Production scheduling on real run rates

You build a schedule that matches what the machines actually do by sizing every job from the rate each machine has measured on that operation, not from its ideal cycle time or the ERP routing standard. Divide the ideal cycle time by the availability, performance and quality you measured, plan with the median of recent runs, load each machine below its planned hours, and correct the rates every week from what was logged.

This guide is for plant managers and planners whose schedule says Tuesday while the floor delivers Thursday. It works one job through by hand, then shows which rate to plan with and how to keep the rates current.

Why a schedule drifts from the floor#

A schedule drifts because it plans each job at a speed the machine never holds for a whole run. Three different numbers carry the name cycle time, and only one of them includes the stops, slow cycles and scrap that every real run contains.

The ideal cycle time is the best cycle the operation sustains in good conditions, the figure you enter on the operation, as setting the ideal cycle time explains. The glossary entry for cycle time separates it from the actual cycle time measured on the machine. For example, a bracket operation on a Haas VF-2 with a 90-second ideal cycle time can still average 135 seconds per good part once stops, short pauses and scrap are counted.

The Lean Enterprise Institute's effective machine cycle time adds load and unload time to the machine cycle, plus the changeover time divided by the pieces between changeovers. A 20-second cycle with 30 seconds of load and unload and a 30-second changeover every 30 pieces is 51 seconds.^[3] Effective cycle time extends the same idea to every stop, slow cycle and scrapped part.

Routing standards drift for a different reason. They belong to the business planning layer: ISA-95 describes Level 4 as all the activities of running a business, ERP included, and places manufacturing execution systems at Level 3.^[2] A standard written at Level 4 for a quote is rarely revisited when the machine, the fixture or the operator changes.

Turn a measured rate into a job duration#

A job's duration is its setup time plus the good quantity multiplied by the effective cycle time. Measure the three OEE factors for the operation on the machine that will run it, as the OEE factor calculations describe, then divide them out of the ideal cycle time.

The quality factor turns the order quantity into the quantity you must start. The availability and performance factors stretch the run from ideal to real.

The shortcut gives the same answer within rounding. Effective cycle time is 90 ÷ (0.80 × 0.85 × 0.98) = 90 ÷ 0.6664 = 135.05 s, and 1,200 × 135.05 s = 45.02 h of run time.

Which rate should you plan with?#

Plan with the median effective cycle time of recent runs of the same operation on the same machine. One run is noisy, the best run is a target rather than a forecast, and a plant-wide OEE hides the slow machine and part pairs.

For example, the last eight runs of OP20 on the Haas VF-2 took 118, 124, 129, 133, 137, 142, 150 and 171 seconds per good part. The median is (133 + 137) ÷ 2 = 135 s. The 80th percentile, the seventh of the eight, is 150 s.

Keep one rate per machine and operation pair. If night shift runs OP20 at 150 s and day shift at 128 s, keep a rate for each shift, or plan with the slower one on any job that spans both.

Load each machine below its planned hours#

Load each machine below its planned hours, even when every duration is right. Run times vary, jobs arrive unevenly, and a machine scheduled to 100% has no room to absorb either, so the queue in front of it grows.

Queueing theory gives the shape of the problem. Kingman's approximation for a single machine estimates the average wait in queue as load ÷ (1 − load), multiplied by a variability term and by the average job time.^[1] When the gaps between job arrivals and the job times each have a coefficient of variation of 1, as in a purely random process, the variability term is 1.

Load the bottleneck to about 85% of its planned hours and keep other machines lower, then adjust from your own lead times. The ideal-rate trap compounds here. A schedule built on ideal cycle times that looks 85% loaded is really 0.85 × (135 ÷ 90) = 127.5% loaded on the machine in the example above.

Spare hours are not free either. An idle hour still carries the machine's fixed costs, so check what an idle machine costs before you add a shift or a machine to protect the margin.

Check the plan against the floor every week#

A rate is only right until the process changes. A new insert grade, a second fixture or a new operator moves the effective cycle time, so check it every week against what the machine logged.

1. Collect each finished task

   Take every task that finished last week: its start, its end, its setup time and its good parts. Use machine data where you have it, because a hand log rounds a 7-minute stop to zero and the stop drops out of availability.

2. Compute each task's effective cycle time

   Divide the planned production time the task took, setup excluded, by the good parts. A task that took 49.5 h for 1,200 good parts ran at 148.5 s per part.

3. Compare it with the planned rate

   Divide actual by planned for each machine and operation pair. A ratio of 1.10 means the job took 10% longer than the schedule allowed.

4. Update rates that drift

   When the median of the last five runs sits more than 10% from the planned rate, replace the rate. For example, five runs of OP20 at 148, 152, 139, 155 and 150 s have a median of 150 s, 11.1% above 135 s, so OP20 moves to 150 s.

5. Track schedule attainment

   Divide the jobs finished by their due date by the jobs due in the week. For example, 34 of 40 jobs on time is 85.0%. Measure in jobs or in hours, and keep the same unit every week.

Read the two numbers together. Rising attainment with stable rates means the schedule is learning. Falling attainment with rates that keep slowing means the process is losing ground, and that is a job for the downtime and performance losses, not for the schedule.

Schedule on live machine data#

SenseAi reads running and stopped time from the outside of each machine and counts parts from its cycle, so every run carries its own availability, performance and actual cycle time.

The Production Scheduler runs on those same machines, so the schedule uses the real part count. A new task's time estimate starts from the ideal cycle time times the quantity; replace it with the quantity times the effective cycle time.

With Working hours on, durations consume shift time only. The optimizer packs against the same estimates, and a strip under each Gantt bar shows what operators logged. If your schedule still lives in a spreadsheet, when to replace a spreadsheet production schedule helps you decide when to move it.

Hugo Benzing plans coiling, heat treatment and packing in the Production Scheduler, with SenseAi counting coils on its coilers while each run is going. Planning its tubes, racks and boxes as tooling brought a 20% efficiency increase, and its planners save more than 10 hours a week.

Checklist for your next schedule#

Run these checks before you release a week's schedule to the floor.

If a job is late with a correct rate and a sensible load, the schedule is not the problem. Look at the stops on that machine, starting with the downtime reasons that cost the most hours.