ICGB cheat sheet

A project charter states the business case, problem statement, goal statement, scope, team, and preliminary timeline; it authorizes the project and is approved by the sponsor.

SMART goal statements specify metric, baseline, target, and deadline; a defect-reduction goal is written as direction (decrease), metric, baseline, target, and date.

Voice of the customer (VOC) is translated into measurable requirements via a Kano analysis or a CTQ (critical-to-quality) tree, which links needs to measurable Y characteristics.

SIPOC maps Suppliers, Inputs, Process, Outputs, Customers at a high level; it defines process boundaries without detailing internal steps.

Kano model classesifies requirements as must-be, performance, attractive, indifferent, or reverse; must-be absence causes dissatisfaction but presence does not delight.

A problem statement is measurable, time-bounded, and free of cause or solution language; it states gap, location, magnitude, and impact.

Stakeholder analysis rates each party by influence and interest to set communication frequency and escalation paths.

Cost of poor quality (COPQ) sums internal failure, external failure, appraisal, and prevention costs; prevention and appraisal are control costs, failures are the cost of nonconformance.

A business case quantifies the financial and strategic justification; projected savings must be validated by Finance before closure.

Measurement system analysis (MSA) for continuous data uses Gage R&R; AIAG guidance treats %GRR under 10% as acceptable, 10-30% as conditionally acceptable, and over 30% as unacceptable.

Bias is the difference between the observed average and the reference value; linearity is how bias changes across the measurement range; stability is variation over time.

For attribute agreement, kappa above 0.75 indicates good agreement, 0.40-0.75 fair, below 0.40 poor; within-appraiser agreement and appraiser-to-standard agreement are reported separately.

Process capability: Cp = (USL - LSL) / (6 sigma); Cpk = min[(USL - mean), (mean - LSL)] / (3 sigma). Cpk accounts for centering, Cp does not.

Ppk uses the overall sample standard deviation; Cpk uses within-subgroup sigma estimated from R-bar/d2 or pooled standard deviation.

For non-normal data, use a transformation (Box-Cox, Johnson) or a non-normal capability index such as Ppk from a fitted distribution before reporting capability.

DPMO = (defects / (units x opportunities)) x 1,000,000; yield = e^(-DPU) for the Poisson-based rolled throughput yield model.

Rolled throughput yield (RTY) is the product of each step's first-pass yield; it exposes hidden factory losses that final inspection yield masks.

Data types: continuous (variable) data supports X-bar/R, X-bar/S, and I-MR charts; discrete (attribute) data supports p, np, c, and u charts.

Hypothesis test logic: H0 assumes no difference or no effect; the p-value is the probability of data at least as extreme given H0; reject H0 when p < alpha.

Type I error (alpha) rejects a true H0; Type II error (beta) fails to reject a false H0; power = 1 - beta and increases with sample size and effect size.

Parametric test selection: 2-sample t for two means, paired t for paired data, ANOVA for three or more means, chi-square for association between categorical variables.

ANOVA F test compares between-group variation to within-group variation; a significant result requires post-hoc pairwise comparison (Tukey) to locate differences.

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

Correlation coefficient r measures linear association only; regression R-squared is the proportion of variation explained; neither establishes causation.

Multi-Vari analysis stratifies variation by within-piece, piece-to-piece, and time-to-time sources to narrow the search for dominant causes.

A cause-and-effect (fishbone) diagram organizes suspected causes into Man, Machine, Material, Method, Measurement, and Environment categories.

FMEA scores severity, occurrence, and detection on 1-10 scales; RPN = S x O x D, and high severity items are addressed regardless of RPN.

DOE terminology: factor is an input variable, level is a setting, response is the measured output; a full factorial with k factors at 2 levels requires 2^k runs.

Randomization protects against nuisance variables and time-order effects; blocking isolates known sources of variation; replication estimates pure error.

Resolution III designs main effects are confounded with two-factor interactions; Resolution IV main effects are clear but two-factor interactions are confounded with each other; Resolution V two-factor interactions are clear.

A screening design (fractional factorial or Plackett-Burman) identifies vital few factors before optimization; steepest ascent or descent moves toward the optimum region.

Response surface methodology (central composite or Box-Behnken) fits quadratic models to locate optimum factor settings.

Lean tools: 5S workplace organization, value stream mapping to separate value-added from non-value-added steps, kanban pull, and poka-yoke error proofing.

SMED reduces changeover time by converting internal setup (machine stopped) to external setup (machine running).

