EXIN LSSA Green Belt Practical cheat sheet

DMAIC phases: Define, Measure, Analyze, Improve, Control; each phase has defined deliverables and gate reviews before the next phase begins.

DPMO = (defects / (units × opportunities per unit)) × 1,000,000; multiply by 1,000,000 after dividing defects by total opportunities.

Sigma level conversion: yield 99.99966% ≈ 6σ, 99.977% ≈ 5σ, 99.38% ≈ 4σ, 93.32% ≈ 3σ, 69.15% ≈ 2σ; use a standard normal table for exact values.

Rolled throughput yield (RTY) = product of each process step's first-pass yield; RTY falls sharply as step count rises even when individual yields are high.

Cp = (USL − LSL) / (6σ); Cpk = min[(USL − mean), (mean − LSL)] / (3σ); Cpk < 1.0 means the process is not capable of meeting specification.

A process is in statistical control when only common-cause variation is present; control chart rules (e.g., a point beyond 3σ, 7 points on one side of the centerline) signal special-cause variation.

Pareto principle: roughly 80% of problems trace to about 20% of causes; rank defect categories by frequency or cost to prioritize improvement work.

MSA: gauge R&R below 10% of tolerance or process variation is generally acceptable; 10–30% may be conditionally acceptable; above 30% requires action on the measurement system.

Waste (muda) categories: overproduction, waiting, transport, overprocessing, inventory, motion, defects, plus unused talent; overproduction is often called the worst waste because it creates the others.

Value stream mapping distinguishes value-added, necessary non-value-added, and non-value-added steps; lead time is measured from customer request to delivery.

Takt time = available production time / customer demand rate; it sets the pace the process must meet to satisfy demand.

Five S: Sort, Set in order, Shine, Standardize, Sustain; each pillar has an audit criterion and is sustained through visual standards.

Kaizen events are short, focused improvement bursts (typically 3–5 days) targeting a defined process area with measurable before/after metrics.

Poka-yoke (error proofing) prevents defects at the source; examples include fixtures that only accept correct orientation and sensors that stop a machine on misload.

SMED separates internal and external setup activities; converting internal setup to external setup is the primary lever for reducing changeover time.

Kanban is a pull signal authorizing replenishment; it limits work-in-process and ties production to downstream consumption rather than forecast.

Hypothesis test selection: continuous data with two groups uses a 2-sample t-test; paired data uses a paired t-test; more than two group means uses ANOVA.

Type I error (alpha) is rejecting a true null hypothesis; Type II error (beta) is failing to reject a false null; power = 1 − beta.

p-value is the probability of observing data at least as extreme as the sample, assuming the null hypothesis is true; reject H0 when p < alpha.

Correlation coefficient r ranges from −1 to +1; r near zero indicates no linear relationship, not necessarily no relationship at all.

Simple linear regression: y = b0 + b1x; R-squared is the proportion of variation in y explained by the model; residual plots check linearity and constant variance.

Chi-square test of independence compares observed and expected counts in a contingency table; expected count = (row total × column total) / grand total.

Confidence interval for a mean = sample mean ± (critical value × standard error); a 95% interval does not contain the true mean in about 5% of repeated samples.

Sample size for a mean estimate rises with variance and desired confidence and falls as the acceptable margin of error grows; larger n narrows the interval.

Design of experiments (DOE) tests factors at set levels; a full factorial with k factors at 2 levels requires 2^k runs before replication.

Screening designs (fractional factorials) sacrifice some interaction information to reduce run count; resolution IV designs confound main effects with two-factor interactions only.

Main effect is the average change in response when a factor moves from low to high level; interaction exists when the effect of one factor depends on the level of another.

Response surface methodology (RSM) fits curvature (e.g., central composite design) to locate optimum factor settings after screening.

FMEA: risk priority number (RPN) = severity × occurrence × detection; higher RPN indicates higher priority for corrective action.

Control plan records the characteristic, specification, measurement method, sample size, frequency, and reaction plan for each controlled parameter.

Mistake-proofing hierarchy: eliminate the error, prevent the error, detect the error, and only last mitigate its effect; design changes outrank inspection.

Standard work documents the current best method with takt time, work sequence, and standard in-process inventory; it is the baseline for further improvement.

Control charts for continuous data: X-bar and R charts for subgrouped data; individuals and moving range (I-MR) charts for single measurements.

Control charts for attribute data: p chart for proportion defective, np chart for number defective with constant sample size, c chart for count of defects per unit, u chart for defects per unit with varying sample size.

