Reliability, Fatigue and Design Under Uncertainty

Introduction and Context

Mechanical components rarely experience perfectly constant loads, identical material properties, or precisely known operating environments. Shafts, gears, springs, fasteners, bearings, and welded structures are subjected to fluctuating stresses, vibration, temperature changes, corrosion, manufacturing variation, and uncertain user behaviour. A design that is safe under one nominal load may fail after millions of cycles or may be unreliable when tolerances and environmental effects are considered.

Reliability-oriented design therefore asks two related questions:

1. What is the probability that a component performs its intended function for a specified time?
2. What design changes reduce the probability and consequences of failure to an acceptable level?

Fatigue analysis addresses failure caused by repeated or fluctuating stress. Reliability analysis treats material properties, loads, dimensions, and operating conditions as uncertain quantities. Risk-based design combines probability of failure with the severity of consequences. These methods do not eliminate uncertainty; they make it visible and manageable.

Learning Objectives

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

• distinguish static failure, fatigue failure, reliability, and risk;
• calculate stress ranges, mean stresses, and stress ratios for fluctuating loads;
• interpret S–N curves and apply fatigue-strength modifiers;
• use the Goodman relation and appropriate fatigue safety factors;
• estimate cumulative fatigue damage using Miner’s rule;
• explain statistical tolerances, distributions, confidence, and reliability targets;
• apply FMEA and fault-tree analysis to mechanical systems;
• make engineering decisions when loads, strengths, and failure consequences are uncertain.

1. Reliability as a Design Requirement

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

where R(t) is dimensionless and lies between 0 and 1. The complementary probability is the unreliability or cumulative failure probability:

These definitions require a clearly specified function, operating environment, maintenance policy, and time interval. “Reliable” has no meaning without these conditions.

For a constant failure rate λ, the exponential model gives

and the mean time to failure is

This model is appropriate mainly during the useful-life period when failures occur approximately randomly and the failure rate is nearly constant. It is not generally appropriate for early manufacturing defects or wear-out failures. The bathtub curve commonly represents three regions: decreasing failure rate during infant mortality, approximately constant failure rate during useful life, and increasing failure rate during wear-out.

For components subject to wear, corrosion, or fatigue crack growth, the failure rate often increases with age. Preventive replacement, inspection, condition monitoring, or redesign may then be more appropriate than assuming exponential reliability.

2. Fluctuating Stress and Fatigue Terminology

For a periodic stress varying between maximum and minimum values, the nominal quantities are

where σ_m is mean stress and σ_a is stress amplitude. The stress range is ∆σ = σ_max - σ_min = 2σ_a, and the stress ratio is

The same definitions apply to shear stress by replacing σ with τ. Tensile stress is normally taken as positive and compressive stress as negative. These equations describe a regular cycle; irregular service histories require cycle counting or another load-spectrum method.

A fully reversed cycle has σ_m = 0 and R = -1. A repeated tensile cycle has σ_min = 0 and R = 0. Mean tensile stress generally reduces fatigue resistance, whereas compressive mean stress often improves it, although residual stresses, yielding, surface damage, and multiaxial effects can modify this generalization.

Fatigue failure is progressive and may occur at stresses below the static yield strength. It usually involves crack initiation, crack propagation, and final fracture. A component can therefore pass a static-strength check and still fail after prolonged service.

3. S–N Behaviour and Fatigue Life

An S–N curve, also called a Wöhler curve, relates cyclic stress level to the number of cycles to failure. The vertical axis may show stress amplitude and the horizontal axis commonly shows cycles on a logarithmic scale. Test data scatter substantially because fatigue is sensitive to surface finish, defects, microstructure, size, residual stress, temperature, and test procedure.

For steels in some laboratory conditions, the curve may approach an endurance limit: below a particular stress amplitude, very long life is possible. This is not universal. Aluminium alloys and many non-ferrous materials generally do not show a sharp endurance limit, so a design life must be specified.

A common idealized high-cycle relationship is Basquin’s equation:

where σ'_f is the fatigue-strength coefficient, b is the fatigue-strength exponent, and N_f is the number of reversals to failure. The exponent is usually negative. The equation is empirical and must be calibrated for the material, stress ratio, surface condition, and life range. It should not be extrapolated far beyond the test data.