Kaizen events are short, focused improvement bursts; PDCA (Plan-Do-Check-Act) is the iterative improvement cycle.

Pilot the improvement on a limited scale and confirm results with a hypothesis test or control chart before full deployment.

Control plan elements: process step, characteristic, specification, measurement method, sample size and frequency, control method, and reaction plan.

Control chart selection: continuous data uses I-MR (individuals), X-bar/R (small subgroups), or X-bar/S (larger subgroups); attribute data uses p, np, c, or u.

A process is in statistical control when all points fall within control limits with no nonrandom patterns; control limits come from process data, specification limits come from the customer.

Western Electric rules flag special cause: one point beyond 3 sigma, 8 or 9 consecutive points on one side of centerline, 6 consecutive trending points, and 2 of 3 points beyond 2 sigma.

Cp and Cpk describe capability, not control; a process can be capable but out of control or in control but incapable.

Standardization sustains gains via documented standard work, updated procedures, training records, and mistake-proofing devices.

A response plan defines the action and owner when a control chart signals special cause; the operator's rule is to stop, contain, and escalate.

A project is closed only after the control plan is transferred to the process owner and the financial benefit is verified.

DMAIC sequence: Define the problem and charter, Measure baseline and validate the measurement system, Analyze root causes, Improve with verified solutions, Control with monitoring and standard work.

Project tollgates require deliverable review by a champion or sponsor; a failed gate returns the team to the prior phase rather than advancing.

The Y = f(X) model treats the output Y as a function of input X variables; Measure quantifies Y, Analyze identifies vital X, Improve optimizes X, Control monitors X and Y.

Sample size for a mean test depends on effect size, sigma, alpha, and power; doubling the detectable effect size roughly quarters the required sample size.

Continuous data generally permits smaller samples than attribute data to detect the same effect; prefer measuring a continuous characteristic when feasible.

A root cause is verified by turning the suspected cause on and off and observing the response, not by consensus alone.

A common exam trap: confusing Cp with Cpk, control limits with specification limits, and RPN with a measure of severity alone.

The ICGB exam is multiple-choice; questions often test which tool applies to a stated data type or phase rather than computation.

When a question states a data type, first classify continuous versus discrete, then select the matching chart or test.

When a question gives USL, LSL, mean, and sigma, compute Cpk; if only Cp is asked, centering is irrelevant.

When a question mentions appraiser variation or repeatability and reproducibility, the answer is MSA Gage R&R.

When a question asks how to sustain a gain, the answer is a control plan, standard work, or poka-yoke, not additional inspection.

When a question asks for the first step in Analyze, the answer is to verify the measurement system or stratify the data, not to run a regression.

When a question asks about causation from observational data, the answer is that correlation does not establish causation.

When a question asks which design screens many factors, the answer is a fractional factorial or Plackett-Burman design.

When a question asks about a signal on a control chart, the answer is to investigate special cause and follow the reaction plan.

Cp = (USL - LSL) / 6 sigma; Cpk = min(USL - mean, mean - LSL) / 3 sigma.

DPMO = defects / (units x opportunities) x 1,000,000; sigma level = 0.8406 + sqrt(29.37 - 2.221 x ln(DPMO)) approximately.

RTY = product of step yields; DPU = defects / units; yield = e^(-DPU).

RPN = Severity x Occurrence x Detection, each rated 1-10.

Gage R&R %GRR = (GRR variation / total variation) x 100; under 10% acceptable, 10-30% conditional, over 30% unacceptable.

Kappa = (Po - Pe) / (1 - Pe), where Po is observed agreement and Pe is chance agreement.

Full factorial 2^k runs; Resolution III, IV, and V describe confounding patterns.

Power = 1 - beta; increasing sample size, effect size, or alpha increases power.

Control limits are typically mean +/- 3 sigma; specification limits are set by the customer or design.

Do not confuse control limits (from process variation) with specification limits (from requirements).

Do not report Cp when the process is off-center; report Cpk or Ppk.

Do not use an X-bar/R chart for individual measurements; use an I-MR chart.

Do not use a p chart for count data with a constant sample size; np, c, or u may apply.

Do not claim a cause from correlation; verify with a designed experiment or a turn-on/turn-off test.

Do not skip MSA before capability analysis; a bad measurement system invalidates both.

Do not treat a high RPN as the only FMEA priority; high severity items are addressed regardless of RPN.

Do not close a project without a control plan and verified financial benefit.

