2.4 Process Capability

Process Capability Introduction Process capability describes how well a stable process can meet specification limits defined by customers or engineering requirements. It compares the natural variability of a process to the allowable variation. Understanding process capability allows you to: - Quantify how “good” a process is relative to specs - Compare different processes or machines on a common scale - Prioritize improvement opportunities - Predict defect rates under stable conditions This article covers the knowledge and skills necessary to understand and apply process capability at an advanced level. --- Fundamental Concepts Process Variation and Distribution Process capability assumes that: - The process is stable (in statistical control) - Data come from a single, consistent process - The key output characteristic is measurable on a continuous scale - The distribution is approximately normal (for standard indices) Key ideas: - Common cause variation: natural, inherent, predictable within a range - Special cause variation: unusual, assignable, causes instability - Stable process: only common causes present; control chart shows no signals - Natural tolerance: typical spread of a stable process, often approximated by ±3 standard deviations from the mean Before calculating capability, verify stability and distributional assumptions. Specifications vs Control Limits Do not confuse: - Specification limits (USL, LSL) - Set by customers, standards, or design - Define what is acceptable or not - External to the process - Control limits - Calculated from process data - Reflect current process performance - Internal to the process, used for monitoring stability Process capability compares process variation to specification limits, not control limits. --- Indices of Process Capability Short-Term vs Long-Term Capability Two time horizons are important: - Short-term (within-subgroup, potential) - Uses within-subgroup variation - Reflects what the process could do under minimal sources of variation - Often uses σ within, or an estimate assuming rational subgrouping - Long-term (overall, actual) - Uses overall variation across all data - Includes more sources of variation (e.g., shifts between shifts, days, operators) - Often uses the overall sample standard deviation Short-term indices: Cp, Cpk Long-term indices: Pp, Ppk Use both perspectives to understand capability potential and actual performance. Cp: Potential Process Capability Cp measures how wide the specification limits are relative to the process spread, ignoring centering. Formula: - Cp = (USL − LSL) / (6σ) Where: - USL = upper specification limit - LSL = lower specification limit - σ = estimated short-term process standard deviation Interpretation: - Cp > 1: specs wider than process spread; potentially capable - Cp = 1: process spread equals spec width (barely capable) - Cp < 1: process spread wider than specs; inherently incapable - Higher Cp indicates less variability relative to specs Limitation: Cp assumes the process mean is centered between LSL and USL. Cpk: Actual Process Capability (Short-Term) Cpk includes both spread and centering. It measures the worst-case capability relative to each spec limit. Formulas: - Cpu = (USL − μ) / (3σ) - Cpl = (μ − LSL) / (3σ) - Cpk = min(Cpu, Cpl) Where: - μ = process mean - σ = estimated short-term standard deviation Interpretation: - Cpk incorporates off-centering - Cpk ≤ Cp (equality only when perfectly centered) - Cpk > 1: majority of process output within specs, under assumptions - Cpk < 1: substantial proportion of output outside specs Use Cp and Cpk together: - Cp high, Cpk low: process capable but off-center - Both low: process too variable and possibly off-center - Both high: process well-centered and low variability Pp and Ppk: Overall Performance Pp and Ppk are analogous to Cp and Cpk but use long-term (overall) variation. Formulas: - Pp = (USL − LSL) / (6 s overall) - Ppu = (USL − μ) / (3 s overall) - Ppl = (μ − LSL) / (3 s overall) - Ppk = min(Ppu, Ppl) Where: - s overall = overall sample standard deviation of all data Interpretation: - Pp, Ppk show actual performance over time, including more variation sources - Typically: Ppk ≤ Cpk and Pp ≤ Cp - Big gaps between Cpk and Ppk indicate significant long-term variation not seen within subgroups Use: - C indices (Cp, Cpk): potential capability under controlled short-term conditions - P indices (Pp, Ppk): real-world performance over time --- Interpreting Capability Indices Capability vs Specification Tightness Common interpretive guidelines (assuming normality and stability): - Index ≈ 1.0 - Process spread roughly equals specs - About 0.27% nonconforming (for two-sided specs, centered, short-term) - Index ≈ 1.33 - Often used as a minimum for capable processes in practice - Provides some margin for variation and shifts - Index ≈ 1.67 or higher - Often targeted for critical characteristics - Lower risk of defects under typical shifts - Index < 1.0 - Process inherently unable to consistently meet specs under current variation level Interpret within context: - Importance of the characteristic - Cost of defects - Legal or safety requirements - Process economics Capability and Nonconformance Rates For a normal, stable, two-sided process: - Transform Cpk into an equivalent Z value: - Z short-term ≈ 3 × Cpk - Then estimate defect proportion beyond spec limits using normal probabilities Illustrative logic (not exact tables): - Cpk = 1 → Z ≈ 3 → about 0.27% outside specs - Cpk = 1.33 → Z ≈ 4 → roughly 0.006% outside specs - Cpk = 1.67 → Z ≈ 5 → roughly 0.00003% outside specs For a one-sided spec (only USL or only LSL), focus on: - Cpu or Cpl (or Ppu or Ppl) - Corresponding one-sided tail probability These estimates assume: - Normal distribution - Stable mean and variance - No significant shifts over time beyond what data capture --- Preconditions and