Quality Engineering and Continuous Improvement

Introduction

Quality engineering is the systematic use of engineering, statistical, and managerial methods to prevent defects, reduce variation, improve reliability, and increase customer value. In mechanical engineering, quality is not limited to dimensional conformity after manufacture. It includes functional performance, safety, durability, maintainability, environmental impact, cost, and consistency throughout the product life cycle.

Traditional inspection detects nonconforming output after it has been produced. Quality engineering seeks instead to understand and control the sources of variation in design, materials, equipment, methods, measurement, people, and environment. Continuous improvement then provides a disciplined mechanism for repeatedly improving those sources.

This chapter integrates root-cause analysis, Failure Mode and Effects Analysis (FMEA), fault-tree analysis, reliability concepts, Design of Experiments (DOE), robust design, PDCA, DMAIC, Lean, and Six Sigma. These methods are complementary rather than interchangeable. A sound improvement project selects the method that matches the problem, uses valid data, and verifies that corrective action produces sustained improvement.

Learning Objectives

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

1. Distinguish symptoms, contributing causes, and verified root causes.
2. Apply structured root-cause analysis to manufacturing and product problems.
3. Construct and interpret process FMEA and fault-tree models.
4. Calculate and interpret basic reliability measures.
5. Design and analyze practical experiments for process improvement.
6. Explain robust design and parameter-design principles.
7. Select and apply PDCA, DMAIC, Lean, and Six Sigma methods.
8. Evaluate improvement projects using technical, economic, ethical, and risk-based criteria.

1. Quality Engineering as a Systems Discipline

A product is created by a system consisting of inputs, transformation processes, controls, outputs, and feedback. Inputs include drawings, materials, energy, tooling, software, operators, and measurement systems. A quality problem may therefore originate far upstream from where it is observed.

Quality engineering has four central principles:

• Prevention: design defects and hazards out rather than relying only on inspection.
• Variation reduction: control both common-cause variation and special causes.
• Evidence-based decisions: use reliable measurements and appropriate statistical reasoning.
• Closed-loop learning: standardize successful changes and continue monitoring performance.

A process characteristic may be represented as:

Here, y is the response or quality characteristic, x₁ … xₖ are controllable or influential factors, and ε represents unexplained variation. The relationship is not necessarily linear. The engineering task is to identify important factors, understand their interactions, and control them economically.

A capable process produces output consistently within specification. However, specifications are not the same as customer requirements. A component can satisfy a tolerance while still failing because of poor fatigue life, difficult assembly, excessive noise, or inadequate corrosion resistance.

2. Root-Cause Analysis

2.1 Symptom, cause, and root cause

A symptom is the observed undesirable condition, such as “shaft runout exceeds specification.” A cause is a condition that contributes to the symptom, such as incorrect chuck alignment. A root cause is a fundamental, verified condition whose removal prevents recurrence within the relevant system. Root causes may involve technical design, process control, maintenance, training, supplier management, or organizational decisions.

Root-cause analysis should not be reduced to assigning blame. Its purpose is learning and prevention. A credible conclusion is supported by evidence, reproduces the failure mechanism, and explains why existing controls did not prevent or detect it.

2.2 A practical investigation sequence

1. Define the problem precisely: what happened, where, when, how often, and with what severity?
2. Contain the immediate risk by segregating product, stopping unsafe operation, or applying temporary controls.
3. Confirm the measurement system and verify that the problem is real.
4. Stratify data by machine, shift, material lot, tool, operator, time, and product variant.
5. Generate possible causes using process maps, cause-and-effect diagrams, and expert knowledge.
6. Test causes using observation, controlled trials, records, and physical analysis.
7. Correct the system condition, not merely the visible consequence.
8. Verify effectiveness and update standards, training, risk analysis, and control plans.

The Five Whys asks “Why?” repeatedly until a controllable systemic cause is identified. It is useful for simple causal chains but can be misleading when causes branch, interact, or involve probability. A fishbone diagram organizes possible causes under categories such as machine, method, material, measurement, manpower, and environment. Neither tool proves causation; both generate hypotheses that require verification.

