Design Quality, Reliability and Failure Prevention

Introduction

Quality is not inspected into a product after it has been made; it is principally designed into the product, service, process and supporting system. Inspection can detect some nonconformities, but it cannot remove the causes of variation, misuse, weak interfaces, inadequate maintenance or inappropriate design decisions. Advanced quality management therefore treats design as a prevention activity and connects customer requirements, technical specifications, process capability, reliability, maintainability, safety and lifecycle cost.

Design quality concerns the degree to which a design is capable of satisfying stated and implied needs. Conformance quality concerns whether production or service delivery consistently follows that design. A well-designed product that is poorly produced will fail in use; a highly capable process cannot compensate for requirements that were misunderstood or for a fundamentally unsafe design.

This chapter develops an integrated approach using quality function deployment (QFD), design reviews, failure mode and effects analysis (FMEA), design of experiments (DOE), robust design, reliability analysis, maintainability and lifecycle quality planning.

Learning Objectives

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

1. Distinguish design quality, conformance quality, reliability and maintainability.
2. Translate customer and stakeholder needs into measurable technical requirements using QFD.
3. Plan effective design reviews and apply FMEA to prevent failures.
4. Select and interpret appropriate DOE and robust-design methods.
5. Calculate and interpret basic reliability measures for common systems.
6. Integrate maintainability, safety, supplier quality and lifecycle evidence into quality planning.
7. Evaluate the limitations, ethical implications and governance requirements of predictive and automated quality methods.

1. Quality as a Design and Lifecycle Responsibility

Design decisions establish much of a product’s eventual cost, risk, usability and quality. Examples include material selection, tolerances, software architecture, ergonomics, energy consumption, inspection access, cybersecurity controls and the availability of replacement parts. These decisions also determine the range of conditions under which the product can operate safely.

A lifecycle perspective includes:

• discovery of customer, legal and societal needs;
• concept selection and feasibility assessment;
• requirements definition;
• architecture and detailed design;
• verification and validation;
• industrialisation and supplier qualification;
• production or service delivery;
• installation, operation, maintenance and support;
• modification, retirement and disposal.

Verification asks, “Was the design output produced according to the specified requirements?” Validation asks, “Does the resulting product or service satisfy the intended use and user needs?” Confusing these questions is a common source of failure. A design may pass a dimensional inspection yet be unusable in real operating conditions.

Important terminology includes:

• Critical-to-quality characteristic (CTQ): A characteristic whose performance materially affects customer satisfaction, safety, compliance or functional success.
• Special characteristic: A product or process characteristic requiring particular control because its variation can affect fit, function, safety or regulation.
• Failure: The loss of ability to perform a required function.
• Fault: An abnormal condition or defect that can lead to failure.
• Reliability: The probability that an item performs its required function for a specified time under stated conditions.
• Maintainability: The probability that an item can be restored to a specified condition within a stated time using specified procedures and resources.
• Robustness: The ability to perform acceptably despite variation in noise factors and operating conditions.

2. Translating Voice of the Customer into Design Requirements

Customer requirements are often qualitative: safe, quiet, fast, easy to clean or dependable. Design teams must convert them into measurable, testable and traceable requirements. A good requirement identifies the function, performance level, operating conditions, verification method and acceptance criterion.

For example, “the device should be quiet” is incomplete. A stronger requirement might specify a maximum sound pressure level, measurement distance, operating mode, environment and instrument method. Requirements should also be checked for conflicts. Increasing structural strength may increase mass; reducing cost may reduce redundancy; improving speed may increase heat generation.

Quality Function Deployment

QFD provides a structured method for translating the voice of the customer into technical characteristics and prioritised design actions. Its familiar first matrix, the House of Quality, generally contains:

1. customer requirements (“Whats”);
2. importance ratings;
3. engineering characteristics (“Hows”);
4. relationships between customer and engineering requirements;
5. technical correlation among engineering characteristics;
6. competitive or benchmark information;
7. target values and priorities.

