20.2 Variable Control Charts: X-bar & R, X-bar & S, and Individual-Moving Range

Variable control charts monitor continuous quality characteristics that can be measured on a numerical scale, such as shaft diameter (mm), tensile strength (MPa), fluid viscosity (cP), or assembly torque (N·m). Because continuous measurements carry far more statistical information than binary go/no-go classifications, variable control charts provide early detection of minute shifts in both the process central tendency (mean) and the process dispersion (variability).

---

1. The Principles of Rational Subgrouping

The mathematical integrity of any variable control chart hinges on the concept of rational subgrouping, formulated by Walter Shewhart:

Rational Subgrouping Principle: Subgroups should be selected such that the opportunity for variation within each subgroup is minimized (reflecting only short-term, common-cause chance variation), while the opportunity for variation between subgroups is maximized (reflecting potential assignable causes over time).

               Rational Subgrouping Mechanics

   Subgroup 1          Subgroup 2          Subgroup 3
  (Time = 8:00)       (Time = 9:00)       (Time = 10:00)
 ┌─────────────┐     ┌─────────────┐     ┌─────────────┐
 │ X1 X2 X3 X4 │     │ X1 X2 X3 X4 │     │ X1 X2 X3 X4 │
 └──────┬──────┘     └──────┬──────┘     └──────┬──────┘
        │                   │                   │
        ▼                   ▼                   ▼
   Within-Sample       Within-Sample       Within-Sample
   Variation:          Variation:          Variation:
   ONLY Common Cause   ONLY Common Cause   ONLY Common Cause
   (Short duration)    (Short duration)    (Short duration)

        ▲                   ▲                   ▲
        └───────────────────┴───────────────────┘
                            │
                            ▼
                 Between-Subgroup Variation:
                 Exposes Assignable Causes
                 (Tool wear, operator change, material lot)

Practical Implementation Rules

1. Consecutive Unit Sampling: Take sample items produced consecutively in immediate temporal succession (e.g., five parts off the press in five seconds). This guarantees that within-sample spread represents only pure machine capability.
2. Avoid Random Shop-Floor Blending: Never sample one part at 8:00 AM, one at 10:00 AM, one at 12:00 PM, and pool them into a single subgroup of $n = 4$. That practice buries process shifts inside the within-subgroup range, inflating control limits and blinding the chart to assignable causes!
3. Subgroup Size ($n$): Typically $n = 4$ or $n = 5$ for $\bar{X}$ and $R$ charts. Small subgroups maximize sampling frequency and minimize inspection costs while invoking the Central Limit Theorem to ensure normality of subgroup means.

---

2. The $\bar{X}$ and $R$ Control Chart Architecture ($n = 2$ to $9$)

When subgroup size is small ($2 \le n \le 9$, most commonly $n = 4$ or $n = 5$), the sample range ($R$) is an efficient and computationally simple proxy for dispersion.

Fundamental Subgroup Statistics

For $m$ subgroups, each containing $n$ individual measurements $X_{ij}$:

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

Control Limits Formulas (from NCEES Handbook)

Using statistical factors tabulated in the NCEES Reference Handbook:

           X-bar & R Chart Formula Architecture

  X-bar Chart (Location)          R Chart (Dispersion)
 ────────────────────────       ────────────────────────
  UCL = X_double_bar + A2 * R_bar  UCL = D4 * R_bar
  CL  = X_double_bar               CL  = R_bar
  LCL = X_double_bar - A2 * R_bar  LCL = D3 * R_bar

• $\bar{X}$ Chart Limits: $$CL_{\bar{X}} = \bar{\bar{X}}$$ $$UCL_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R}$$ $$LCL_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R}$$
• $R$ Chart Limits: $$CL_R = \bar{R}$$ $$UCL_R = D_4 \bar{R}$$ $$LCL_R = D_3 \bar{R}$$

Important Note on $D_3$: For subgroup sizes $n \le 6$, the factor $D_3 = 0$ (or is non-existent). Consequently, $LCL_R = 0$ for sample sizes from 2 to 6. A lower range of zero simply means perfect uniformity; it cannot be an out-of-control condition.

Estimating the Process Standard Deviation ($\hat{\sigma}$)

The sample range $\bar{R}$ is a biased estimator of population standard deviation $\sigma$. To convert $\bar{R}$ into an unbiased estimate of $\sigma$, divide by the NCEES factor $d_2$:

$$\hat{\sigma} = \frac{\bar{R}}{d_2}$$

Note on Derivation: The factor $A_2$ in the NCEES Handbook is mathematically defined as:

$$A_2 = \frac{3}{d_2 \sqrt{n}}$$

Therefore: $A_2 \bar{R} = \frac{3 \bar{R}}{d_2 \sqrt{n}} = 3 \frac{\hat{\sigma}}{\sqrt{n}} = 3 \sigma_{\bar{X}}$, which confirms exact equivalence to 3-sigma limits!