2.3 Distinguishing common and special causes

Common-cause variation is inherent in the current process system. Special-cause variation arises from an identifiable, unusual condition, such as a broken tool, incorrect material, or sensor failure. Statistical process control helps distinguish these conditions, but a control chart signal is an indication for investigation, not automatic proof of a particular cause.

3. Failure Mode and Effects Analysis

FMEA is a structured, prospective method for identifying how a product or process might fail, why it might fail, what effect failure would have, and how risk can be reduced. A design FMEA focuses on product functions and design weaknesses; a process FMEA focuses on manufacturing and assembly operations.

Typical columns include function, requirement, failure mode, effect, severity, cause, occurrence, current prevention controls, current detection controls, detection rating, recommended action, responsibility, and completion evidence.

Historically, organizations calculated a Risk Priority Number:

S is severity, O occurrence, and D detection, each assigned an ordinal rating. RPN is useful as a screening aid, but it has serious limitations: ordinal ratings are not truly interval measurements, different combinations can yield the same RPN, and a low-occurrence but catastrophic hazard may be understated. Modern FMEA practice therefore emphasizes action priority, severity, prevention, detection, and explicit treatment of high-severity hazards rather than ranking solely by RPN.

FMEA is strongest when conducted by a cross-functional team using actual field, test, process, and supplier data. It should be revised after design changes, process changes, complaints, incidents, and lessons learned. A completed worksheet without implemented action is not risk reduction.

4. Fault-Tree Analysis

Fault-tree analysis (FTA) is a deductive, top-down technique. It begins with a defined top event, such as “hydraulic brake fails to generate pressure,” and decomposes that event into contributing failures using logical gates.

For statistically independent events A and B:

For an AND combination:

The equations assume clearly defined events and independence where multiplication is used. Common-cause failures, dependent events, and conditional probabilities require a more careful model. FTA is particularly valuable for safety-critical systems because it reveals single-point failures, redundant paths, and combinations of faults. FMEA is generally bottom-up, asking how each item can fail; FTA is top-down, asking what combinations can produce a specified system-level event. They are complementary.

5. Reliability Engineering

Reliability is the probability that an item performs its required function for a stated time under stated conditions. It is therefore incomplete to say that a component is “reliable” without defining its mission, environment, load, and duration.

For a nonrepairable item, the reliability function is:

T is the random failure time and F(t) is the cumulative distribution function. The failure probability by time t is F(t).

The failure-rate or hazard function is:

where f(t) is the failure-time probability density. The hazard is an instantaneous conditional rate; it is not generally the same as the fraction of units that fail during a time interval.

For an exponential lifetime distribution with constant failure rate λ:

The mean time to failure is:

These exponential expressions apply only when a constant hazard is a reasonable approximation. Mechanical products often exhibit infant mortality, a useful-life region, and wear-out; a Weibull model may represent these stages more realistically:

Here, η is the characteristic life and β is the shape parameter. β < 1 suggests decreasing hazard, β ≈ 1 approximates constant hazard, and β > 1 suggests increasing hazard.

For a series system, all components must operate:

For independent parallel components where the system works if at least one operates:

Independence and the stated system logic are essential assumptions. Redundancy may not improve reliability if components share a power supply, environment, software, or common manufacturing defect.

Worked example. Three independent series components have reliabilities 0.98, 0.97, and 0.99 for a specified mission. Thus:

The system reliability is approximately 94.1%, lower than every component reliability because every component must succeed.

6. Design of Experiments and Robust Design

Changing one factor at a time is easy to understand but can be inefficient and can conceal interactions. DOE varies factors systematically so that their effects and interactions can be estimated with fewer, more informative trials.

For two factors, a first-order model may be written:

The coefficients describe the intercept, main effects, and interaction in coded factor units. The model is an approximation over the selected experimental region. It should not be extrapolated far beyond that region without evidence.