A simple weighted priority for engineering characteristic j can be expressed as:

where Pⱼ is the unnormalised priority of characteristic j, wᵢ is the importance weight of customer requirement i, and rᵢⱼ is the relationship score between requirement i and characteristic j. The relationship scale must be defined by the organisation, commonly using values such as 0, 1, 3 and 9 for none, weak, moderate and strong relationships. The calculation is meaningful only when customer weights and relationship definitions are applied consistently. It is a prioritisation aid, not an objective substitute for engineering judgement.

If a characteristic has a high priority but poor current performance, it deserves early design attention. A strong negative correlation in the roof of the matrix indicates a trade-off requiring experimentation or architectural reconsideration.

Requirements traceability

A traceability matrix links customer needs to system requirements, subsystem requirements, design outputs, verification tests, validation evidence and change records. Traceability supports configuration control and prevents “orphan” requirements that are specified but never tested. It also identifies tests that lack a clear requirement justification.

3. Design Reviews and Prevention Gates

A design review is a formal, evidence-based decision point, not merely a presentation. Participants should include design engineering, manufacturing or operations, quality, procurement, service, safety, regulatory specialists and representative users where appropriate.

Typical review gates include:

Effective reviews define entry criteria, independent reviewers, decision authority, open actions, due dates and escalation rules. A review that records concerns without assigning owners is an observation exercise rather than a control mechanism.

4. Failure Mode and Effects Analysis

FMEA is a structured, prospective method for identifying how a product or process might fail, what effects may result, what causes are plausible and what controls are required. It is most valuable early, when changes are less expensive.

A product or design FMEA examines functions, failure modes, effects, causes, prevention controls and detection controls. A process FMEA examines operations such as machining, assembly, packaging or service delivery. The analysis should begin with the intended function and operating context rather than with a list of components.

A conventional FMEA evaluates:

• Severity (S): seriousness of the effect if failure occurs;
• Occurrence (O): likelihood of the cause or failure mode;
• Detection (D): likelihood that current controls will fail to detect the problem before release or use.

Historically, teams multiplied ordinal ratings to obtain a risk priority number:

The ratings are ordinal, not necessarily interval-scale measurements. Consequently, an RPN of 120 is not meaningfully “twice as risky” as an RPN of 60, and different combinations can produce the same value. Modern practice therefore uses action-priority logic, severity thresholds, prevention emphasis and explicit review of high-severity hazards rather than relying on RPN alone.

FMEA procedure

1. Define scope, boundaries, intended function and assumptions.
2. Decompose the system into functions, interfaces and process steps.
3. Identify credible failure modes.
4. Describe local, next-level and end effects.
5. Identify causes and mechanisms, not merely symptoms.
6. Record existing prevention and detection controls.
7. Assess risk using the organisation’s defined criteria.
8. Select actions, prioritising elimination or reduction of causes.
9. Assign owners and completion dates.
10. Reassess residual risk and retain objective evidence.

Prevention is stronger than detection. For example, a keyed connector that prevents incorrect assembly is generally more effective than a final inspection intended to discover reversed connections. FMEA should be updated after design changes, field failures, supplier changes, process deviations and newly available evidence.

5. Design of Experiments and Robust Design

Changing one factor at a time is inefficient when factors interact. DOE varies selected factors systematically so that main effects, interactions and, where appropriate, curvature can be estimated with fewer trials than an unstructured test programme.

For a two-factor first-order model, a useful representation is:

Here y is the measured response, x₁ and x₂ are coded factor levels, β coefficients represent effects, β₁₂ represents interaction, and ε is random experimental error. The model is appropriate only as an approximation over the investigated region and requires suitable randomisation, replication or another basis for estimating error, and attention to residual behaviour.

A positive interaction means the effect of one factor depends on the level of the other. For example, adhesive cure time may have little effect at low humidity but a substantial effect at high humidity. A one-factor-at-a-time experiment could miss this relationship.

Planning a sound experiment

The experimenter should:

• state the engineering question and response variables;
• identify controllable factors and noise factors;
• define practical factor ranges;
• select a design such as full factorial, fractional factorial, response-surface or mixture design;
• randomise run order where feasible;
• block known nuisance variation;
• include replication or repeated measurements;
• specify measurement-system requirements;
• analyse residuals and practical significance, not only p-values;
• confirm the selected settings in a separate confirmation run.

