In a DMAIC project, automation belongs in the Improve phase, after Measure and Analyze have shown the root cause and proved that a manual fix won’t hold. It then needs a Control phase of its own: the automated check or process has to be verified, monitored and owned, or the gains fade like any other countermeasure.
This article is part of our industrial automation guide. It is written for continuous improvement, lean and six sigma practitioners deciding whether an automation project fits their improvement work.
In short
- Automate a proven root cause, not a symptom.
- Use your Measure data to size the automation and to set acceptance criteria.
- A pilot or proof of concept is the Improve-phase experiment.
- Control means a verification routine, a reaction plan and an owner.
Where automation fits in DMAIC
| Phase | Question | What automation needs from it |
|---|---|---|
| Define | What problem, for whom, worth how much? | A problem statement with a cost, such as the cost of poor quality |
| Measure | How big is it, and can we trust the data? | Baseline defect, yield or cycle data, and a capable measurement system |
| Analyze | What causes it? | A verified root cause that a person or fixture can’t reliably control |
| Improve | What fixes it? | Automation options, a proof of concept with real parts, and a business case |
| Control | How do we keep it fixed? | Verification checks, alarms, a reaction plan and an owner |
Signs automation is the right countermeasure
- The root cause is human variation in a repetitive task: inspection, counting, assembly checks.
- Training, standard work and visual management have been tried and the defect returns.
- The check must happen on every part, faster than a person can do it reliably.
- Data has to be captured for every part or batch.
- The fix is needed on several lines or shifts, so a repeatable system beats repeated training.
If the root cause is a process setting, a supplier issue or a design margin, automation is usually the wrong first step. Fix the cause, then decide whether to automate the check.
Running the Improve phase as an experiment
- Write acceptance criteria from your Measure data: the defect rate, cycle time and false-reject rate the system must achieve.
- Test real parts first. A feasibility test or proof of concept with good, bad and borderline samples answers “can this be automated?” before money is committed. See engineering feasibility studies.
- Pilot on one line and compare against the baseline with the same metric.
- Accept against the criteria in a factory and site acceptance test.
The Control phase for an automated fix
- Verification: a challenge part or reference check each shift.
- Monitoring: charts of the system’s own results, such as reject rate or measured values, so drift is visible.
- Reaction plan: what operators do when the system rejects, alarms or fails its verification.
- Ownership: who maintains it, who can change settings, and how changes are approved.
- Control plan update: the automated check listed with its frequency and reaction plan.
Automated measurements are also a ready source for statistical process control, which turns the system into a permanent monitoring tool, not just a gate.
Bringing an automation idea into a CI project?
Sciotex runs feasibility tests on your parts at no charge and designs automation that fits your improvement plan, from proof of concept to supported production system. No subscription or support contract required.
Frequently asked questions
Automation is an Improve-phase countermeasure, chosen after Measure and Analyze have identified a root cause that manual methods can’t reliably control, and it needs its own Control-phase plan.
Use the baseline from the Measure phase to quantify the cost of the problem, then compare the cost removed by the automated fix with the cost of the system.
Yes. Testing real good, bad and borderline parts is the Improve-phase experiment that shows whether an automated fix can meet the target before full investment.
A verification routine such as a challenge part each shift, monitoring of the system’s own results, a reaction plan, an owner and an updated control plan.
Yes, especially for repetitive checks, counting and data capture, as long as the root cause is understood and the automated fix is verified and owned.