# Control Charts: Which One Should I Use?

## You can perfectly model a process’s statistical personality—as long as you choose the right control chart.

Control charts, ushered in by Walter Shewhart in 1928, continue to provide real-time benefits in today’s modern factories. When they were first introduced, there were seven basic types of control charts, divided into two categories: variable and attribute.

**Variable Data Charts**

- IX-MR (individual X and moving range)
- Xbar-R (averages and ranges)
- Xbar-s (averages and sample standard deviation)

**Attribute Data Charts**

- p (proportion defective for subgroup sizes that vary)
- np (number of defectives in a fixed subgroup size)
- u (defects per unit for subgroup sizes that vary)
- c (defect counts in a fixed subgroup size)

Those who make control charts their business know that there have been significant contributions to chart offerings since the original seven were introduced. Today, you can choose from hundreds of control charts. You can perfectly model a process’s statistical personality as long as you choose the right control chart.

**Do You Know Your Control Charts?**

Picking the right chart for your purpose starts with knowing the factors that define the chart type. But before we get into the details of chart type combinations, let’s define, at a high level, what control charts are and what they are not.

**Control charts ARE:**

**Real-time**graphical process feedback tools- Designed to tell the operator to do
**something**or do**nothing** - Time-ordered representations of process
**personalities**or**behaviors** - Designed to separate
**signals**from**noise** - Able to detect changes in the
**process mean**and**standard deviation** - Used to determine whether a process is
**stable**(predictable) or**out of control**(not predictable)

**Control charts are NOT:**

- A substitute for capability analysis
- Useful in receiving inspection (time order is lost)
- Efficient comparative analysis tools
- To be confused with Run charts or PRE-control charts (Run charts are time-ordered but not statistically based limits; PRE-control charts compare plot points to specification limits)

Control charts utilize *control limits* to help identify when a process has significantly changed or to isolate an unusual event. Because control limits are derived from data, you can’t know what the limits are until after you’ve collected a representative series of data.

If used for the wrong reasons, control limits can cause confusion and counterproductive actions by those asked to use charts to monitor and improve their processes.

**Control limits ARE:**

- Limits based on
**expected**plot point variation - Calculated from
**mean**and**standard deviation**(derived after representative plot points have been gathered) - Typically expressed as +/- 3 standard deviations of the plot points (not the standard deviation of the underlying distribution)

Control limits should be updated when a process improvement has been verified.

**Control limits are NOT:**

- Based on a percentage of the specification limits
- 75% of the specification limits
- Production limits
- Anything to do with specification limits or desired limits

**Choose the Right Control Chart to Answer the Right Question**

Selecting the right control chart starts with knowing something about what you want the chart to say about the process. What questions do you want the chart answer?

Another way to look at this is to ask, “Why am I collecting data on this part?”

The answers to these questions will provide the information you need to determine the sampling strategy, sample size, and any special needs that would require implementing special processing options that extend the function of traditional charts. Ultimately, your choice will be influenced by multiple considerations and data type. However, here we’ll address sample size, target charting, and multiple process streams with variables data.

**Sample Size**

The sample size is the number of measurement values for a given test feature that you will gather to represent a single “snapshot of time.” For example, if weights are taken from three consecutive filled bottles every 30 minutes, the sample size is three and the sampling interval is 30 minutes.

The sample size does not represent the number of plot points on a chart. The common symbol used for sample size is *n.* There are three sample size considerations:

- sample sizes of 1 (
*n*=1) - sample sizes between 2 and 9 (2 ≤
*n*≤ 9) - sample sizes of 10 or larger (
*n*≥ 10)

Most variables-charting techniques are rooted in one of the three core variables control charts.

- When sample sizes are 1, the
**Individual X and Moving Range (IX-MR)**chart is used. - For sample sizes of 2 through 9, the
**Xbar-Range (Xbar-R)**chart is used. - For sample sizes of 10 or greater, the
**Xbar-Sigma (Xbar-**chart is used.*s*)

**Number of Process Streams**

Using InfinityQS terminology, a *process stream* is characterized by Part, Process, and Test. A single process stream generally represents a series of plot points from one part, one process, and one test.

For example, 50ml bottle weights from fill nozzle A would be one process stream; 50ml bottle weights from fill nozzle B would be another process stream. Because fill nozzle A could have a unique statistical personality—different from fill nozzle B—you wouldn’t want to combine (confound) the data from both nozzles in a single subgroup. The better sampling strategy would be to treat the data from each fill nozzle as separate streams of data.

When challenged with a process that generates multiple process streams, you have the option of using one control chart for each process stream or using a specialized chart that allows all process streams to coexist on the same chart. Charts for multiple process streams are called *Group charts.*

**Same Feature but Different Targets**

Processes are commonly used to produce different products. In many cases a product changeover means changing process set points in order to produce the different product. When you want to monitor a process’s ability to hold a set point, regardless of the product, the data can be combined across multiple set points by simply subtracting the set point from the actual output result.

Continuing with the fill nozzle example, when the line changes from a 50ml bottle to a 100ml bottle, the same nozzles are used but are programmed to fill to 100ml. A fill of 100.3 would be represented on the chart as 0.3. When you take out the target values, a single chart can be used to monitor—in time order—a process’s ability to hold a set point regardless of the specification of the product being produced at the time. These types of charts are called *Target charts**.* As long as the combined products share similar variation, multiple parts can be represented on the same chart. Target charts are especially useful in short-production-run environments.

Control charts are designed for specific purposes; using a control chart that isn’t sensitive enough for your process can produce false positives. But today’s manufacturing environments produce an increasing amount of data, so selecting the right control chart for a given situation can be overwhelming.

When you take the time to learn about the control charts available to you, you’ll have a rich toolset that can help you discover transformational insights about your products and processes.

*Visit the InfinityQS **Definitive Guide to SPC Charts** to learn more about the most popular SPC control charts and how to use them. And to learn more about how to choose the right chart for your needs, download our free white paper **A Practical Guide to Selecting the Right Control Chart**.*