9. Design Quality, Reliability and Risk Engineering

Introduction and Context

Quality is not created only on the production line. A large proportion of product performance, customer satisfaction, warranty cost, safety exposure, and service difficulty is determined much earlier—when customer needs are translated into specifications, architectures, materials, interfaces, and control logic. Once a design has been released, correcting weaknesses becomes progressively more expensive. Design quality therefore concerns the ability of a product or process design to satisfy intended requirements consistently, economically, safely, and throughout its useful life.

Reliability and risk engineering extend this idea across time and uncertainty. Reliability asks whether an item will perform its required function for a specified period under stated conditions. Risk engineering asks what can go wrong, how often it may occur, what consequences may result, and how risks can be reduced or controlled. Together, these disciplines move quality management from detection and correction toward prevention.

For production managers, the subject has particular importance. A product may conform to drawing dimensions and still fail in the field because of thermal cycling, vibration, contamination, software interaction, misuse, or inadequate maintenance. Conversely, an extremely robust design may be commercially unsuitable if it is unnecessarily costly or difficult to manufacture. The managerial task is to balance performance, reliability, safety, manufacturability, maintainability, cost, and environmental responsibility.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the relationship among design quality, reliability, maintainability, availability, and risk.

2. Apply quality function deployment, robust design, design of experiments, and design-for-quality principles.

3. Construct and interpret process and design FMEAs, including risk-priority decisions.

4. Calculate and interpret basic reliability measures and distinguish reliability from maintainability and availability.

5. Explain life-data concepts, failure distributions, censored data, and accelerated testing.

6. Select appropriate mistake-proofing, verification, validation, and control strategies.

7. Evaluate the managerial, ethical, regulatory, and digital implications of reliability decisions.

Design Quality as a Life-Cycle Discipline

Design quality begins with the voice of the customer, but it cannot be reduced to customer satisfaction alone. Requirements may arise from users, regulators, maintainers, production personnel, suppliers, owners, and communities affected by the product. These requirements must be converted into measurable characteristics. For example, “quiet operation” may become a maximum sound-pressure level under defined operating conditions; “safe braking” may become stopping distance, response time, and performance under specified loads and temperatures.

A useful distinction is between quality of design and quality of conformance. Quality of design refers to how well the selected product concept and specifications address legitimate needs. Quality of conformance refers to how consistently production matches the approved design. A well-controlled process cannot compensate for a fundamentally inadequate design, while an excellent design will not perform reliably if manufacturing variation is excessive.

Design decisions should be made through a life-cycle perspective. The relevant stages generally include need identification, concept development, detailed design, supplier development, process design, pilot production, launch, operation, maintenance, retirement, and disposal. Feedback from later stages should inform earlier decisions. Field failures, service records, warranty claims, and end-of-life observations are valuable design knowledge rather than merely after-sales information.

Translating Requirements into Design Characteristics

Quality function deployment (QFD) provides a structured means of translating customer requirements into technical responses. In its familiar “house of quality,” customer requirements are weighted according to importance and compared with engineering characteristics. The relationship matrix identifies how strongly each technical characteristic influences each customer requirement, while a correlation roof highlights reinforcing or conflicting engineering choices.

QFD is not a substitute for engineering judgment. Its value lies in making assumptions visible, encouraging cross-functional discussion, and preventing a technically attractive feature from displacing a more important customer need. A production manager may use QFD to identify characteristics requiring special process capability, supplier controls, inspection, or mistake-proofing.

Design for Manufacture, Assembly, and Service

Design for manufacture and assembly (DFMA) seeks to reduce unnecessary complexity, variation, and cost while protecting function. Typical considerations include reducing part count, standardizing components, providing clear orientation, avoiding inaccessible fasteners, and designing tolerances that are achievable with capable processes. Design for service adds access, diagnosis, replacement time, human factors, and safe isolation of energy sources.

The principle is not simply to make products cheaper. A design that is easy to assemble incorrectly is a quality risk; a design that requires extensive inspection may indicate that the process or design itself should be improved. Cross-functional design reviews should therefore include operators, maintenance personnel, quality specialists, suppliers, and users where feasible.

