Production scheduling: deciding what runs next on which machine

What the scheduler actually settles before the shift starts

Three decisions get made every day: the sequence at each work centre, which jobs are released to the floor at all, and what to do about the work that cannot start because something is missing. The third is where schedulers earn their keep. A job short of one component, one operator or one fixture will sit on the floor consuming space and attention, so pulling it back and promoting a runnable job is usually better than leaving it queued in hope. Good schedulers keep a short list of substitutable work for exactly this, agreed in advance with the planner so the substitution does not create a new shortage next week.

Dispatch rules and what each one quietly sacrifices

Earliest due date protects delivery but ignores setup, so it can shred a day into changeovers. Shortest processing time flatters throughput counts and starves large jobs. Critical ratio balances remaining work against remaining time and behaves reasonably in mixed shops, but it needs accurate remaining-work data to mean anything. Where setups depend on sequence — colour, grade, tooling family, allergen — the sequence itself carries most of the cost, and a rule that ignores it will lose more capacity than it saves. Most plants use a family sequence as the outer loop and a due-date rule inside each family, which is a compromise the scheduler should be able to explain.

The data a schedule needs before anyone will believe it

Schedules fail on stale masters more often than on bad logic. Routings that list a machine retired last year, run rates copied from the equipment brochure rather than measured on the floor, setup times recorded before the tooling was improved, and shift calendars that ignore planned maintenance all produce a schedule that looks credible and is not. Before adopting any scheduling discipline, take a sample of routings for the highest-volume items and check the run rates against actual recorded output. Where the numbers disagree, fix the master data first. Supervisors detect a fictional schedule quickly, and once they distrust it they revert to their own list.

Rescheduling versus absorbing the disruption

When a machine goes down or a batch is quarantined, the scheduler chooses between reissuing the schedule and letting the floor absorb the hit. Reissuing everything on every disturbance produces nervousness: supervisors see a new sequence each hour, stop believing any of them, and revert to working from paper. Absorbing everything produces a schedule that silently detaches from reality. A workable rule is to reschedule when the disruption pushes a customer commitment or unbalances the constraint, and otherwise to let the sequence stand and record the variance. Whichever rule you adopt, publish it, so the floor knows whether the list in front of them is current.

Handing the sequence to the floor without losing it

The schedule only exists once it reaches the people who execute it, in a form they can act on: a dispatch list per work centre showing job, quantity, setup required and the next job so the setter can prepare. Where that handover is verbal or buried in a screen nobody opens, the sequence decays within a shift. The reciprocal duty is feedback: the floor must report starts, completions and stoppages promptly enough for the next sequence to be built on facts. A schedule that receives no confirmation is a forecast. Supervisors and schedulers should agree what counts as a completed operation, because ambiguity there corrupts every downstream calculation.

Frequently asked questions

Should the supervisor be allowed to change the sequence?

Yes, within stated limits, because the supervisor sees conditions the scheduler cannot. Define what may be changed locally — swapping two jobs within the same setup family, promoting a job when material arrives late — and what may not, typically anything that breaks a customer commitment or moves work away from the constraint. Require the change to be recorded with a reason so the pattern of local overrides becomes visible and can be designed out.

How often should the schedule be regenerated?

Match the regeneration rhythm to how fast conditions actually move. A plant with long runs and stable equipment can work from a schedule refreshed once a shift or once a day. A job shop with frequent setups and variable material arrival needs a shorter loop, but even there the published horizon should stay short and firm, with anything beyond it treated as indicative. Frequent regeneration without a firm near-term window produces churn rather than accuracy.

Why does our schedule always look achievable and never get achieved?

Usually because the model of the plant is more capable than the plant. Check three things: whether run rates in the routing reflect measured output including minor stoppages, whether setup times include the fetch, fit, first-off and approval steps rather than only the machine work, and whether the calendar deducts breaks, meetings, planned maintenance and known absence. Each of these individually looks like a rounding matter, and together they turn a full shift into a fictional one.