Four cavities, one chart
An injection moulding tool has four cavities. Every hour, a quality technician measures the wall thickness of four parts and enters the values. The control chart shows a wide spread, and a few values are close to the lower limit. Is the process unstable, or is one cavity different from the others?
With only the values, nobody can tell. If each value also carries its cavity number, the question takes a minute to answer: plot the cavities separately and the one that runs thin stands out.
The cavity number in this example is an annotation.
What an annotation is
An annotation is arbitrary information bound to a specific item. Typical examples:
- an operator name bound to a measured value
- a cavity, spindle or fixture number bound to a measured value
- a serial number, batch or order number bound to a part
- a drawing number and revision bound to a product
- a comment such as “new tool fitted” bound to a point in time
Other words with the same meaning are metadata (data about data) and trace field (a field that makes it possible to trace something). The terms are used interchangeably in practice.
What annotations make possible
Separating sources of variation. If values are annotated with machine, cavity, operator or material batch, the spread can be broken down by each. A wide spread may turn out to be two narrow ones side by side. This is also why statistical process control relies on rational subgroups: you cannot form them without knowing where each value came from.
Explaining signals. The AIAG SPC manual recommends recording process events, such as tool changes, adjustments and new material, alongside the chart. A signal on a chart next to the note “new material batch” points the investigation in the right direction.
Traceability. ISO 9001:2015 (clause 8.5.2) requires organisations to identify outputs and, where traceability is a requirement, to control their unique identification. If a customer asks about part 2417 from order 50832, annotations let you find exactly which values, gauge and operator belong to it.
Worked example
Illustrative values from one shift, wall thickness in mm, nominal 2.00 with LSL 1.90:
- cavity 1: 1.99, 2.01, 2.00, 1.98
- cavity 2: 2.00, 2.02, 1.99, 2.01
- cavity 3: 1.93, 1.94, 1.92, 1.94
- cavity 4: 2.01, 1.99, 2.00, 2.02
Taken together, the sixteen values range from 1.92 to 2.02, and the process looks unstable. Split by cavity, three cavities sit around 2.00 with a range of 0.03, and cavity 3 sits around 1.93. The problem is one cavity, not the process, and the action is on the tool, not the machine settings.
Good practice
- Record at the source. An annotation added later from memory is unreliable. Capture it when the value is measured, ideally automatically from the machine, the order system or the gauge.
- Use fixed choices. “Cav 3”, “cavity 3” and “K3” are three different values to a computer. Use lists, not free text, for anything you want to filter on.
- Decide what matters. Annotate the factors that can explain variation or that the customer may ask about. Too many fields that nobody fills in are worse than a few that are always filled in.
- Keep the link. If values are copied between systems, the annotations must travel with them. This is one of the main arguments for digitalisation: a value that is retyped often loses its context.
For automated measurements, such as those from a coordinate measuring machine, the article, order and serial number are often available in the measuring program or the order system and can be attached without manual entry.
Limitations
Annotations show associations, not causes. A cavity that runs thin may be the cause, or it may be measured on a different fixture. Use the annotation to point the investigation, then confirm the cause.
In short
A measured value tells you how large something was. Annotations tell you what it was, where it came from and what was happening at the time. That context is what turns data into something you can act on and answer questions with.