Robust Design and Variation Management

Robust design seeks performance that is relatively insensitive to sources of variation. These sources may be controllable factors, such as material grade or assembly torque, or uncontrollable noise factors, such as temperature, humidity, vibration, aging, operator behavior, and supply variation.

The Taguchi perspective distinguishes a nominal-the-best, smaller-the-better, and larger-the-better quality characteristic. The quality-loss concept argues that loss often increases continuously as performance moves away from target, even when the item remains within specification. This encourages organizations to aim at target values and reduce variation rather than merely sorting conforming from nonconforming units.

Robust design commonly uses three stages:

• System design: selecting the basic technology and architecture.
• Parameter design: choosing settings that reduce sensitivity to noise.
• Tolerance design: assigning economical tolerances after the design has been made as insensitive as possible.

This sequence matters. Tightening tolerances is often an expensive response to a design that is inherently sensitive. A better solution may be a different material, geometry, control algorithm, operating point, or compensation method.

Design of Experiments for Engineering Decisions

Design of experiments (DOE) is a structured method for learning how input factors affect one or more responses. Unlike one-factor-at-a-time experimentation, DOE can reveal interactions and estimate effects efficiently. A typical engineering study proceeds through problem definition, response selection, factor identification, experimental design, randomization, execution, analysis, confirmation, and implementation.

Consider a sealing process in which leakage is affected by temperature, pressure, dwell time, and material batch. A factorial experiment can reveal not only whether temperature matters but also whether its effect depends on pressure. Analysis may use effects plots, analysis of variance, residual analysis, and prediction intervals. Statistical significance alone is insufficient: engineers must also assess practical significance, measurement-system adequacy, process feasibility, and confirmation results.

Screening designs are useful when many factors may be important. Response-surface methods are useful for optimizing significant factors and locating a stable operating region. Experiments should be conducted with attention to blocking, randomization, replication, and control of nuisance variables. Poorly planned DOE can generate false confidence through confounding, inadequate measurement, or extrapolation beyond the experimental region.

FMEA: Anticipating Failure Before It Occurs

Failure mode and effects analysis (FMEA) is a disciplined, team-based method for identifying how a product or process might fail, why it might fail, what effects may result, and how the risk should be controlled. A design FMEA focuses on product functions and design weaknesses; a process FMEA focuses on manufacturing or service-process steps. Both should be living documents, revised when requirements, materials, suppliers, equipment, controls, or field knowledge change.

A sound FMEA begins with the intended function and boundary of the analysis. For each function, the team identifies potential failure modes, local and end effects, causes or mechanisms, existing prevention controls, existing detection controls, and recommended actions. Controls that prevent a cause are generally stronger than controls that detect a defective output after it has been produced.

Traditional FMEA scoring uses severity (S), occurrence (O), and detection (D), often on ordinal scales from 1 to 10. The risk priority number is calculated as:

RPN = S × O × D

RPN is a prioritization aid, not a measure of probability or expected loss. A low-occurrence failure with catastrophic severity may deserve immediate action even if its RPN is lower than that of a frequent minor defect. Modern practice therefore supplements or replaces simple RPN ranking with action-priority tables, severity rules, criticality analysis, fault-tree analysis, expected-loss reasoning, and explicit escalation criteria.

FMEA quality depends on the quality of the team’s thinking. Generic statements such as “part fails” are inadequate. The failure mode should describe the manner of failure, such as “seal leaks at operating temperature,” while the cause may be “insufficient compression due to tolerance stack-up.” The team should distinguish cause, failure mode, and effect, and should verify that proposed actions have owners, due dates, and evidence of completion.

Reliability: Concepts, Measures, and Models

Reliability is the probability that an item performs its required function without failure for a specified time, in a specified environment, under specified conditions. The phrase “specified conditions” is essential: reliability has no meaningful interpretation without a mission profile, load, duty cycle, maintenance policy, and definition of failure.