Fractional factorial designs reduce runs but may confound effects. Resolution, alias structure and the plausibility of higher-order interactions must be understood before conclusions are drawn. Statistical significance does not guarantee engineering importance; a tiny improvement may be statistically detectable but economically irrelevant.

Robust design

Robust design seeks settings that reduce sensitivity to noise factors such as temperature, vibration, material variation, operator technique or supply voltage. The objective is not necessarily to control every noise factor tightly; it may be cheaper to choose a design that is intrinsically less sensitive.

A signal-to-noise measure is sometimes used in Taguchi-style studies. For a “smaller-is-better” response with observations y₁, …, yₙ, one common definition is:

The response must be positive and measured consistently; the logarithm is dimensionally problematic unless a reference or consistent unit convention is understood, so comparisons should use the same units and measurement definition. Other forms apply to larger-is-better or nominal-is-best objectives. The S/N statistic should supplement, not replace, direct analysis of mean performance, variability and customer risk.

6. Reliability Analysis

Reliability analysis quantifies the likelihood of successful operation over time. Let T be the random time to failure. The reliability function is:

where R(t) is the probability of surviving beyond time t. The cumulative distribution function is F(t) = P(T ≤ t) = 1 − R(t), assuming failure eventually occurs within the modelled population.

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

It represents the conditional failure rate at time t, given survival to t. It is not simply the proportion failing at that instant.

Exponential model

If the failure rate λ is constant and failures are independent under stable conditions, then:

The mean time to failure is:

These relationships apply to non-repairable items under the constant-rate exponential assumption. They should not be used automatically for products with wear-out, early-life defects or changing environments.

For a component with λ = 0.0002 failures per operating hour, reliability at 1,000 hours is:

Thus approximately 81.9% survive to 1,000 hours under the stated model, and MTTF = 5,000 hours. This is a model-based estimate, not a guarantee for an individual item.

Series and parallel systems

For a series system, every component must operate. If component failures are independent:

For two independent parallel-redundant components, the system works if at least one works:

Independence is a major assumption. Shared power, environment, software, maintenance errors or common manufacturing causes can defeat apparent redundancy. Common-cause failures must therefore be considered explicitly.

Bathtub curve and evidence

The bathtub concept describes early-life failure, a relatively stable useful-life period and wear-out failure. It is a conceptual model, not a universal law. Reliability evidence may come from accelerated testing, field returns, warranty data, life testing, supplier records and physics-of-failure analysis. Accelerated tests are useful only when the acceleration mechanism does not change the failure mode; extrapolation outside validated conditions can be misleading.

7. Maintainability, Availability and Service Design

Reliability alone is insufficient for repairable assets. Maintainability concerns how quickly and consistently a failed item can be restored. Design features include modular replacement, access, diagnostic information, standard fasteners, safe isolation, error-proof connectors and clear work instructions.

A commonly used steady-state availability approximation is:

where MTBF is mean time between failures and MTTR is mean time to repair. The approximation assumes a repairable system with stable average rates and includes the relevant repair time definition. If logistics, waiting for parts, administrative delays or preventive maintenance are excluded from MTTR, the resulting availability may be overstated.

For MTBF = 2,000 hours and MTTR = 10 hours:

Availability is therefore approximately 99.50% under the specified assumptions. Improving maintainability can increase availability without changing intrinsic component reliability.

Service design should also consider human factors. Alarms must be actionable, maintenance tasks must minimise foreseeable error, and access should not require unsafe postures or bypassing protective devices. A design that depends on perfect operator attention is not robust.

8. Lifecycle Quality Planning

Lifecycle quality planning integrates design outputs with procurement, production, service and end-of-life controls. A practical plan links each CTQ to its specification, measurement method, process owner, control method, reaction plan and evidence location.

Key elements include:

• design and process FMEA;
• drawings, models, software and configuration baselines;
• tolerance and interface analysis;
• supplier capability and change-control requirements;
• measurement-system analysis;
• control plans and process-monitoring methods;
• verification and validation protocols;
• packaging, transport and environmental controls;
• maintenance and spare-parts strategy;
• warranty and field-feedback loops;
• corrective and preventive action;
• disposal, recycling and environmental obligations.

