Bottleneck & Flow Lab.
A product moves through five stations. The slowest one sets the pace for the whole line. That’s the bottleneck (also called the constraint). Speed up the others and no more gets out the door. Start work faster than the bottleneck can handle and the piles just grow. Change station speeds, how often work starts and how much gets redone. Then watch where the piles form and how long each unit takes.
Find the bottleneck before you fix anything
New work starts 69.2 times an hour, but Weld, the bottleneck, can finish only 58.1 an hour. The extra 11.2 units an hour pile up in the line. Speeding up any other station won’t raise output. Try Weld’s − button below, or start work less often.
Try a scenario
How new work starts
Variability (how much times swing)
Playback speed
| Station | Avg time | Load per unit | Busy | Waiting |
|---|---|---|---|---|
| Cut | 48 s | 48 s | 0% | 0 |
| WeldBOTTLENECK | 62 s | 62 s | 0% | 0 |
| Assemble | 52 s | 52 s | 0% | 0 |
| Test | 57 s | 57 s | 0% | 0 |
| Pack | 40 s | 40 s | 0% | 0 |
Load per unit is how much of a station’s time each shipped unit really uses: average time × visits ÷ uptime. Redone units add visits. The station with the biggest load is the bottleneck. The whole line can ship at most 3600 ÷ that load units an hour (there are 3600 seconds in an hour).
How the model works, how it’s checked and what it leaves out
How it works
Five stations in a row, each with one machine. In front of each one is a waiting pile with no size limit, worked first come, first served (FIFO). Job times and the gaps between new starts are random. They follow a gamma distribution, a math pattern for random times that never go below zero. The variability setting picks how much they swing (the coefficient of variation). On a timer (push), new units start on the clock until 360 are in the line. Then the start dock is full, and starts are skipped until the work in progress (WIP) drops. With CONWIP, a new unit starts only when one ships and WIP is under the cap.
Uptime is the share of time a station runs instead of being broken down. Here it stretches every job time by 1 ÷ uptime. Rework sends a unit that fails Test back to the Assemble pile. If a share r of units fail, both stations see 1 ÷ (1 − r) visits per shipped unit. The numbers on screen average the last 90 simulated minutes.
How it’s checked
Throughput, the number of units shipped per hour, can never stay above 3600 ÷ (largest load per unit) for long. Speeding up a station that isn’t the bottleneck must not raise that ceiling. More rework or less uptime must not raise it either. When work starts slower than the bottleneck can handle, throughput ≈ the start rate.
When the line is steady, WIP ≈ TH × CT: work in progress equals throughput times cycle time. That’s Little’s Law. While WIP is still growing, the two sides drift apart, and that’s the point. Each scenario uses a fixed random seed, so it plays out the same way every time.
What it leaves out
Piles never run out of room, so a full pile never blocks the station before it. There’s no changeover time between jobs and no sudden breakdowns (here uptime is just a steady slowdown). There’s one product, no shift schedule and no time to move parts between stations. Each of these matters on a real factory floor. Each one would be added only when a decision depends on it.