A full two-level factorial with k factors requires 2ᵏ treatment combinations before replication or center points. Replication estimates experimental error; randomization protects against time-related bias; blocking controls known nuisance variation. Analysis commonly uses effects, confidence intervals, analysis of variance, residual plots, and confirmation runs. Statistical significance does not automatically imply engineering importance: a small effect may be statistically detectable but economically irrelevant.

Response-surface methods add curvature terms, for example:

They are useful for locating a process operating region rather than merely identifying important factors.

Robust design

Robust design seeks settings that make performance less sensitive to noise factors such as temperature, material variation, wear, or operator technique. The objective is not necessarily to eliminate every noise source, which may be expensive or impossible, but to reduce the sensitivity of the response.

A signal-to-noise measure for a “smaller-is-better” response is often expressed as:

where n is the number of observations and yᵢ are nonnegative measured responses. Other objectives—larger-is-better and nominal-is-best—require different definitions. These measures are useful screening tools, not substitutes for understanding physical mechanisms and customer loss.

7. PDCA, DMAIC, Lean, and Six Sigma

PDCA

Plan–Do–Check–Act is a general learning cycle. Plan defines the problem, objective, method, risks, and measures. Do tests the change, preferably on a limited scale. Check compares results with the baseline and examines unintended effects. Act standardizes successful practice or revises the hypothesis. PDCA is flexible and appropriate for incremental improvement and learning.

DMAIC

DMAIC is a more structured Six Sigma framework:

• Define: establish the customer-critical problem, scope, team, and business case.
• Measure: define operational measures, validate the measurement system, and establish baseline performance.
• Analyze: identify and verify relationships between causes and responses.
• Improve: develop, test, optimize, and implement solutions.
• Control: maintain gains using process controls, reaction plans, audits, and monitoring.

DMAIC is suited to an existing process with an incompletely understood performance gap. For a new product or process, DFSS or DMADV approaches may be more appropriate.

Lean

Lean focuses on value from the customer’s perspective and the removal of waste. Common waste categories include defects, overproduction, waiting, unused talent, transportation, inventory, motion, and extra processing. Value-stream mapping distinguishes value-adding, necessary but non-value-adding, and avoidable activity. Lean tools include standard work, 5S, visual management, pull systems, quick changeover, mistake-proofing, and total productive maintenance.

Lean speed without process capability can accelerate defects. Six Sigma analysis without attention to flow can optimize an isolated operation while increasing total lead time. Integrated Lean Six Sigma therefore combines flow, waste reduction, variation reduction, and disciplined problem solving.

8. Measurement, Capability, and Control

Improvement conclusions are only as credible as the data. A measurement-system analysis should consider bias, repeatability, reproducibility, stability, resolution, and suitability for the tolerance and decision being made. If a gage is inconsistent, apparent process variation may actually be measurement variation.

For a stable, approximately normal process with specification limits LSL and USL, potential capability is often summarized as:

Capability relative to centering is:

μ is the process mean and σ is the within-process standard deviation. These indices assume a stable process and meaningful estimates of variation. A high index calculated from an unstable process is misleading. Capability indices also do not prove that the product meets functional requirements if the specification itself is inadequate.

Control charts distinguish process monitoring from product acceptance. A control limit is calculated from process behavior; a specification limit comes from engineering or customer requirements. They must not be confused.

9. Integrated Applied Case: Illustrative CNC Shaft Runout Problem

Background

An illustrative manufacturer produces rotating shafts for a small industrial pump. The drawing specifies maximum total indicated runout of 0.050 mm at a bearing seat. Customer returns have increased, and assembly technicians report vibration.

Problem definition and containment

A cross-functional team confirms that the measurement method is suitable and repeats measurements using a calibrated indicator and controlled fixturing. Twenty-four rejected shafts are segregated, and shipments are temporarily subjected to enhanced verification. The team stratifies results by machine, chuck, tool, material lot, and shift.

Analysis

The highest defect concentration occurs on one machine and after chuck changes. A fishbone diagram identifies possible causes: chuck jaw wear, incorrect seating, thermal growth, material distortion, indicator misalignment, and programming error. A Five Whys discussion suggests “operator did not tighten the chuck sufficiently,” but observation shows that tightening torque varies because no controlled method exists and worn jaws can produce runout even at the correct torque.

