5.3 X-bar and R Charts

5.3 X-bar and R Charts

When quality technicians monitor continuous variables data—such as machined shaft diameters, stamped bracket thicknesses, chemical bath temperatures, or electrical resistance—the $\bar{X}$ and $R$ control chart pair is the foundational statistical tool of industry. Developed by Dr. Walter Shewhart, this paired chart system monitors both the central tendency (location) and the dispersion (spread) of a manufacturing process over time.

To pass the ASQ CQT examination and successfully establish SPC on the shop floor, technicians must be able to calculate subgroup means and ranges, compute trial centerlines and control limits using standard factor tables, execute the mandatory sequence of evaluating process spread before process location, eliminate confirmed assignable causes, and estimate the process standard deviation ($\hat{\sigma}$).

---

1. Application Scope: Why Variables Data and Paired Charts?

Continuous Variables Data

Variables data consists of quantitative measurements along a continuous numerical scale (e.g., millimeters, inches, pounds, foot-pounds, degrees Celsius, ohms). Variables data contains substantially more statistical information than attribute data (pass/fail or go/no-go). While an attribute check sheet only tells you whether a part is acceptable, variables data reveals where the process is centered within the tolerance band and how tightly the measurements cluster.

Subgroup Size ($n$)

The $\bar{X}$ and $R$ chart is designed specifically for subgroup sizes from $n = 2$ to $n = 9$, with $n = 4$ or $n = 5$ representing the industry standard. Subgroup sizes larger than 9 cause the range to lose statistical efficiency, requiring the sample standard deviation ($s$) to replace the range ($R$).

The Necessity of Paired Charts

Any continuous probability distribution is fundamentally characterized by two independent parameters: its central tendency ($\mu$) and its dispersion ($\sigma$). Monitoring only one parameter leaves the technician blind to half of all potential manufacturing failures:

1. Mean Shifts with Constant Dispersion: Tool wear, thermal drift, or incorrect machine offsets shift the average part diameter toward the Upper Specification Limit, while piece-to-piece repeatability remains unchanged. This is detected by the $\bar{X}$ chart.
2. Dispersion Expands with Constant Mean: Bearing looseness, spindle runout, inconsistent raw material hardness, or operator clamping inconsistency widens the overall spread of parts, while the average remains centered on nominal. This is detected by the $R$ chart.

   CASE 1: Mean Shifts, Spread Constant      CASE 2: Spread Expands, Mean Constant
             (Caught by X-bar)                           (Caught by R)

          Shifted Mean                                    Expanded Spread
              |                                            <----------->
           .  |  .                                       .       |       .
         .    |    .                                   .         |         .
       .      |      .                               .           |           .
     ---------+---------                           --------------+--------------
            Target                                             Target

---

2. Core Mathematical Formulas

For a baseline study consisting of $k$ rational subgroups, each containing $n$ measured observations:

Step 1: Subgroup Statistics

For each individual subgroup $i$ (where $i = 1, 2, \dots, k$):

• Subgroup Average ($\bar{X}_i$): $\bar{X}_i = \frac{1}{n} \sum_{j=1}^n X_{ij} = \frac{X_{i1} + X_{i2} + \dots + X_{in}}{n}$
• Subgroup Range ($R_i$): $R_i = X_{i,\max} - X_{i,\min}$

Step 2: Centerlines (Grand Mean and Average Range)

• Grand Mean ($\bar{\bar{X}}$) (Centerline for the $\bar{X}$ chart): $\bar{\bar{X}} = \frac{1}{k} \sum_{i=1}^k \bar{X}_i = \frac{\bar{X}_1 + \bar{X}_2 + \dots + \bar{X}_k}{k}$
• Average Range ($\bar{R}$) (Centerline for the $R$ chart): $\bar{R} = \frac{1}{k} \sum_{i=1}^k R_i = \frac{R_1 + R_2 + \dots + R_k}{k}$

Step 3: Range ($R$) Chart Control Limits

Control limits for process dispersion are computed first: $UCL_R = D_4 \bar{R}$ $CL_R = \bar{R}$ $LCL_R = D_3 \bar{R}$

[!IMPORTANT] Critical CQT Fact: $D_3 = 0$ for $n \le 6$ For subgroup sizes of $n = 2, 3, 4, 5,$ and $6$, the factor $D_3$ does not exist or is mathematically defined as 0. Therefore, for $n \le 6$, $LCL_R = 0$. There is no lower control limit for the range chart because obtaining a sample range of zero in small samples is within normal common cause variation.