Low-cycle fatigue involves significant cyclic plastic strain. In that regime, a strain-life model such as the Coffin–Manson–Basquin relation is more suitable:

where ε_a is total strain amplitude, E is Young’s modulus, ε'_f is the fatigue-ductility coefficient, and c is the fatigue-ductility exponent. The first term represents elastic strain and the second plastic strain. Bachelor-level design calculations should use this model only when reliable material parameters and local strain estimates are available.

Fatigue Strength Modifiers

Laboratory endurance strengths must be adjusted for real components. A commonly used modified endurance strength is

where S'_e is the laboratory specimen endurance limit and the factors represent, as applicable, surface finish, size, load type, temperature, reliability, and miscellaneous effects. Not every handbook uses the same symbols or factor definitions; the designer must use one internally consistent standard.

Typical effects include:

A geometric stress concentration factor is

For fatigue, the effective fatigue concentration factor is commonly represented by

where q is notch sensitivity, between approximately 0 and 1 for many engineering situations. K_f should not automatically be replaced by K_t: ductile materials may blunt a notch, while brittle materials and sharp defects may behave more severely.

4. Mean-Stress Correction and Fatigue Safety Factor

For a component with alternating stress σ_a and mean stress σ_m, a widely used linear Goodman relation is

where S_e is corrected endurance strength, S_ut is ultimate tensile strength, and n_f is fatigue safety factor. This relation is an engineering approximation for high-cycle fatigue, particularly when mean stress is tensile. It is not a universal failure law.

For completely reversed stress, σ_m = 0, and the condition becomes σ_a ≤ S_e/n_f. For significant compressive mean stress, Goodman may be conservative or may be inappropriate if yielding, buckling, or local contact governs. Alternative criteria include Gerber, Soderberg, and ASME-type relations. Soderberg uses yield strength instead of ultimate strength and is generally more conservative:

The selected criterion should reflect material ductility, design standard, loading uncertainty, and consequences of failure.

Worked Example: Goodman Check

A steel shaft has corrected endurance strength S_e = 180 MPa, ultimate strength S_ut = 600 MPa, mean bending stress σ_m = 80 MPa, and alternating bending stress σ_a = 120 MPa. Using Goodman:

Therefore,

The result is dimensionless. Whether n_f = 1.25 is acceptable depends on the design code, load uncertainty, inspection capability, and failure consequence. It is not automatically adequate merely because it exceeds one.

5. Variable-Amplitude Loading and Cumulative Damage

Real machines rarely operate at one stress amplitude. A vehicle axle, for example, experiences different loads during starting, cruising, braking, and road shocks. The load record must be converted into cycles, often using rainflow counting. Each cycle is assigned a stress amplitude, mean stress, and number of occurrences.

Miner’s linear damage rule estimates cumulative damage as

where n_i is the number of applied cycles at stress level i, N_i is the constant-amplitude life at that level, and k is the number of stress classes. Failure is conventionally predicted near D = 1.

Worked Example: Miner Damage

A component experiences 100,000 cycles at a stress level for which the S–N curve predicts N₁ = 1,000,000 cycles to failure, and 20,000 cycles at a second level for which N₂ = 50,000 cycles to failure. Then

The estimated consumed life is 50% according to the linear rule. If this spectrum repeats unchanged, two blocks would produce D = 1.0. The estimate does not prove that failure will occur precisely at the second block. Damage sequence, crack closure, overloads, residual stresses, and environmental effects can make the real result higher or lower.

Miner’s rule is useful for preliminary design and spectrum comparison, but it assumes linear damage accumulation and does not explicitly model crack size. Crack-growth mechanics, inspection data, or full-scale testing may be required for critical components.

6. Statistical Variation, Tolerances, and Reliability Targets

Mechanical design variables are distributions rather than single values. Diameter, hardness, surface roughness, load, material strength, and alignment vary between parts and over time. A nominal calculation hides this variation.

For a normal random variable X, the mean is μ and standard deviation is σ. A tolerance written as μ ± 3σ contains approximately 99.73% of values if the normal model is valid. This is a statistical statement, not a guarantee that no value lies outside the interval.

For independent random variables, the variance of a sum or difference is

and therefore

The covariance term must be included when variables are correlated:

Ignoring correlation can either overestimate or underestimate uncertainty.

A simple limit-state formulation defines safety margin

where S is available strength and L is applied demand. Failure occurs when g(X) ≤ 0. Reliability is

If strength and load are independent normal variables, then g is normal with

and the reliability index is

The failure probability is P_f = Φ(-β), where Φ is the standard normal cumulative distribution function. This closed-form result requires normality and independence; nonlinear limit states and non-normal variables require transformation, simulation, or reliability methods such as FORM.