The process FMEA identifies chuck condition and seating as high-priority risks. An FTA for “runout exceeds limit” shows alternative paths involving chuck misalignment OR incorrect seating OR thermal distortion, with a common measurement branch that could create false rejection. Physical checks confirm worn jaws and inadequate seating verification.

Improvement approach

The team replaces worn jaws, introduces a defined cleaning and seating procedure, adds a torque-controlled chucking method where appropriate, and establishes a setup verification using a reference artefact. A small factorial experiment examines chucking method, warm-up condition, and cutting speed. The experiment is randomized within practical blocks and includes replicated center conditions. Results indicate that chucking method has the largest effect and that its interaction with warm-up condition is important.

A revised control plan requires first-piece verification, periodic reference checks, jaw-life tracking, and a reaction plan for runout signals. Standard work is updated, and operators are trained. The improvement is not accepted merely because the average runout decreases; the team verifies distribution, stability, measurement adequacy, cycle time, and unintended effects such as jaw damage and increased setup cost.

Outcome and learning points

In this illustrative case, the important lesson is that a visible dimensional defect arose from an interaction among equipment condition, setup method, and process control. Replacing parts alone would have provided temporary relief. FMEA helped prioritize prevention, FTA clarified combinations leading to the top event, DOE revealed interaction, and PDCA/DMAIC provided the improvement and control structure.

10. Challenges, Governance, and Professional Practice

Quality improvement can fail through poor problem definition, biased sampling, inadequate replication, invalid assumptions, measurement error, confounding, or premature closure on a favored cause. Teams should distinguish correlation from causation and document both supporting and contradictory evidence.

Ethically, engineers must not conceal failures, manipulate data, redefine specifications to avoid nonconformance, or release unsafe products because of schedule pressure. Safety-critical risks require escalation and independent review. Traceability, change control, calibration, record integrity, supplier oversight, and competence management support compliance with quality-management and sector-specific requirements.

Artificial intelligence can assist with anomaly detection, image inspection, predictive maintenance, and analysis of large sensor datasets. It does not remove the need for validated measurement, representative data, explainable decisions, cybersecurity, human oversight, and physical confirmation. A model trained on historical acceptance decisions may reproduce past bias or miss novel failure modes. AI should be treated as a decision-support technology within a controlled engineering process.

Best practice is to connect customer requirements to measurable critical-to-quality characteristics; validate measurement systems; use risk analysis early; select experiments deliberately; verify changes with independent data; standardize successful methods; and monitor for regression. Improvement is complete only when the new performance is sustained and the organization has learned from the change.

Chapter Summary

Quality engineering treats quality as a property of the whole product and process system. Root-cause analysis separates symptoms from verified causes and uses evidence rather than blame. FMEA is a prospective, bottom-up risk-analysis method, while FTA is a deductive, top-down analysis of combinations producing a top event. Reliability measures quantify the probability of successful operation under stated conditions, and system reliability depends on architecture and assumptions.

DOE provides efficient information about factor effects and interactions; robust design reduces sensitivity to noise. PDCA supports iterative learning, DMAIC structures data-driven improvement of existing processes, Lean reduces waste and improves flow, and Six Sigma reduces variation using statistical evidence. Measurement-system adequacy, process stability, ethical conduct, and control plans are essential to credible and lasting improvement.

Key Terms

Root cause: Fundamental verified condition whose removal prevents recurrence.

Common cause: Variation inherent in the current process system.

Special cause: Identifiable unusual source of process variation.

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

Fault-tree analysis: Top-down logical analysis of causes of a defined top event.

Severity: Consequence level associated with a failure effect.

Occurrence: Likelihood or frequency rating for a failure cause.

Detection: Ability of current controls to detect a failure or cause.

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

Failure rate: Conditional instantaneous rate of failure.

MTTF: Mean time to failure for a nonrepairable item.

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

Design of experiments: Planned investigation of factor effects and interactions.

