Control charts are a widely used tool of statistical process control (SPC). They allow monitoring of and identification of "out of control" states for values in repetitive processes that follow a known mathematical distribution (e.g. are normally distributed).
From a mathematical perspective, identifying an "out of control" process means identifying that the values occurring (e.g. the data being measured on the shopfloor) do not fit the expectations based on the statistical distribution previously collected data.
Some common use cases for control charts are monitoring the following:
A measured dimension of individual parts or the mean value of measuring a sample of identical parts
The duration of a repeated process
The number of defective items in each batch
Control Chart Rules
In the control chart layer one or multiple rules are evaluated for the data. There are different rule sets used around industry (e.g. Nelson and Western Electric). Tulip currently supports the two rules that are statistically most effective in detecting out of control states without leading to a high number of false detection ("false positives"). These are:
Any single data point above the Upper Control Limit (UCL) or below the Lower Control Limit (LCL).
Nelson Rule 1
Western Electric Rule 1
Single value is significantly out of control
9 or more consecutive data points either entirely above or entirely below the center line.
Nelson Rule 2
Western Electric Rule 4
Prolonged offset in on direction hinting at potential shift of process mean
As described in the article on What is a Control Chart, layers are available for specific chart configurations. In the case of the Control Chart Layer these are the following:
Date Source: Tulip Table
Template: One Operation
View: Line Chart
y-Axis: Numeric value (Number, Integer, Interval)
Once a chart configuration is fulfilling these requirements, the Control Chart Layer will be available in the Chart Layers panel. This can be opened by clicking on the Chart Layers button in the top right corner of the analytics editor.
To enable the control chart layer, click on the toggle next to "Control Chart" in the Layers side panel. This will open up the layer settings in which the following can be set:
Upper control limit (UCL) and lower control limit (LCL) - mandatory
Custom center line value - optional
As default, the center line will be calculated as the average of UCL and LCL [(UCL + LCL)/2]
Active rules - at least one
Tulip control charts are implemented as "manual limit" control charts at the moment. This means that the control limits are input manually by the user upon initial chart configuration. (Note: This is a one-time action for a chart, as control limits should not be changed while using the chart to monitor a process). In combination with the analytics editor settings, this allows to implement the following standard control chart types:
There are different ways to identify the control limits for each of the chart types for entering them in the configuration:
If you used another tool for SPC before switching to Tulip and your process has not changed, you can enter the same boundaries as before
If you have already collected data within Tulip you can calculate the statistically correct limits using the equations relevant for the desired chart type based on that data.
If you want to use the control chart more freely you can set any limits you like and use the rules with those limits of your choice to detect outliers and shifts
A custom center line value can optionally be configured to accommodate for skewed distributions.
Once the control chart layer is successfully configured, you can save the analysis and use it the same way you use any other analysis. You can:
Add it to a dashboard
Embed it into a Tulip App
Make it available via a sharing link
If data in the chart triggers any of the rules selected in the configuration, the relevant data points will be highlighted in the graph as depicted below.