Do not confuse Type I and Type II errors; alpha is the producer's risk, beta is the consumer's risk.

Define: charter, VOC, CTQ tree, SIPOC, stakeholder analysis, Kano model, COPQ.

Measure: data collection plan, MSA Gage R&R, capability analysis (Cp, Cpk, Ppk), DPMO, RTY, control chart baseline.

Analyze: hypothesis tests (t, ANOVA, chi-square), regression, multi-Vari, fishbone, FMEA, Pareto analysis.

Improve: DOE (full factorial, fractional factorial, response surface), lean tools (5S, VSM, SMED, poka-yoke), kaizen, PDCA.

Control: control plan, control charts, standard work, mistake-proofing, response plan, project closure.

Pareto analysis applies the 80/20 rule to prioritize the vital few causes.

A run chart detects trends and shifts before control limits are established.

Central limit theorem: sample means approach normality as sample size grows, enabling parametric tests on non-normal populations.

Confidence interval width decreases as sample size increases and as variability decreases; a 95% interval excludes the true mean 5% of the time.

A p-value is not the probability that H0 is true; it is the probability of the observed data given H0.

Statistical significance does not imply practical significance; compare effect size to the CTQ requirement.

Randomization, replication, and blocking are the three pillars of experimental design.

A confounded effect cannot be separated from the variable it is aliased with without additional runs.

Normal probability plots assess normality; a straight line indicates a normal distribution.

The seven wastes: overproduction, waiting, transport, overprocessing, inventory, motion, and defects.

Takt time = available production time / customer demand.

Little's Law: average inventory = average throughput x average flow time.

Value-added steps transform the product in a way the customer pays for; non-value-added steps are waste.

A pull system (kanban) replenishes based on consumption rather than forecast.

Poka-yoke prevents errors at the source rather than detecting them downstream.

Standard work documents the current best method, takt time, and required inventory.

A champion removes barriers and secures resources; the sponsor approves the charter and funding.

A green belt leads projects within their own area; a black belt leads cross-functional projects and mentors green belts.

RACI assigns Responsible, Accountable, Consulted, and Informed roles for each deliverable.

Team stages: forming, storming, norming, performing, and adjourning.

A tollgate review verifies deliverables and data before the next phase begins.

Change management addresses stakeholder resistance through communication, training, and visible sponsor support.

Hard savings are verified reductions in cost or increases in revenue; soft savings are avoided costs or capacity freed.

COPQ categories: prevention, appraisal, internal failure, and external failure.

A cost-benefit analysis compares project cost to projected benefit over the payback period.

Finance validates savings before project closure; unverified savings are not counted.

A business case links the improvement to strategic objectives and quantifies expected return.

A data collection plan specifies what to measure, how, by whom, when, and the sample size.

Random sampling reduces bias; stratified sampling ensures representation of subgroups.

A check sheet records defect locations and frequencies during collection.

Measurement error adds variation; MSA quantifies repeatability and reproducibility.

Operational definitions ensure all collectors measure the same characteristic the same way.

Sampling frequency should reflect the rate of process change; too infrequent sampling misses shifts.

Histograms show distribution shape, center, and spread; they do not show time order.

Run charts and control charts show time-ordered behavior and detect trends and shifts.

Scatter plots reveal relationships between two continuous variables.

Pareto charts rank causes by frequency or cost to prioritize action.

Box plots display median, quartiles, and outliers compactly.

Multi-Vari charts stratify variation by source to narrow root cause search.

Simple linear regression fits y = a + bx; the slope b estimates the change in y per unit change in x.

R-squared is the proportion of variation in y explained by the model; adjusted R-squared penalizes extra predictors.

Residual plots check linearity, constant variance, and independence assumptions.

Multicollinearity inflates coefficient variance; variance inflation factor above 10 signals a problem.

Logistic regression models a binary response; it predicts probability rather than a continuous value.

X-bar/R charts use subgroup ranges to estimate sigma; X-bar/S charts use subgroup standard deviations for larger subgroups (n > 9).

I-MR charts use moving ranges of individuals; they suit slow processes or single measurements.

p charts track proportion defective with variable sample size; np charts require constant sample size.

c charts count defects per unit with constant opportunity; u charts allow variable opportunity.

Control limits are recalculated only after a verified process change, not after every signal.

One-tailed tests place alpha in one tail; two-tailed tests split alpha across both tails.

Paired t tests analyze before-and-after measurements on the same units.