Worked Example: Probabilistic Margin

Suppose strength is normally distributed with mean 500 MPa and standard deviation 40 MPa. Load is independent and normally distributed with mean 350 MPa and standard deviation 20 MPa. Then

Thus

A standard normal table gives a failure probability of approximately 0.0004, or 0.04%, under the stated model. This result is conditional on the assumed distributions, independence, and accurate parameter estimates. It should not be confused with a universal guarantee of reliability.

7. Safety Factors and Reliability-Based Design

A deterministic safety factor may be defined as

This is useful when values are known with reasonable confidence, but a single factor can conceal different uncertainty sources. A load factor, material factor, fatigue factor, and consequence-based requirement may be more informative than one arbitrary multiplier.

Reliability targets should be connected to consequences. A replaceable low-cost guard may tolerate a greater failure probability than a rotating component whose fracture can injure personnel. The target should also reflect detectability, maintenance access, redundancy, and regulatory requirements.

Good practice includes:

• specifying reliability at a stated mission time;
• separating aleatory variability from epistemic uncertainty;
• using test data and field data to update assumptions;
• checking sensitivity to dominant variables;
• selecting safety factors consistent with a recognized code;
• documenting assumptions, model validity, and residual risk.

Aleatory uncertainty is inherent variation, such as cycle-to-cycle load variation. Epistemic uncertainty results from incomplete knowledge, such as sparse fatigue data. More testing can reduce epistemic uncertainty but usually cannot eliminate inherent variability.

8. FMEA and Fault-Tree Reasoning

Failure Modes and Effects Analysis (FMEA) is a bottom-up method. For each component or function, the team identifies possible failure modes, causes, local effects, system effects, existing controls, and recommended actions. Typical fields include severity, occurrence, and detectability. Some organizations multiply ordinal ratings to form a risk-priority number, but such products can be misleading because ordinal scores do not have equal numerical meaning. A risk matrix, action thresholds, and explicit justification are often better governance.

A practical FMEA sequence is:

1. Define system boundaries and intended functions.
2. Identify component failure modes.
3. Record causes and effects.
4. Assess severity, likelihood, and detectability.
5. Identify preventive and detection controls.
6. Assign owners and due dates for corrective actions.
7. Reassess residual risk after actions.

Fault-tree analysis is top-down. It begins with a top event, such as “loss of braking,” and decomposes it using logical gates. An AND gate means all listed events are required; an OR gate means any one may produce the output. For independent events A and B:

The independence assumption must be justified. Common-cause failures, shared power supplies, common environments, and maintenance errors can invalidate simple multiplication.

FMEA asks, “What can fail, and what are the effects?” Fault-tree analysis asks, “What combinations can produce this undesired top event?” Used together, they expose both local weaknesses and system-level combinations.

9. Illustrative Case: Fatigue Risk in a Conveyor Drive Shaft

Background

A conveyor in a mineral-processing plant uses a steel drive shaft. The shaft transmits torque, experiences bending from chain tension, and operates in a dusty, humid environment. A previous shaft developed a crack near a keyway after approximately 18 months.

Problem

The nominal static torque calculation showed a safety factor above two, yet the shaft still failed. The design team must determine whether the cause is fatigue, overload, corrosion, manufacturing variation, or maintenance practice.

Analysis

Inspection reveals a sharp keyway corner, machining marks, and evidence of fretting. Operating data show frequent start–stop cycles and occasional jam-clearing events. The shaft therefore experiences variable-amplitude torsion and bending, tensile residual stress near the keyway, and a corrosive environment.

The team first calculates nominal alternating and mean stresses, applies a fatigue concentration factor rather than only a geometric factor, and corrects the endurance strength for surface, size, reliability, and environment. A Goodman check is performed for representative cycles. Rainflow counting of the measured torque history produces three main cycle groups. Miner damage is then estimated, with jam-clearing events treated separately because overloads may accelerate crack growth.

An FMEA identifies keyway cracking as a high-severity mode. A fault tree for “unplanned conveyor shutdown” includes shaft fracture OR coupling failure OR motor protection trip. The shaft-fracture branch includes fatigue crack growth AND inadequate detection, while common maintenance practices are examined as possible shared causes.

Approach and Solution

The redesign uses a larger fillet radius, improved keyway geometry, a polished surface, and a coupling arrangement that reduces bending. The shaft material and heat treatment are verified by certificates and hardness testing. A guarded torque limiter reduces jam-clearing overloads. Vibration and torque monitoring are introduced, and inspection intervals are based on calculated damage and observed crack-growth risk rather than calendar convenience.

