Production line simulation example: a three-step cell with a machine that breaks down

A manufacturing cell with cutting, two assembly stations and a test, where the cutting machine fails on average every 240 minutes and takes 20 to repair. Breakdowns raise the cutting wait from 3.3 min to 15.2 min and the time a part spends in the cell from 15.5 min to 32 min, with throughput almost unchanged.

Cutting wait with breakdowns15.2 min3.3 min without breakdowns
Time in the cell32 min15.5 min if the machine never failed
Throughput27.3/h27.3/h without breakdowns: the cost is in time, not output
Engine run: busiest is Test at 88% (8 replications, seed 2026)

The process

3 stations, as in the template: Cutting (1 worker, Triangular(1.2, 1.6, 2.4) min, fails about every 240 min, repair 20 min); Assembly (2 workers, Normal(3.6, 0.8) min); Test (1 worker, Triangular(1.4, 1.8, 2.6) min).

Results of the run

Utilization, average wait and longest line at each station of the Manufacturing cell template
StationWorkersBusyAverage waitLongest line (avg of runs)
Cutting185%15.2 min44.2
Assembly282%2.5 min10.1
Test188%6.9 min17.9

Highlighted: the busiest station. Throughput 27.3 an hour (95% interval 27.0 to 27.5); time in the process 32 min.

What a breakdown costs

Parts arrive every 2.2 minutes at random, 27.3 an hour. The cutting machine takes Triangular(1.2, 1.6, 2.4) minutes per part, and fails after an exponential up time of 240 minutes on average, then takes an exponential 20 minutes to repair. That makes it available 92% of the time, so its 79% load on paper becomes 85% busy, and its wait grows from 3.3 min to 15.2 min (longest line 12.8 parts without failures, 44.2 with). Throughput barely moves (27.3 to 27.3 an hour) because the cell has slack to catch up between failures. What the failures cost is time and work in process: a part spends 15.5 min in the cell without breakdowns and 32 min with them.

Which fix is worth more

The test station is the busiest in the base run at 88% (wait 6.9 min), with assembly at 82%. A second tester takes the time in the cell from 32 min to 25.1 min, a gain of 6.9 min. Removing the cutting breakdowns takes it to 15.5 min, a gain of 16.5 min. Under these inputs, maintenance on the cutting machine is worth more than extra test capacity, and the simulation puts a number on the comparison that a utilization table alone does not.

To tune the takt and number of stations for a line like this one, use the takt time calculator and the line balancing calculator.

The what-if

Throughput and time in the process for the template and its what-if runs
RunUnits per hourTime in the process (95% interval)Busiest station
As in the template27.332 min (26.7 min to 37.3 min)Test, 88% busy
A second tester (Test)27.325.1 min (20.8 min to 29.5 min)Cutting, 85% busy
No cutting breakdowns27.315.5 min (14.8 min to 16.2 min)Test, 88% busy

Three runs on the same seed: as in the template, one more worker at the busiest station, and the same cell with the failure settings removed.

Run it yourself

This template has 3 stations and fits the Free plan, which simulates up to 5. Opening it needs a free account so your copy can be saved. Change any number to match your site and press play. The inputs on this page are illustrative, so treat the results as a demonstration of behaviour and not as a forecast.

Open this exact case in SimflowlySee all templates

Questions

How does Simflowly model a machine breakdown?

Each station can have a mean time between failures and a mean time to repair, both exponential. While it is down, no new work starts and the work in progress waits. Here that is 240 and 20 minutes at cutting, an availability of 92%. Breaks and shared crews are separate settings.

Why do breakdowns hardly change throughput?

Because the cell can catch up. Cutting is 79% loaded without failures, so after a repair it works through the backlog, and over days the output matches the arrivals: 27.3 against 27.3 an hour. The damage appears as longer waits, larger work in process and later parts.

Are the failure numbers realistic?

They are illustrative, not benchmarks. A 240-minute time between failures and a 20-minute repair are starting points to show the effect. Use your own maintenance records for both, as exponential averages, and test other values by changing them in the app: the sensitivity of the cell to the repair time is the useful part.

Keep going