Control limits (typically mean ± 3σ) come from the process data, not from specification limits; specification limits belong to the customer.

Control phase deliverables include a control plan, updated standard work, monitoring charts, response plan for out-of-control signals, and documented project handoff.

A process capability study requires a stable process first; computing Cpk on an unstable process produces misleading results.

Sustaining gains: mistake-proofing, visual management, layered audits, and ownership transfer to the process owner with defined reaction plans.

Project closure documents verified results against the business case, lessons learned, and replication opportunities for other areas.

Project charter elements: problem statement, business case, goal statement (SMART), scope and boundaries, team roles, milestones, and expected benefits.

SIPOC maps Suppliers, Inputs, Process, Outputs, Customers at a high level; it defines scope boundaries before detailed process mapping.

Voice of the customer (VOC) is translated into critical-to-quality (CTQ) characteristics that are measurable and tied to specifications.

Stakeholder analysis classifies stakeholders by influence and interest; high-influence, high-interest parties require active engagement.

A well-formed problem statement describes the gap (actual vs. desired), location, timing, and magnitude without stating a cause or a solution.

Green Belt roles: lead small-scope projects, support Black Belts on larger projects, apply DMAIC tools, and coach team members in basic statistics.

Cost of poor quality categories: prevention, appraisal, internal failure, external failure; external failure costs are typically the most damaging.

Measurement system analysis evaluates bias, linearity, stability, repeatability, and reproducibility; repeatability is equipment variation, reproducibility is appraiser variation.

Data types: continuous (variable) data supports more powerful tests; discrete (attribute) data requires larger samples for the same confidence.

Sampling plan choice: random sampling reduces bias; stratified sampling ensures representation across subgroups; rational subgrouping captures common-cause variation within a subgroup.

Operational definition states what is measured, how, by whom, and the decision rule (e.g., pass/fail criterion) so measurements are repeatable across appraisers.

Measurement error propagates into capability indices; a gauge with poor R&R inflates or deflates observed variation and distorts Cpk.

Data collection plan specifies metric, data type, source, sampling method, sample size, frequency, and owner before data are gathered.

Root cause analysis tools: 5 Whys, fishbone (Ishikawa) diagram, fault tree analysis, and cause-and-effect matrix; verify causes with data, not consensus.

5 Whys traces a problem to a systemic cause by repeatedly asking why; stop when the answer is actionable and within the team's control.

Fishbone categories commonly used: Man, Machine, Material, Method, Measurement, Environment (6M).

Multi-voting and nominal group technique narrow candidate causes before data collection; they prioritize but do not validate.

Scatter plots visualize the relationship between two continuous variables; a pattern suggests correlation, not causation.

Hypothesis testing confirms suspected causes: compare group means or proportions, and report confidence intervals alongside p-values.

Failure mode and effects analysis (FMEA) identifies potential failure modes, their effects, and current controls before failures occur.

Improve phase selects and pilots solutions, then verifies results with data before full implementation.

Solution selection matrix scores options against criteria such as impact, cost, effort, and risk; weighted scoring makes trade-offs explicit.

Pilot studies use a small-scale trial with defined success metrics; compare pilot results to baseline using the same measurement method.

Implementation planning includes owners, dates, resources, communication, and training; a Gantt chart or action plan tracks execution.

Before/after comparison uses the same metric definition; a change in measurement method invalidates the comparison.

Poka-yoke and standard work are preferred over training and admonition for sustaining improvement; design changes outlast reminders.

Cost-benefit analysis compares hard savings (labor, material, scrap) with soft savings (capacity, satisfaction) and accounts for implementation cost.

Lean tools: 5S, value stream mapping, kanban, SMED, standard work, andon, and heijunka (level scheduling).

Heijunka levels production volume and mix over time to smooth demand variation and reduce inventory buffers.

Andon signals an abnormality so help arrives before a defect or stoppage escalates; it supports stop-the-line authority.

Theory of constraints: identify the bottleneck, exploit it, subordinate other processes to it, elevate it, then repeat; improvement elsewhere does not raise throughput.

Little's Law: average work-in-process = throughput × average flow time; reducing WIP shortens lead time at constant throughput.

Overall equipment effectiveness (OEE) = availability × performance × quality; world-class OEE is often cited near 85%.

Cycle time is the time to complete one unit; it differs from lead time, which includes waiting between steps.

Team stages: forming, storming, norming, performing, and adjourning; conflict typically peaks in storming and resolves as norms form.

RACI matrix assigns Responsible, Accountable, Consulted, and Informed roles; exactly one person should be accountable per task.