Quality planning must distinguish inspection limits from customer specification limits. A control limit is calculated from process behaviour; a specification limit is imposed by requirements. A process can be statistically stable but incapable of meeting specification, or apparently capable during a short study but unstable over time.

For a normally distributed characteristic with process mean μ and standard deviation σ, a common capability index is:

where USL and LSL are upper and lower specification limits. The index measures potential capability if centred. A centred capability of 1.33 corresponds to a six-standard-deviation process spread occupying 75% of the specification width, under the normality and stability assumptions. Actual performance also depends on centring; therefore Cₚ alone is insufficient. A centred-performance index is commonly:

These indices should not be used when the process is unstable, the measurement system is inadequate or the distributional assumption is inappropriate without suitable adaptation.

9. Integrated Application: Illustrative Electric Pump Case

Background

An industrial electric coolant pump is being redesigned for a high-temperature manufacturing line. Customers require dependable flow, low noise, simple servicing and safe operation. Field history from the previous model shows intermittent leakage, premature bearing wear and incorrect replacement of a connector during maintenance.

Problem analysis

The QFD team translates “dependable flow” into flow-rate stability across temperature and supply-voltage ranges. “Easy servicing” becomes a maximum replacement time, defined access clearance and a keyed connection. Preliminary FMEA identifies three high-consequence risks: seal leakage causing equipment contamination, bearing seizure causing loss of cooling, and incorrect wiring causing reverse rotation.

Reliability analysis indicates that the bearing is exposed to vibration and temperature combinations not represented in the previous qualification test. The design review also finds that the maintenance instruction assumes a tool unavailable at several customer sites.

Approach and solution

The team takes the following actions:

1. QFD: prioritises leakage prevention, stable flow and service accessibility.
2. Design review: adds manufacturing, field-service and safety personnel before design freeze.
3. FMEA: replaces a visual connector check with a mechanically keyed connector and electrical interlock.
4. DOE: studies seal material, surface finish, compression and temperature. Interaction analysis shows that the best seal material depends on surface finish.
5. Robust design: selects a geometry less sensitive to thermal expansion rather than relying only on tighter assembly control.
6. Reliability testing: introduces temperature-vibration combinations and confirms that the intended failure mode remains the same under accelerated conditions.
7. Maintainability: redesigns the bearing cartridge for modular replacement and verifies the task with representative technicians.
8. Lifecycle planning: adds supplier process controls for seal dimensions, a reaction plan for drift and field-data review at monthly intervals.

Outcome

Confirmation testing demonstrates stable flow across the specified environment, no leakage in the qualification sample and a substantial reduction in service time. The final decision is not based solely on a lower FMEA score; it is supported by objective tests, controlled design changes, traceability and explicit residual-risk acceptance.

Learning points

The case illustrates that failure prevention is systemic. A connector problem can be solved through design error-proofing, not merely retraining. Reliability improvement requires realistic environments and attention to interactions. Maintainability affects operational quality and must be designed alongside performance. Finally, FMEA, DOE and testing are complementary: none is sufficient in isolation.

10. Governance, Ethics and Contemporary Practice

Design organisations have a duty to protect users, workers, communities and the environment. Concealing adverse test results, weakening acceptance criteria to meet a launch date, or treating foreseeable misuse as solely the customer’s responsibility is unethical and potentially unlawful. Safety-critical decisions require competent review, independent challenge and documented acceptance of residual risk.

Regulatory requirements vary by sector, but common governance expectations include controlled records, competent personnel, configuration management, change impact assessment, supplier oversight, cybersecurity and post-market surveillance. Software and connected products add risks involving data integrity, algorithmic failure, remote updates and privacy.

AI and predictive analytics can identify patterns in warranty, sensor or process data, but models may reproduce historical bias, miss rare failures and generate false confidence when operating conditions change. A responsible system documents training data, validation boundaries, uncertainty, human decision authority and escalation rules. AI recommendations should not override mandatory safety controls without accountable engineering review.

