5.2.4 U Chart

U Chart Introduction A U Chart is a control chart used to monitor defects per unit when the sample size (number of units inspected) can vary from subgroup to subgroup. It is part of the family of attribute control charts and is specifically designed for count data normalized by a unit of opportunity. Typical uses include: - Defects per invoice, order, form, shipment, or patient. - Number of errors per page, per line of code, or per unit produced. - Complaints per customer, per day, or per batch. Understanding and correctly applying a U Chart requires clarity on units, defects, assumptions, calculations, and interpretation. --- When to Use a U Chart Defects vs Nonconformities A U Chart is for defects, not just defective items. - Defective unit: A unit is either conforming or nonconforming (pass/fail). - Defect (nonconformity): A single instance where a requirement is not met; one unit can have multiple defects. Use a U Chart when: - One unit can contain multiple defects. - You want to track the average number of defects per unit. - Sample size (number of units checked) is not constant. If recording only whether a unit is defective or not, a different chart (e.g., P or NP) is more appropriate. U Chart vs C Chart A U Chart is closely related to the C Chart. The distinction is critical. - C Chart: - Tracks count of defects per sample (C). - Requires constant sample size (same number of units or same area inspected each time). - Control limits are constant across subgroups. - U Chart: - Tracks defects per unit (U = C / n). - Allows varying sample size (n can change each subgroup). - Control limits change with sample size. Use a U Chart when the number of units inspected, or the area/time of opportunity, is not constant. Conditions and Assumptions A U Chart is based on a Poisson model for counts of defects: - Defects are counted on each unit (or set of units). - Defects occur independently within and across units. - The chance of a defect on one small portion of a unit is relatively small. - The average defect rate is assumed stable during the baseline period used to compute control limits. If these assumptions are heavily violated (e.g., strong clustering, very high defect rates), interpretation becomes more difficult and other approaches may be needed. --- Key Definitions and Notation To work with a U Chart, use consistent notation: - i: Subgroup index (e.g., day i, lot i, week i). - Cᵢ: Number of defects counted in subgroup i. - nᵢ: Number of units (or standardized opportunities) inspected in subgroup i. - Uᵢ: Defects per unit in subgroup i. - Uᵢ = Cᵢ / nᵢ - Ū: Overall average defects per unit across m subgroups (baseline period). - Ū = (Σ Cᵢ) / (Σ nᵢ) - CL: Center line of the chart, equal to Ū. - UCLᵢ: Upper control limit for subgroup i. - LCLᵢ: Lower control limit for subgroup i. Control limits depend on sample size: - UCLᵢ = Ū + 3 × √(Ū / nᵢ) - LCLᵢ = Ū − 3 × √(Ū / nᵢ), but never less than 0. Because nᵢ appears in the denominator under the square root, the limits narrow with larger sample sizes and widen with smaller sample sizes. --- Constructing a U Chart Step 1: Define the Unit and the Defect Clarity at the start avoids misinterpretation later. - Define the unit: - What is one unit? - Example: one order, one invoice, one patient visit, one product. - Is the unit stable in nature and scope over time? - Define what counts as a defect: - Specific, objective criteria. - Examples: - Missing field on an order. - Incorrect price on an invoice. - Software bug found in testing. Keep the definitions consistent throughout data collection. Step 2: Collect the Data For each subgroup i (time period or batch): - Record: - Cᵢ: total number of defects found. - nᵢ: number of units inspected. - Compute: - Uᵢ = Cᵢ / nᵢ. Considerations: - Subgroups should represent logically consistent time intervals or production batches. - If possible, use at least 20–25 subgroups to establish reliable control limits. - Check for data entry errors, miscounts, or inconsistent definitions before proceeding. Step 3: Calculate Overall Average Ū Use the entire baseline dataset to compute Ū: - Ū = (Σ Cᵢ) / (Σ nᵢ) Important: - Do not simply average Uᵢ values if sample sizes differ. - The weighted formula above correctly accounts for varying nᵢ. Step 4: Calculate Control Limits For each subgroup i: - Center Line: - CL = Ū (same for all subgroups). - Control Limits: - UCLᵢ = Ū + 3 × √(Ū / nᵢ) - LCLᵢ = Ū − 3 × √(Ū / nᵢ) - If LCLᵢ < 0, set LCLᵢ = 0. Notes: - Larger nᵢ → smaller √(Ū / nᵢ) → tighter control limits. - Smaller nᵢ → wider limits, making signals less sensitive. Step 5: Plot and Annotate the Chart On the U Chart: - Horizontal axis: subgroup index (e.g., time order). - Vertical axis: defects per unit (Uᵢ). - Plot: - Points at (i, Uᵢ). - The center line CL at Ū. - UCLᵢ and LCLᵢ as varying lines across subgroups (since they depend on nᵢ). Label: - Chart title (e.g., “U Chart of Defects per Invoice per Day”). - Time range. - Definition of defect and unit (briefly in notes or legend). --- Interpreting a U Chart Common Cause vs Special Cause A U Chart helps distinguish: - Common cause variation: - Natural, random fluctuations in the process. - Points vary within control limits with no unusual patterns. - Special cause variation: - Signals of unusual, non-random influences. - Requires investigation and, if confirmed, corrective or preventive action. Basic Control Chart Signals Key signals on a U Chart: - Point outside control limits: - Uᵢ above UCLᵢ: unusually high defects per unit; investigate for assignable causes. - Uᵢ below LCLᵢ: unusually low defects per unit; may indicate process improvement or data issue; also investigate. - Shift in level: - Several consecutive points on one side of CL (for example, 7 or more on the same side). - Suggests a change in the underlying defect rate. - Trend: - Several consecutive points consistently increasing or decreasing. - May indicate a drift due to wear, learning, or changing conditions. - Systematic patterns: - Cycles linked to time of day, shift, supplier, etc. - Points systematically higher during certain conditions. Investigate signals by looking for changes in: - Methods, materials, machines, environments. - People, training, inspection methods. - Workload, complexity, or customer mix. Impact of Varying Sample Size Because UCLᵢ and LCLᵢ depend on nᵢ: - Subgroups with much larger nᵢ have tighter limits, so smaller changes in Uᵢ can signal special causes. - Subgroups with small nᵢ have wider limits, making the chart less sensitive. Implications: - Large swings in Uᵢ with very small sample sizes may still fall inside wide control limits. - When possible, avoid extremely small nᵢ; aim for a reasonably consistent and sufficiently large number of units per subgroup. --- Practical Considerations and Pitfalls Consistency of Unit and Opportunity The U Chart relies on a meaningful and stable definition of a “unit”: - If the definition of a unit changes over time (e.g., mixing different product families with very different complexity) without adjustment, the chart may mislead. - Consider standardizing to a meaningful denominator (e.g., “per 1000 lines of code,” “per 100 claims”) when units vary in underlying opportunity. In such cases: - Define nᵢ as the number of standardized units of opportunity in subgroup i. - Cᵢ remains the total defects; Uᵢ = Cᵢ / nᵢ. Overdispersion and Underdispersion The U Chart assumes variance in counts is roughly equal to the mean (Poisson logic). Watch for: - Overdispersion: - Observed variation larger than expected from the Poisson model. - May show many points near limits or frequent out-of-control points without clear causes. - Often due to clustering of defects or unmodeled sources of variation. - Underdispersion: - Observed variation smaller than expected. - Points unusually close to the center line. - May indicate issues like data smoothing, averaging, or ineffective inspection. When such issues appear: - Recheck data definitions and collection methods. - Examine whether subgroups mix different conditions that should be charted separately. - Consider stratifying the data (e.g., by product type, shift) if justified. Rational Subgrouping Rational subgrouping means grouping data so that: - Variation within a subgroup reflects only common cause variation. - Special causes show up between subgroups. For a U Chart, typical subgrouping choices: - By time period (day, shift, batch). - By lot, supplier, or line, if that aligns with how the process operates. Avoid mixing very different conditions (e.g., multiple lines, very different products) in one subgroup unless there is a clear reason and the underlying defect rate is expected to be comparable. --- Using the U Chart for Process Improvement Baseline, Improvement, and Re-Baselining A common pattern: - Use historical data to create a baseline U Chart. - Identify whether the process is in control. - Implement improvements aimed at reducing defects. - Monitor the chart for a new, stable level. When sustained changes occur: - If there is clear evidence of a lasting shift in Uᵢ (e.g., multiple points stably below the old CL and within the new reduced range), consider re-baselining: - Use data from the improved, stable period. - Recalculate Ū and control limits based only on that period. - Use the new chart to monitor ongoing performance against the new standard. Comparing Performance Over Time The U Chart allows comparison of defect rates even when sample sizes vary: - Compare Uᵢ values before and after a change, observing whether: - Points after an improvement consistently lie lower. - Control chart signals confirm a shift or trend. Be cautious: - Do not compare raw counts Cᵢ without accounting for nᵢ. - Use Uᵢ when judging differences across time periods with different nᵢ. --- Worked Example (Conceptual) Suppose a process inspects customer orders for errors each day. - For day i: - Cᵢ = number of defects (errors) found on all orders. - nᵢ = number of orders inspected. - Uᵢ = Cᵢ / nᵢ = defects per order. Steps: - Collect data for, say, 25 days: - Each day’s Cᵢ and nᵢ. - Compute: - Total defects: Σ Cᵢ. - Total orders: Σ nᵢ. - Ū = (Σ Cᵢ) / (Σ nᵢ). - For each day i: - Compute UCLᵢ and LCLᵢ using nᵢ. - Plot Uᵢ, CL, UCLᵢ, and LCLᵢ. Interpret: - Check for days where Uᵢ > UCLᵢ or Uᵢ < LCLᵢ. - Look for patterns: clusters of high days, trends, or shifts. - Investigate special causes (e.g., new staff, system changes). --- Common Questions About U Charts What if LCLᵢ is Negative? By convention: - If LCLᵢ < 0, set LCLᵢ = 0. A negative lower limit has no practical meaning, since a count of defects cannot be negative. Can U Chart Handle Zero-Defect Subgroups? Yes: - If Cᵢ = 0, then Uᵢ = 0. - Such points are plotted at 0 and interpreted just like others. - Many zero-defect subgroups may indicate excellent performance, but still check model assumptions (very low defect rates can make Poisson approximations sensitive). Can Multiple Types of Defects Be Combined? Yes, if appropriate: - Cᵢ can be total combined defects of all types. - Make sure: - All counted defects are relevant to the same process. - The total count is meaningful for your improvement objective. If defect types behave differently or you need separate insights: - Consider separate U Charts, one per major defect category. --- Summary A U Chart monitors defects per unit when the number of units inspected varies across subgroups. It is based on a Poisson model and normalizes the count of defects by the size of the inspected sample, allowing meaningful comparison over time despite differing sample sizes. Key points: - Use a U Chart for defects, not just defective units, and when sample sizes vary. - Compute Uᵢ = Cᵢ / nᵢ, and Ū = (Σ Cᵢ) / (Σ nᵢ). - Control limits depend on sample size: UCLᵢ = Ū + 3 × √(Ū / nᵢ), LCLᵢ = max[0, Ū − 3 × √(Ū / nᵢ)]. - Interpret signals such as points beyond limits, shifts, trends, and patterns to distinguish common and special causes. - Ensure consistent definitions of unit and defect, apply rational subgrouping, and be aware of overdispersion or underdispersion. - Use the U Chart to establish a baseline, monitor changes, and verify the stability and capability of the process in terms of defects per unit. Mastering these concepts enables rigorous use of the U Chart to detect unusual variation and support focused, data-driven process improvement.