If a population has a lifetime random variable T, the reliability function is:

R(t) = P(T > t)

The cumulative distribution function is F(t) = P(T ≤ t), so R(t) = 1 − F(t). The probability density is f(t), and the hazard or instantaneous failure rate is:

h(t) = f(t) / R(t)

For a constant failure rate λ, the exponential model gives:

R(t) = e^(−λt)

and the mean time to failure is MTTF = 1/λ. For repairable equipment, mean time between failures (MTBF) is often used, but it must be interpreted carefully. MTBF is an average interval and is not a guaranteed life, nor does it imply a constant failure rate unless the model supports that assumption.

The Weibull distribution is widely used because it can represent different failure behaviors. Its shape parameter β indicates the trend of the hazard rate. β less than 1 is associated with early failures, β approximately 1 with a roughly constant hazard, and β greater than 1 with wear-out behavior. The scale parameter represents a characteristic life. Weibull analysis can therefore support decisions about burn-in, design improvement, preventive replacement, and warranty policy.

Reliability of a system depends on configuration. For independent components in series, all components must function, so:

Rₛ(t) = R₁(t)R₂(t)…Rₙ(t)

For independent, identical components in parallel, the system functions if at least one functions:

Rₚ(t) = 1 − [1 − R(t)]ⁿ

Redundancy can improve reliability, but it adds cost, weight, interfaces, failure opportunities, and sometimes common-cause vulnerabilities. A redundant system is not automatically reliable if redundant elements share a power supply, software defect, environmental exposure, or maintenance error.

Life Data, Testing, and Reliability Demonstration

Life-data analysis uses observed times to failure and information about units that have not failed by the end of observation. Such observations are right-censored and should not simply be discarded. Field data may also contain left-censoring, interval-censoring, repair effects, changing environments, and incomplete failure records.

Reliability testing can be developmental, comparative, qualification, or demonstration-oriented. Accelerated life testing applies increased stress—such as temperature, voltage, pressure, or vibration—to obtain information in less time. The acceleration model must be physically credible. For temperature-dependent mechanisms, an Arrhenius relationship may be appropriate; for mechanical fatigue, stress-life or strain-life relationships may be more relevant. An acceleration test that activates a different failure mechanism can produce dangerously misleading results.

Reliability growth is the systematic reduction of failure intensity through test, failure analysis, corrective action, and retest. Duane and Crow-AMSAA models are commonly used to describe growth trends, but models do not replace engineering investigation. Test planning should define failure criteria, sample size, confidence level, censoring rules, environmental conditions, and the decision to be made from results.

Maintainability and Availability

Reliability concerns failure-free performance; maintainability concerns the ability to restore function after failure through repair or replacement. Maintainability is influenced by access, modularity, diagnostic information, tools, skills, documentation, spare parts, safety isolation, and human factors. Mean time to repair (MTTR) includes more than hands-on repair: waiting, diagnosis, access, testing, and administrative delay may dominate total downtime.

Availability combines uptime and downtime. A simple steady-state expression is:

A = MTBF / (MTBF + MTTR)

This equation shows why improving maintainability can substantially increase availability even when reliability cannot be immediately improved. Operational availability may additionally include logistics delays, preventive maintenance, supply shortages, and scheduled downtime. Designers should therefore specify not only reliability targets but also diagnostic coverage, restoration time, maintenance intervals, and support requirements.

Mistake-Proofing, Verification, and Validation

Mistake-proofing, or poka-yoke, prevents an error or makes it immediately evident. Examples include keyed connectors, asymmetric locating features, interlocks, presence sensors, torque-controlled fastening, barcode verification, and software validation rules. A warning light is usually weaker than a physical prevention device because it depends on attention and response.

Verification asks whether the design output satisfies specified requirements: dimensions, calculations, code reviews, simulations, inspections, and tests are typical verification activities. Validation asks whether the resulting product satisfies intended use and user needs in the real or representative context. A product can pass verification and still fail validation—for example, a medical device may meet dimensional specifications but be confusing or unsafe for its intended users.

A robust validation strategy includes representative users, realistic operating environments, boundary conditions, foreseeable misuse, maintenance activities, and acceptance criteria established before testing. Safety-critical claims require traceable evidence, controlled configuration, competent reviewers, and documented disposition of deviations.

