Statistical process control Study Guide

📖 Core Concepts Statistical Process Control (SPC) – Use of statistical methods to monitor and control a production (or service) process so it stays within specification limits. Common (natural) variation – Random fluctuations that are always present when the process runs under the same conditions. Special (assignable) variation – Unexpected changes caused by atypical sources; they signal that the process is out of statistical control. Control Chart – Graph that plots a process characteristic with a center line (CL) = in‑control mean and upper/lower control limits (UCL/LCL) = $CL \pm 3\sigma$; separates common from special variation. Stable process – No control‑chart detection rules are triggered; variation stays within the calculated limits. Process capability – Once a process is stable, capability analysis (e.g., $Cp$, $C{pk}$) predicts how often products will meet specifications. --- 📌 Must Remember Purpose of SPC: Detect problems early, prevent waste, and keep production efficient. Key historical figures: Walter Shewhart (invented the control chart, 1924) and W. Edwards Deming (propagated SPC during WWII). Control‑limit formula: $UCL/LCL = CL \pm 3\sigma$ (where $\sigma$ = in‑control standard deviation). Two SPC phases: 1) Establish the process (identify/specify limits, eliminate assignable causes). 2) Monitor the established process (continuous control‑chart use). Common vs. Special variation: Common → expected, does not trigger alarms. Special → unexpected, produces points outside limits → corrective action needed. Main SPC tools: Run charts, control charts, Ishikawa (fishbone) diagrams, Pareto charts, designed experiments. Advanced chart types: CUSUM and EWMA charts give higher sensitivity to small or persistent shifts. --- 🔄 Key Processes Define process & specification limits – Know the product/service requirements and measurable characteristics. Eliminate assignable causes – Identify and remove special sources of variation to reach stability. Select appropriate control chart – Choose based on data type (individuals, averages, ranges, etc.). Calculate control limits – $UCL/LCL = CL \pm 3\sigma$. Plot ongoing observations. Apply detection rules – Watch for points beyond limits or patterns that violate Western Electric rules. Diagnose excessive variation – If rules are violated, use fishbone diagrams, Pareto analysis, or designed experiments to locate root causes. Implement corrective actions – Standardize work, train staff, error‑proof, adjust inputs/processes. Re‑assess stability – After changes, re‑plot control chart; confirm no violations before declaring the process stable. --- 🔍 Key Comparisons Common vs. Special Variation Common: natural, always present, stays within limits. Special: atypical, creates out‑of‑limit points, requires corrective action. SPC vs. Traditional Inspection SPC: monitors the process continuously; prevents defects. Inspection: checks after production; catches defects after they occur. Shewhart (classic) Chart vs. CUSUM/EWMA Shewhart: good for large, abrupt shifts; limits at $±3\sigma$. CUSUM: accumulates small deviations → sensitive to modest, sustained shifts. EWMA: weights recent data more heavily → detects gradual drifts. Manufacturing vs. Non‑Manufacturing SPC Manufacturing: repetitive, measurable outputs (e.g., dimensions). Non‑manufacturing: repetitive service/administrative tasks (e.g., billing cycles, IT operations). --- ⚠️ Common Misunderstandings “A point outside the limits means a defective product.” It signals process out‑of‑control; the individual item may still be OK. “If the chart looks steady, no action is ever needed.” Even a stable chart can hide drifts that CUSUM/EWMA detect better. “Special causes are always bad.” Some special causes are opportunities for improvement once identified. “SPC can’t be used in knowledge‑intensive work.” It can be applied to repetitive parts of such work; the controversy is only for truly non‑repetitive tasks. --- 🧠 Mental Models / Intuition River Analogy: Think of a process as a river flowing within banks (control limits). Common variation is the natural ripple of water; special variation is a sudden rock that throws the water outside the banks. Signal vs. Noise: SPC treats the signal (special causes) as a red flag; common variation is just background noise. Stability as a Baseline: Once the river’s flow is steady (stable), you can reliably predict where the water will be (process capability). --- 🚩 Exceptions & Edge Cases Knowledge‑intensive, non‑repetitive activities (e.g., R&D, system design) – SPC may be inappropriate because essential variation cannot be removed. Highly variable processes where common variation dominates and control limits become too wide → capability analysis may show the process can’t meet specs without redesign. Small sample sizes – Estimating $\sigma$ reliably may be difficult; alternative charts (e.g., $X$‑chart with $R$‑chart) are recommended. --- 📍 When to Use Which Shewhart chart → When you expect large, abrupt shifts and have ample data per subgroup. CUSUM → When detecting small, persistent shifts (e.g., a 1‑σ drift). EWMA → When you need to catch gradual trends and give more weight to recent observations. Ishikawa (fishbone) diagram → First step after a control‑chart violation to brainstorm possible causes. Designed experiments → When you need quantitative evidence of which factors most affect variation. SPC vs. Inspection → Use SPC for high‑volume, repetitive processes; rely on inspection only when the process cannot be monitored continuously (e.g., low‑volume custom parts). --- 👀 Patterns to Recognize Single point beyond UCL/LCL → Immediate special cause. Run of 7+ points on one side of the CL → Indicates a shift in the mean. Trend of 6+ points steadily increasing or decreasing → Possible drift. Two out of three consecutive points beyond 2σ on same side → Early warning of a shift. Increasing variability (wider spread) over time → May signal loss of control even if points stay within limits. --- 🗂️ Exam Traps Choosing the wrong chart type: Selecting a Shewhart chart for a tiny 0.5‑σ shift will miss the problem; the exam may present a scenario where CUSUM/EWMA is the correct answer. Confusing “common” with “acceptable”: A process can be in control (common variation) yet still be incapable of meeting specifications—capability analysis is a separate step. Misreading the control‑limit formula: Remember it is $CL \pm 3\sigma$, not $±2\sigma$ (that's for Shewhart’s warning limits, not the official control limits). Attributing a violation to “measurement error” automatically: The first assumption should be a special cause in the process; measurement error is only considered after ruling out process issues. Assuming SPC eliminates the need for any inspection: Even with SPC, final product verification may still be required for safety‑critical items; the exam may test this nuance.