12.2 Variable Control Charts: I-MR, X-bar & R, X-bar & S

12.2 Variable Control Charts: I-MR, X-bar & R, X-bar & S

Quick Summary: Variable control charts are designed for continuous, quantitative data. Because continuous processes can experience shifts in central tendency, changes in spread, or both, variable SPC always pairs a location chart (tracking the mean or individual observation) with a dispersion chart (tracking the range or standard deviation). The selection of the proper chart depends entirely on the subgroup size ($n$): the I-MR chart for $n = 1$, the $\bar{X}-R$ chart for small subgroups ($2 \le n \le 10$), and the $\bar{X}-s$ chart for larger subgroups ($n > 10$).

---

The Dual-Chart Requirement for Continuous Data

Unlike attribute data (which simply counts pass/fail items or flaws), continuous variable data possesses both a location (where the distribution is centered) and a dispersion (how widely the data spreads around that center).

A process can fail in two independent ways:

1. The process mean can drift while variation remains constant (e.g., tool wear shifting shaft diameter outward).
2. Process variation can explode while the process mean remains exactly on target (e.g., loose machine bearings causing erratic chatter).

Therefore, variable SPC always operates in tandem pairs:

• Top Chart (Location): Monitors the central tendency of the process ($\bar{X}$ or Individual $X$).
• Bottom Chart (Dispersion): Monitors the variability of the process ($R$, $s$, or Moving Range $MR$).

The Golden Rule of Variable Control Charts: Always evaluate the dispersion chart FIRST. The formulas used to compute the control limits on the location chart ($\bar{X}$ or $I$) depend directly on the average dispersion ($\bar{R}$, $\bar{s}$, or $\overline{MR}$). If the dispersion chart is out of control, within-subgroup variation is unstable. This invalidates the calculated standard error, rendering the control limits on the $\bar{X}$ chart statistically meaningless.

---

Rational Subgrouping: The Foundation of SPC Sampling

The integrity of any control chart depends entirely on the concept of rational subgrouping, introduced by Walter Shewhart.

                    Rational Subgrouping Architecture

   Subgroup 1: [x1, x2, x3, x4, x5]  ──► Collected at 08:00 (Homogeneous: Same Shift/Tool)
        │
        ▼ Within-Subgroup Variation (Noise) = Short-Term Common Cause

        ▲ Between-Subgroup Variation (Signal) = Detects Long-Term Special Causes
        │
   Subgroup 2: [x1, x2, x3, x4, x5]  ──► Collected at 09:00 (Homogeneous: Same Shift/Tool)

The Two Rules of Rational Subgrouping:

1. Minimize Within-Subgroup Variation: Samples comprising an individual subgroup must be produced as close together in time as possible under identical operational conditions (same machine, same operator, same tool, same batch of raw material). Under this condition, variation within the subgroup is strictly due to inherent, short-term common cause noise.
2. Maximize Between-Subgroup Opportunity: Subgroups should be spaced across time (e.g., once per hour, once per shift, or between batches) so that if a special cause occurs (e.g., operator handover, material change, thermal expansion), it appears as a detectable shift between subgroups on the $\bar{X}$ chart.

Violation Example: If an inspector gathers a subgroup of $n = 5$ by taking 1 part from Shift A, 2 parts from Shift B, and 2 parts from Shift C, the within-subgroup variation now contains shift-to-shift differences. This inflates the subgroup range $\bar{R}$, widens the $\bar{X}$ control limits, and masks true out-of-control conditions.

---

Variable Control Chart Selection Guide

Selecting the proper variable control chart is dictated strictly by the rational subgroup sample size $n$:

---

Individuals and Moving Range (I-MR) Chart ($n = 1$)

When units are produced slowly, testing is destructive (e.g., tensile pull testing where the specimen is destroyed), or automated inline systems record individual measurements one at a time, grouping parts into rational subgroups is impossible or economically unfeasible. In this scenario ($n = 1$), the Individuals and Moving Range (I-MR) chart (also called an $X-MR$ chart) is used.

Moving Range Computation

Because each subgroup consists of a single observation, within-subgroup variation cannot be calculated directly. Instead, dispersion is estimated by computing the Moving Range ($MR$) between consecutive chronological observations:

$$MR_i = |X_i - X_{i-1}|$$

For a dataset of $k$ individual observations, there are $k - 1$ moving range values. The average moving range is:

$$\overline{MR} = \frac{\sum_{i=2}^k MR_i}{k - 1}$$

Estimating Population Standard Deviation ($\hat{\sigma}$)

To calculate $3\sigma$ control limits from the moving range, an unbiasing constant ($d_2$) is required. For a moving range of span $w = 2$, the constant is $d_2 = 1.128$:

$$\hat{\sigma} = \frac{\overline{MR}}{d_2} = \frac{\overline{MR}}{1.128}$$

Control Limit Formulas for I-MR:

