Pareto chart
Ranks categories so the team can focus on the few contributors creating most of the observed effect.
Use count, cost, downtime, or another consequence that matches the project goal.North Rivet Technical Library | Quality, Six Sigma & Process Improvement
A DMAIC Yellow Belt field guide for people who must improve real processes - not merely memorize terminology. Learn how to define a useful problem, collect trustworthy data, verify causes, test improvements, and hold the gain.
Lean and Six Sigma overlap, but they attack different forms of loss. Lean improves flow by removing waste. Six Sigma improves consistency by reducing variation and defects. A strong project often uses both.
Yellow Belt: supports data collection, process mapping, tool use, brainstorming, and local improvements. Green Belt: leads moderate projects, often part-time. Black Belt: leads complex projects and deeper analysis. Master Black Belt: coaches, governs methods, and develops capability. Champion or sponsor: removes barriers and aligns the work with business priorities. Process owner: accepts and sustains the new process.
Teams commonly move through forming, storming, norming, performing, and adjourning. Conflict during storming is not automatically failure; it becomes useful when the team has a clear charter, data, respectful facilitation, and a decision method.
| Tool | Use | Plant-floor caution |
|---|---|---|
| Brainstorming | Generate many possibilities without early judgment. | Do not confuse a long list with verified causes. |
| Multivoting | Reduce a large list to a manageable set for further work. | Popularity is not evidence. |
| Nominal group technique | Collect ideas independently, clarify them, then rank or vote. | Useful when louder voices dominate normal discussion. |
| Agenda and minutes | Define purpose, decisions, owners, dates, and unresolved items. | Record commitments, not a transcript of conversation. |
| Status report | Communicate progress, risk, data, decisions, and support needed. | Use the same metric definitions every time. |
The tools are simple by design. Their value comes from selecting the right tool, using defensible definitions, and interpreting the result in process context.
Ranks categories so the team can focus on the few contributors creating most of the observed effect.
Use count, cost, downtime, or another consequence that matches the project goal.Organizes possible causes under logical branches so the team can investigate rather than guess.
A fishbone produces hypotheses, not proof.Shows the actual sequence, decisions, loops, handoffs, and rework paths.
Map what really happens, not only what the procedure says.Plots a measure in time order to reveal shifts, trends, cycles, and process changes.
Do not scramble the time sequence.Creates a consistent method for recording events at the point where they occur.
Define categories before data collection begins.Displays paired observations to look for a relationship between two variables.
Correlation suggests association; it does not prove cause.Shows the shape, center, spread, skew, gaps, and possible multiple populations in measured data.
A histogram hides time order, so pair it with a run or control chart.Defects per unit allows more than one defect on a unit. Ten defects found on 100 units gives DPU = 0.10.
Defects per million opportunities requires a valid, consistently defined opportunity count.
Rolled throughput yield is the probability of passing every process step without defect or rework.
Define the start and stop events. Machine cycle, operator cycle, queue time, and total lead time are not interchangeable.
Cost of poor quality may include scrap, rework, sorting, downtime, premium freight, returns, warranty, and lost capacity.
Range is easy to understand but uses only the two extreme values. Standard deviation uses every observation.
A line produces 500 assemblies. Inspectors record 38 total defects across 4 defined defect opportunities per assembly. Step yields are 98%, 96%, and 99%.
| Metric | Calculation | Result | Interpretation |
|---|---|---|---|
| DPU | 38 / 500 | 0.076 | 7.6 defects per 100 assemblies. |
| DPMO | 38 / (500 x 4) x 1,000,000 | 19,000 | Opportunity definition must remain stable for comparison. |
| RTY | 0.98 x 0.96 x 0.99 | 93.14% | Only about 93 of 100 units are expected to pass all three steps first time. |
The calculations run locally in the browser; no data is transmitted.
Customer language is often broad: fast, reliable, easy, quiet, accurate. A CTQ translates that need into a measurable characteristic with a unit, target, specification, and method.
A SIPOC establishes high-level boundaries before detailed mapping. Keep the process to roughly four to seven steps. It is a framing tool, not a work instruction.
| Weak statement | Stronger statement | Why stronger |
|---|---|---|
| Operators keep making bad parts. | From May 1 through June 15, Line 3 produced 8.7% assemblies above the 25.40 mm height limit, compared with a 1.5% internal target, causing 47 hours of sorting and rework. | Defines location, period, metric, baseline, requirement, and consequence without assigning an unverified cause. |
| The press is unreliable. | Press A experienced 14 unplanned stops exceeding 10 minutes during the last 30 production days, totaling 19.6 hours of lost scheduled time. | Creates an operational definition for a stop and a measurable baseline. |
A useful charter contains the problem statement, business case, baseline, goal, scope, primary metric, team, owner, milestones, and known constraints. It prevents scope drift and creates an explicit agreement about success.
| Stakeholder | Need | Influence |
|---|---|---|
| Operators | Usable method, clear standard | High process knowledge |
| Maintenance | Service access, fault visibility | High technical influence |
| Quality | Valid measurement and records | Release authority |
| Process owner | Stable output and ownership | Sustains control plan |
Mean is the arithmetic average. Median is the middle ordered value and resists extreme values. Mode is the most frequent value. Range, variance, and standard deviation describe spread in different ways.