Outcome

The revised design increases calculated fatigue margin and reduces stress concentration. The monitoring system detects abnormal torque before the shaft reaches the previous overload condition. After a validation period, no crack indications are found and the maintenance team reports fewer emergency stops. The reliability improvement comes from combining geometry, process control, overload prevention, inspection, and operational discipline—not from increasing one nominal safety factor alone.

Learning Points

• Static strength does not demonstrate fatigue reliability.
• Small geometric details can dominate local fatigue performance.
• Field load data are more valuable than assumed duty cycles.
• Detection and maintenance controls are part of the reliability design.
• Risk reduction should address both probability and consequence.

10. Limitations, Ethics, Standards, and Contemporary Practice

Fatigue data are statistical and specimen-specific. A polished laboratory specimen is not equivalent to a corroded, welded, misaligned production assembly. Designers should consult applicable standards, material certificates, validated fatigue databases, and company procedures. For safety-critical machinery, conformity assessment, traceability, inspection records, and change control are essential.

Ethically, engineers must not select optimistic distributions, hide scatter, or report only the best-case safety factor. Uncertainty should be communicated to managers and operators in language that supports informed decisions. If data are inadequate, the correct response may be additional testing, a conservative design, redundancy, or restricted operating conditions.

Digital monitoring and machine learning can support predictive maintenance by identifying changes in vibration, temperature, torque, or acoustic emission. However, a model trained on limited historical failures may miss new failure modes. AI-based alerts require sensor validation, cybersecurity, explainable thresholds, human oversight, and a safe fallback when data are missing. A prediction system supplements, rather than replaces, sound mechanical design and inspection.

Best-Practice Design Workflow

1. Define function, mission time, environment, and failure consequences.
2. Obtain realistic load histories rather than relying only on nominal ratings.
3. Identify stress concentrations, surface conditions, residual stresses, and likely damage mechanisms.
4. Select an appropriate fatigue model and corrected material data.
5. Check static strength, fatigue strength, deflection, wear, thermal effects, and stability separately.
6. Quantify variation in load, strength, dimensions, and model parameters.
7. Set a reliability target and select a defensible safety factor or probabilistic criterion.
8. Use FMEA and fault-tree analysis to examine local and system-level failures.
9. Validate through testing, inspection, monitoring, and field feedback.
10. Document assumptions, uncertainty, acceptance criteria, and residual risk.

Chapter Summary

Reliability is the probability of successful function for a specified time and condition. Fatigue is a progressive failure process caused by repeated or fluctuating stress and may occur below static yield strength. Stress amplitude, mean stress, stress ratio, S–N curves, fatigue modifiers, mean-stress corrections, and Miner’s rule provide practical tools for design, but each has assumptions and limits.

Uncertainty should be represented through distributions, tolerances, confidence, reliability targets, and limit-state functions. For simple independent normal variables, the reliability index provides a useful first estimate, but non-normality, correlation, nonlinear behaviour, and sparse data require more advanced analysis. FMEA and fault-tree analysis complement numerical calculations by revealing failure modes, effects, combinations, controls, and consequences. Reliable design is ultimately a decision process that combines mechanics, statistics, testing, monitoring, standards, and ethical judgement.

Key Terms

Reliability: Probability that an item performs its required function for a specified time and condition.

Fatigue: Progressive damage and failure caused by repeated or fluctuating loading.

Stress amplitude: Half the difference between maximum and minimum cyclic stress.

Mean stress: Average of maximum and minimum cyclic stress.

Stress ratio: Ratio of minimum stress to maximum stress.

S–N curve: Relationship between cyclic stress level and cycles to failure.

Endurance limit: Stress amplitude below which very long life may be obtained for some materials.

Fatigue concentration factor: Effective stress concentration factor for fatigue loading.

Goodman relation: Linear mean-stress correction using endurance and ultimate strengths.

Miner’s rule: Linear cumulative fatigue-damage approximation.

Reliability target: Required probability of successful performance over a stated mission.

Uncertainty: Lack of complete certainty about loads, properties, models, or conditions.

Aleatory variability: Inherent random variation in a process or population.

Epistemic uncertainty: Uncertainty caused by incomplete knowledge or limited data.

Limit state: Boundary separating acceptable performance from failure.

Reliability index: Standardized distance between mean safety margin and failure boundary.