Step 4: Average ($\bar{X}$) Chart Control Limits

Control limits for process location are computed using factor $A_2$: $UCL_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R}$ $CL_{\bar{X}} = \bar{\bar{X}}$ $LCL_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R}$

Derivation of Factor $A_2$

Where does factor $A_2$ originate? Shewhart derived $A_2$ directly from statistical theory. The true $3\sigma$ control limits for subgroup averages are $\bar{\bar{X}} \pm 3\sigma_{\bar{X}} = \bar{\bar{X}} \pm 3(\sigma / \sqrt{n})$. Because the population standard deviation $\sigma$ is unknown, it is estimated from the average range as $\hat{\sigma} = \bar{R} / d_2$. Substituting $\hat{\sigma}$ into the control limit equation yields: $\text{3-Sigma Margin} = 3\left(\frac{\bar{R}}{d_2 \sqrt{n}}\right) = \left(\frac{3}{d_2 \sqrt{n}}\right) \bar{R} = A_2 \bar{R}$ $A_2 = \frac{3}{d_2 \sqrt{n}}$

Step 5: Estimating Process Standard Deviation ($\hat{\sigma}$)

To calculate process capability ($C_p, C_{pk}$) from an $\bar{X}$ and $R$ chart, the within-subgroup population standard deviation $\hat{\sigma}$ is estimated using the standard factor $d_2$: $\hat{\sigma} = \frac{\bar{R}}{d_2}$

---

3. Standard Factor Table for $\bar{X}$ and $R$ Charts

The standard factors established by the American Society for Quality (ASQ) and ASTM are provided in the reference table below for subgroup sizes $n = 2$ through $n = 10$:

Key Patterns to Memorize for the CQT Exam:

• As sample size $n$ increases, $A_2$ decreases because larger samples have a smaller standard error ($\sigma / \sqrt{n}$), pulling the control limits closer to the grand mean.
• As sample size $n$ increases, $D_4$ decreases because larger samples have a higher probability of capturing extreme values, stabilizing the expected sample range.
• $D_3$ is zero for $n \le 6$, and becomes a positive decimal starting at $n = 7$ ($D_3 = 0.076$).

---

4. Step-by-Step Construction and Limit Revision Workflow

Constructing an operational control chart requires an orderly engineering sequence:

  [Phase 1: Setup]        Select critical-to-quality (CTQ) characteristic;
                          Verify Gage R&R < 10% on measurement system.
                                 |
                                 v
  [Phase 2: Baseline]     Collect 20 to 25 rational subgroups (k = 20-25,
                          typically n = 4 or 5; 100+ total pieces).
                                 |
                                 v
  [Phase 3: Compute]      Calculate X-bar_i and R_i for each subgroup;
                          Calculate centerlines X-double-bar and R-bar;
                          Calculate trial control limits (A2, D3, D4).
                                 |
                                 v
  [Phase 4: R Chart]      CRITICAL: Evaluate the R Chart FIRST!
                          Are all range points in statistical control?
                             /          \
                           YES           NO --> Investigate assignable cause.
                            |                   Found? Discard subgroup & recalculate.
                            |                   Not found? Retain point; resolve cause.
                            v
  [Phase 5: X-bar Chart]  Evaluate the X-bar Chart against stable R-bar.
                          Investigate out-of-control means for assignable causes.
                                 |
                                 v
  [Phase 6: Release]      Lock in final standard limits for live shop-floor monitoring.

Why Evaluate the $R$ Chart FIRST?

The most critical procedural rule in variables control charting is: Always evaluate the $R$ chart before interpreting the $\bar{X}$ chart.