The Cardinal Operational Rule: Evaluate the $R$ Chart FIRST!

On both real manufacturing floors and the FE exam, the sequence of interpretation is inviolable:

1. Step 1: Check the $R$ Chart: Verify that all subgroup ranges fall within $UCL_R$ and $LCL_R$ without non-random patterns.
2. Step 2: Check the $\bar{X}$ Chart: Only if the $R$ chart is in control can the $\bar{X}$ chart be interpreted.

Why? The control limits of the $\bar{X}$ chart are calculated directly from $\bar{R}$ ($UCL = \bar{\bar{X}} \pm A_2 \bar{R}$). If process dispersion is out of control, the estimate of within-subgroup variance is unstable, making the $\bar{X}$ control limits meaningless. An out-of-control range chart invalidates the entire $\bar{X}$ chart!

---

3. The $\bar{X}$ and $S$ Control Chart Architecture ($n \ge 10$)

When subgroup size is moderately large ($n \ge 10$), the range ($R$) loses statistical efficiency because it depends only on the two extreme values ($X_{\max}$ and $X_{\min}$) and ignores the intermediate $n - 2$ data points. Furthermore, extreme values in large samples exhibit high variance. Therefore, for $n \ge 10$, the sample standard deviation ($S$) replaces the range.

Subgroup Standard Deviation Statistics

• Subgroup Standard Deviation ($S_i$): $$S_i = \sqrt{\frac{1}{n-1} \sum_{j=1}^n (X_{ij} - \bar{X}_i)^2}$$
• Average Standard Deviation ($\bar{S}$): $$\bar{S} = \frac{1}{m} \sum_{i=1}^m S_i$$

Control Limits Formulas (from NCEES Handbook)

• $\bar{X}$ Chart Limits: $$CL_{\bar{X}} = \bar{\bar{X}}$$ $$UCL_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{S}$$ $$LCL_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{S}$$
• $S$ Chart Limits: $$CL_S = \bar{S}$$ $$UCL_S = B_4 \bar{S}$$ $$LCL_S = B_3 \bar{S}$$ (Note: For $n \le 5$, $B_3 = 0$; for $n \ge 6$, $B_3 > 0$).

Estimating Process Standard Deviation from $S$

The sample standard deviation $S$ is also slightly biased. Dividing $\bar{S}$ by the NCEES bias-correction factor $c_4$ yields an unbiased estimate:

$$\hat{\sigma} = \frac{\bar{S}}{c_4}$$

Note on Derivation: $A_3 = \frac{3}{c_4 \sqrt{n}}$, establishing that $A_3 \bar{S} = 3 \frac{\hat{\sigma}}{\sqrt{n}}$.

---

4. Individual and Moving Range ($I-MR$ / $X-mR$) Charts ($n = 1$)

In many modern engineering environments, subgrouping multiple parts ($n > 1$) is physically impossible, prohibitively expensive, or meaningless:

1. Destructive Testing: Testing the tensile failure strength of missile casing steel or explosion caps destroys the unit.
2. Automated Continuous Sensing: In chemical bioreactors, automated probes log pH, temperature, or dissolved oxygen every 10 seconds. Grouping consecutive seconds into subgroups masks fluid dynamics.
3. Low Production Volume / Long Cycle Times: When producing heavy gas turbines or commercial aircraft, output may only be 1 or 2 units per week.

In these scenarios, the sample size is $n = 1$. We construct an Individuals and Moving Range ($I-MR$) chart.

Two-Point Moving Range Mechanics

Because $n = 1$, within-sample range cannot be calculated directly. Instead, artificial subgroups are formed by taking the absolute difference between consecutive observations (span $w = 2$):

$$MR_i = |X_i - X_{i-1}| \quad \text{for } i = 2, 3, \dots, m$$

For $m$ individual measurements, there are $m - 1$ moving ranges. The average moving range is:

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

Control Limits Formulas (from NCEES Handbook)

For $w = 2$, the NCEES factors are: $d_2 = 1.128$, $D_3 = 0$, $D_4 = 3.267$, and $E_2 = \frac{3}{d_2} = \frac{3}{1.128} \approx 2.660$.

