5.2 Statistical Process Control (SPC)

Statistical Process Control (SPC) Introduction to SPC Statistical Process Control (SPC) is a structured way to monitor, control, and improve processes using data. It focuses on detecting and understanding variation so that processes can be kept stable, predictable, and capable of meeting requirements. SPC does not remove all variation. Instead, it distinguishes between normal, inherent variation and unusual, special variation that signals potential problems or changes in the process. A strong understanding of SPC includes: - Types and sources of variation - Control charts for variables and attributes data - Rational subgrouping and sampling - Interpretation of control charts, including Western Electric rules - Process capability and its relationship to control - Practical actions based on SPC results --- Variation and Process Stability Common Cause vs Special Cause Every process shows variation. SPC categorizes sources of variation into: - Common cause variation - Natural, random variation inherent in the process design. - Many small sources acting together. - Stable over time unless the system changes. - Special cause variation - Unusual, identifiable events or changes. - Often intermittent, sudden, or large shifts. - Indicates something changed: equipment, materials, methods, environment, or people. For SPC: - A process with only common cause variation is statistically stable (in control). - A process with special cause variation is unstable (out of control). Control charts are designed to detect special causes. --- Basics of Control Charts Purpose of Control Charts Control charts display process data over time with statistically calculated limits. They help to: - Separate signal (special cause) from noise (common cause). - Detect shifts, trends, and patterns early. - Support decisions about when to intervene and when to leave the process alone. Control charts are not specification charts. They describe what the process is doing, not what it should do according to customer requirements. Control Limits vs Specification Limits - Control limits (UCL, LCL) - Calculated from process data. - Reflect actual process variation. - Typically set at ±3 standard deviations from the process mean. - Used to judge statistical stability. - Specification limits (USL, LSL) - Defined by customers, standards, or design. - Represent acceptable performance targets. - Independent of current process performance. Key ideas: - Control limits are about process behavior. - Specification limits are about requirements. - Do not put spec limits on control charts when interpreting control; keep the concepts distinct. Rational Subgrouping and Sampling Rational subgrouping is the method of forming subgroups so that: - Variation within a subgroup reflects short-term, common cause variation. - Variation between subgroups reflects potential changes over time. Guidelines: - Group items produced under similar conditions into the same subgroup (same machine, operator, time window). - Avoid mixing different conditions within one subgroup. - Use consistent subgroup size whenever possible. Sampling considerations: - Frequency: Frequent enough to detect meaningful changes without overburdening operations. - Subgroup size: Large enough to provide stable estimates, small enough to be practical. - Timing: Representative of normal operation; avoid biased times (e.g., only when the best operator works). --- Components of a Control Chart Basic Structure A control chart typically has: - A time-ordered horizontal axis (subgroup number or time). - A vertical axis representing the statistic plotted (mean, range, proportion, etc.). - A center line (CL): expected value of the statistic if the process is stable. - Upper control limit (UCL) and lower control limit (LCL): thresholds that define expected variation. Control limits are calculated based on: - The distribution of the statistic. - Sample size or subgroup size. - An assumed stable process during the baseline period. Types of Control Charts Overview Control charts can be grouped into: - Variables charts (data measured on a continuous scale, such as length, time, temperature). - Attributes charts (data counted, such as defects or defectives). Selection depends on: - Type of data (continuous vs discrete). - Whether subgroup sizes are constant or varying. - Whether you count items with defects (defectives) or count defects per unit. --- Variables Control Charts Variables charts monitor measured characteristics such as weight, diameter, cycle time, or temperature. X̄–R Chart (Mean and Range) Use X̄–R charts when: - Data are continuous. - Subgroup size n is small (commonly 2–10, typical is 4 or 5). - Subgroup size is constant. Components: - X̄ chart: plots subgroup means, monitors shifts in process average. - R chart: plots subgroup ranges, monitors short-term variation. Conceptual formulas: - Center line for X̄ chart: overall mean of subgroup means. - UCL/LCL for X̄ chart: based on overall mean and average range, using constants (A2). - Center line for R chart: average range. - UCL/LCL for R chart: based on average range, using constants (D3, D4). Interpretation: - Check R chart first: if R chart is out of control, estimates of variation are not reliable. - If R chart is stable, then interpret X̄ chart for shifts or trends in the process mean. X̄–s Chart (Mean and Standard Deviation) Use X̄–s charts when: - Data are continuous. - Subgroup size is