1. Moving Range (MR) Chart (Dispersion):

$$\text{CL}_{MR} = \overline{MR}$$ $$\text{UCL}_{MR} = D_4 \cdot \overline{MR} = 3.267 \cdot \overline{MR} \quad (\text{for span } 2, D_4 = 3.267)$$ $$\text{LCL}_{MR} = D_3 \cdot \overline{MR} = 0 \quad (\text{for span } 2, D_3 = 0)$$

2. Individuals (I) Chart (Location):

$$\text{CL}_I = \bar{X} = \frac{\sum X_i}{k}$$ $$\text{UCL}_I = \bar{X} + 3\left(\frac{\overline{MR}}{d_2}\right) = \bar{X} + 3\left(\frac{\overline{MR}}{1.128}\right) = \bar{X} + 2.660 \cdot \overline{MR}$$ $$\text{LCL}_I = \bar{X} - 3\left(\frac{\overline{MR}}{d_2}\right) = \bar{X} - 2.660 \cdot \overline{MR}$$

Exam Constant Shortcut: Remember the number $2.66$. On the CSSC exam, $3 / d_2 = 3 / 1.128 = 2.660$. The control limits for an Individuals chart are simply $\bar{X} \pm 2.66 \cdot \overline{MR}$.

---

The $\bar{X}$ and $R$ Chart ($2 \le n \le 10$)

The $\bar{X}$ and $R$ chart is the classic workhorse of industrial SPC. Each rational subgroup contains $n$ items (typically 4 or 5), collected over $k$ inspection periods (recommended $k \ge 20$ to $25$ subgroups).

Metric Calculations:

• Subgroup Mean ($\bar{X}_i$): $\bar{X}_i = \frac{\sum_{j=1}^n X_{ij}}{n}$
• Overall Process Mean ($\bar{\bar{X}}$): $\bar{\bar{X}} = \frac{\sum_{i=1}^k \bar{X}_i}{k}$
• Subgroup Range ($R_i$): $R_i = X_{i,\max} - X_{i,\min}$
• Average Range ($\bar{R}$): $\bar{R} = \frac{\sum_{i=1}^k R_i}{k}$

Control Limit Formulas for $\bar{X}-R$:

Standard control chart factor tables eliminate the need to calculate standard deviations manually:

1. Range (R) Chart (Dispersion - Evaluated First):

$$\text{CL}_R = \bar{R}$$ $$\text{UCL}_R = D_4 \bar{R}$$ $$\text{LCL}_R = D_3 \bar{R}$$

(Note: For sample sizes $n \le 6$, $D_3 = 0$, meaning the Lower Control Limit for the Range chart is zero).

2. $\bar{X}$ Chart (Location - Evaluated Second):

$$\text{CL}_{\bar{X}} = \bar{\bar{X}}$$ $$\text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R}$$ $$\text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R}$$

(Note: The factor $A_2$ incorporates both $3 / (d_2 \sqrt{n})$, establishing true $3$-sigma limits for the distribution of subgroup averages).

Standard SPC Factors Reference Table (CSSC Exam Essential)

---

The $\bar{X}$ and $s$ Chart ($n > 10$)

Why does the $\bar{X}-R$ chart lose validity when $n > 10$? The sample range ($R$) considers only the two extreme observations ($X_{\max}$ and $X_{\min}$) and completely ignores all intermediate data points. As sample size grows past 10, the range becomes an increasingly inefficient and unstable estimator of population dispersion, heavily distorted by extreme outliers.

For large subgroups ($n > 10$), the $\bar{X}$ and $s$ chart replaces the range with the sample standard deviation ($s$):

$$s_i = \sqrt{\frac{\sum_{j=1}^n (X_{ij} - \bar{X}_i)^2}{n - 1}}$$ $$\bar{s} = \frac{\sum_{i=1}^k s_i}{k}$$

Control Limit Formulas for $\bar{X}-s$:

Using standard statistical factors $B_3, B_4,$ and $A_3$:

• $s$ Chart (Dispersion): $$\text{CL}_s = \bar{s}$$ $$\text{UCL}_s = B_4 \bar{s}$$ $$\text{LCL}_s = B_3 \bar{s}$$
• $\bar{X}$ Chart (Location): $$\text{CL}_{\bar{X}} = \bar{\bar{X}}$$ $$\text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{s}$$ $$\text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{s}$$

(When estimating population standard deviation from $\bar{s}$, use unbiasing factor $c_4$: $\hat{\sigma} = \bar{s} / c_4$).

