Control Charts for Process Monitoring

1. Constructing Control Charts for Variable Data

Let's begin with variable (or continuous) data, which involves measurements like length, weight, or time. As you'll recall, the most common charts here are the X-bar & R chart and the I-MR chart.

The X-bar & R Chart

This chart is a workhorse in manufacturing, used when you can collect data in small, rational subgroups (typically 2 to 9 observations). It simultaneously monitors the process average ($\bar{X}$) and its variability (R).

The construction involves these key steps:

1. Collect Data: Gather your data in subgroups over time. Let's say we have k subgroups, each of size n.
2. Calculate Subgroup Statistics: For each subgroup, calculate the average ($\bar{X}$) and the range (R = Max value - Min value).
3. Calculate Centerlines:
   • The centerline for the X-bar chart is the "grand average" ($\overline{\stackrel{ˉ}{X}}$), which is the average of all the subgroup averages.
   • The centerline for the R chart is the average range ($\bar{R}$).
4. Calculate Control Limits: The limits are calculated using the average range and standard constants (A₂, D₃, D₄) found in tables in any SPC reference. These constants are derived from the statistical properties of the range and are dependent on the subgroup size, n.
   • X-bar Chart Limits: $UC{L}_{\bar{X}}=\overline{\stackrel{ˉ}{X}}+A_2\bar{R}$ and $LC{L}_{\bar{X}}=\overline{\stackrel{ˉ}{X}}-A_2\bar{R}$
   • R Chart Limits: $UC{L}_R=D_4\bar{R}$ and $LC{L}_R=D_3\bar{R}$
5. Plot the Charts: Plot the subgroup averages on the X-bar chart and the subgroup ranges on the R chart, along with their respective centerlines and control limits.

To see this process in detail, including the calculation steps and a software demonstration, the following video is an excellent guide.

A crucial takeaway from the video is the standard practice of analyzing the R chart first. The control limits on the X-bar chart are derived from the process variability ($\bar{R}$). If the R chart shows that variability is unstable and out of control, the control limits on the X-bar chart are meaningless. You must first achieve stability in process variation before you can properly assess the process average.

For a static, step-by-step reference guide that you can save, the resource below provides a thorough breakdown of these calculations.

The Individuals and Moving Range (I-MR) Chart

When you can only collect one data point per time period (subgroup size n=1), the I-MR chart is your tool. This is common in destructive testing, chemical processes, or when automated measurements are taken one at a time.

The construction is similar in principle:

1. Calculate the Moving Range (MR): Since you can't calculate a range within a subgroup of one, you calculate the range between consecutive points: $M{R}_i=|X_i-X_{i-1}|$.
2. Calculate Centerlines: The centerline for the Individuals (I) chart is the average of all individual values ($\bar{X}$). The centerline for the MR chart is the average of the moving ranges ($\overline{MR}$).
3. Calculate Control Limits: Again, standard formulas are used. A common method is:
   • I Chart Limits: $UC{L}_I=\bar{X}+2.66\cdot \overline{MR}$ and $LC{L}_I=\bar{X}-2.66\cdot \overline{MR}$
   • MR Chart Limits: $UC{L}_{MR}=3.268\cdot \overline{MR}$ and $LC{L}_{MR}=0$

2. Constructing Control Charts for Attribute Data

Now, let's turn to attribute data, where we are counting things rather than measuring them. As we covered in the last lesson, the chart choice depends on whether we're counting defectives (bad units) or defects (flaws), and whether the sample size is constant or variable.

The following video provides an excellent and comprehensive guide to constructing all four main attribute charts.

Let's briefly summarize the centerline and control limit concepts for each chart you saw in the video:

• np-Chart (Number of Defectives, Constant Sample Size):

  • Centerline ($\overline{np}$): Average number of defectives per subgroup.
  • Control Limits: Based on the binomial distribution.

• p-Chart (Proportion of Defectives, Variable Sample Size):

  • Centerline ($\bar{p}$): Overall proportion of defectives across all samples.
  • Control Limits: Also based on the binomial distribution, but they will vary for each subgroup if the sample size n changes. Larger samples result in tighter limits.

• c-Chart (Number of Defects, Constant Opportunity):

  • Centerline ($\bar{c}$): Average number of defects per inspection unit.
  • Control Limits: Based on the Poisson distribution.

• u-Chart (Rate of Defects, Variable Opportunity):

  • Centerline ($\bar{u}$): Overall rate of defects (e.g., defects per unit, defects per square meter).
  • Control Limits: Based on the Poisson distribution. Like the p-chart, the limits can vary if the area of opportunity changes between samples.

3. Interpreting Control Charts: The Voice of the Process

Constructing a chart is only half the battle. The real value comes from interpreting it. A control chart helps us distinguish between two types of variation:

• Common Cause Variation: This is the natural, inherent "noise" or randomness in a process. It is predictable and characterized by data points fluctuating randomly within the control limits.
• Special Cause Variation: This is variation that comes from an external, assignable cause. It represents a fundamental change in the process. It is unpredictable and is what we aim to detect and eliminate.

A process is said to be in statistical control when only common cause variation is present. Our goal is to use the control chart to detect signals of special cause variation.

Rules for Detecting Special Causes

While there are many rules, a few core ones are used universally to detect non-random patterns. An out-of-control signal is a trigger to investigate the root cause.

The most critical and common rules illustrated above are:

1. Rule 1: One point outside the control limits. This is the clearest signal of a significant event or shift.
2. Rule 2 (Shift): A run of 9 (or 7, or 8) consecutive points on the same side of the centerline. This suggests the process average has shifted.
3. Rule 3 (Trend): 6 consecutive points steadily increasing or decreasing. This often indicates gradual change, like tool wear or operator fatigue.
4. Rule 4 (Systematic Variation): 14 consecutive points alternating up and down. This can point to a systematic issue, such as alternating between two different suppliers or machines.

When you detect a signal, the correct response is not to adjust the process immediately, but to investigate. This is where your skills with root cause analysis tools like the 5 Whys and Fishbone diagrams become essential. The chart tells you that something happened at a specific time; your job is to find out what and why.

Conclusion

In this lesson, we transitioned from selecting control charts to actively building and interpreting them. We've seen that constructing a chart involves a clear, statistical process of calculating centerlines and control limits, a task readily handled by software but crucial to understand conceptually.

Most importantly, we've learned to listen to the "voice of the process" by interpreting these charts. By distinguishing between common and special cause variation, we can avoid tampering with a stable process and know exactly when to launch an investigation to find and eliminate root causes of instability.

Key Takeaways:

• Control chart construction involves calculating centerlines (the process average) and control limits (the bounds of expected variation).
• Always analyze the variability chart (R, S, or MR) before the location chart (X-bar or I).
• A process in statistical control exhibits only predictable, common cause variation.
• Special cause variation is signaled by non-random patterns, such as points outside the limits, long runs, or trends.
• An out-of-control signal is a call to investigate, not to make knee-jerk adjustments.

Preview of the Next Lesson:

We've covered the basics of chart interpretation and the most common rules. In our next lesson, we will take a deeper dive into this topic, focusing on the learning outcome: Identify common out-of-control signals and patterns on a control chart using standard industry rules (e.g., Nelson Rules). We will explore more nuanced patterns and discuss what these specific patterns often imply in a real-world manufacturing environment, further sharpening your diagnostic skills.