• Individuals ($X$) Chart Limits: $$CL_X = \bar{X}$$ $$UCL_X = \bar{X} + E_2 \overline{MR} = \bar{X} + 2.660 \overline{MR}$$ $$LCL_X = \bar{X} - E_2 \overline{MR} = \bar{X} - 2.660 \overline{MR}$$
• Moving Range ($MR$) Chart Limits: $$CL_{MR} = \overline{MR}$$ $$UCL_{MR} = D_4 \overline{MR} = 3.267 \overline{MR}$$ $$LCL_{MR} = D_3 \overline{MR} = 0$$

Estimating Process Standard Deviation for $n = 1$

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

Caution on Normality: While $\bar{X}$ charts are protected by the Central Limit Theorem (subgroup means are approximately normal even if raw data is non-normal), the Individuals ($X$) chart plots raw individual data points ($n = 1$). It is highly sensitive to departures from normality. If the underlying distribution is skewed or heavy-tailed, false alarm rates on an $I-MR$ chart can deviate substantially from $\alpha = 0.0027$.

---

5. Master NCEES Control Chart Factors Table

The following table compiles the exact factors published in the NCEES Reference Handbook for subgroup sizes $n = 2$ through $n = 10$:

---

6. Step-by-Step Worked Engineering Calculations

Worked Example 20.2.1: $\bar{X}$ and $R$ Chart Control Limits and Capability Estimation

Problem Statement: A manufacturing plant turns titanium alloy fasteners on a high-precision CNC lathe. The quality department samples $n = 5$ consecutive fasteners every 30 minutes over a two-shift production run, compiling $m = 25$ subgroups. The summary calculations across all 25 subgroups yield:

$$\sum_{i=1}^{25} \bar{X}_i = 312.550 \text{ mm}, \qquad \sum_{i=1}^{25} R_i = 1.150 \text{ mm}$$

Engineering blueprint tolerances for this diameter are specified as $12.500 \pm 0.050$ mm ($USL = 12.550$ mm, $LSL = 12.450$ mm).

1. Calculate the grand average $\bar{\bar{X}}$ and average range $\bar{R}$.
2. Determine the Upper and Lower Control Limits for both the $R$ chart and the $\bar{X}$ chart.
3. Estimate the true process standard deviation $\hat{\sigma}$.
4. If Subgroup 14 has an average of $\bar{X}_{14} = 12.535$ mm and a range of $R_{14} = 0.095$ mm, evaluate whether this specific subgroup indicates an out-of-control condition.

Solution:

Step 1: Compute Baseline Grand Averages

• Grand Average: $\bar{\bar{X}} = \frac{312.550}{25} = 12.5020$ mm
• Average Range: $\bar{R} = \frac{1.150}{25} = 0.0460$ mm

Step 2: Calculate Control Limits for $R$ and $\bar{X}$ Charts From the NCEES table for $n = 5$: $A_2 = 0.577, \quad d_2 = 2.326, \quad D_3 = 0, \quad D_4 = 2.114$

R Chart Limits (Dispersion):

• $CL_R = \bar{R} = 0.0460$ mm
• $UCL_R = D_4 \bar{R} = (2.114)(0.0460) = 0.09724 \approx 0.0972$ mm
• $LCL_R = D_3 \bar{R} = (0)(0.0460) = 0.000$ mm

$\bar{X}$ Chart Limits (Location):

• $CL_{\bar{X}} = \bar{\bar{X}} = 12.5020$ mm
• $UCL_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R} = 12.5020 + (0.577)(0.0460) = 12.5020 + 0.02654 = 12.5285$ mm
• $LCL_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R} = 12.5020 - 0.02654 = 12.4755$ mm

Step 3: Estimate True Process Standard Deviation ($\hat{\sigma}$)

$$\hat{\sigma} = \frac{\bar{R}}{d_2} = \frac{0.0460}{2.326} = 0.019776 \approx 0.0198 \text{ mm}$$

Step 4: Evaluate Subgroup 14

• Check dispersion first: $R_{14} = 0.095$ mm. Comparing to $R$ chart limits: $LCL_R (0) \le 0.095 \le UCL_R (0.0972)$. The range is in control.
• Check location: $\bar{X}_{14} = 12.535$ mm. Comparing to $\bar{X}$ chart limits: $UCL_{\bar{X}} = 12.5285$ mm.
• Conclusion: Because $\bar{X}_{14} = 12.535 > 12.5285$, Subgroup 14 is out of control! An assignable cause (such as a chipped insert or thermal tool growth) has shifted the process mean upward.
• Crucial Exam Trap Insight: Notice that $\bar{X}_{14} = 12.535$ mm is still below $USL = 12.550$ mm! A novice would say "the parts are good because they meet spec." An industrial engineer recognizes that the process has experienced a special-cause disturbance requiring immediate root-cause correction before parts violate $USL$.