The mathematical justification is absolute: The control limits for the $\bar{X}$ chart are calculated directly from $\bar{R}$ ($UCL, LCL = \bar{\bar{X}} \pm A_2 \bar{R}$). If the $R$ chart is out of control, process dispersion is unstable. An unstable dispersion means that $\bar{R}$ is a distorted, mathematically invalid estimate of process variability. If the estimate of variation is invalid, the control limits on the $ar{X}$ chart are completely fictitious! Evaluating $ar{X}$ limits when the $R$ chart is out of control is an exercise in futility.

Protocol for Handling Out-of-Control Trial Points

When a point on the initial trial $R$ or $\bar{X}$ chart falls outside control limits:

1. Investigate for Assignable Causes: The technician must immediately examine machine logs, setup sheets, and maintenance records for that specific time window (e.g., tool chipped, broken coolant nozzle, wrong material heat loaded).
2. If an Assignable Cause is Identified and Corrected: The out-of-control subgroup is discarded from the baseline dataset. The centerlines ($\bar{\bar{X}}$ and $\bar{R}$) and control limits are recalculated using the remaining $k - 1$ subgroups.
3. If No Assignable Cause Can Be Found: The point cannot be discarded! Arbitrarily scrubbing out-of-control points without an engineering assignable cause is known as 'data scrubbing' or 'cherry-picking,' which artificially narrows limits and creates false confidence in an unstable system.

---

5. Fully Worked Shop-Floor Numerical Example: CNC Shaft Turning

Scenario

A precision machine shop is establishing an $\bar{X}$ and $R$ control chart for the outer diameter of a hydraulic actuator shaft turned on a CNC lathe. Engineering drawing callout is $35.000 \pm 0.050\text{ mm}$ ($LSL = 34.950\text{ mm}, USL = 35.050\text{ mm}$).

The quality technician collects $k = 20$ baseline rational subgroups of size $n = 5$ consecutive shafts across two production shifts (100 total shafts measured with an optical shaft gage).

Baseline Dataset Summary

Sum of all 20 subgroup means: $\sum_{i=1}^{20} \bar{X}_i = 700.400\text{ mm}$

Sum of all 20 subgroup ranges: $\sum_{i=1}^{20} R_i = 0.560\text{ mm}$

Subgroup 8 recorded the following statistics: $\bar{X}_8 = 35.034\text{ mm}$, $R_8 = 0.065\text{ mm}$.

Step 1: Calculate Initial Trial Centerlines

$\bar{\bar{X}} = \frac{\sum \bar{X}_i}{k} = \frac{700.400}{20} = 35.0200\text{ mm}$ $\bar{R} = \frac{\sum R_i}{k} = \frac{0.560}{20} = 0.0280\text{ mm}$

Step 2: Retrieve Standard Factors for $n = 5$

From the factor table for $n = 5$: $A_2 = 0.577, \quad D_3 = 0, \quad D_4 = 2.114, \quad d_2 = 2.326$

Step 3: Compute Initial Trial Limits for the $R$ Chart

$UCL_R = D_4 \bar{R} = 2.114 \times 0.0280 = 0.05919 \approx 0.0592\text{ mm}$ $CL_R = \bar{R} = 0.0280\text{ mm}$ $LCL_R = D_3 \bar{R} = 0 \times 0.0280 = 0\text{ mm}$

Step 4: Evaluate the $R$ Chart

The technician examines all 20 subgroup ranges against $UCL_R = 0.0592\text{ mm}$:

• Subgroup 8 recorded a range of $R_8 = 0.0650\text{ mm}$.
• Since $0.0650 > 0.0592$, Subgroup 8 is out of control on the $R$ chart!

Step 5: Assignable Cause Investigation

The technician reviews the CNC maintenance log for Subgroup 8 (produced at 10:15 AM). The log reveals that the lathe's secondary coolant nozzle became clogged with aluminum swarf, causing thermal shocking and severe chipping on the carbide insert. The operator noticed the smoke, cleared the swarf, and indexed to a fresh carbide insert.