moderate to large (typically n ≥ 9). - Standard deviation is more appropriate than range to estimate variation. Components: - X̄ chart: monitors process mean. - s chart: monitors subgroup standard deviation. Logic parallels X̄–R: - s chart assesses variability. - X̄ chart assesses central tendency. Individual and Moving Range (I–MR) Chart Use I–MR charts when: - Subgroup size n = 1 (individual measurements). - Data are continuous. - Rational subgrouping is not possible or not meaningful. Components: - Individuals (I) chart: plots each individual observation. - Moving range (MR) chart: plots the absolute difference between consecutive observations (often MR of 2). Interpretation: - MR chart reflects short-term variability between consecutive values. - I chart shows shifts in level over time. - Check MR chart first; then interpret the I chart. --- Attributes Control Charts Attributes charts monitor counts of defectives or defects. Data are discrete. p Chart (Fraction Defective) Use p charts when: - You count the number of nonconforming units in a sample. - Sample size can vary. - You are interested in the fraction (or proportion) defective. Key points: - The statistic is p = (number of defectives) / (sample size). - Center line is the average proportion defective. - Control limits depend on p and sample size n; limits may vary with n. np Chart (Number Defective) Use np charts when: - You count the number of defective units. - Sample size is constant. - You prefer to track count directly instead of the proportion. Key points: - The statistic is np = number of defectives per sample. - Center line is average number defective. - Control limits are constant when n is constant. c Chart (Number of Defects per Unit) Use c charts when: - You count the number of defects on a single unit or a constant area, length, or item. - The opportunity area or unit size is constant. Key points: - The statistic is c = number of defects per inspection unit. - Center line is the average count of defects. - Control limits are based on the assumption of a Poisson distribution. u Chart (Defects per Unit with Varying Area or Size) Use u charts when: - You count defects, but the size of the inspection unit varies (e.g., different lengths, areas, or times). - You measure defects per unit or per standardized area. Key points: - The statistic is u = (number of defects) / (number of units or standardized area). - Center line is the average defects per unit. - Control limits adjust for the varying denominator. --- Interpreting Control Charts Basic Out-of-Control Signals When a process is stable, points fall randomly around the center line within control limits. Statistical rules help identify special causes. Common rules include: - Single point beyond control limits - A point above UCL or below LCL is a strong signal of special cause. - Run of points on one side of the center line - Several consecutive points (for example, 7 or more) all above or all below the center line indicate a shift in the process mean. - Trend of increasing or decreasing points - A sequence of points all moving upward or downward suggests a trend, possibly due to wear, learning, or environmental change. - Too few or too many points near the center line or near the limits - Unusual patterns of clustering may indicate incorrect subgrouping, data manipulation, or multiple underlying distributions. - Cyclic or systematic patterns - Repeating cycles can indicate external factors such as shifts, days of week, or maintenance cycles. The exact number of points used in these rules (e.g., 7 points in a row) is chosen to balance sensitivity and false alarms. The underlying idea is to identify patterns that are very unlikely to occur by chance alone. Western Electric Rules (Conceptual) Western Electric rules combine several pattern-detection criteria. The intent is to: - Improve sensitivity to small and moderate shifts. - Control the overall false alarm rate. Examples of patterns these rules consider: - One point beyond 3 standard deviations from the center line. - Two of three consecutive points beyond 2 standard deviations on the same side. - Four of five consecutive points beyond 1 standard deviation on the same side. - A specified number of consecutive points on the same side of the center line. These rules formalize pattern recognition on control charts. Applying them consistently reduces subjective interpretation. Distinguishing Signals from Noise Interpreting control charts requires balancing: - Sensitivity: detecting meaningful shifts quickly. - Specificity: avoiding overreaction to random noise. Overreacting to noise causes unnecessary adjustments, which can increase variation. Underreacting to true signals allows problems to persist. Principles: - Treat clear rule violations as signals of special cause. - Avoid adjusting the process for random point-to-point fluctuations within control limits. - Always look for process-based explanations before changing the system. --- Linking SPC to Process Capability Control vs Capability Two different questions: - Control: Is the process stable and predictable over time? - Answered by control charts and out-of-control tests. - Capability: How well does the stable process meet specification limits? - Answered by process capability indices and performance metrics. Key relationships: - Capability calculations are meaningful only when the process is stable. - An unstable process does not have a single consistent distribution, so