Effective meetings have a stated purpose, agenda, timebox, assigned facilitator, and documented actions with owners and due dates.

Change management: communicate the why, engage affected staff early, provide training, and address resistance with data and listening.

Green Belt project reviews check DMAIC gate deliverables, data integrity, and alignment with the business case rather than presentation polish.

Lessons learned capture what worked, what failed, and what to replicate; they feed the organization's knowledge base.

Baseline capability is established before improvement; compare post-improvement Cpk and DPMO against the same specification limits.

Statistical significance (p < alpha) does not guarantee practical significance; effect size and business impact must be assessed separately.

Sustained results require control charts showing stability over time after the improvement, not just an immediate post-pilot measurement.

Financial validation converts process metrics (scrap rate, cycle time) into currency using agreed cost standards and is signed off by finance.

Project closure includes updating documentation, transferring ownership, archiving data, and recording replication candidates.

A control plan reaction rule specifies who does what when a control chart signals out of control, including containment and escalation.

Define phase outputs: charter, SIPOC, VOC/CTQ, stakeholder analysis, team selection, and a high-level project plan.

Measure phase outputs: operational definitions, data collection plan, baseline capability, MSA results, and a validated measurement system.

Analyze phase outputs: verified root causes supported by data, prioritized with impact evidence and statistical tests.

Improve phase outputs: selected solutions, pilot results, implementation plan, and updated standard work.

Control phase outputs: control plan, control charts, response plan, documented savings, and handoff to the process owner.

A tollgate review at each phase boundary confirms deliverables are complete and data are trustworthy before resources are committed to the next phase.

Scope creep is controlled by restating the charter's in-scope and out-of-scope boundaries at each tollgate.

Descriptive statistics: mean, median, mode, range, variance, standard deviation, and interquartile range; the median resists outliers better than the mean.

Normal distribution: about 68% of values within 1σ of the mean, 95% within 2σ, 99.7% within 3σ.

Central limit theorem: sample means approach normality as sample size grows, enabling normal-based tests even for non-normal populations with adequate n.

Standard deviation measures spread around the mean; variance is its square; both are sensitive to outliers.

Box plots show median, quartiles, and outliers; overlapping boxes suggest groups may not differ, but a formal test is required.

Histograms reveal shape, centering, and spread; bimodal shapes often indicate mixed streams or two populations in one data set.

Stability and capability are separate questions: a stable process can still be incapable, and an unstable process cannot be reliably assessed for capability.

Cost of poor quality includes scrap, rework, warranty, inspection, and expediting; quantify baseline COPQ to justify the project.

Hard savings reduce actual spend or increase revenue; soft savings free capacity or avoid cost but may not appear in the budget.

Project benefit calculation must state the baseline, the improvement delta, the unit cost, and the annualized volume used.

Payback period = implementation cost / annual net benefit; shorter payback strengthens the business case.

Finance sign-off validates claimed savings against accounting standards before project closure claims benefits.

Prioritization matrices weight criteria (impact, feasibility, cost, risk) and score candidate projects to allocate Green Belt resources.

Process mapping levels: SIPOC (high level), swimlane flowchart (roles and handoffs), and detailed flowchart (decision points and rework loops).

Swimlane maps expose handoffs and waiting time between functions, which are frequent sources of delay and defects.

Spaghetti diagram traces physical movement; it quantifies travel distance and motion waste in a layout.

Value-added ratio = value-added time / total lead time; low ratios indicate large waiting and transport content.

Bottleneck identification: the step with the highest cycle time relative to takt time constrains output; balance work across steps to approach takt.

Process cycle efficiency improvement comes from removing non-value-added steps and reducing queues, not from speeding up value-added work alone.

Kaizen blitz scope: one process, one week, measurable target, empowered team, immediate implementation of low-cost changes.

Rapid improvement events end with implemented changes and an action list, not a report; follow-up audits verify sustainment at 30 and 90 days.

Gemba walks observe the actual process where work happens; data collected at the gemba often contradicts assumptions made in meeting rooms.

Standard work combination sheets show operator motion, machine time, and walking time against takt.

Visual management makes normal versus abnormal obvious at a glance; examples include floor markings, kanban boards, and andon lights.

Daily management systems use short tiered meetings to review metrics, escalate abnormalities, and assign countermeasures with owners.

Project selection criteria: linked to strategic goals, measurable gap, manageable scope (typically 3–6 months for a Green Belt), data availability, and process owner support.

Avoid projects whose root cause is already known and whose solution is a capital purchase; those are implementation tasks, not DMAIC projects.

