Process Behavior Charts: Separating Behavior from Specification
Right now there is a long running debate on whether specification limits should be shown on Process Behavior Charts (control charts).
Those following Walter A. Shewhart and Donald J. Wheeler are clear. If the purpose is to understand the process, specification limits do not belong on the chart.
A Process Behavior Chart in this context answers one question. What is the process doing over time. It separates routine variation from signals that require action.
Specifications answer a different question. Is the result acceptable - Yes or No our specifications are often imposed on a process reflecting quality expected by the patient.
Control limits serve a different role. They are derived from the process and act as action limits. They indicate when there is an economic benefit to investigate or intervene. When the process remains within these limits, the most effective economical action is to do nothing and our biggest failure is not understanding this.
Problems begin when both specifications and control limits are shown without a defined purpose. The focus often shifts to only pass or fail. Decisions are driven by proximity to specification rather than signals in the data.
The question is not whether specifications should be shown.
The question is what you are trying to learn.
If you want to understand the process, specifications distract
If you want to communicate risk or conformance, specifications add context
The objective defines whether process behavior chart should have specification limits present.
What Process Behavior Charts are used for?
Control charts are used to:
Monitor and control process behaviour
Understand variation and enable prediction
Support investigation and root cause analysis
Verify improvement
Communication - Report state of control to work colleagues, customers, suppliers, regulators with OPV/CPV and APR/PQR
Compare variation against specification
Define rational subgrouping and sampling
The objective determines the Process Behavior Charts design
For operators, charts must be simple and actionable. Use:
Target
Upper and lower control limits
Mechanism to capture Actions taken
These reflect real process variation and define when action is justified.
Do not include specifications. They do not help the operator run the process and certainly do not release the control limits with Specifications.
For reporting or review, include both when needed:
Control limits to show behavior
Specification limits to show acceptability, but make sure the audience understands the difference otherwise they will default to the widest limits shown.
Table 1: Historical values of Parameter X (mg)
Table 1, data often sits in rows and columns. The only check applied is whether values fall within specification.
In this example, all batches pass.
From a compliance perspective, this is often presented as individuals argue that the specifications are consistently met.
It is argued that a state of control is demonstrated.
Under ICH Q10 Pharmaceutical Quality System, a state of control requires that the process consistently provides assurance of performance and product quality
That should mean:
Stable behavior
Predictable output
Evidence over time
A table cannot show this. It hides sequence, variation, and change.
All results can pass while the process remains unknown.
Figure 1: Specifications only
Figure 1 now plots the data and a pattern appears.
A shift is visible.
Without control limits, the change can be seen but not evaluated. There is no rule for action, only interpretation.
With more subtle step changes or less obvious non-random patterns, rule-based decision making becomes necessary.
It is only when we move to Figure 2 and apply a rule-based method that judgement is replaced.
Figure 2. Both specification and control limits
With Figure 2 the addition of control limits and the picture changes.
The process shifts. Clearly.
This is the type of chart used in reporting. It often leads to a familiar conclusion:
Everything is in specification
Therefore the process is in control
Despite this, it is often argued that the process remains acceptable because all points lie within specification and the control limits sit within those limits.
This interpretation misses the point.
Reality is:
The chart shows two different process states.
The mean blends them. It represents neither - it does not represent anything
Cpk around 1.08 suggests acceptable capability. That calculation assumes a single stable process. That assumption is false.
You are averaging two behaviors and reporting one number.
The message is misleading.
The process and its variation meets specification.
A StepChange is present, yet everything remains within specification.
No action is taken
Figure 3. Subgrouping by process state
Split the data into Pre and Post.
Now the process is clear.
Two states. Two behaviors. Two different risks.
Conclusions
1. Control limits and Cpk across the full dataset are misleading. They describe an average that does not exist in practice.
2. In the Pre Era, control limits fall below the LSL. The process produces values below specification even though none were observed. Cpk of 0.66 confirms the risk.
3. The process is off target. Meeting specification hides the loss associated with this position.
4. The step change is real. It must be explained. This is process knowledge, not an inconvenience.
5. In the Post Era, the process is centred and stable. Control limits sit within specification. Cpk improves to around 1.39.
6. Specification limits support communication. Control limits support control. Use them accordingly.
Perspective
At Verto we assess a process by the position of the control limits relative to specification.
Cpk and Ppk can mislead when the process is not stable or when used without the chart. The timeline shows what the process is doing. We will use these indices but only when they reflect the process relationship with the specifications.
When designing process behavior charts for manufacturing control, we avoid including specifications. They add complexity without supporting decision making.
Rational subgrouping is critical. If you group data correctly, the truth appears.
Specifications on histograms support analysis but are insensitive to time and that matters.
Summary
– when to include specifications on process behaviour charts
Know the objective
Know your audience
Decide whether specifications support or distract from the objective and the audience
Apply consistent honest thinking