Validity Requirement: Statistical Stability Process capability results are only meaningful when the process is stable. Check stability using: - Appropriate control charts for the data type - Rational subgrouping (grouping data so that variation within subgroups reflects only short-term common causes) If control charts show: - Trends - Shifts - Cycles - Out-of-control points then: - The process is unstable - Capability indices may be misleading - Address special causes first, then recalculate capability Normality and Data Distribution Standard capability indices assume approximate normality. Key steps: - Visually inspect the distribution (histogram, probability plot) - Use tests for normality as supporting evidence (not as the sole criterion) - Consider the physical nature of the characteristic (bounded, skewed, integer, etc.) If data are not approximately normal: - Simple Cp/Cpk may not reflect true defect rates - Tail behavior particularly important for specs near the extremes Non-normality options (core awareness only): - Transform the data (for example, log, Box-Cox) to approximate normality, then analyze - Use non-normal capability methods based on the fitted distribution Regardless of method, ensure: - Selected approach matches data behavior - Interpretation reflects the actual risk at the specification limits Sample Size and Data Quality Capability analysis depends heavily on data quality. Consider: - Sample size - Very small samples give unstable estimates of σ and μ - Larger samples provide more reliable indices - Measurement system - High measurement error inflates apparent variation - Inadequate measurement systems distort capability - Data representativeness - Data must represent typical operating conditions - Avoid biased sampling (e.g., only best shifts, only certain machines) When possible: - Confirm the measurement system is acceptable - Collect data across relevant sources of variation for long-term indices --- Centering, Spread, and Improvement Centering vs Reducing Variation Cp and Cpk help distinguish two improvement levers: - Centering (adjust mean) - When Cp is high but Cpk is low - Process is capable but off-center - Shifting the mean toward the target or center of specs improves Cpk without changing Cp - Reducing variation - When Cp itself is low - Need to reduce inherent variability via process improvement - Both Cp and Cpk can improve Guidelines: - First, ensure the process is reasonably centered to avoid one-sided defect issues - Then, reduce variation to increase both Cp and Cpk Linking Capability to Targets Sometimes there is a target value T (nominal) within the specification limits. Key ideas: - A process can be within specs but still far from target - Loss to the customer may increase with distance from target, not just spec violations - Capability indices (Cp, Cpk) do not directly measure closeness to target When targets matter: - Monitor the mean against the target - Use capability indices alongside measures of on-target performance --- Special Situations in Capability Analysis One-Sided Specifications When only one spec limit exists: - Example: Only a maximum is specified (USL); lower values are all acceptable - Use one-sided indices: - Cpu = (USL − μ) / (3σ) or Ppu - Cpk = Cpu (if no LSL) - For only LSL, use Cpl or Ppl Interpretation: - Focus on the risk at the specified side - Convert the one-sided Cpk (Cpu or Cpl) to Z and estimate tail probability as needed Non-Normal or Bounded Data When data are clearly non-normal: - Strong skew (e.g., time data, positive-only measures) - Bounded by zero or other physical limits - Discrete, but treated as continuous (e.g., large counts) Options include: - Applying transformations to approximate normality, then using standard indices - Using distribution fitting and non-normal capability methods Key considerations: - Ensure the chosen distribution or transformation adequately represents the data, especially in the tails - Communicate that indices and defect estimates are based on these modeling choices --- Using Capability in Practice Capability Studies A typical capability study involves: - Clarifying the characteristic and its specs (USL, LSL, target) - Verifying that the measurement system is adequate - Collecting data under typical operating conditions - Establishing process stability using control charts - Checking distributional assumptions - Calculating Cp, Cpk (short-term) and Pp, Ppk (long-term) when appropriate - Interpreting indices in terms of: - Spread vs specs - Centering - Short-term vs long-term differences - Estimated nonconformance rates Decision Making with Capability Results Use capability indices to: - Decide if a process can consistently meet requirements - Compare alternative processes, machines, or methods - Support supplier qualification or process approval - Prioritize improvement projects: - Low Cp/Cpk processes first - Processes with high gap between potential (Cp) and actual (Pp, Ppk) Capability is not static. Recalculate after significant process changes or improvements. --- Summary Process capability quantifies how well a stable process can meet specified limits, using indices that compare process variation and centering to customer or design requirements. Key points: - Capability analysis requires a stable process and an appropriate understanding of the data distribution. - Cp measures potential capability based on spread only; Cpk adds centering. Pp and Ppk provide long-term performance views. - Indices greater than 1 indicate specs wider than process spread; higher values represent better capability, under underlying assumptions. - Interpret capability together with estimated nonconformance rates, considering the importance of the characteristic and context. - Use capability results to distinguish between issues of centering and variability, guide improvement actions, and monitor process performance over time.