Continuous: measured values such as diameter, time, temperature, force, and pressure. Discrete: counts such as defects or stops. Nominal: categories without order. Ordinal: ordered categories such as low, medium, high.
| Element | Question to answer | Example |
|---|---|---|
| Metric | What will be measured? | Final assembly height in millimeters. |
| Operational definition | What exactly qualifies? | Maximum height measured within 30 seconds of press release. |
| Source and method | Where and how? | Digital indicator in fixed nest; automatic timestamp. |
| Sampling | How often and which units? | First five after changeover, then one every 30 minutes. |
| Stratification | Which factors travel with the value? | Press, die set, operator, material lot, shift, temperature. |
| Ownership | Who records and audits? | Operator records; quality audits first shift daily. |
Accuracy describes closeness to a reference. Precision describes closeness among repeated results. Bias is a systematic offset. Linearity asks whether bias changes across the range. Stability asks whether the system changes over time.
Repeatability is variation when the same operator measures the same part repeatedly with the same gauge. Reproducibility is variation between operators, fixtures, stations, or other measurement conditions. Gauge R&R evaluates whether measurement variation is small enough for the intended decision.
Failure mode and effects analysis asks how a process can fail, what the effect would be, why it could happen, what controls exist, and what action should reduce risk. Severity, occurrence, and detection may be combined into an RPN, but a low arithmetic rank must not hide a severe safety or customer risk.
Use process mapping, 5 Whys, fishbone, 8D, force-field analysis, relations diagrams, and matrices to structure investigation. Then design a test that can distinguish among competing explanations.
Common causes are built into the current process system. Special causes are unusual, identifiable influences that change the process. Adjusting the process after every common-cause fluctuation can increase variation; ignoring a special cause allows instability to persist.
Correlation describes the direction and strength of association. Regression models an outcome as a function of one or more predictors. Neither automatically proves causation; time order, mechanism, confounding variables, and experimental evidence still matter.
| Concept | Meaning | Common mistake |
|---|---|---|
| Normal distribution | Continuous, symmetric, bell-shaped model described by mean and standard deviation. | Assuming every manufacturing distribution is normal. |
| Binomial distribution | Counts successes in a fixed number of independent trials with constant probability. | Using it when probability changes or observations are dependent. |
| Skewed distribution | One tail extends farther than the other. | Using mean alone when extreme values dominate. |
| Bimodal distribution | Two peaks suggest mixed populations or operating states. | Combining shifts, machines, tools, or materials without stratification. |
| Null hypothesis | Default statement, commonly no difference or no effect. | Treating failure to reject as proof of equality. |
| Type I error | Rejecting a true null - a false alarm. | Ignoring the selected significance level. |
| Type II error | Failing to reject a false null - a missed detection. | Using too little data or low test power. |
| p-value | Probability of data at least this extreme assuming the null model is true. | Calling it the probability that the null is true. |
| Power | Probability of detecting an effect of a specified size when it exists. | Discussing significance without practical effect size. |
PDCA is a compact learning cycle: plan the change, do it at controlled scale, check the result, and act by standardizing or revising. Kaizen supports ongoing incremental improvement. A kaizen blitz compresses focused improvement into a short, intensive event.
Generate multiple solutions, define criteria, and compare safety, customer effect, expected impact, cost, timing, complexity, maintainability, and risk. Test the solution at controlled scale before full release whenever practical.
| Item | Annual value |
|---|---|
| Scrap reduction | $18,600 |
| Rework labor reduction | $12,400 |
| Recovered production capacity | $9,800 |
| Fixture and sensor implementation | ($14,500) |
| Training and validation | ($2,300) |
| First-year net benefit | $24,000 |
Cost-benefit analysis should include risk, uncertainty, recurring maintenance, implementation downtime, and whether the savings are cash, avoided cost, or recovered capacity. Do not present capacity as cash savings unless the organization can actually use or sell it.
A control plan states what is critical, how it is checked, frequency, sample, method, owner, record, limits, and reaction. A reaction plan must say what to do when the process signals trouble - not merely who to notify.
An X-bar chart monitors subgroup averages while an R chart monitors within-subgroup spread. Both are needed because a process can change in center, variation, or both. Control limits describe expected process behavior; specification limits describe requirements and are not interchangeable.
| Control element | Evidence of completion |
|---|---|
| Standard work / SOP | Approved revision, clear sequence, limits, abnormal conditions, and visual aids. |
| Training | Named roles trained with demonstrated competence, not attendance alone. |
| Document control | Obsolete copies removed; current revision available where work occurs. |
| Monitoring | Named metric, chart or report, review frequency, owner, and reaction path. |
| Maintenance | PM, calibration, spare parts, inspection, and service instructions updated. |
| Ownership | Process owner accepts the control plan and knows escalation requirements. |
| Effectiveness review | Scheduled date and criteria for confirming sustained performance. |
This fictional but realistic example shows how the phases connect. The numbers are instructional and do not represent a North Rivet or customer production process.