Risk Engineering and Decision-Making

Risk is commonly considered as a combination of the likelihood of an event and the severity of its consequences. In formal analysis, risk may be represented through a risk matrix, expected monetary loss, fault-tree analysis, event-tree analysis, or quantitative simulation. A risk matrix is useful for communication but can hide important differences within broad categories and should not be treated as a precise probability model.

Fault-tree analysis begins with a top event, such as loss of braking, and logically develops combinations of contributing events using AND and OR gates. Event-tree analysis begins with an initiating event and examines the success or failure of protective barriers. Bow-tie analysis combines the two perspectives, showing preventive controls on the left and consequence-mitigation controls on the right.

The hierarchy of controls favors elimination and substitution, followed by engineering controls, administrative controls, and personal protective equipment. In product design, inherent safety and fail-safe behavior are preferable to warnings alone. Risk acceptance must be governed, documented, and reviewed by authorized personnel; commercial pressure must not silently redefine an unacceptable safety risk as a tolerable quality issue.

An Illustrative Case: Reliability Improvement in an Industrial Pump

Background

A fictional manufacturer, Meridian Process Equipment, produces centrifugal pumps for chemical-processing customers. The pump meets rated flow and pressure requirements during final inspection, but customers report premature seal leakage within the first year. The resulting downtime, field service, and reputational damage are increasing.

Situation and Analysis

The cross-functional team maps the product function: transfer fluid without leakage under specified pressure, temperature, and shaft-speed conditions. A design FMEA identifies several potential failure modes: seal-face distortion, inadequate compression, shaft misalignment, elastomer degradation, and assembly contamination. Field data are incomplete because many customers replace seals without returning failed parts. The team therefore introduces a standardized failure-report form and preserves failed components for analysis.

A tolerance-stack study indicates that the combined variation in shaft runout, housing alignment, and seal-seat position can reduce effective seal compression. DOE is then used with alignment, compression, temperature, and contamination as factors. The experiment shows a significant interaction between temperature and elastomer type, as well as a strong effect of alignment. A purely tighter inspection plan would detect some failures but would not remove the underlying sensitivity.

Strategy and Solution

The design team changes the seal material for the validated temperature range, adds a locating feature that controls alignment, and redesigns the assembly fixture to prevent angular insertion. A poka-yoke device blocks assembly unless the correct seal orientation is detected. The process FMEA is updated, and the critical alignment characteristic receives a capable measurement system and periodic process monitoring.

Reliability testing uses representative pressure and temperature cycles, with both failed and surviving units recorded. The test plan includes right-censored observations and a defined failure criterion. Validation is conducted with customer-representative fluid and maintenance procedures rather than clean laboratory water alone.

Outcome

The revised design reduces sensitivity to alignment variation and provides stronger prevention controls. Before release, the organization confirms the improvement through a controlled pilot, reviews serviceability, updates work instructions, and establishes feedback from early field installations. The important result is not a single test number but a closed learning loop connecting field evidence, FMEA, DOE, design change, process control, and validation.

Learning Points

First, final inspection could not compensate for an unstable design interaction. Second, incomplete field data create a risk of solving the wrong problem. Third, robust design is often more economical than extreme tolerancing. Fourth, FMEA is valuable only when it leads to verified actions. Finally, reliability improvement requires cooperation among design, production, suppliers, service, and customers.

Managerial and Professional Considerations

Reliability work is organizational as well as technical. Senior managers must allocate time for design reviews, testing, failure analysis, and corrective action rather than rewarding only rapid launch. Metrics should distinguish field failure rate, warranty cost, first-time-through quality, test escapes, mean restoration time, and closure effectiveness. Teams should avoid gaming RPN scores or selecting test conditions that make success likely but irrelevant.

Ethically, engineers and managers have a duty to communicate uncertainty, disclose significant failure modes, preserve traceability, and escalate safety concerns. Regulatory compliance is a minimum requirement, not proof of adequate quality. Products affecting health, transport, energy, or public safety require disciplined configuration control, independent review where appropriate, cybersecurity consideration, and evidence that foreseeable misuse has been addressed.