---

Step-by-Step Worked Calculation Example: Setting up an $\bar{X}-R$ Chart

Scenario: A Green Belt oversees a precision hydraulic valve grinding line. To establish baseline control limits, the belt collects $k = 25$ rational subgroups of size $n = 5$ valve spools. The aggregate summary data is:

• Total sum of subgroup means: $\sum_{i=1}^{25} \bar{X}_i = 625.000 \text{ mm}$
• Total sum of subgroup ranges: $\sum_{i=1}^{25} R_i = 12.500 \text{ mm}$

Step 1: Compute the Center Lines

$$\bar{\bar{X}} = \frac{625.000}{25} = 25.000 \text{ mm}$$ $$\bar{R} = \frac{12.500}{25} = 0.500 \text{ mm}$$

Step 2: Retrieve Factors for $n = 5$

From the factors table for $n = 5$:

$$A_2 = 0.577, \quad D_3 = 0, \quad D_4 = 2.114$$

Step 3: Calculate Range ($R$) Chart Control Limits (Dispersion First!)

$$\text{CL}_R = \bar{R} = 0.500 \text{ mm}$$ $$\text{UCL}_R = D_4 \bar{R} = 2.114 \times 0.500 = 1.057 \text{ mm}$$ $$\text{LCL}_R = D_3 \bar{R} = 0 \times 0.500 = 0 \text{ mm}$$

Result: The Range chart operates between $0$ and $1.057 \text{ mm}$. If all 25 sample ranges fall below $1.057$, the dispersion is stable.

Step 4: Calculate $\bar{X}$ Chart Control Limits

$$\text{CL}_{\bar{X}} = \bar{\bar{X}} = 25.000 \text{ mm}$$ $$\text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R} = 25.000 + (0.577 \times 0.500) = 25.000 + 0.2885 = 25.2885 \text{ mm}$$ $$\text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R} = 25.000 - (0.577 \times 0.500) = 25.000 - 0.2885 = 24.7115 \text{ mm}$$

---

Step-by-Step Worked Calculation Example: I-MR Chart

Scenario: A bio-pharmaceutical process tests the purity percentage of a high-value chemical batch ($n = 1$). Over 20 consecutive batches, the grand average is $\bar{X} = 94.20\%$ and the average moving range is $\overline{MR} = 1.50\%$.

1. Moving Range Chart Limits:
   • $\text{CL}_{MR} = \overline{MR} = 1.50\%$
   • $\text{UCL}_{MR} = D_4 \cdot \overline{MR} = 3.267 \times 1.50 = 4.9005\%$
   • $\text{LCL}_{MR} = D_3 \cdot \overline{MR} = 0\%$
2. Individuals Chart Limits:
   • $\text{CL}_I = \bar{X} = 94.20\%$
   • $\text{UCL}_I = \bar{X} + 2.660 \cdot \overline{MR} = 94.20 + (2.660 \times 1.50) = 94.20 + 3.990 = 98.19\%$
   • $\text{LCL}_I = \bar{X} - 2.660 \cdot \overline{MR} = 94.20 - 3.990 = 90.21\%$

---

Step-by-Step Worked Calculation Example: $\bar{X}-s$ Chart ($n > 10$)

Scenario: An automated test bench measures the burst pressure of automotive brake lines. Subgroups of size $n = 12$ are sampled across $k = 20$ test cycles. The summary data yields:

• Grand process mean: $\bar{\bar{X}} = 120.00\text{ psi}$
• Average sample standard deviation: $\bar{s} = 3.60\text{ psi}$
• Factors for $n = 12$: $A_3 = 0.886, B_3 = 0.283, B_4 = 1.717$

1. Standard Deviation ($s$) Chart Limits (Dispersion First!):
   • $\text{CL}_s = \bar{s} = 3.60\text{ psi}$
   • $\text{UCL}_s = B_4 \bar{s} = 1.717 \times 3.60 = 6.1812 \approx 6.18\text{ psi}$
   • $\text{LCL}_s = B_3 \bar{s} = 0.283 \times 3.60 = 1.0188 \approx 1.02\text{ psi}$
2. $\bar{X}$ Chart Limits (Location):
   • $\text{CL}_{\bar{X}} = \bar{\bar{X}} = 120.00\text{ psi}$
   • $\text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{s} = 120.00 + (0.886 \times 3.60) = 120.00 + 3.1896 = 123.19\text{ psi}$
   • $\text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{s} = 120.00 - (0.886 \times 3.60) = 120.00 - 3.1896 = 116.81\text{ psi}$

---

Critical CSSC Exam Traps

• Trap 1: Evaluating the $\bar{X}$ Chart When the $R$ Chart is Out of Control — If an exam question describes an $\bar{X}-R$ chart where several points exceed the UCL on the $R$ chart, the correct action is never to analyze the $\bar{X}$ chart or compute $C_{pk}$. The correct action is to stop and investigate the special causes causing unstable dispersion on the $R$ chart.
• Trap 2: Using the Wrong Subgroup Size Factor — In manual calculations, candidates often select factors corresponding to $k$ (the number of subgroups, e.g., 25) rather than $n$ (the sample size within each subgroup, e.g., 5). Always look up $A_2, D_3, D_4$ using the subgroup size $n$.
• Trap 3: Using $\bar{X}-R$ When $n > 10$ — The exam tests your ability to select the right tool. If a problem states that automated machinery inspects $n = 15$ parts every hour, selecting an $\bar{X}-R$ chart is incorrect; an $\bar{X}-s$ chart must be selected.
• Trap 4: Forgetting that $D_3 = 0$ for $n \le 6$ — Candidates often expect a non-zero lower control limit on Range charts. For sample sizes of 2, 3, 4, 5, and 6, $D_3$ is zero, making $\text{LCL}_R = 0$.