Assembly Cell 4 produced 8.7% units above the 25.40 mm upper limit over six weeks, creating 47 hours of sorting and rework. Goal: reduce defects below 2.0% within 10 weeks without increasing cycle time or safety risk. Scope: loading through final height verification; upstream component design is initially out of scope.
A check sheet shows final height is the dominant defect. The team validates the fixed measurement nest, confirms operator repeatability, and stratifies 1,200 observations by press, die set, material lot, shift, temperature, and changeover status.
Run charts show defects cluster during the first 30 cycles after changeover. A fishbone identifies die seating, component stack height, press stop position, debris, and measurement delay. Controlled trials reproduce the defect only when the locating surface contains a specific burr pattern and the die clamp is below a verified seating force. Operator identity does not predict the result after these conditions are controlled.
The team adds a keyed die seating feature, a clamp-force verification sensor, a defined cleaning method, and first-piece height confirmation. A two-week pilot reduces defects to 1.3% with no cycle-time increase. The team checks for sensor nuisance faults, maintenance access, and alternate die compatibility.
The control plan requires clamp-force verification every cycle, first-five measurement after changeover, hourly height sampling, a stop-and-hold reaction for out-of-control signals, and preventive inspection of the locating surface. Results remain below 2.0% for eight weeks across all shifts.
The project separates correlation from cause. Changeovers and operators were initially associated, but the verified mechanism was a physical seating condition plus inadequate clamp force. The permanent actions changed the process and detection system rather than relying on reminders.
| Measure | Baseline | After control | Interpretation |
|---|---|---|---|
| Height defect rate | 8.7% | 1.3% | 85% relative reduction. |
| Sorting and rework | 47 hr / 6 weeks | 6 hr / 6 weeks | 41 hours of labor and capacity recovered. |
| Changeover first-pass yield | 82% | 97% | Improved startup stability. |
| Cycle time | 41.2 s | 41.0 s | No meaningful penalty. |
The current ASQ Certified Six Sigma Yellow Belt Body of Knowledge groups the exam content into five areas. This primer follows that structure while adding manufacturing application and cautionary guidance. It is independent educational material and is not affiliated with, endorsed by, or a substitute for ASQ's official handbook, study guide, question bank, or Body of Knowledge.
| Area | Current ASQ weighting | What to know and apply |
|---|---|---|
| Six Sigma Fundamentals | 20 questions | Six Sigma and Lean principles, roles, teams, decision methods, seven quality tools, DPU, DPMO, RTY, cycle time, and COPQ. |
| Define | 14 questions | VOC, CTQs, project selection, stakeholders, SIPOC, supply chain, charter, communication, project planning tools, and tollgates. |
| Measure | 15 questions | Mean, median, mode, range, variance, standard deviation, data types, collection plans, surveys, check sheets, measurement-system terms, and Gauge R&R concepts. |
| Analyze | 17 questions | 5S, value analysis, FMEA, root cause tools, corrective and preventive action, distributions, variation, correlation, regression, and hypothesis terms. |
| Improve and Control | 14 questions | Kaizen, PDCA, cost-benefit analysis, control plans, X-bar and R charts, document control, work instructions, and SOPs. |
No. The fishbone organizes hypotheses. The team still needs evidence or a controlled test that confirms the suspected cause produces the observed effect.
No. Specification limits describe requirements. Stability is determined from time-ordered behavior and control limits based on the process.
It embeds an unverified cause and lacks a measurable baseline, time window, location, requirement, and consequence.
When data are strongly skewed or contain extreme values, such as repair duration with a few very long events.
Repeatability is variation under the same measurement conditions, commonly the same operator and gauge. Reproducibility is variation between measurement conditions, commonly operators or stations.
A third variable, time trend, selection effect, or common cause may drive both variables. Correlation alone does not establish mechanism or causation.
0.97 x 0.96 x 0.98 = 0.9126, or approximately 91.26%.
A trigger, immediate containment, authority to stop or hold product, diagnostic or escalation steps, disposition, required records, and criteria for restart.
Prevent the error or make it physically impossible. Detection and warning are generally weaker because they still allow the error to occur.
Containment protects the customer from the current effect. Corrective action removes or controls the verified cause to prevent recurrence.
Original technical content and illustrations: Validus Group Inc. This web edition is maintained in the North Rivet Technical Library.
Author: Fred Fisher - President, Validus Group Inc.; Founder & Principal Engineer, North Rivet.
North Rivet develops practical engineering software for real equipment. Validus Group Inc. provides industrial engineering, automation, quality, and precision manufacturing expertise.