• Conclusion: A verified, documented assignable cause exists and has been physically eliminated.
• Action: Discard Subgroup 8 from the baseline study and recalculate revised limits with $k = 19$ subgroups.

Step 6: Recalculate Revised Centerlines ($k = 19$)

Subtract Subgroup 8 values from the original sums: $\sum \bar{X}_{\text{rev}} = 700.400 - 35.034 = 665.366\text{ mm}$ $\bar{\bar{X}}_{\text{rev}} = \frac{665.366}{19} = 35.01926 \approx 35.0193\text{ mm}$

$\sum R_{\text{rev}} = 0.560 - 0.065 = 0.495\text{ mm}$ $\bar{R}_{\text{rev}} = \frac{0.495}{19} = 0.026053 \approx 0.02605\text{ mm}$

Step 7: Recalculate Revised $R$ Chart Limits

$UCL_{R,\text{rev}} = D_4 \bar{R}_{\text{rev}} = 2.114 \times 0.026053 = 0.05508 \approx 0.0551\text{ mm}$ $CL_{R,\text{rev}} = 0.02605\text{ mm}$ $LCL_{R,\text{rev}} = 0\text{ mm}$

Checking the remaining 19 subgroup ranges against the revised $UCL_{R,\text{rev}} = 0.0551\text{ mm}$: the highest remaining range is $R_3 = 0.048\text{ mm}$, which is well below $UCL_R$. The $R$ chart is now in statistical control!

Step 8: Compute Final Control Limits for the $\bar{X}$ Chart

Now that dispersion is stable, compute limits for process location: $\text{3-Sigma Margin} = A_2 \bar{R}_{\text{rev}} = 0.577 \times 0.026053 = 0.01503\text{ mm}$ $UCL_{\bar{X}} = \bar{\bar{X}}_{\text{rev}} + A_2 \bar{R}_{\text{rev}} = 35.0193 + 0.01503 = 35.0343\text{ mm}$ $CL_{\bar{X}} = 35.0193\text{ mm}$ $LCL_{\bar{X}} = \bar{\bar{X}}_{\text{rev}} - A_2 \bar{R}_{\text{rev}} = 35.0193 - 0.01503 = 35.0043\text{ mm}$

Checking all 19 subgroup averages against these limits confirms that all points fall safely between $35.0043\text{ mm}$ and $35.0343\text{ mm}$. The process is in a verified state of statistical control.

Step 9: Estimate Population Standard Deviation ($\hat{\sigma}$)

$\hat{\sigma} = \frac{\bar{R}_{\text{rev}}}{d_2} = \frac{0.026053}{2.326} = 0.01120\text{ mm}$

This estimated $\hat{\sigma} = 0.01120\text{ mm}$ will now serve as the official denominator for calculating process capability ($C_p, C_{pk}$).

---

6. Common Technician Exam Pitfalls

Be prepared for these classic traps on the ASQ CQT exam:

1. Exam Trap: Calculating $\bar{X}$ Limits Before Checking $R$: Questions often present a dataset with an out-of-control range point and ask for the $\bar{X}$ limits. If an assignable cause is present, you must discard the subgroup and recalculate $\bar{R}$ before computing $\bar{X}$ limits.
2. Exam Trap: Confusing $d_2$ with $\sqrt{n}$: When estimating process standard deviation, remember: $\hat{\sigma} = \bar{R} / d_2$. Never divide by $\sqrt{n}$! Dividing by $\sqrt{n}$ gives the standard error of the mean, not the standard deviation of individual pieces.
3. Exam Trap: Forgetting $D_3 = 0$: Exam questions love to test whether candidates remember that $LCL_R$ is 0 for $n = 2, 3, 4, 5, 6$. If a calculation gives a negative number, $LCL_R$ is automatically set to zero.
4. Exam Trap: Confusing $A_2$ with $A_3$: Factor $A_2$ is used exclusively with the Average Range ($\bar{R}$). Factor $A_3$ is used exclusively with the Average Sample Standard Deviation ($\bar{s}$).