Chi-square goodness-of-fit tests whether observed frequencies match expected frequencies.

Chi-square test of independence tests association between two categorical variables.

ANOVA assumes normality, independence, and equal variances; violations may require a nonparametric test.

Equivalence of variances is tested with Levene's or Bartlett's test before ANOVA.

Main effect is the average change in response when a factor moves between levels.

Interaction occurs when the effect of one factor depends on the level of another.

Center points detect curvature and estimate pure error in a two-level design.

A confounding structure (alias) defines which effects are indistinguishable in a fractional design.

A response surface design fits a quadratic model to find optimum settings.

Repeatability is variation from repeated measurements by one appraiser on one part.

Reproducibility is variation between appraisers measuring the same parts.

Part-to-part variation is the true process variation the study should capture.

Adequate part selection spans the process range; parts clustered near the mean hide variation.

Attribute agreement analysis reports within-appraiser and between-appraiser agreement plus agreement with the standard.

Capability requires a stable, in-control process; computing Cpk on an unstable process is invalid.

Cpk of 1.33 corresponds to about 63 DPMO for a centered process; 1.67 corresponds to about 0.6 DPMO.

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

A process can be in control but incapable; capability compares spread to specification.

Non-normal data requires transformation or a non-normal capability index before reporting.

Severity rates the effect of failure on the customer, 1 (none) to 10 (hazardous).

Occurrence rates the likelihood of the cause producing the failure, 1 to 10.

Detection rates the ability to detect the failure before it reaches the customer, 1 (certain detection) to 10 (no detection).

RPN prioritizes actions; high severity items are addressed even when RPN is low.

Actions reduce occurrence or improve detection; severity is reduced only by design change.

5S: sort, set in order, shine, standardize, sustain.

Value stream mapping distinguishes value-added, necessary non-value-added, and pure waste steps.

SMED converts internal setup to external setup to cut changeover time.

Kanban signals replenishment based on consumption; it limits work-in-process.

Poka-yoke devices prevent or detect errors at the source.

A charter defines scope, goal, team, timeline, and business case; scope creep is controlled through change approval.

A Gantt chart schedules tasks and dependencies; a critical path is the longest sequence of dependent tasks.

A risk register lists risks, likelihood, impact, and mitigation owner.

Lessons learned are captured at project closure and shared with the organization.

A tollgate review verifies phase deliverables and data quality before advancing.

Deming's 14 points emphasize constancy of purpose and removing fear; his System of Profound Knowledge covers systems, variation, knowledge, and psychology.

Juran's quality trilogy: quality planning, quality control, and quality improvement.

Crosby defined quality as conformance to requirements and promoted zero defects.

Ishikawa introduced the fishbone diagram and quality circles.

Taguchi emphasized robust design and loss function rather than conformance to specification alone.

Acceptance sampling uses a sampling plan with sample size n and acceptance number c.

Producer's risk (alpha) is the probability of rejecting a good lot; consumer's risk (beta) is accepting a bad lot.

An operating characteristic curve plots probability of acceptance against lot quality.

AQL is the acceptable quality level; LTPD is the lot tolerance percent defective.

100% inspection is not a substitute for process control and does not guarantee zero defects.

Nominal data are categories with no order (defect type, color).

Ordinal data have order but unequal intervals (survey ratings, rankings).

Interval data have equal intervals but no true zero (temperature in Celsius).

Ratio data have equal intervals and a true zero (weight, cycle time).

Parametric tests require interval or ratio data; nominal and ordinal data use nonparametric methods.

Common cause variation is inherent to the process and requires management action to reduce.

Special cause variation arises from assignable sources and is detected by control chart signals.

Tampering with a stable process increases variation; Deming's funnel experiment illustrates this.

Process variation is the sum of within-piece, piece-to-piece, and time-to-time variation.

Reducing variation is the primary route to improving capability and yield.

DMAIC applies to improving existing processes; DMADV (define, measure, analyze, design, verify) applies to designing new processes.

A green belt leads projects part-time within their function; a black belt leads full-time cross-functional projects.

A master black belt provides technical leadership and mentors belts.

A champion selects projects, removes barriers, and reports to leadership.

A tollgate review is a formal checkpoint with defined exit criteria.

Verify improvement with a hypothesis test comparing before and after data or with a control chart shift.

A pilot limits risk and confirms results before full deployment.

Financial benefit is validated by Finance before project closure.

Standard work and a control plan sustain the gain after the team disbands.