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

Fault-tree analysis: Top-down logical analysis of combinations leading to a top event.

Common-cause failure: Multiple failures produced by one shared cause.

Residual risk: Risk remaining after controls and corrective actions.

Review Questions

5 Short Answer Questions

1. Define reliability and explain why mission time must be stated.
2. Distinguish mean stress, stress amplitude, and stress ratio.
3. Why can a component pass a static-strength test and still fail by fatigue?
4. State the assumptions behind the simple normal-variable reliability-index calculation.
5. What is the difference between FMEA and fault-tree analysis?

5 Long Answer Questions

1. Explain the construction, interpretation, and limitations of an S–N curve. Include the effects of surface finish, size, environment, and reliability.
2. Derive the stress quantities for a fluctuating cycle and discuss how tensile and compressive mean stresses influence fatigue design.
3. Explain Miner’s rule, show how it is applied to a variable-amplitude spectrum, and discuss why its prediction may differ from actual failure life.
4. Compare deterministic safety factors with reliability-based design using a limit-state formulation.
5. Describe a complete FMEA and fault-tree workflow for a safety-critical mechanical drive.

3 Analytical / Application Questions

1. A shaft experiences a maximum bending stress of 220 MPa and a minimum stress of −60 MPa. Calculate the mean stress, alternating stress, stress range, and stress ratio. Explain which fatigue criterion would be appropriate for an initial assessment.
2. A component has strength with mean 480 MPa and standard deviation 30 MPa. Its independent load has mean 360 MPa and standard deviation 25 MPa. Calculate the reliability index using the normal limit-state model and discuss at least two reasons why the result may be uncertain.
3. Develop a brief FMEA and fault tree for a conveyor drive shaft that fractures during operation. Identify at least three failure causes, two detection controls, one common-cause possibility, and two risk-reduction actions.

Multiple Choice Questions with Answers

Question 1

For a cyclic stress varying from −40 MPa to 160 MPa, what is the stress amplitude?

A. 60 MPa

B. 100 MPa

C. 120 MPa

D. 200 MPa

Correct Answer: B. 100 MPa

Question 2

In the Goodman relation, which material property commonly appears in the mean-stress term?

A. Ultimate tensile strength

B. Poisson’s ratio

C. Thermal conductivity

D. Density

Correct Answer: A. Ultimate tensile strength

Question 3

If a component experiences 20,000 cycles and its predicted constant-amplitude life at that level is 100,000 cycles, what damage fraction is assigned by Miner’s rule?

A. 0.02

B. 0.05

C. 0.20

D. 5.00

Correct Answer: C. 0.20

Question 4

Which condition corresponds to a fully reversed stress cycle?

A. R = 0

B. R = 1

C. R = −1

D. R = 2

Correct Answer: C. R = −1

Question 5

A corrected endurance strength is obtained by multiplying the laboratory endurance strength by surface, size, load, temperature, reliability, and miscellaneous factors. What is the purpose of this procedure?

A. To account for real-component conditions

B. To convert stress into strain

C. To remove all statistical scatter

D. To calculate shaft torque directly

Correct Answer: A. To account for real-component conditions

Question 6

For independent normal strength and load variables, the standard deviation of the safety margin S − L is calculated as:

A. σS − σL

B. σS + σL

C. √(σS² + σL²)

D. σS² − σL²

Correct Answer: C. √(σS² + σL²)

Question 7

Which statement best describes FMEA?

A. It begins with a top event and uses logic gates

B. It identifies component failure modes, causes, effects, and controls

C. It applies only to fatigue crack growth

D. It assumes all failures are statistically independent

Correct Answer: B. It identifies component failure modes, causes, effects, and controls

Question 8

In a fault tree, an AND gate indicates that:

A. Any one input event is sufficient

B. No input event can cause the output

C. All specified input events are required for the output event

D. The input probabilities must be equal

Correct Answer: C. All specified input events are required for the output event

Question 9

If the mean safety margin is 150 MPa and its standard deviation is 50 MPa, what is the reliability index under the simple normal model?

A. 0.33

B. 2.0

C. 3.0

D. 7,500

Correct Answer: C. 3.0

Question 10

Which is an example of epistemic uncertainty?

A. Natural cycle-to-cycle variation in operating load

B. Random variation in grain structure among parts

C. Limited knowledge caused by sparse fatigue test data

D. Normal variation in ambient temperature during service

Correct Answer: C. Limited knowledge caused by sparse fatigue test data