Digital Tools and Emerging Practice

Digital engineering increasingly links requirements, CAD models, simulation, bills of material, process plans, FMEAs, test results, and field records through a digital thread. Reliability dashboards can combine warranty data, sensor streams, maintenance logs, and operating conditions. Predictive maintenance models may estimate remaining useful life, but their output is only as trustworthy as the sensor quality, failure labels, assumptions, and operating-context data.

Artificial intelligence can assist in failure classification, anomaly detection, document analysis, and prioritization of test cases. It should not replace engineering accountability. Models may inherit biased field data, confuse correlation with causation, or fail when operating conditions change. Human review, explainability appropriate to the risk, cybersecurity, data governance, and controlled model updates are essential.

Emerging practice is also moving toward reliability allocation across complex cyber-physical systems, digital twins for design and service decisions, sustainability-informed life-cycle risk, and greater attention to software updates and common-cause failures. Connected products create new opportunities for early detection but also introduce dependence on communication networks, cloud services, data integrity, and cybersecurity controls.

Best-Practice Guidelines

Effective organizations generally:

• define functions, requirements, environments, and failure criteria precisely;
• involve production, service, suppliers, users, and quality specialists early;
• use FMEA to drive preventive actions rather than merely complete documentation;
• reduce sensitivity through robust design before tightening tolerances;
• verify measurement systems before relying on experimental or field data;
• plan reliability tests around credible mechanisms and explicit decisions;
• distinguish reliability, maintainability, availability, and safety metrics;
• treat censored data and uncertainty honestly;
• validate products under realistic use, maintenance, and misuse conditions;
• maintain traceability from requirement to design feature, process control, and evidence;
• review assumptions after field experience, design changes, and supplier changes; and
• escalate high-severity risks regardless of a favorable numerical ranking.

Chapter Summary

Design quality is established through the translation of needs into effective, manufacturable, serviceable, and safe product and process designs. Robust design reduces sensitivity to noise, while DOE provides evidence about influential factors and interactions. FMEA anticipates failure modes and supports preventive action, but numerical rankings must be interpreted with engineering judgment and severity-based governance.

Reliability is the probability of functioning without failure for a defined time and condition. Reliability functions, hazard rates, exponential and Weibull models, system configurations, life-data analysis, and accelerated testing provide analytical foundations for engineering decisions. Maintainability concerns restoration, and availability combines operating time with repair and support performance. Mistake-proofing, verification, validation, fault-tree analysis, and event-tree analysis extend prevention across the life cycle.

The central managerial lesson is that quality assurance is strongest when design, process capability, reliability evidence, field learning, ethics, and digital information are connected. Prevention requires not only tools but also disciplined decisions, cross-functional ownership, and willingness to act on inconvenient evidence.

Key Terms

Design quality: The degree to which a selected design satisfies stakeholder needs, performance requirements, safety expectations, and life-cycle objectives.

Quality of conformance: The degree to which produced items and processes conform to the approved design and specifications.

Quality Function Deployment (QFD): A structured method for translating customer requirements into measurable technical characteristics and priorities.

DFMA: Design for manufacture and assembly; designing products for efficient, consistent production and assembly.

Robust design: Designing performance to be relatively insensitive to controllable and uncontrollable sources of variation.

Noise factor: A variable that affects performance but is difficult or uneconomical to control during normal use.

Design of experiments (DOE): A planned method for studying the effects and interactions of input factors on responses.

FMEA: A systematic method for identifying failure modes, causes, effects, controls, and corrective actions.

Failure mode: The specific manner in which a function or component can fail.

Risk Priority Number (RPN): The product of severity, occurrence, and detection ratings in traditional FMEA.

Reliability: The probability of performing a required function without failure for a specified time and condition.

Hazard rate: The instantaneous conditional rate of failure among units surviving to a given time.

Weibull distribution: A flexible life distribution used to model early-life, random, and wear-out failure behavior.

Censored data: Observations for which the exact failure time is not known, often because a unit survives the test end.

MTTF: Mean time to failure for non-repairable items.

MTBF: Mean time between failures, commonly used for repairable equipment but dependent on operating and repair assumptions.

Maintainability: The ability to restore an item to functional condition within specified time and resource conditions.

Availability: The proportion of time an item is capable of performing its function, considering failures and restoration.

Mistake-proofing: Design or process provisions that prevent errors or make them immediately detectable.