capability indices are misleading. Capability Indices and Control Charts Once a process is in control: - Use the variation observed under stable conditions to estimate short-term and long-term performance. - Compare the process spread (e.g., 6 standard deviations) with specification width (USL – LSL). Concepts relevant to SPC: - Control charts establish the baseline behavior of the process. - Capability indices (such as Cp, Cpk, Pp, Ppk) rely on that baseline to measure how well the process fits within specs. - Control charts continue to monitor the process to ensure the capability remains valid. --- Implementing SPC in Practice Selecting the Right Chart Choosing the correct chart is essential: - Variables data - Subgroup size >1 and reasonably small: X̄–R chart. - Subgroup size >1 and moderate/large: X̄–s chart. - Subgroup size = 1: I–MR chart. - Attributes data - Counting defectives, varying sample size: p chart. - Counting defectives, constant sample size: np chart. - Counting defects, constant area or unit: c chart. - Counting defects, varying area or unit: u chart. Confirm: - That data collection supports rational subgrouping. - That assumptions (e.g., independence of observations) are reasonable. Establishing Baseline Control Limits To set initial control limits: - Collect data during a period believed to be free of known special causes. - Use consistent sampling and subgrouping. - Calculate center lines and limits from this baseline data. - Verify that no strong out-of-control signals occur in this baseline period; if they do, remove special causes and recalculate as needed. Do not continually recalculate limits whenever special causes occur; this hides real problems. Recalculate only when the process has clearly changed to a new stable level. Responding to Out-of-Control Signals When a control chart signals potential special cause: - Stop and investigate, when practical and appropriate for the context. - Look for assignable causes such as: - Equipment failures or adjustments - Material changes - Method changes or rework - Operator changes or new training - Environmental changes (temperature, humidity, time of day) - Document: - What happened - When it started and ended - Corrective or preventive actions taken After action: - Continue monitoring to confirm the process returns to stable behavior. - If a deliberate process improvement permanently changes the mean or variation, establish new control limits based on the improved process. --- Advanced Considerations in SPC Short-Run and Multiple-Product SPC Standard control charts assume a single product or characteristic with enough data to estimate its behavior. For short-run or multi-product cases, consider approaches that still respect SPC principles: - Transform data to a common scale (e.g., standardized distances from target). - Use charts that group multiple similar products or features while preserving rational subgrouping. - Maintain clarity about what each chart represents and how limits were derived. The core idea remains: - Use statistics to distinguish typical variability from unusual events, even when volume per product is low. Non-Normal Data Many control chart formulas assume underlying normality of subgroup means or counts with appropriate distributions (binomial, Poisson). When data are strongly non-normal: - For variables data, the central limit theorem often justifies normal-based charts for subgroup means if subgroup size is reasonably large. - For counts (attributes), p, np, c, and u charts are derived from binomial or Poisson assumptions; they typically work under common practical conditions. - When distributions are highly skewed or boundaries exist (e.g., data cannot be negative), interpret patterns carefully, especially near limits. If normal-based assumptions are severely violated, consider: - Transforming the data (for example, using a suitable statistical transformation). - Using alternative chart designs specifically adapted to the distribution, while keeping the SPC logic of distinguishing special from common cause. Integration with Ongoing Improvement SPC is not only a monitoring tool but also a diagnostic aid: - Stable but incapable processes require changes to the process design (shifting mean, reducing variation). - Unstable processes require elimination or control of special causes before capability analysis. - Sustained improvement shows up as new, better control chart baselines with reduced variation, fewer signals, and improved position relative to specifications. --- Summary Statistical Process Control (SPC) provides a disciplined, data-based way to understand and manage process behavior over time. It focuses on: - Distinguishing common cause from special cause variation. - Using control charts to monitor key process characteristics. - Applying rational subgrouping and appropriate sampling plans. - Selecting and interpreting the correct control chart for variables and attributes data. - Using rules such as Western Electric patterns to detect shifts, trends, and unusual behavior. - Ensuring process capability assessments are based on stable processes. - Taking structured action in response to out-of-control signals and sustaining improved performance. Mastery of SPC means using these tools and concepts to make reliable decisions about when to adjust, when not to adjust, and where to focus improvement efforts so that processes remain stable, predictable, and aligned with their requirements.