Engineer Quality Management common mistakes
Engineer Quality Management · Understand the traps and how to avoid them.
Computing Cpk as (USL − LSL)/(6σ) and calling it process capability, instead of dividing the smaller of (USL − x̄) and (x̄ − LSL) by 3σ.
Candidates memorize a single 'capability formula' and confuse Cp (spread only, centered assumption) with Cpk (spread plus centering). Under time pressure they grab whichever formula surfaces first and never check whether the process mean is off-center, which is exactly the case most exam items construct.
What to do instead
Before computing, always write the mean and both specification limits, then compute both one-sided indices and take the minimum. Practice with a process whose mean sits at 1.5σ from target so the Cp/Cpk gap is obvious. If a question asks for 'capability' without specifying, compute Cp, Cpk, and Ppk and state which answers the question.
Treating all seven Western Electric / Nelson rules as interchangeable alarms and counting a single point beyond 3σ as a 'trend' or a run of eight as a 'shift', without matching the rule number to the pattern.
The rules are learned as a list rather than as distinct pattern detectors, so candidates pattern-match on keywords like 'consecutive' or 'outside limits' and select the first plausible rule. Items frequently show a run of nine points on one side of the centerline and offer 'Rule 1' as a distractor.
What to do instead
Build a one-page table pairing each rule with a hand-drawn pattern and the minimum number of points it requires (1 beyond 3σ; 2 of 3 beyond 2σ; 4 of 5 beyond 1σ; 8 or 9 in a row on one side; 6 trending). Drill by sketching patterns and naming the rule, not the reverse.
Choosing an R chart over an s chart (or vice versa) without checking subgroup size, and applying X̄-R constants to n = 12.
Most textbook examples use n = 4 or 5, so the R chart becomes the default mental model. Candidates forget the rule of thumb that R charts lose efficiency above roughly n = 9–10 and that the constants d2, A2, D3, D4 are tabulated per n.
What to do instead
Memorize the decision rule: n ≤ 8–9 use X̄-R; n > 9 use X̄-s; n = 1 use individuals and moving range. When a question gives n, circle it first and let it drive chart selection before touching any constant table.
Reporting %GRR against the wrong reference (total variation vs. tolerance) and applying the 10%/30% AIAG acceptance bands to a tolerance-based study.
Candidates learn 'under 10% good, over 30% bad' as an absolute rule and do not notice that %GRR can be expressed as a share of study variation (TV) or of tolerance. The two give different verdicts on the same data, and items deliberately pick a gauge that is acceptable on one basis and not the other.
What to do instead
Always compute both %GRR(TV) and %GRR(tolerance) and state the basis in the answer. Apply the tolerance-based bands when the gauge is used for part conformance and the TV-based bands when used for process control. Sanity-check with ndc = 1.41 × (PV/GRR), which should be ≥ 5.
Confusing repeatability with reproducibility in a Gage R&R item, and attributing an operator-by-part interaction to equipment variation.
Both terms start with 're' and are defined in one line each, so they are recalled as synonyms. Candidates also skip the interaction term because ANOVA output looks intimidating, then misassign the largest variance component.
What to do instead
Anchor the definitions physically: repeatability is the same operator measuring the same part twice (equipment); reproducibility is different operators measuring the same part (appraiser). Read the variance components table top-down (repeatability, reproducibility, interaction, part-to-part) and confirm the percentages sum sensibly before answering.
Designing or interpreting a screening experiment as a full factorial when the item describes many factors, or failing to recognize that a resolution III design aliases main effects with two-factor interactions.
Candidates default to full factorials because they are conceptually simple, and they treat 'resolution' as a vocabulary word rather than an alias-structure statement. Items that ask 'what can you estimate?' hinge entirely on resolution.
What to do instead
Memorize the resolution ladder: III = main effects clear, two-factor interactions aliased with main effects; IV = main effects clear, two-factor interactions aliased with each other; V = main effects and two-factor interactions clear. For k ≥ 5 factors with a screening goal, expect a fractional factorial or Plackett-Burman and check the generator before answering.
Stopping root cause analysis at the operator who made the error, and writing a corrective action that is really a containment (retrain, re-inspect, sort).
The 5-Why chain is often run only until a human is named, because human error feels like an explanation. Candidates also conflate the immediate fix with the systemic fix, and exam items reward the action that changes the system, not the person.
What to do instead
Force the chain past the person: for every 'operator error' ask what made the error possible or undetectable (ambiguous work instruction, no poka-yoke, missing feedback). Classify each proposed action explicitly as containment, correction, or corrective action, and select the one that removes the cause. Verify effectiveness with a defined metric and date.
Selecting a sampling plan by AQL alone and assuming a lot accepted at AQL has near-zero defectives, or confusing AQL with LTPD and AOQL.
AQL is the most familiar term on an OC curve and gets treated as a target quality level. Candidates forget that AQL is the quality level with a defined (not zero) acceptance probability, typically 95%, so a small fraction of lots at AQL is still rejected and accepted lots can contain defectives.
What to do instead
Sketch the OC curve for every sampling item and mark AQL (producer's risk, α), LTPD/RQL (consumer's risk, β), and AOQ. Remember the acceptance number c and sample size n jointly define the plan; changing n at fixed c changes both risks. Read whether the question asks about producer risk, consumer risk, or average outgoing quality before choosing.
Answering audit and compliance items by citing a clause number or 'ISO 9001 requires' without identifying the specific requirement (e.g., documented information, risk-based thinking, control of nonconforming outputs) that the scenario violates.
Candidates memorize clause numbers as a shortcut and then cannot map a scenario to the underlying requirement. Standards are also revised, so clause numbers from one edition do not match another, and items are written around the requirement, not the numbering.
What to do instead
Study by requirement theme rather than clause number: context and risk, documented information, competence, control of externally provided processes, nonconforming output, monitoring and measurement. For each practice scenario, write the requirement in your own words first, then name the clause as a secondary check.
Treating Lean tools as interchangeable - proposing 5S for a flow problem, kanban for a setup-time problem, or value stream mapping for a single-station defect issue - and mislabeling the eight wastes.
The tools are learned as a menu rather than as diagnoses, and the eight wastes overlap (waiting vs. delay, inventory vs. overproduction). Candidates pick the tool with the most familiar name instead of the one matched to the observed symptom.
What to do instead
Build a symptom-to-tool map: long lead time → value stream map; excess WIP → kanban/pull; long changeover → SMED; unbalanced line → line balancing; defects → root cause and poka-yoke. Practice classifying wastes from short scenario sentences and justify why the neighboring waste does not apply.