Interaction: Condition in which the effect of one factor depends on another factor.

Robust design: Design that reduces sensitivity to noise factors.

PDCA: Plan–Do–Check–Act improvement cycle.

DMAIC: Define–Measure–Analyze–Improve–Control methodology.

Lean: Improvement philosophy focused on customer value and waste reduction.

Process capability: Comparison of stable process variation with specification width and centering.

Control plan: Documented method for maintaining process performance and responding to signals.

Review Questions

Short Answer Questions

1. Distinguish a symptom, a contributing cause, and a root cause.
2. Why must a measurement system be evaluated before analyzing process data?
3. State one major difference between FMEA and fault-tree analysis.
4. What is meant by a robust process?
5. Why are control limits different from specification limits?

Long Answer Questions

1. Explain a complete root-cause analysis procedure for a recurring machining defect.
2. Compare design FMEA, process FMEA, and fault-tree analysis, including suitable applications.
3. Explain the assumptions and limitations of exponential reliability models.
4. Describe how a factorial DOE can reveal interactions that one-factor-at-a-time testing may miss.
5. Compare PDCA, DMAIC, Lean, and Six Sigma in terms of purpose, structure, and appropriate use.

Analytical / Application Questions

1. A series system contains three independent components with reliabilities 0.98, 0.97, and 0.99. Calculate system reliability and explain why it is lower than each component reliability.
2. A process has USL = 10.12 mm, LSL = 9.88 mm, mean μ = 10.08 mm, and standard deviation σ = 0.020 mm. Calculate Cₚ and Cₚₖ, then interpret the effect of process centering.
3. A plant has recurring pump leakage after assembly. Develop an integrated investigation using measurement validation, stratification, FMEA, FTA, DOE, and a control plan. Explain what evidence would justify each major decision.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes root cause from symptom?

A. A root cause is always the first event in chronological order.

B. A root cause is a verified condition whose removal prevents recurrence.

C. A root cause is the most visible defect in the product.

D. A root cause is any opinion offered by an experienced operator.

Correct Answer: B. A root cause is a verified condition whose removal prevents recurrence.

Question 2

For independent series components, system reliability is calculated by:

A. Adding all component reliabilities

B. Multiplying all component reliabilities

C. Subtracting the lowest reliability from one

D. Averaging the component reliabilities

Correct Answer: B. Multiplying all component reliabilities

Question 3

Which method is primarily deductive and begins with a defined top event?

A. Process FMEA

B. Five Whys

C. Fault-tree analysis

D. Value-stream mapping

Correct Answer: C. Fault-tree analysis

Question 4

In an exponential reliability model, the failure rate is assumed to be:

A. Constant

B. Always increasing

C. Always decreasing

D. Equal to the process mean

Correct Answer: A. Constant

Question 5

The main purpose of randomization in DOE is to:

A. Increase the specification width

B. Reduce the number of factors

C. Protect against bias from time-related or uncontrolled effects

D. Guarantee a statistically significant result

Correct Answer: C. Protect against bias from time-related or uncontrolled effects

Question 6

Which DMAIC phase establishes the baseline and validates operational measurement?

A. Define

B. Measure

C. Improve

D. Control

Correct Answer: B. Measure

Question 7

A high Cp but substantially lower Cpk most directly indicates that the process is:

A. Highly centered

B. Unstable by definition

C. Potentially capable but poorly centered

D. Measured with zero error

Correct Answer: C. Potentially capable but poorly centered

Question 8

Which is an example of a Lean waste category?

A. Customer value

B. Overproduction

C. Process capability

D. Reliability allocation

Correct Answer: B. Overproduction

Question 9

For a two-level full factorial experiment with three factors, the number of treatment combinations is:

A. 3

B. 6

C. 8

D. 9

Correct Answer: C. 8

Question 10

A control limit is primarily based on:

A. Customer tolerance only

B. Observed process behavior and statistical variation

C. The product selling price

D. The maximum possible machine speed

Correct Answer: B. Observed process behavior and statistical variation