Validation: Evidence that a product satisfies intended use and user needs in representative conditions.

Review Questions

Short Answer Questions

1. Distinguish quality of design from quality of conformance.

2. What is the purpose of a noise factor in robust design?

3. Define reliability and explain why operating conditions must be specified.

4. Why should right-censored life data not simply be discarded?

5. Distinguish verification from validation.

Long Answer Questions

1. Explain how QFD, DFMA, robust design, and FMEA can be integrated during product development.

2. Describe the reliability bathtub concept using hazard-rate behavior and explain its managerial implications.

3. Explain the stages of a well-designed DOE and discuss why interactions matter.

4. Compare reliability, maintainability, and availability, including suitable measures for each.

5. Discuss the strengths and limitations of RPN as a basis for FMEA prioritization.

Analytical / Application Questions

1. A repairable machine has an MTBF of 900 hours and an MTTR of 6 hours. Calculate its steady-state availability and explain one practical way to improve it.

2. An FMEA team assigns severity 10, occurrence 2, and detection 2 to a safety-related failure, while another defect receives scores of 5, 7, and 6. Which should receive attention first, and why might RPN alone be misleading?

3. A product passes all dimensional verification tests but receives repeated complaints that users cannot safely replace a consumable part. Determine whether this is primarily a verification or validation failure and propose a study to address it.

22. Multiple Choice Questions with Answers

Question 1

A design team discovers that a product response is highly sensitive to temperature variation. Which action best reflects robust design?

A. Tighten every dimensional tolerance immediately

B. Identify a design parameter or material that reduces temperature sensitivity

C. Increase final inspection frequency only

D. Remove temperature from the product specification

Correct Answer: B. Identify a design parameter or material that reduces temperature sensitivity

Question 2

In an FMEA, which control is generally strongest?

A. A warning issued after the defect is produced

B. A final visual inspection

C. A design feature that prevents the cause of failure

D. A customer complaint review

Correct Answer: C. A design feature that prevents the cause of failure

Question 3

For a non-repairable item, R(t) = P(T > t) represents:

A. The probability of failure before time t

B. The probability of surviving beyond time t

C. The average repair time

D. The instantaneous failure rate

Correct Answer: B. The probability of surviving beyond time t

Question 4

A Weibull shape parameter greater than one generally indicates:

A. Decreasing hazard associated with early failures

B. A constant hazard rate

C. Increasing hazard associated with wear-out

D. No relationship between age and failure

Correct Answer: C. Increasing hazard associated with wear-out

Question 5

Why is one-factor-at-a-time experimentation often inferior to factorial DOE?

A. It cannot use quantitative responses

B. It cannot reveal interactions efficiently

C. It always requires more observations

D. It eliminates randomization

Correct Answer: B. It cannot reveal interactions efficiently

Question 6

A system has two independent identical components in parallel, each with reliability 0.90 at a specified time. What is the system reliability?

A. 0.09

B. 0.81

C. 0.90

D. 0.99

Correct Answer: D. 0.99

Question 7

Which statement best distinguishes validation from verification?

A. Verification tests intended use; validation checks drawings

B. Verification checks specified outputs; validation checks suitability for intended use

C. Verification is performed only after launch; validation only before design

D. They are identical activities with different names

Correct Answer: B. Verification checks specified outputs; validation checks suitability for intended use

Question 8

A machine has MTBF of 100 hours and MTTR of 5 hours. Its simple steady-state availability is closest to:

A. 0.050

B. 0.500

C. 0.952

D. 1.050

Correct Answer: C. 0.952

Question 9

Why can a low RPN fail to justify postponing action on a safety-critical failure?

A. RPN does not include any numerical ratings

B. Low occurrence or good detection can mask extremely high severity

C. RPN applies only to process FMEA

D. Severity is excluded from the calculation

Correct Answer: B. Low occurrence or good detection can mask extremely high severity

Question 10

A reliability test ends with several units still functioning. These observations are best treated as:

A. Invalid data that must be removed

B. Right-censored life data

C. Proof that failure is impossible

D. Automatic evidence of constant hazard

Correct Answer: B. Right-censored life data