2.4.1 Capability Analysis

Capability Analysis Introduction to Capability Analysis Capability analysis evaluates how well a process can meet customer requirements, expressed as specification limits. It compares the natural variation of a stable process to the allowable variation defined by specs. Key ideas: - Goal: Quantify how capable a process is of producing output within specification limits. - Focus: Relationship between process behavior (mean and variation) and customer requirements (USL/LSL). - Output: Capability indices and performance indices that guide decisions on process improvement or control. Capability analysis is meaningful only when the process is statistically stable and measurements are appropriate. --- Foundations: Specs, Tolerance, and Distribution Specification Limits and Tolerances Capability analysis always references explicit customer or engineering requirements. - LSL (Lower Specification Limit): Minimum acceptable value. - USL (Upper Specification Limit): Maximum acceptable value. - Target (T): Desired value, often midpoint of spec range but may differ. The tolerance is: - Two-sided: USL − LSL - One-sided: Distance from target or spec in one direction only Capability indices are undefined or misleading if specs are absent, arbitrary, or not linked to a real requirement. Process Distribution and Normality Many capability formulas assume a normal distribution of process output. - If data are approximately normal: - Spread is well described by the standard deviation. - ±3σ around the mean captures most of the data. - If data are non-normal: - Normal-based capability indices can be inaccurate. - Alternative approaches (e.g., transformations, percentile-based methods) may be required. Normality is assessed using: - Visual checks (histogram, probability plot) aligned with the fitted normal curve. - Tests for normality when sample size and context justify them. Process Stability and Control Capability requires a stable process. - Stable process: Only common cause variation present; no significant shifts or trends. - Unstable process: Special causes present; capability calculations can be misleading. Check stability before capability: - Review control charts (e.g., X̄-R, X̄-S, I-MR) for: - Points outside control limits - Trends, runs, or patterns - If unstable: - Identify and remove special causes. - Re-establish stability, then re-evaluate capability. --- Short-Term vs Long-Term Measures Within-Subgroup Variation vs Overall Variation Capability analysis distinguishes: - Within-subgroup (short-term) variation: - Variation within small, rational subgroups collected over short intervals. - Reflects short-term, inherent process noise. - Estimated by R̄/d2, S̄/c4, or pooled subgroup standard deviations. - Overall (long-term) variation: - Variation across all data over a longer period. - Includes shifts, drifts, and other sources of variation. - Estimated by the standard deviation of all individual data points. The choice affects which indices are appropriate and how to interpret them. Capability vs Performance Indices - Capability indices use within-subgroup (short-term) variation. - Typical symbols: Cp, Cpk, Cpm. - Performance indices use overall (long-term) variation. - Typical symbols: Pp, Ppk, Ppm. They answer different questions: - Capability indices: “What can this process do under short-term stable conditions?” - Performance indices: “What has this process actually been doing over time?” --- Key Capability Indices Cp: Potential Capability (Centered Assumption) Cp measures the potential capability assuming the process is perfectly centered between the specs. Formula (for two-sided specs): - Cp = (USL − LSL) / (6σ_within) Interpretation: - Cp > 1: Tolerance is wider than the 6σ spread; process can potentially fit within specs. - Cp = 1: 6σ spread equals tolerance width. - Cp < 1: Natural spread exceeds tolerance; defects are expected, even if centered. Limitations: - Ignores process mean location. - Overestimates capability if the process is off-center. Cpk: Actual Capability (Accounts for Centering) Cpk measures capability considering both spread and centering. Formula: - Cpu = (USL − μ) / (3σ_within) - Cpl = (μ − LSL) / (3σ_within) - Cpk = min(Cpu, Cpl) Interpretation: - Cpk reflects the “worst side” relative to specs. - If Cpk ≈ Cp: - Process is reasonably centered. - If Cpk << Cp: - Process mean is off-center; centering issues dominate. Guidance (conceptual, not prescriptive): - Higher Cpk implies fewer expected defects. - Compare to required capability levels defined by the organization or customer. Cpm: Capability with Target Deviation Penalized Cpm accounts for deviation from a specific target value, not just from spec limits. Formula (Taguchi-based): - Cpm = (USL − LSL) / [6 * √(σ² + (μ − T)²)] Where: - T is the target value. - μ is the process mean. - σ is the within-subgroup standard deviation. Interpretation: - When μ = T, Cpm ≈ Cp. - When μ is far from T, Cpm is significantly lower than Cp and Cpk. - Useful when being near target is critical, not just being within specs. --- Performance Indices (Long-Term) Pp and Ppk Performance indices use overall standard deviation (σ_overall) to reflect long-term conditions. Formulas: - Pp = (USL − LSL) / (6σ_overall) - Ppu = (USL − μ) / (3σ_overall) - Ppl = (μ − LSL) / (3σ_overall) - Ppk = min(Ppu, Ppl) Interpretation: - Pp and Ppk typically ≤ Cp and Cpk, because long-term variation is usually higher. - Pp focuses on potential, ignoring centering. - Ppk captures long-term performance including both spread and centering. Comparison of indices: - Cp vs Pp: Short-term vs long-term spread. - Cpk vs Ppk: Short-term vs long-term capability including centering. - Large gaps between Cpk and Ppk may indicate: - Shifts, drifts, or seasonal patterns. - Differences between controlled “study” conditions and everyday operation. Ppm: Performance Relative to Target Analogous to Cpm but using overall variation: - Ppm = (USL − LSL) / [6 * √(σ_overall² + (μ − T)²)] Use when: - Long-term adherence to target is important. - You want to include both long-term variation and long-term offset from target. --- One-Sided Capability Upper or Lower Specs Only Many processes have only one relevant specification: - Only an upper spec: e.g., maximum impurity, maximum defect size. - Only a lower spec: e.g., minimum strength, minimum yield. For one-sided capability: - Use Cpu (upper) or Cpl (lower) as the primary indicator. - Cpk reduces to the single side of interest. Examples: - Upper spec only: - Cpu = (USL − μ) / (3σ) - Lower spec only: - Cpl = (μ − LSL) / (3σ) Interpretation: - Larger Cpu/Cpl indicates better protection against violating the one-sided requirement. - Balance between process mean and variation is still critical. --- Practical Steps for Capability Analysis Step 1: Clarify Requirements and Data Ensure the basis of capability analysis is sound. - Confirm: - LS L/USL (or one-sided spec) are clearly defined and justified. - Target value (if relevant) is known. - Verify: - Measurement system is suitable: - Adequate resolution. - Acceptable accuracy and repeatability. - Choose: - Rational subgrouping strategy that reflects natural short-term conditions. Step 2: Assess Stability and Distribution Before computing indices: - Plot control charts: - Confirm no special causes. - If unstable: - Identify and remove special cause data or fix the process and recollect data. - Check distribution: - Use histograms and probability plots. - Decide whether normal-based indices are reasonable or if an alternative approach is needed. Step 3: Estimate Variation Compute: - Within-subgroup standard deviation (σ_within): - From R̄/d2, S̄/c4, or pooled methods depending on chart type and subgroup size. - Overall standard deviation (σ_overall): - From all individual observations. Use: - σ_within for Cp, Cpk, Cpm. - σ_overall for Pp, Ppk, Ppm. Step 4: Calculate Capability and Performance Indices For two-sided specs: - Compute Cp, Cpk, (optionally Cpm). - Compute Pp, Ppk, (optionally Ppm). For one-sided specs: - Focus on Cpu or Cpl and corresponding performance index. Document: - Mean (μ). - Standard deviations (σwithin, σoverall). - Indices with appropriate units/labels. Step 5: Interpret Results in Context Interpretation must consider: - Specification tightness relative to natural variation. - Centering of the process mean. - Differences between short-term and long-term behavior. Ask: - Is short-term capability satisfactory? - Does long-term performance match short-term capability? - Is the process mean appropriately centered relative to target and specs? - Are there practical limits to shifting or tightening variation? Use the results to guide: - Centering adjustments (e.g., mean shifts). - Variation reduction projects (e.g., process improvements). - Ongoing monitoring strategies. --- Defect Rates and Capability Linking Capability to Defects Capability indices are proxies for the expected defect rate, assuming: - Stable process. - Appropriate distribution model (often normal). Conceptually: - Higher Cpk or Ppk ⇒ fewer units outside specs. - Lower Cpk or Ppk ⇒ greater risk of nonconformance. To connect indices to defect rates: - Convert distances from mean to specs into Z-values: - Z_upper = (USL − μ) / σ - Z_lower = (μ − LSL) / σ - Zbench = min(Zupper, Z_lower) - Use Z_bench with the normal distribution to estimate: - Probability of exceeding USL or falling below LSL. - Expected nonconforming rate (PPM or %). Important: - Performance-based predictions (using σ_overall) often better reflect actual defect levels. - Capability-based predictions (using σ_within) describe potential under tightly controlled conditions. --- Common Pitfalls and Misinterpretations Treating Indices as Absolute Truth - Capability indices are summaries, not complete descriptions. - A single index cannot capture: - Multimodal distributions. - Batch-to-batch shifts. - Non-normal tails. Avoid: - Comparing indices across processes without understanding context, data quality, and stability. - Assuming that a specific threshold (e.g., a single capability value) guarantees zero defects. Ignoring Measurement System Issues An inadequate measurement system can: - Inflate apparent variation ⇒ underestimate capability. - Mask true variation ⇒ overestimate capability. Always ensure: - Measurement resolution and repeatability are sufficient for the tolerance being evaluated. - Any known bias or linearity issues are addressed. Using Capability on Unstable Processes Capability analysis on unstable processes: - Represents a mixture of different process states. - Gives indices that are not predictive of future performance. - Can hide the real improvement opportunities in addressing special causes. Ensure: - Stability is verified first. - Data with known special causes are handled appropriately. --- Summary Capability analysis quantifies how well a stable process meets defined specification limits by comparing process variation and centering to customer requirements. It relies on: - Clear and justified specification limits and targets. - A stable process and appropriate distribution assumptions. - Distinction between short-term (within-subgroup) and long-term (overall) variation. Key indices: - Cp, Cpk, Cpm: Capability based on short-term variation. - Pp, Ppk, Ppm: Performance based on long-term variation. - Cpu/Cpl: One-sided capability when only an upper or lower spec applies. Effective use of capability analysis involves: - Proper data collection and verification of stability and normality. - Accurate estimation of both within and overall variation. - Careful interpretation of indices in context, including links to expected defect rates. - Awareness of common pitfalls such as unstable processes and poor measurement systems. Together, these concepts provide a complete foundation for evaluating and improving process capability in a rigorous, data-driven way.