---

Worked Example 20.2.2: Destructive Pull-Testing on an $I-MR$ Chart

Problem Statement: A composites lab performs destructive tensile pull-to-failure testing on carbon-fiber structural coupons. Because testing completely destroys each specimen, only one coupon ($n = 1$) is evaluated per cure batch. The pull strengths (in MPa) recorded across 10 consecutive batches are:

1. Calculate the moving ranges $MR_i$ and the average moving range $\overline{MR}$.
2. Determine the Center Line and Control Limits for both the Moving Range chart and the Individuals chart.
3. Estimate the process standard deviation $\hat{\sigma}$.

Solution:

Step 1: Calculate Moving Ranges Compute consecutive two-point absolute differences:

• $MR_2 = |435 - 420| = 15$
• $MR_3 = |428 - 435| = 7$
• $MR_4 = |415 - 428| = 13$
• $MR_5 = |440 - 415| = 25$
• $MR_6 = |452 - 440| = 12$
• $MR_7 = |430 - 452| = 22$
• $MR_8 = |418 - 430| = 12$
• $MR_9 = |422 - 418| = 4$
• $MR_{10} = |438 - 422| = 16$

Sum of moving ranges: $\sum MR_i = 15 + 7 + 13 + 25 + 12 + 22 + 12 + 4 + 16 = 126$ MPa. Number of moving ranges: $m - 1 = 10 - 1 = 9$.

$$\overline{MR} = \frac{126}{9} = 14.00 \text{ MPa}$$

Average of individual values $\bar{X}$:

$$\bar{X} = \frac{420 + 435 + 428 + 415 + 440 + 452 + 430 + 418 + 422 + 438}{10} = \frac{4,298}{10} = 429.80 \text{ MPa}$$

Step 2: Control Limits for Moving Range and Individuals Charts For span $w = 2$: $d_2 = 1.128, \quad D_3 = 0, \quad D_4 = 3.267, \quad E_2 = 2.660$

Moving Range ($MR$) Chart:

• $CL_{MR} = \overline{MR} = 14.00$ MPa
• $UCL_{MR} = D_4 \overline{MR} = (3.267)(14.00) = 45.74$ MPa
• $LCL_{MR} = D_3 \overline{MR} = 0$ MPa (All moving ranges fall between 4 and 25 MPa, well below 45.74 MPa; dispersion is in control).

Individuals ($X$) Chart:

• $CL_X = \bar{X} = 429.80$ MPa
• $UCL_X = \bar{X} + E_2 \overline{MR} = 429.80 + (2.660)(14.00) = 429.80 + 37.24 = 467.04$ MPa
• $LCL_X = \bar{X} - E_2 \overline{MR} = 429.80 - 37.24 = 392.56$ MPa (All individual coupon strengths fall between 415 and 452 MPa, safely inside [392.56, 467.04]; process mean is in control).

Step 3: Estimate Process Standard Deviation

$$\hat{\sigma} = \frac{\overline{MR}}{d_2} = \frac{14.00}{1.128} = 12.41 \text{ MPa}$$

---

7. NCEES Reference Handbook Tips & Realistic Exam Traps

• Formula Selection by Subgroup Size ($n$):
  • $n = 1$: Use $I-MR$ chart with factors $E_2 = 2.660$ and $D_4 = 3.267$.
  • $2 \le n \le 9$: Use $\bar{X}$ and $R$ chart with factors $A_2, D_3, D_4$.
  • $n \ge 10$: Use $\bar{X}$ and $S$ chart with factors $A_3, B_3, B_4$.
• Estimating Sigma Trap: Be careful not to confuse $\hat{\sigma}$ with $\sigma_{\bar{X}}$. The true process standard deviation is $\hat{\sigma} = \bar{R} / d_2$ or $\bar{S} / c_4$. The standard deviation of the plotted averages is $\sigma_{\bar{X}} = \hat{\sigma} / \sqrt{n} = \bar{R} / (d_2 \sqrt{n})$. NCEES questions may ask for either "the estimated process standard deviation" ($\hat{\sigma}$) or "the standard deviation of the subgroup means" ($\sigma_{\bar{X}}$).
• $D_3$ is Zero for $n \le 6$: If asked for the Lower Control Limit of an $R$ chart where $n = 4$ or $n = 5$, the answer is immediately $0$. Do not waste time hunting for non-existent negative limits.
• Denominator of $\overline{MR}$ is $m - 1$: When calculating the average moving range across $m$ observations, remember there are only $m - 1$ consecutive pairs. Dividing by $m$ instead of $m - 1$ is a classic distractor on numerical exam problems.