Contemporary quality practice increasingly combines digital twins, model-based systems engineering, in-line sensing, prognostics and closed-loop field feedback. These technologies are valuable when measurement validity, cybersecurity, interoperability and change control are addressed. More data does not automatically mean better quality; the data must represent the operating population and support a decision.

Challenges and Limitations

Several limitations require professional judgement:

• QFD priorities depend on subjective weights and relationship scores.
• FMEA can become a paperwork exercise if the team is inexperienced or detached from actual use.
• RPN multiplication can obscure high-severity hazards and treats ordinal ratings inadequately.
• DOE conclusions may not generalise outside the tested factor region.
• Reliability models are only as credible as their assumptions and failure data.
• Accelerated testing can create unrealistic failure mechanisms.
• Capability indices are misleading for unstable or non-normal processes.
• Redundancy may introduce common-cause failures, complexity and maintenance burden.
• Tight tolerances may increase cost without improving customer value.
• Predictive models can fail when products, suppliers or environments change.

The appropriate response is not to abandon analytical tools, but to state assumptions, triangulate evidence, involve diverse expertise and maintain feedback from real operation.

Best-Practice Checklist

A mature design-quality system should:

1. Begin with verified customer, regulatory and lifecycle requirements.
2. Define CTQs and measurable acceptance criteria.
3. Maintain bidirectional requirements traceability.
4. Use cross-functional, evidence-based design reviews.
5. Apply FMEA early and update it with new evidence.
6. Prefer prevention and inherently safe design over detection.
7. Use DOE when interactions or noise factors matter.
8. Confirm measurement-system adequacy before analysing data.
9. Select reliability models that match the failure mechanism and data.
10. Design for maintenance, human factors and safe error recovery.
11. Link design outputs to process controls and supplier controls.
12. Monitor field performance and feed learning into future designs.
13. Document assumptions, uncertainty and residual-risk decisions.
14. Govern automated and AI-supported decisions with human accountability.

Chapter Summary

Design quality establishes the capability of a product, service or process to satisfy needs throughout its lifecycle. QFD translates customer requirements into prioritised technical characteristics, while traceability connects needs to verification and validation evidence. Design reviews provide structured prevention gates and should involve all functions that influence or experience quality.

FMEA identifies potential failure modes, effects and causes so that risks can be reduced before release. Its ratings support prioritisation but do not replace engineering judgement. DOE reveals main effects and interactions, and robust design reduces sensitivity to noise factors. Reliability analysis uses functions such as R(t) = P(T > t) and hazard rate to model survival, while system configurations require careful treatment of series, parallel and common-cause behaviour. Maintainability and availability extend quality analysis to repairable systems.

Lifecycle quality planning links design, suppliers, production, service, maintenance, field data and end-of-life responsibilities. Effective practice is evidence-based, ethically governed and conscious of assumptions. Digital and AI technologies can improve prediction and feedback, but only when data quality, uncertainty, cybersecurity and human accountability are controlled.

Key Terms

Design quality: The capability designed into a product, service or process to satisfy requirements.

Conformance quality: The degree to which delivered outputs conform to specified requirements.

Reliability: Probability of performing a required function for a stated time and conditions.

Maintainability: Probability of restoring an item within a specified time under stated conditions.

Robustness: Ability to perform acceptably despite noise and environmental variation.

CTQ: Critical-to-quality characteristic affecting customer, safety or compliance outcomes.

QFD: Method for translating customer needs into technical and design priorities.

House of Quality: The first QFD matrix linking customer needs with engineering characteristics.

Design review: Formal evidence-based evaluation of design maturity and risk.

FMEA: Prospective analysis of failure modes, effects, causes and controls.

Severity: Seriousness of a failure effect.

Occurrence: Likelihood of a failure cause or mode.

Detection: Likelihood that existing controls will fail to detect a problem.

DOE: Planned experimentation used to estimate factor effects and interactions.

Noise factor: A variable that affects performance but is difficult or undesirable to control.

Reliability function: Probability that an item survives beyond a specified time.

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

MTBF: Mean time between failures for a repairable item or system.

MTTR: Mean time to repair or restore a failed item.

Traceability: Ability to link requirements, design outputs, tests, changes and evidence.

