms

B. Hidden protective functions

C. Newly installed components only

D. Production scheduling systems

Correct Answer: B. Hidden protective functions

Question 5

If a component has reliability 0.90 and two identical independent components operate in parallel, nominal system reliability is:

A. 0.81

B. 0.90

C. 0.99

D. 1.80

Correct Answer: C. 0.99

Question 6

In an FMEA, contamination causing bearing seizure is best classified as the:

A. Failure effect

B. Failure cause or mechanism

C. Top event

D. Availability measure

Correct Answer: B. Failure cause or mechanism

Question 7

Age-based replacement is most defensible when:

A. The failure hazard is known to increase with age

B. All failures are random

C. Replacement always creates no risk

D. No failure data exist

Correct Answer: A. The failure hazard is known to increase with age

Question 8

The P–F interval is the time between:

A. Purchase and commissioning

B. Failure and repair completion

C. Potential failure detection and functional failure

D. Two preventive maintenance tasks

Correct Answer: C. Potential failure detection and functional failure

Question 9

Which issue can make nominal redundancy ineffective?

A. Independent power supplies

B. Common-cause failure

C. Separate physical locations

D. Verified standby testing

Correct Answer: B. Common-cause failure

Question 10

The principal purpose of RCM is to:

A. Maximize the number of preventive tasks

B. Select maintenance based on functions, failure consequences, and feasible tasks

C. Replace all components at fixed ages

D. Eliminate the need for operator involvement

Correct Answer: B. Select maintenance based on functions, failure consequences, and feasible tasks