Problem statement quality check: no cause, no solution, no blame, quantified gap, clear location and time frame.

Goal statement should be specific, measurable, achievable, relevant, and time-bound, and should state the target metric and date.

Business case states the cost of the problem and the expected benefit; without it, the project lacks justification for resources.

Process ownership: the process owner sustains gains after handoff; the Green Belt leads the project but does not own the process long term.

Kano model classifies features as must-be, performance, and delighter; must-be absence causes dissatisfaction while presence does not delight.

Quality function deployment (QFD) translates customer requirements into technical characteristics via a relationship matrix (house of quality).

CTQ tree breaks a vague customer need into measurable requirements with specifications and targets.

Customer complaint data, surveys, interviews, and observation are VOC sources; each has different bias and coverage characteristics.

Critical-to-quality metrics must be measurable with the existing or improved measurement system; otherwise the project cannot verify improvement.

Delighters erode over time and become performance or must-be expectations; VOC must be refreshed, not treated as static.

Sample versus population: statistics describe samples, parameters describe populations; inference estimates parameters from statistics.

Random sampling reduces selection bias; convenience samples can mislead regardless of size.

Sample size for a proportion: larger n is required as the expected proportion approaches 0.5 and as the desired margin of error shrinks.

Stratified sampling divides the population into homogeneous strata and samples within each; it improves precision when strata differ.

Systematic sampling selects every k-th item; it fails if the list has a periodic pattern aligned with k.

Data integrity checks: verify units, timestamps, duplicates, missing values, and outliers before analysis; document any exclusions and rationale.

Normal probability plot: points close to a straight line support normality; curvature or S-shape suggests skew or heavy tails.

Anderson-Darling and Shapiro-Wilk tests assess normality; with large samples, trivial departures can be flagged as significant, so pair tests with plots.

Transformations (log, square root, Box-Cox) can stabilize variance or normalize skewed data before parametric tests.

Nonparametric alternatives: Mann-Whitney for two independent groups, Wilcoxon signed-rank for paired data, Kruskal-Wallis for multiple groups.

Homogeneity of variance matters for ANOVA and t-tests; Levene's or Bartlett's test checks it, and Welch's t-test handles unequal variances.

Independence of observations is assumed by most tests; time-ordered or clustered data violate it and require different methods.

ANOVA tests equality of three or more means; a significant F test indicates at least one mean differs, not which one.

Post-hoc tests (Tukey, Bonferroni) identify differing pairs while controlling family-wise error.

Two-way ANOVA tests two factors and their interaction; an interaction means the effect of one factor depends on the level of the other.

Regression assumptions: linearity, independence, constant variance, and normal residuals; check with residual plots and normal probability plots.

Multicollinearity among predictors inflates standard errors; variance inflation factors above roughly 10 signal a problem.

Adjusted R-squared penalizes added predictors; use it rather than R-squared when comparing models with different predictor counts.

Logistic regression models a binary outcome; coefficients are interpreted on the odds scale via exponentiation.

Statistical process control distinguishes common-cause from special-cause variation; only special causes justify local intervention.

Tampering with a stable process (adjusting on every common-cause fluctuation) increases variation; this is over-adjustment.

X-bar and R charts: R chart monitors within-subgroup variation and is checked first; X-bar chart monitors subgroup means.

I-MR chart uses moving ranges for individual measurements where subgrouping is impractical; it is less sensitive than subgrouped charts.

Attribute charts: p and np for defective units, c and u for defect counts; choose based on constant versus varying sample size.

Process capability requires a stable process and a normal (or transformed) distribution; Ppk uses overall variation while Cpk uses within-subgroup variation.

Reaction plan defines containment, root cause investigation, and escalation when a control chart rule is violated.

Kanban card count is set from demand during replenishment lead time plus a safety buffer; too few cards starve downstream, too many create inventory.

Pull systems authorize production from downstream consumption; push systems schedule from forecast and tend to build inventory.

Supermarket pull, sequential pull, and CONWIP are pull variants; CONWIP caps total WIP in a line.

Heijunka box levels production mix and volume across a time interval to dampen demand swings.

Line balancing assigns tasks to stations so each station's cycle time is at or below takt time, minimizing idle time.

Changeover reduction (SMED) enables smaller lot sizes, which reduce lead time and inventory while supporting pull.

5S audits score each pillar against defined criteria; scores without follow-up actions do not sustain gains.

Visual workplace standards: labeled locations, min/max markings, shadow boards, and color coding make abnormalities visible.

