Every morning eight orders wait at the CNC machining center. Someone decides which one runs first. If nobody has written down how, the decision depends on who asks loudest, which setup is convenient and which order the operator likes.
A priority rule replaces that with a decision everyone can predict.
Three rules, one queue
The eight orders, with processing time and due time in hours from now:
J1 12 h due 40, J2 6 h due 18, J3 20 h due 60, J4 4 h due 14, J5 10 h due 30, J6 8 h due 50, J7 14 h due 36, J8 5 h due 24.
Sequenced three ways:
| Rule | Avg flow time | Avg lateness | Orders late | Max lateness |
|---|---|---|---|---|
| First in, first out | 46.9 h | 19.1 h | 5 | 55 h |
| Shortest processing time | 33.4 h | 6.3 h | 4 | 23 h |
| Earliest due date | 35.3 h | 5.3 h | 4 | 19 h |
The total work is the same 79 hours in every case. Only the order changes, and with it the average lateness falls from 19 to 5 hours.
What each rule is good at
- First in, first out is fair, simple and predictable. It ignores due dates, which makes it a poor choice at the bottleneck. It works well at non-bottlenecks with short queues.
- Shortest processing time minimizes average flow time. Long orders can starve, so use it only with an age limit.
- Earliest due date minimizes lateness, both the number of late orders and the largest delay. It is the natural choice at the bottleneck and before shipping.
- Critical ratio, time left divided by work left, works well in job shops with good feedback data.
The rule for the example
Earliest due date at the bottleneck, first in first out everywhere else. The rule goes on the board next to the dispatch list. Deviations are allowed, for example to combine two orders with the same setup, but they are recorded with the reason.
Rules need clean dates
Earliest due date is only as good as the due dates. If every order is marked urgent, or dates were promised without a capacity check, the rule sorts noise. Clean dates come from release and promising, not from the sequencing rule.
A game worth playing
In a workshop, give the team eight real orders from their own queue and let them sequence them three times. Calculating flow time and lateness takes twenty minutes. The discussion that follows usually ends the debate about priorities faster than any presentation: the numbers show what each habit costs.
A rule everybody follows beats a clever rule nobody knows. Choose one per work center, write it down, and review it when the numbers change.