Review Questions

Short Answer Questions

1. Distinguish design quality from conformance quality.
2. What is the purpose of a House of Quality?
3. Why is prevention generally stronger than detection in FMEA?
4. Define reliability and maintainability.
5. State two assumptions required when using the exponential reliability model.

Long Answer Questions

1. Explain how QFD, requirements traceability and design reviews work together to prevent design deficiencies.
2. Describe the complete FMEA process and critically evaluate the limitations of RPN.
3. Explain how DOE can reveal interactions that one-factor-at-a-time testing may miss.
4. Compare verification, validation, reliability testing and maintainability demonstration.
5. Discuss how lifecycle quality planning connects design decisions with suppliers, operations and field performance.

Analytical / Application Questions

1. A repairable system has an MTBF of 1,500 hours and an MTTR of 15 hours. Calculate approximate steady-state availability and state the assumptions behind the calculation.
2. Two independent components in a parallel arrangement each have reliability 0.90 at a specified mission time. Calculate system reliability and discuss why common-cause failure could invalidate the result.
3. A product team identifies a high-severity failure that has a low RPN because occurrence is rare and detection is strong. Recommend an action-prioritisation approach and justify why the RPN alone should not determine the decision.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes verification from validation?

A. Verification evaluates intended user satisfaction, while validation checks dimensions.

B. Verification checks conformity to specified requirements, while validation evaluates intended use and user needs.

C. Verification is performed only after launch, while validation is performed before design.

D. Verification concerns suppliers, while validation concerns only customers.

Correct Answer: B. Verification checks conformity to specified requirements, while validation evaluates intended use and user needs.

Question 2

In QFD, the primary purpose of relationship scores is to show the strength of the link between:

A. Failure modes and maintenance tasks

B. Customer requirements and engineering characteristics

C. Control limits and specification limits

D. Suppliers and warranty claims

Correct Answer: B. Customer requirements and engineering characteristics

Question 3

Which FMEA action is generally the strongest form of risk reduction?

A. Adding a final visual inspection

B. Increasing the number of inspectors

C. Eliminating the failure cause through design prevention

D. Recording the failure in a lessons-learned database

Correct Answer: C. Eliminating the failure cause through design prevention

Question 4

A component has a constant failure rate of 0.0002 per operating hour. Under the exponential model, its reliability at 1,000 hours is approximately:

A. 0.181

B. 0.500

C. 0.819

D. 1.200

Correct Answer: C. 0.819

Question 5

For independent components connected in series, system reliability is:

A. The sum of component reliabilities

B. The product of component reliabilities

C. One minus the product of component reliabilities

D. Always equal to the highest component reliability

Correct Answer: B. The product of component reliabilities

Question 6

In a DOE model, an interaction indicates that:

A. The response has no measurement error

B. One factor’s effect depends on the level of another factor

C. All factors have identical effects

D. The experiment requires no randomisation

Correct Answer: B. One factor’s effect depends on the level of another factor

Question 7

For MTBF = 2,000 hours and MTTR = 10 hours, approximate availability is:

A. 0.5000

B. 0.9500

C. 0.9950

D. 1.0050

Correct Answer: C. 0.9950

Question 8

Which statement about Cₚ is correct?

A. It measures centring only.

B. It measures potential spread capability assuming a stable process and appropriate distributional conditions.

C. It is identical to the defect percentage in every process.

D. It can be calculated without specification limits.

Correct Answer: B. It measures potential spread capability assuming a stable process and appropriate distributional conditions.

Question 9

Why can redundancy fail to provide expected reliability improvement?

A. Redundant systems never require maintenance.

B. Parallel components always have identical failure times.

C. Common-cause failures can disable multiple redundant components.

D. Reliability is unrelated to system architecture.

Correct Answer: C. Common-cause failures can disable multiple redundant components.

Question 10

Which is an appropriate governance control for AI-supported quality prediction?

A. Allowing the model to override safety controls automatically

B. Ignoring model uncertainty when accuracy is high on historical data

C. Documenting data boundaries, uncertainty and human accountability

D. Removing configuration control to accelerate updates

Correct Answer: C. Documenting data boundaries, uncertainty and human accountability