Standard work includes takt time, work sequence, and standard in-process stock; it is the baseline against which improvement is measured.

Layered process audits sample compliance with standard work at defined frequencies and escalate deviations.

Total productive maintenance (TPM) uses autonomous maintenance, planned maintenance, and equipment improvement to raise OEE.

Sustainment failure modes: no owner, no audit cadence, no reaction plan, and reversion when the project team leaves.

Hypothesis test steps: state H0 and H1, choose alpha, select the test, check assumptions, compute the statistic and p-value, and interpret in context.

Two-sided versus one-sided tests: a one-sided test places all alpha in one tail and requires justification before data collection.

Confidence interval and hypothesis test agree for two-sided tests at the same alpha; the interval shows the estimated effect size and precision.

Practical significance: report the effect size (difference in means, proportion difference) alongside the p-value.

Multiple comparisons inflate Type I error; apply Bonferroni (alpha divided by number of tests) or use an omnibus test first.

Statistical power increases with larger sample size, larger effect, and lower variability; underpowered studies risk false negatives.

Pareto chart orders categories by frequency or cost with a cumulative percentage line; it identifies the vital few.

Cause-and-effect matrix scores process inputs against customer outputs to prioritize which inputs to investigate.

Scatter plot and correlation quantify association; regression quantifies the size of the relationship and enables prediction.

Multi-vari study examines variation within a unit, between units, and over time to locate the dominant source.

Stratification splits data by machine, shift, operator, or lot to reveal hidden patterns masked in aggregate data.

Hypothesis tests confirm suspected causes; a cause supported only by team opinion remains unverified.

Pilot design: define success metric, baseline, sample size, duration, and decision rule before starting.

Run charts show time-ordered data with a centerline; a shift of 7 or more points on one side signals a non-random pattern.

Before/after comparison must use the same operational definition and measurement system; otherwise the delta is confounded.

Small pilots may lack power to detect real improvement; report confidence intervals rather than only point estimates.

Scale-up decision criteria should be agreed in advance, including cost, capacity, training, and risk thresholds.

Documentation of pilot conditions enables replication and reveals whether results depend on specific people or shifts.

FMEA steps: list process steps, identify failure modes, rate severity, occurrence, and detection, compute RPN, and act on high-RPN items.

Severity ratings should not be reduced by detection controls; only design or process changes reduce severity.

Detection ratings improve with better controls but do not remove the failure mode; prevention outranks detection.

Recommended actions in FMEA require owners and due dates; the FMEA is a living document updated after changes.

Control plan links each critical characteristic to its measurement method, sample size, frequency, and reaction plan.

Mistake-proofing devices should be validated to confirm they actually prevent or detect the error under real conditions.

Control charts are chosen by data type and subgroup size: I-MR for n=1, X-bar/R for small subgroups, X-bar/S for larger subgroups.

Control limit calculation uses the average moving range or average subgroup range multiplied by standard factors (e.g., 2.66 for I-MR limits).

Out-of-control rules include a point beyond control limits, 7 points on one side, 6 trending points, and 2 of 3 points beyond 2σ.

Specification limits are set by the customer or design; control limits are computed from process data and are not interchangeable with them.

Ppk uses overall standard deviation and is preferred for initial capability studies; Cpk uses within-subgroup variation for ongoing capability.

Capability indices assume a stable, approximately normal process; for non-normal data use transformation or non-normal capability methods.

Project storyboard structure mirrors DMAIC: problem, baseline data, root causes, solutions, results, and control plan.

Results reporting must state baseline, target, actual, and verification method with dates and data sources.

Lessons learned include what worked, what did not, and recommendations for replication; they are recorded before team disbanding.

Knowledge transfer: update standard work, training materials, and control plans so gains do not depend on the project team.

Financial validation requires finance sign-off on savings methodology and annualized volume assumptions.

Closure criteria: control plan in place, process owner acceptance, documentation archived, and open actions assigned with dates.

Green Belt project scope is typically a sub-process within one function or a focused cross-functional problem solvable in 3–6 months.

Black Belts lead broader cross-functional projects and mentor Green Belts; Master Black Belts own methodology and deployment strategy.

Sponsor/champion removes barriers, secures resources, and approves tollgates; without an engaged sponsor projects stall.

Team selection balances process knowledge, data skills, and authority to implement changes.

DMAIC is used for improving existing processes; DMADV (define, measure, analyze, design, verify) is used for designing new processes or products.

Project risk register tracks scope, data, resource, and implementation risks with owners and mitigation actions.