CSSC MBB cheat sheet

DMAIC phases: Define (charter, VOC, CTQ, SIPOC), Measure (baseline capability, MSA), Analyze (root cause verification), Improve (solution selection, pilot), Control (SPC, response plan, handover).

DPMO = (defects / (units × opportunities)) × 1,000,000; process sigma is the Z-value of the defect rate shifted by 1.5 sigma in the classic Six Sigma convention.

Rolled throughput yield RTY = product of each process step's first-pass yield; a 5-step process at 95% each yields RTY ≈ 0.77.

COPQ categories: prevention, appraisal, internal failure, external failure; only prevention and appraisal are cost-of-good-quality investments.

Lean's eight wastes (TIMWOODS): Transport, Inventory, Motion, Waiting, Overproduction, Over-processing, Defects, Skills underuse.

Kaizen, 5S, kanban, poka-yoke and value stream mapping are Lean tools used inside DMAIC Improve/Control, not replacements for it.

CTQ tree translates a vague VOC statement into measurable, specified characteristics with target and specification limits.

Project charter scope rule: one problem, one metric, one owner, one timebox (typically 3–6 months for a Black Belt project).

Charter elements: business case, problem/opportunity statement, goal statement (SMART), scope with in/out lists, milestones, team roles, sponsor sign-off.

RACI assigns Responsible, Accountable, Consulted, Informed; exactly one Accountable per deliverable prevents diffusion of ownership.

Stakeholder analysis grid: power vs. interest; high-power/high-interest stakeholders get manage-closely engagement, low/low get monitor.

Tuckman stages - Forming, Storming, Norming, Performing, Adjourning - predict team conflict timing and required leader behavior.

Gate reviews at D-M-A-I-C boundaries require deliverable evidence; a failed gate returns the team to the prior phase rather than proceeding.

Work breakdown structure decomposes scope into work packages of 8–80 hours; the critical path is the longest duration path with zero float.

Black Belt role: full-time change agent coaching Green Belts; Champion removes barriers and owns resources; Master Black Belt governs methodology and portfolio.

VOC sources: complaints, surveys, warranty data, interviews, field observation; Kano model classifies must-be, performance and delighter requirements.

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

Problem statement rules: describes observed gap with data, no cause, no solution, no blame; quantified magnitude and time frame included.

Goal statement must be specific, measurable, achievable, relevant, time-bound, and tied to the business case (often COPQ or customer metric).

Affinity diagram and multi-voting organize raw customer statements into themes before CTQ selection.

Project selection matrix scores candidate projects on customer impact, COPQ, feasibility, and strategic alignment to rank the portfolio.

Define-phase tollgate evidence: signed charter, SIPOC, VOC/CTQ list, stakeholder map, baseline data plan.

Data types: continuous (variable) vs. discrete (attribute: nominal, ordinal); continuous data needs far smaller samples for the same power.

Measurement system analysis: repeatability = equipment variation, reproducibility = appraiser variation; %GRR under 10% acceptable, over 30% unacceptable.

Gage R&R %GRR = (GRR standard deviation / total or tolerance) × 100; use tolerance-based when the spec matters, study-based when process variation matters.

Attribute agreement analysis reports within-appraiser and between-appraiser agreement; kappa near 1 means agreement beyond chance.

Cp = (USL − LSL) / 6σ; Cpk = min(USL − μ, μ − LSL) / 3σ; Cpk penalizes centering, Cp does not.

Pp/Ppk use overall (long-term) standard deviation; Cp/Cpk use within-subgroup (short-term) sigma.

Sample size for a mean: n = (Z × σ / E)²; for a proportion: n = Z²p(1−p)/E².

Baseline capability and DPMO are locked in Measure; they are the reference for proving improvement in Control.

Hypothesis test logic: H0 assumes no difference; p-value < α rejects H0; α is the producer's risk of false alarm (Type I), β is the consumer's risk of missed effect (Type II).

Test selection by data: 2-level means t-test, multiple means ANOVA, variances F-test or Levene, proportions chi-square, paired data paired t-test.

Normality check precedes parametric tests: Anderson-Darling or Shapiro-Wilk p > 0.05 supports normality; otherwise use nonparametric alternatives (Mann-Whitney, Kruskal-Wallis).

Correlation r measures linear association (−1 to 1); regression y = b0 + b1x quantifies effect; R² is the fraction of variation explained, not causation.

Multi-vari study separates positional, cyclical and temporal variation; it narrows the search before formal testing.

FMEA: RPN = Severity × Occurrence × Detection; high severity items are acted on regardless of RPN score.

5 Whys and fishbone (Ishikawa) generate candidate causes; causes are verified with data, never accepted on opinion alone.

OFAT (one-factor-at-a-time) cannot detect interactions; factorial designs estimate main effects and interactions with fewer runs.

Full 2^k factorial has 2^k runs; a half-fraction 2^(k−1) uses the defining relation to alias effects, so resolution IV or higher is preferred when interactions matter.

Randomization, replication and blocking control nuisance variation; blocking removes known sources such as shift or lot.

DOE sequence: screening (fractional factorial) → characterization (response surface) → optimization (steepest ascent, then confirm at predicted optimum).

Main effect = change in mean response as a factor moves between levels; interaction means the effect of one factor depends on the level of another.

Pilot the chosen solution on a limited scale; confirm the improvement exceeds the measurement error before full deployment.

Solution selection matrix weights impact, cost, risk, and implementation time; error-proofing (poka-yoke) ranks above training and procedure as a control.

Control plan lists each CTQ, its specification, measurement method, sample size/frequency, control method, and reaction plan with owner.

SPC chart choice: Xbar-R for subgroup n = 2–9, Xbar-S for n ≥ 10, individuals (I-MR) for n = 1; p, np, c, u charts for attribute data.

Control limits are ±3 sigma of the plotted statistic from process data, not specification limits; a point outside a limit or a nonrandom run signals special cause.

Western Electric rules: 1 point beyond 3σ, 2 of 3 beyond 2σ, 4 of 5 beyond 1σ, 8 in a row on one side - all indicate special-cause variation.

Process capability must be re-verified after improvement; Cpk ≥ 1.33 is a common minimum, ≥ 1.67 for critical characteristics.

Standardize and document: updated work instructions, training records, control charts, and a response plan define who acts when a signal fires.

Project closure transfers ownership to the process owner, verifies realized savings against the business case, and captures lessons learned.

DFSS methodologies: DMADV (Define, Measure, Analyze, Design, Verify) for new designs; IDOV (Identify, Design, Optimize, Verify) is an alternative roadmap.

QFD house of quality converts customer requirements into design characteristics, ranking them by competitive importance and target values.

Tolerance design and robust design (Taguchi) minimize sensitivity to noise factors; the loss function is quadratic around the target, not a step at the limits.

FMEA in design (DFMEA) precedes process FMEA (PFMEA); both are living documents revised when design or process changes.

Quality system standards: ISO 9001 requirements, ISO 14001 environment, IATF 16949 automotive; ISO 9001 does not prescribe Six Sigma methods.

Cost of quality rises steeply downstream: a defect found in design costs least, in the field costs most; prevention spend reduces total COQ.

Audit types: first-party (internal), second-party (supplier), third-party (certification body); corrective action closes findings with root-cause verification.