Document updated procedures and train affected staff before handover.

A root cause is verified by turning the suspected cause on and off and observing the response.

The five whys drills from symptom to underlying cause through repeated questioning.

A fishbone diagram organizes suspected causes but does not verify them.

Multi-Vari analysis narrows the search by stratifying variation by source.

Correlation between a suspected cause and the effect is not proof of causation.

Process step and characteristic identify what is controlled.

Specification and measurement method define the requirement and how it is checked.

Sample size and frequency define the inspection schedule.

Control method identifies the chart, poka-yoke, or inspection used.

Reaction plan names the action and owner when a signal occurs.

Operational definitions ensure consistent measurement across collectors.

Calibration ensures instruments measure accurately against a reference.

Data collection plans prevent missing or biased data.

MSA quantifies measurement system variation before capability analysis.

Data entry errors are detected through verification and double-checking.

A SIPOC defines process boundaries at a high level.

A flowchart details step sequence and decision points.

A value stream map adds time and inventory data to identify waste.

A swimlane map assigns steps to owners across functions.

A deployment flowchart clarifies handoffs and responsibility.

Prevention costs include training, design reviews, and quality planning.

Appraisal costs include inspection, testing, and audits.

Internal failure costs include scrap, rework, and downtime.

External failure costs include warranty, returns, and liability.

COPQ quantifies the financial impact of poor quality and prioritizes improvement.

Benchmarking compares process performance to internal, competitive, or best-in-class references.

Gap analysis quantifies the difference between current and target performance.

Best practice adoption requires adaptation to the local process context.

Benchmarking data must be comparable in definition and measurement method.

Targets set from benchmarks must be achievable and tied to CTQ requirements.

Stakeholder analysis rates influence and interest to plan communication.

Resistance is addressed through communication, training, and visible sponsor support.

Early wins build momentum for the change.

Standard work and training sustain the change after deployment.

Communication plans specify audience, message, channel, and frequency.

A risk register lists risks, likelihood, impact, and mitigation owner.

FMEA prioritizes failure modes by severity, occurrence, and detection.

Poka-yoke reduces risk by preventing errors at the source.

A control plan defines monitoring and reaction for residual risk.

Risk review occurs at each tollgate.

Projects are selected for strategic alignment, financial impact, and feasibility.

A Pareto analysis identifies the vital few problems worth solving.

A project charter is approved by the sponsor before work begins.

Scope must be defined tightly to avoid scope creep.

Quick wins build credibility for larger projects.

Minitab, JMP, and R are common tools for capability, MSA, and DOE analysis.

Control charts, capability analysis, and hypothesis tests are standard software outputs.

Software output must be interpreted against the study design, not blindly.

Assumption checks (normality, equal variance) precede parametric tests.

Graphical analysis complements numerical output.

Read the question for the data type and phase before selecting a tool.

Eliminate answers that confuse control limits with specification limits.

When asked to sustain a gain, choose control plan or standard work.

When asked to verify a cause, choose a designed test, not consensus.

When asked about correlation and causation, choose the caution answer.

CTQ: a measurable characteristic critical to customer satisfaction.

DPU: defects per unit.

RTY: rolled throughput yield, the product of step yields.

Cpk: capability index accounting for centering.

RPN: risk priority number, S x O x D.

MSA: measurement system analysis.

DOE: design of experiments.

DMAIC phases and their primary tools are the core of the exam.

Capability formulas and control chart selection are high-frequency topics.

MSA and DOE concepts are tested conceptually and numerically.

Lean tools and waste categories appear in Improve questions.

Control plan and standard work appear in Control questions.

Define the problem, measure the baseline, analyze causes, improve with verified solutions, control with monitoring.

Select tools by data type and phase.

Verify causes and improvements with data, not opinion.

Sustain gains with control plans and standard work.

Validate financial benefit before closure.

The exam tests application, not memorization alone.

Practice selecting tools for stated scenarios.

Check formulas for centering and sample size effects.

Confirm control versus capability distinctions.

Confirm special cause rules and reaction plans.

Formulas: Cp, Cpk, Ppk, DPMO, RTY, RPN, kappa.

Charts: I-MR, X-bar/R, X-bar/S, p, np, c, u.

Tests: t, paired t, ANOVA, chi-square, Mann-Whitney, Kruskal-Wallis.

Tools: SIPOC, VOC, CTQ, FMEA, DOE, MSA.

Phases: Define, Measure, Analyze, Improve, Control.